Datasets:

Hendrik-a

/

arxiv

like 0

strongly interacting fermi gases provide a clean and controllable laboratory system for modeling strong interparticle interactions between fermions in nature , from high temperature superconductors to neutron matter and quark - gluon plasmas . model - independent thermodynamic measurements , which do not require theoretical models for calibrations , are very important for exploring this important system experimentally , as they enable direct tests of predictions based on the best current non - perturbative many - body theories . at duke university , we use all - optical methods to produce a strongly interacting fermi gas of spin-1/2-up and spin-1/2-downli atoms that is magnetically tuned near a collisional ( feshbach ) resonance . we conduct a series of measurements on the thermodynamic properties of this unique quantum gas , including the energy , entropy , and sound velocity . our model - independent measurements of and enable a precision study of the finite temperature thermodynamics . the data are directly compared to several recent predictions . the temperature in both the superfluid and normal fluid regime is obtained from the fundamental thermodynamic relation by parameterizing the data using two different power laws that are joined with continuous and at a certain entropy , where the fit is optimized . we observe a significant change in the scaling of with above and below . taking the fitted value of as an estimate of the critical entropy for a superfluid - normal fluid phase transition in the strongly interacting fermi gas , we estimate the critical parameters . our data are also used to experimentally calibrate the endpoint temperatures obtained for adiabatic sweeps of the magnetic field between the ideal and strongly interacting regimes . this enables the first experimental calibration of the temperature scale used in experiments on fermionic pair condensation , where the ideal fermi gas temperature is measured before sweeping the magnetic field to the strongly interacting regime . our calibration shows that the ideal gas temperature measured for the onset of pair condensation corresponds closely to the critical temperature estimated in the strongly interacting regime from the fits to our data . we also calibrate the empirical temperature employed in studies of the heat capacity and obtain nearly the same . we determine the ground state energy by three different methods , using sound velocity measurements , by extrapolating to and by measuring the ratio of the cloud sizes in the strongly and weakly interacting regimes . the results are in very good agreement with recent predictions . finally , using universal thermodynamic relations , we estimate the chemical potential and heat capacity of the trapped gas from the data . pacs numbers : 03.75.ss

0811.1159

interacting fermionic particles play a central role in the structure of matter and exist over a very broad range of energies , from extremely low temperature trapped atomic fermi gases , where @xmath10 k @xcite , to very high temperature primordial matter , like quark - gluon plasmas , where @xmath11 k @xcite . for all of these systems , the most intriguing physics is related to very strong interactions between fermionic particles , such as the strong coupling between electrons in high-@xmath8 superconductors and the strong interactions between neutrons in neutron matter . current many - body quantum theories face great challenges in solving problems for strongly interacting fermi systems , due to the lack of a small coupling parameter . for example , the critical temperature of a superfluid - normal fluid transition in a strongly interacting fermi gas has been controversial for many years . the critical temperature @xmath12 has been predicted to have values in the range between 0.15 and 0.35 by different theoretical methods @xcite . a complete understanding of the physics of strongly interacting systems can not yet be obtained from a theoretical point of view . there is a pressing need to investigate strongly interacting fermions experimentally . in recent years , based on progress in optical cooling and trapping of fermionic atoms , a clean and controllable strongly interacting fermi system , comprising a degenerate , strongly interacting fermi gas @xcite , is now of interest to the whole physics community . strongly interacting fermi gases are produced near a broad feshbach resonance @xcite , where the zero energy s - wave scattering length @xmath13 is large compared to the interparticle spacing , while the interparticle spacing is large compared to the range of the two - body interaction . in this regime , the system is known as a unitary fermi gas , where the properties are universal and independent of the details of the two - body scattering interaction @xcite . in contrast to other strongly interacting fermi systems , in atomic gases , the interactions , energy , and spin population can be precisely adjusted , enabling a variety of experiments for exploring this model system . intense studies of strongly interacting fermi gases have been implemented over the past several years from a variety of perspectives . some of the first experiments observed the expansion hydrodynamics of the strongly interacting cloud @xcite . evidence for superfluid hydrodynamics was first observed in collective modes @xcite . collective modes were later used to study the @xmath14 equation of state throughout the crossover regime @xcite . recently , measurements of sound velocity have also been used to explore the @xmath14 equation of state @xcite . below a feshbach resonance , fermionic atoms join to form stable molecules and molecular bose - einstein condensates @xcite . fermionic pair condensation has been observed by projection experiments using fast magnetic field sweeps @xcite . above resonance , strongly bound pairs have been probed by radio frequency and optical spectroscopy @xcite . phase separation has been observed in spin polarized samples @xcite . rotating fermi gases have revealed vortex lattices in the superfluid regime @xcite as well as irrotational flow in both the superfluid and normal fluid regimes @xcite . measurement of the thermodynamic properties of a strongly interacting fermi gas was first accomplished by adding a known energy to the gas , and then determining an empirical temperature that was calibrated using a pseudogap theory @xcite . recent model - independent measurements of the energy and entropy @xcite provide a very important piece of the puzzle , because they enable direct and precision tests that distinguish predictions from recent many - body theories , without invoking any specific theoretical model @xcite . one of the major challenges for the experiments in strongly interacting fermi gases is the lack of a precise model - independent thermometry . two widely - used thermometry methods are model - dependent , in that they rely on theoretical models for calibration . the first relies on adiabatic magnetic field sweeps between the molecular bec regime and the strongly interacting regime @xcite . subsequently , the temperature of the strongly interacting gas is estimated from the measured temperature in the bec regime using a theoretical model of the entropy @xcite . the second method , used by our group @xcite , is based on determining an empirical temperature from the cloud profiles that is calibrated by comparing the measured density distribution with a theoretical model for the density profiles . currently two model - independent thermometry methods have been reported for strongly - interacting gases . one is the technique employed by the mit group @xcite , which is only applicable to imbalanced mixtures of spin - up and spin - down atoms . that method is based on fitting the noninteracting edge for the majority spin after phase separation . another model - independent method is demonstrated in ref . @xcite , which is applicable to both balanced and imbalanced mixtures of spin - up and spin - down fermions . the energy @xmath1 and entropy @xmath2 are measured and then parameterized to determine a smooth curve @xmath4 . then the temperature in both the superfluid and normal fluid regime is obtained from the fundamental thermodynamic relation @xmath5 . in this paper , we will describe our model - independent thermodynamic experiments on a strongly interacting fermi gas of @xmath0li , which we have conducted at duke university . first , we will describe our measurements of both the total energy @xmath1 and the total entropy @xmath2 of a trapped strongly - interacting fermi gas tuned near a feshbach resonance . then , we determine the temperature @xmath5 after showing that the @xmath4 data are very well parameterized by using two different power laws that are joined with continuous @xmath1 and @xmath6 at a certain entropy @xmath7 that gives the best fit . to examine the sensitivity of the temperature to the form of the fit function , we employ two different fit functions that allow for a heat capacity jump or for a continuous heat capacity at @xmath7 . we find that the @xmath6 values closely agree for both cases . we find a significant change in the scaling of @xmath1 with @xmath2 above and below @xmath7 , in contrast to the behavior for an ideal fermi gas , where a single power - law well parameterizes @xmath4 over the same energy range . by interpreting @xmath7 as the critical entropy for a superfluid - normal fluid transition in the strongly interacting fermi gas , we estimate the critical energy @xmath15 and critical temperature @xmath8 . both the model - independent @xmath4 data and the estimated critical parameters are compared with several recent many - body theories based on both analytic and quantum monte carlo methods . we also show how parameterizing the @xmath4 data provides experimental temperature calibrations , which helps to unify , in a model - independent way , the results obtained by several groups @xcite . first we relate the endpoint temperatures for adiabatic sweeps of the bias magnetic field between the strongly interacting and ideal noninteracting regimes , as used in the jila experiments to characterize the condensed pair fraction @xcite . this enables the ideal gas temperature observed for the onset of pair condensation @xcite to be related to the critical temperature of the strongly interacting fermi gas . the temperature obtained by parameterizing the strongly interacting gas data also calibrates the empirical temperature based on the cloud profiles , as used in our previous studies of the heat capacity @xcite . these temperature calibrations yield values of @xmath8 close to that estimated from our @xmath4 data . next , we discuss three different methods for determining the universal many - body parameter , @xmath16 @xcite , where @xmath17 is the energy per particle in a uniform strongly interacting fermi gas at @xmath14 in units of the energy per particle of an ideal fermi gas at the same density . first , we describe the measurement of the sound velocity at resonance and its relationship to @xmath16 . then , we determine @xmath16 from the ground state energy @xmath18 of the trapped gas . here , @xmath18 is obtained by extrapolating the @xmath4 data to @xmath9 , as suggested by hu et al . this avoids a systematic error in the sound velocity experiments arising from the unknown finite temperature . finally , to explore the systematic error arising from the measurement of the number of atoms , @xmath16 is determined in a number - independent manner from the ratio of the cloud sizes in the strongly and weakly interacting regimes . all three results are found to be in very good agreement with each other and with recent predictions . finally , we obtain three universal thermodynamic functions from the parameterized @xmath4 data , the energy @xmath19 , heat capacity @xmath20 , and global chemical potential @xmath21 . our experiments begin with an optically - trapped highly degenerate , strongly interacting fermi gas of @xmath0li @xcite . a 50:50 mixture of the two lowest hyperfine states of @xmath0li atoms is confined in an ultrastable co@xmath22 laser trap with a bias magnetic field of 840 g , just above a broad feshbach resonance at @xmath23 g @xcite . at 840 g , the gas is cooled close to the ground state by lowering the trap depth @xmath24 @xcite . then @xmath24 is recompressed to a final trap depth of @xmath25k , which is much larger than the energy per particle of the gas , for the highest energies employed in the experiments . this suppresses evaporation during the time scale of the measurements . the shallow trap yields a low density that suppresses three body loss and heating . the low density also yields a weakly interacting sample when the bias magnetic field is swept to 1200 g , although the scattering length is @xmath26 bohr , as discussed in detail in [ sec : entropy ] . the shape of the trapping potential is that of a gaussian laser beam , with a transverse gaussian profile determined by the spot size and an axial lorentzian profile determined by the rayleigh length . to simplify the calculations of the ideal gas properties in subsequent sections , as well as the theoretical modelling , we take the trap potential to be approximated by a three dimensional gaussian profile , @xmath27 where @xmath28 is the @xmath29 width of trap for each direction . here , we take the zero of energy to be at @xmath30 . when the cold atoms stay in the deepest portion of the optical trap , where @xmath31 , the gaussian potential can be well approximated as a harmonic trap with transverse frequencies @xmath32 , @xmath33 and axial frequency @xmath34 , where @xmath35 here @xmath36 is the @xmath0li mass . at our final trap depth @xmath37 , the measured transverse frequencies are @xmath38 hz and @xmath39 hz . the axial frequency is weakly magnetic field dependent since the total axial frequency has both an optical potential contribution @xmath40 determined by eq . [ eq : trapfreqwelldepth ] and a magnetic potential contribution arising from magnetic field curvature , @xmath41 . the net axial frequency is then @xmath42 . we find @xmath43 hz at 840 g and @xmath44 hz at 1200 g. the total number of atoms is @xmath45 . the corresponding fermi energy @xmath46 and fermi temperature @xmath47 at the trap center for an ideal noninteracting harmonically trapped gas are @xmath48 , where @xmath49 . for our trap conditions , we obtain @xmath50k . using @xmath51 , we can rewrite eq . [ eq : udipole ] as a symmetric effective potential , @xmath52 where @xmath53 is the scaled position vector . here , @xmath54 with @xmath55 . to obtain the anharmonic corrections for the gaussian trap , we expand eq . [ eq : reducedtruepotential ] in a taylor series up to second order in @xmath56 , @xmath57 model - independent energy measurement is based on a virial theorem , which for an ideal gas in a harmonic confining potential @xmath58 yields @xmath59 . since the harmonic potential energy is proportional to the mean square cloud size , measurement of the cloud profile determines the total energy . remarkably , a trapped unitary fermi gas at a broad feshbach resonance obeys the same virial theorem as an ideal gas , although it contains superfluid pairs , noncondensed pairs , and unpaired atoms , all strongly interacting . this has been demonstrated both theoretically and experimentally @xcite . the virial theorem shows that the total energy of the gas at all temperatures can be measured from the cloud profile using @xmath60 where @xmath24 is the trapping potential and @xmath61 is the position vector . . [ eq : energymeas ] can be shown to be valid for any trapping potential @xmath24 and for any spin mixture , without assuming either the local density approximation or harmonic confinement @xcite . using eq . [ eq : truepotentialquartic ] in eq . [ eq : energymeas ] and keeping the lowest order anharmonic corrections , we obtain the energy per atom in terms of the axial mean square size , @xmath62 . \label{eq : energygaussian}\ ] ] here , we have used the local density approximation with a scalar pressure , which ensures that @xmath63 . for the ground state , where the spatial profile is a zero temperature thomas - fermi profile , we have @xmath64 . for energies @xmath65 , where the spatial profile is approximately gaussian , we have @xmath66 . since the anharmonic correction is small at low temperatures where the cloud size is small , we use the gaussian approximation over the whole range of energies explored in our experiments . for the conditions of our experiments , there is no evidence that the local density approximation breaks down for a 50:50 spin mixture . in this case , measurement of the mean square size in any one direction determines the total energy . from eq . [ eq : energygaussian ] , we see that by simply measuring the axial mean square size @xmath67 at 840 g and measuring the axial trap frequency by parametric resonance , we actually measure @xmath68 , the total energy per particle of the strongly interacting fermi gas at 840 g. this determines the total energy per particle in a model - independent way @xcite . the entropy @xmath2 of the strongly interacting gas at 840 g is determined by adiabatically sweeping the bias magnetic field from 840 g to 1200 g , where the gas is weakly interacting @xcite . the entropy @xmath69 of the weakly interacting gas is essentially the entropy of an ideal fermi gas in a harmonic trap , which can be calculated in terms of the mean square axial cloud size @xmath70 measured after the sweep . since the sweep is adiabatic , we have @xmath71 the adiabaticity of the magnetic field sweep is verified by employing a round - trip - sweep : the mean square size of the cloud at 840 g after a round - trip - sweep lasting 2s is found to be within 3% of mean square size of a cloud that remains at 840 g for a hold time of 2s . the nearly unchanged atom number and mean square size proves the sweep does not cause any significant atom loss or heating , which ensures entropy conservation for the sweep . the background heating rate is the same with and without the sweep and increases the mean square size by about 2% over 2s . the mean square size data are corrected by subtracting the increase arising from background heating over the 1 s sweep time @xcite . at 1200 g in our shallow trap , we have @xmath72 , where the fermi wavevector @xmath73 and the s - wave scattering length @xmath74 bohr @xcite . we find that the gas is weakly interacting : for the lowest temperatures attained in our experiments , the gas at 1200 g is a normal fluid that we observe to expand ballistically . we have calculated the ground state mean square size at 1200 g in our gaussian trap , based on a mean - field theory , @xmath75 @xcite , which is close to that of an ideal harmonically trapped gas , @xmath76 . here , @xmath77 is the mean square size corresponding to the fermi energy of an ideal noninteracting fermi gas at magnetic field @xmath78 , which includes the magnetic field dependence of the axial trapping frequency : @xmath79 . we expect that the entropy of the gas at 1200 g is close to that of an ideal gas , except for a mean field shift of the energy . we therefore assume that a reasonable approximation to the entropy is that of an ideal fermi gas , @xmath80 , where @xmath81 is the ground state mean square size of an ideal fermi gas in the gaussian trapping potential of eq . [ eq : reducedtruepotential ] . here , we apply an elementary calculation based on integrating the density of states for the gaussian trap with the entropy per orbital @xmath82 $ ] , where @xmath83 is the ideal fermi gas occupation number at temperature @xmath6 for an orbital of energy @xmath84 . by calculating @xmath85 as a function of the _ difference _ between the finite temperature and ground state mean square cloud sizes , we reduce the error arising from the mean field shift at 1200 g , and ensure that @xmath86 for the ground state . the exact entropy of a weakly interacting gas @xmath69 at 1200 g , @xmath87 , has been calculated using many - body theories @xcite for the gaussian potential of eq . [ eq : reducedtruepotential ] . in the experiments , we determine the value of @xmath88 , where we take @xmath89 , the value measured at our lowest energy at 1200 g by extrapolation to @xmath14 using the sommerfeld expansion for the spatial profile of an ideal gas . this result is close to the theoretical value , @xmath90 . the entropy versus cloud size curve for an ideal noninteracting fermi gas and the exact value for a weakly interacting gas @xmath69 at 1200 g are plotted in fig . [ fig : entropysizeoverlap ] . we find that the entropies @xmath91 and @xmath92 ) , agree within a few percent over most of the energy range we studied , except at the point of lowest measured energy , where they differ by 10% . the results show clearly that the shape of the entropy curve of a weakly interacting fermi gas is nearly identical to that of an ideal gas when the mean field shift of the ground state size is included by referring the mean square cloud size to that of the ground state . so we have to a good approximation , @xmath93 since the corrections to ideal gas behavior are small , the determination of @xmath94 by measuring the axial mean square size @xmath70 relative to the ground state provides an essentially model - independent estimate of the entropy of the strongly interacting gas . sound velocity measurements have been implemented for fermi gases that are nearly in the ground state , from the molecular bec regime to the weakly interacting fermi gas regime @xcite . a sound wave is excited in the sample by using a thin slice of green light that bisects the cigar - shaped cloud . the green light at 532 nm is blue detuned from the 671 nm transition in lithium , creating a knife that locally repels the atoms . the laser knife is pulsed on for 280 @xmath95s , much shorter than typical sound propagation times @xmath96 ms and excites a ripple in the density consisting of low density valleys and high density peaks . after excitation , the density ripple propagates outward along the axial direction @xmath97 . after a variable amount of propagation time , we release the cloud , let it expand , and image destructively . in the strongly interacting regime , we use zero - temperature thomas - fermi profiles for a non - interacting fermi gas to fit the density profiles , and locate the positions of the density valley and peak . by recording the position of the density ripple versus the propagation time , the sound velocity is determined . a detailed discussion of potential sources of systematic error is given by joseph et al . @xcite . for a strongly interacting fermi gas in the unitary limit , the sound velocity @xmath98 at the trap center for the ground state is determined by the fermi velocity of an ideal gas at the trap center , @xmath99 and the universal constant @xmath16 , @xmath100 a precision measurement of the sound speed therefore enables a determination of @xmath16 @xcite . as discussed below , the values of @xmath16 determined from the @xmath4 data and the sound velocity data are in very good agreement . in the experiments , the raw data consists of the measured mean square cloud sizes at 840 g and after an adiabatic sweep of the magnetic field to 1200 g. using this data , we determine both the energy and the entropy of the strongly interacting gas . the data is then compared to several recent predictions . we begin by determining the axial mean square cloud sizes at 840 g and after the adiabatic sweep to 1200 g. since the atom number can vary between different runs by up to 20% , it is important to make the comparison independent of the atom number and trap parameters . for this purpose , the mean square sizes are given in units of @xmath77 , as defined above . the measured mean square sizes are listed in table [ tbl : energyentropy ] . in the experiments , evaporative cooling is used to produce an atom cloud near the ground state . energy is controllably added by releasing the cloud and then recapturing it after a short time @xmath101 as described previously @xcite . for a series of different values of @xmath101 , the energy at 840 g is directly measured from the axial cloud size according to eq . [ eq : energygaussian ] . then the same sequence is repeated , but the cloud size is measured after an adiabatic sweep to 1200 g. in each case , the systematic increase in mean square size arising from background heating rate is determined and subtracted . the total data comprise about 900 individual measurements of the cloud size at 840 g and 900 similar measurements of the cloud size after a sweep to 1200 g. to estimate the measurement error , we split the energy scale at 840 g into bins with a width of @xmath102 . measured data points within the width of the energy bin are used to calculate the average measured values of the cloud sizes and the corresponding standard deviation at both 840 g and 1200 g. ' '' '' & @xmath103&@xmath104&@xmath105&@xmath106&@xmath107&@xmath108 + ' '' '' 1&0.568(4)&0.743(6)&0.548(4)&0.63(8)&0.91(23)&0.97(5 ) + ' '' '' 2&0.612(5)&0.776(13)&0.589(5)&0.99(11)&1.18(22)&1.24(9 ) + ' '' '' 3&0.661(5)&0.803(11)&0.634(5)&1.22(8)&1.36(20)&1.42(7 ) + ' '' '' 4&0.697(9)&0.814(15)&0.667(8)&1.30(10)&1.43(18)&1.49(8 ) + ' '' '' 5&0.74(1)&0.87(4)&0.71(1)&1.6(2)&1.72(18)&1.8(2 ) + ' '' '' 6&0.79(1)&0.89(2)&0.75(1)&1.7(1)&1.79(15)&1.9(1 ) + ' '' '' 7&0.83(1)&0.94(2)&0.79(1)&2.0(1)&2.03(16)&2.1(1 ) + ' '' '' 8&0.89(2)&1.02(2)&0.84(2)&2.3(1)&2.32(18)&2.4(1 ) + ' '' '' 9&0.91(1)&1.02(3)&0.86(1)&2.3(1)&2.31(16)&2.4(1 ) + ' '' '' 10&0.97(1)&1.10(1)&0.91(1)&2.55(4)&2.57(17)&2.64(4 ) + ' '' '' 11&1.01(1)&1.17(2)&0.94(1)&2.74(7)&2.75(19)&2.82(7 ) + ' '' '' 12&1.05(1)&1.18(1)&0.98(1)&2.78(4)&2.80(17)&2.87(4 ) + ' '' '' 13&1.10(1)&1.22(1)&1.03(1)&2.89(2)&2.90(15)&2.97(2 ) + ' '' '' 14&1.25(2)&1.35(5)&1.15(1)&3.21(12)&3.20(14)&3.27(12 ) + ' '' '' 15&1.28(1)&1.39(3)&1.18(1)&3.28(6)&3.28(14)&3.35(6 ) + ' '' '' 16&1.44(2)&1.49(2)&1.31(2)&3.49(4)&3.48(9)&3.55(4 ) + ' '' '' 17&1.53(2)&1.62(6)&1.39(2)&3.74(10)&3.73(11)&3.80(10 ) + ' '' '' 18&1.58(1)&1.63(1)&1.42(1)&3.76(2)&3.74(8)&3.81(2 ) + ' '' '' 19&1.70(2)&1.73(6)&1.52(1)&3.94(9)&3.92(7)&3.99(9 ) + ' '' '' 20&1.83(5)&1.79(2)&1.62(4)&4.03(3)&4.01(1)&4.08(3 ) + ' '' '' 21&1.93(3)&1.96(3)&1.70(2)&4.28(4)&4.26(6)&4.32(4 ) + ' '' '' 22&2.11(5)&2.17(3)&1.83(4)&4.55(3)&4.53(7)&4.59(3 ) + the ratio of the mean square axial cloud size at 1200 g ( measured after the sweep ) to that at 840 g ( measured prior to the sweep ) is plotted in fig . [ fig : sizeratio ] as a function of the energy of a strongly interacting gas at 840 g. the ratio is @xmath109 , since for an adiabatic sweep of the magnetic field from the strongly interacting regime to the weakly interacting regime , the total entropy in the system is conserved but the energy increases : the strongly interacting gas is more attractive than the weakly interacting gas . a similar method was used to measure the potential energy change in a fermi gas of @xmath110k , where the bias magnetic field was adiabatically swept between the strongly interacting regime at the feshbach resonance and a noninteracting regime above resonance @xcite . the resulting potential energy ratios are given as a function of the temperature of the noninteracting gas @xcite . in contrast , by exploiting the virial theorem which holds for the unitary gas , we determine both the energy and entropy of the strongly interacting gas , as described below . [ fig : sizeconvertentropy ] shows the entropy which is obtained from the mean square size at 1200 g @xmath111 as listed in table [ tbl : energyentropy ] . first , we find the mean square size relative to that of the ground state , @xmath112 . we use the measured value @xmath113 for the lowest energy state that we obtained at 1200 g , as determined by extrapolation to @xmath14 using a sommerfeld expansion for the spatial profile of an ideal fermi gas . then we determine the entropy in the noninteracting ideal fermi gas approximation : @xmath114 $ ] , where we have replaced @xmath81 by the ground state value at 1200 g. as discussed above , this method automatically ensures that @xmath9 corresponds to the measured ground state @xmath115 at 1200 g , and compensates for the mean field shift between the measured @xmath116 for a weakly interacting fermi gas and that calculated @xmath117 for an ideal fermi gas in our gaussian trapping potential . as shown in fig . [ fig : entropysizeoverlap ] , the entropy obtained from a more precise many - body calculations is in close agreement with the ideal gas entropy calculated in the ideal gas approximation . the energy is determined from the cloud profiles at 840 g using eq . [ eq : energygaussian ] . finally , we generate the energy - entropy curve for a strongly interacting fermi gas , as shown in fig . [ fig : energyentropy ] . here , the energy @xmath1 measured from the mean square axial cloud size at 840 g is plotted as a function of the entropy @xmath2 measured at 1200 g after an adiabatic sweep of the magnetic field . we note that above @xmath118 ( @xmath119 ) the @xmath4 data ( blue dots ) for the strongly interacting gas appear to merge smoothly to the ideal gas curve ( dashed green ) . for an ideal fermi gas . for this figure , the ideal gas approximation to the entropy is used , @xmath106 of table [ tbl : energyentropy ] . [ fig : energyentropy],width=384 ] in addition to the entropy calculated in the ideal gas approximation , table [ tbl : energyentropy ] also provides a more precise entropy @xmath120 versus the axial mean square cloud size . these results are obtained by hu et al . @xcite using a many - body calculation for @xmath121 at 1200 g in the gaussian trap of eq . [ eq : reducedtruepotential ] . perhaps the most important application of the energy - entropy measurements is to test strong coupling many - body theories and simulations . since the energy and entropy are obtained in absolute units without invoking any specific theoretical model , the data can be used to distinguish recent predictions for a trapped strongly interacting fermi gas . fig . [ fig : comparetheory ] shows how four different predictions compare to the measured energy and entropy data . these include a pseudogap theory @xcite , a combined luttinger - ward - de dominicis - martin ( lw - ddm ) variational formalism @xcite , a t - matrix calculation using a modified nozires and schmitt - rink ( nsr ) approximation @xcite , and a quantum monte carlo simulation @xcite . the most significant deviations appear to occur near the ground state , where the precise determination of the energy seems most difficult . the pseudogap theory predicts a ground state energy that is above the measured value while the prediction of ref . @xcite is somewhat low compared to the measurement . all of the different theories appear to converge at the higher energies . the temperature @xmath6 is determined from the measured @xmath4 data using the fundamental relation , @xmath5 . to implement this method , we need to parameterize the data to obtain a smooth differentiable curve . at low temperatures , one expects the energy to increase from the ground state according to a power law in @xmath6 and a corresponding power law in @xmath2 , i.e. , @xmath122 . for a harmonically trapped ideal fermi gas , we have in the sommerfeld approximation an energy per particle in units of @xmath46 given by @xmath123 . the corresponding entropy per particle in units of @xmath124 is @xmath125 , so that @xmath126 . we attempt to use a single power law to fit the @xmath127 curve for a noninteracting fermi gas in a gaussian trapping potential , with @xmath128 , as in our experiments . the energy and entropy are calculated in the energy range @xmath129 and displayed as dots in fig . [ fig : idealgasfit ] . we find that a single power law @xmath130 fits the curve very well over this energy range . note that the power law exponent is @xmath131 , close to the low temperature value . using the fit function , we can extract the reduced temperature @xmath132 as a function of @xmath85 and compare it to the theoretical reduced temperature @xmath133 at the same @xmath85 . the results are shown as the green dashed line in fig . [ fig : idealgasfit ] . we see that the agreement is quite good except below @xmath134 and above @xmath135 , where the deviation is @xmath136% . to improve the fit and to make a more precise determination of the temperature , we employ a fit function comprising two power laws that are joined at a certain entropy @xmath7 , which gives the best fit . when used to fit the data for the strongly interacting fermi gas , we consider two types of fits that incorporate either a jump in heat capacity or a continuous heat capacity at @xmath7 . in this way , we are able determine the sensitivity of the temperature and critical parameters to the form of the fit function . the two types of fits yield nearly identical temperatures , but different values of @xmath7 and hence of the critical parameters , as discussed below . we take the energy per particle @xmath1 in units of @xmath46 to be given in terms of the entropy per particle in units of @xmath124 in the form @xmath137 we constrain the values of @xmath138 and @xmath3 by demanding that energy and temperature be continuous at the joining point @xmath7 : @xmath139 by construction , the value of @xmath140 does not affect these constraints and is chosen in one of two ways . fixing @xmath141 , the fit incorporates a heat capacity jump at @xmath7 , which arises from the change in the power law exponents at @xmath7 . alternatively , we choose @xmath140 so that the second derivative @xmath142 is continuous at @xmath7 , making the heat capacity continuous . the final fit function has 5 independent parameters @xmath143 , and takes the form @xmath144+e(s - s_c)^2 ; \,\,s\geq s_c . \label{eq : evsslowengtwopower}\end{aligned}\ ] ] here , when @xmath140 is not constrained to be zero , it is given by @xmath145 fig . [ fig : idealgasfit ] shows the improved fit to the calculated energy versus entropy of a noninteracting fermi gas in a gaussian trap for @xmath128 , using eq . [ eq : evsslowengtwopower ] with @xmath146 , since the ideal gas has no heat capacity jump . in this case , both power law exponents @xmath147 and @xmath148 are close to @xmath149 as for the single power law fit . the temperature determined from the fit agrees very closely with the exact temperature , as shown in fig . [ fig : idealgasfit ] ( red solid line ) . in contrast to the noninteracting case , we have found that the energy - entropy data of a strongly interacting fermi gas is not well fit by a single power law function @xcite . however , the two power - law function fits quite well , with a factor of two smaller value of @xmath150 than for the single power - law fit . here , we use @xmath151 , where @xmath152 ( @xmath153 ) is the fitted ( data ) value for the @xmath154 point , and @xmath155 is the corresponding the standard error . motivated by the good fits of the two power - law function to the ideal gas energy versus entropy curve and the good agreement between the fitted and exact temperature , we apply the two power - law fit function to the data for the strongly interacting fermi gas . ) . for comparison , the dot - dashed green curve shows @xmath4 for an ideal fermi gas . for this figure , the ideal gas approximation to the entropy is used , @xmath106 of table [ tbl : energyentropy ] . [ fig : energyentropyfit],width=384 ] fig . [ fig : energyentropyfit ] shows the fit ( red solid curve ) obtained with a heat capacity jump using eq . [ eq : evsslowengtwopower ] with @xmath141 and @xmath156 , @xmath157 , @xmath158 , the ground state energy @xmath159 , and the critical entropy @xmath160 . also shown is the fit ( blue dashed curve ) with continuous heat capacity ( @xmath146 ) and @xmath161 , @xmath162 , @xmath163 , the ground state energy @xmath164 , and the critical entropy @xmath165 . the fit functions for the @xmath4 data for the strongly interacting fermi gas exhibit a significant change in the scaling of @xmath1 with @xmath2 below and above @xmath7 . the dramatic change in the power law exponents for the strongly interacting gas suggests a transition in the thermodynamic properties . the power law exponent is @xmath166 above @xmath7 , comparable to that obtained for the ideal gas , where @xmath167 . the power law exponent below @xmath7 is @xmath168 , which corresponds to the low temperature dependence @xmath169 , close to that obtained in measurements of the heat capacity , where the observed power law was @xmath170 after the model - dependent calibration of the empirical temperature @xcite , see [ sec : ttwiddle ] . if we interpret @xmath7 as the critical entropy for a superfluid - normal fluid transition in the strongly interacting fermi gas , then we can estimate the critical energy @xmath15 and the critical temperature @xmath171 . for the fits of eq . [ eq : evsslowengtwopower ] with a heat capacity jump @xmath172 ) or with continuous heat capacity ( @xmath146 ) , we obtain @xmath173 using the fit parameters in eq . [ eq : criticalparam ] yields critical parameters of the strongly interacting fermi gas , which are summarized in table [ tbl : criticalparam ] . the statistical error estimates are from the fit , and do not include systematic errors arising from the form of the fit function . ' '' '' & @xmath174 & @xmath175 & @xmath176 + ' '' '' expt @xmath4 fit@xmath177 & 2.2(1 ) & 0.83(2 ) & 0.21(1 ) + ' '' '' expt @xmath4 fit@xmath178 & 1.6(3 ) & 0.70(5 ) & 0.185(15 ) + ' '' '' heat capacity experiment@xmath179 & & 0.85 & 0.20 + ' '' '' theory ref . @xcite & & & 0.30 + ' '' '' theory ref . @xcite & & & 0.31 + ' '' '' theory ref . @xcite & & & 0.27 + ' '' '' theory ref . @xcite & & & 0.29 + ' '' '' theory ref . @xcite & 2.15 & 0.82 & 0.27 + ' '' '' theory ref . @xcite & 1.61(5 ) & 0.667(10 ) & 0.214(7 ) + we note that the fit function for @xmath180 previously used in ref . @xcite to determine the temperature was continuous in @xmath2 and @xmath1 , but intentionally ignored the continuous temperature constraint in order to determine the entropy as a power of @xmath181 both above and below the joining energy @xmath15 . as the continuous temperature constraint is a physical requirement , we consider the present estimate of the temperature @xmath6 to be more useful for temperature calibrations and for characterizing the physical properties of the gas than the estimate of ref . @xcite . in contrast to the temperature @xmath6 , the estimate of @xmath8 depends on the value of the joining entropy @xmath7 that optimizes the fit and is more sensitive to the form of fit function than the temperature that is determined from the @xmath1 and @xmath2 data . for the fit function @xmath180 used in ref . @xcite , the temperatures determined by the fit function just above @xmath15 , @xmath182 , and below @xmath15 , @xmath183 , were different . an average of the slopes @xmath184 and @xmath185 was used to estimate the critical temperature . from those fits , the critical energy was found to be @xmath186 , the critical entropy per particle was @xmath187 . the estimated critical temperature obtained from the average was @xmath188 , significantly higher than than the value @xmath189 obtained using eq . [ eq : evsslowengtwopower ] , which incorporates continuous temperature . we are able to substantiate the critical temperature @xmath189 by using our data to experimentally calibrate the temperature scales in two other experiments . in [ sec : onsetpaircond ] , we find that this value is in very good agreement with the estimate we obtain by calibrating the ideal gas temperature observed for the onset of pair condensation . nearly the same transition temperature is obtained in [ sec : ttwiddle ] by using the @xmath4 data to calibrate the empirical transition temperature measured in heat capacity experiments @xcite . table [ tbl : criticalparam ] compares the critical parameters estimated from the power - law fits to the @xmath4 data with the predictions for a trapped unitary fermi gas from several theoretical groups . we note that calculations for a uniform strongly interacting fermi gas at unitarity @xcite yield a lower critical temperature , @xmath190 , than that of the trapped gas , where @xmath191 is the fermi temperature corresponding to the uniform density @xmath192 . extrapolation of the uniform gas critical temperature to that of the trapped gas shows that the results are consistent @xcite . using the parameters from the fits and eq . [ eq : evsslowengtwopower ] , the temperature of the strongly interacting fermi gas , in units of @xmath47 can be determined as a function of the entropy per particle , in units of @xmath124 , @xmath193 here @xmath7 is given in table [ tbl : criticalparam ] from the fits to the @xmath4 data for the strongly interacting gas , eq . [ eq : criticalparam ] gives @xmath8 . [ fig : tempvsentropy ] shows the temperature as a function of entropy according to eq . [ eq : temperaturestrongint ] for fits with a heat capacity jump and for continuous heat capacity . the estimates of the temperature of the strongly interacting fermi gas as a function of the entropy can be used to experimentally calibrate the temperatures measured in other experiments , without invoking any specific theoretical models . the jila group measures the pair condensate fraction in a strongly interacting fermi gas of @xmath110k as a function of the initial temperature @xmath194 in the noninteracting regime above the feshbach resonance @xcite . in these experiments , a downward adiabatic sweep of the bias magnetic field to resonance produces a strongly interacting sample . using our @xmath4 data , we relate the endpoint temperatures for adiabatic sweeps of the bias magnetic field between the ideal and strongly interacting fermi gas regimes . we therefore obtain the critical temperature for the onset of pair condensation in the strongly interacting fermi gas , and find very good agreement with our estimates based on entropy - energy measurement . in addition , we calibrate the empirical temperature based on the cloud profiles , which was employed in our previous measurements of the heat capacity @xcite . we relate the endpoint temperatures for an adiabatic sweep between the strongly interacting and ideal fermi gas regimes . [ eq : temperaturestrongint ] gives the temperature of the strongly interacting gas as a function of entropy , i.e. , @xmath195 . next , we calculate the entropy per particle @xmath196 for an ideal fermi gas in our gaussian trap , in units of @xmath124 , with @xmath197 in units of @xmath47 , as used in [ sec : idealgas ] to determine @xmath198 . for an adiabatic sweep between the strongly interacting and ideal fermi gas regimes , where @xmath199 , the temperature of the strongly interacting gas is related to that for the ideal fermi gas by @xmath200 , \label{eq : adiabactictempcal}\ ] ] which is shown in fig . [ fig : tempsivni ] . + for an adiabatic sweep from the ideal fermi gas regime to the strongly interacting fermi gas regime at low temperature @xmath201 , the reduced temperature of the strongly interacting gas is greater than or equal to that of the ideal gas . this arises because the entropy of the strongly interacting gas scales as a higher power of the temperature than that of the ideal gas . in our present experiments , we could not take data at high enough temperatures to properly characterize the approach of the temperature to the ideal gas regime . above @xmath8 , our @xmath4 data are obtained over a limited range of energies @xmath202 to avoid evaporation in our shallow trap . in this energy range , our data are reasonably well fit by a single power law . however , such a power law fit can not completely describe the higher temperature regime . we expect that the temperatures of the strongly interacting gas and ideal gas must start to merge in the region @xmath203 , where the @xmath4 data for the strongly interacting gas nearly overlaps with the @xmath4 curve for an ideal gas , as shown in fig . [ fig : energyentropy ] . from fig . [ fig : tempvsentropy ] , @xmath204 corresponds to @xmath205 , approximately the place where the calibrations from the two different power law fits ( for @xmath141 and @xmath146 ) begin to differ in fig . [ fig : tempsivni ] . we therefore expect that the single power law fit overestimates the temperature @xmath6 of the strongly interacting gas for @xmath206 , yielding a trend away from ideal gas temperature , in contrast to the expected merging at high temperature . in ref . @xcite , projection experiments measure the ideal fermi gas temperature @xmath194 where pair condensation first appears . in those experiments , @xmath194 is estimated to be @xmath207 @xcite . from the calibration , [ fig : tempsivni ] , we see that for @xmath208 , the corresponding temperature of the strongly interacting gas is @xmath209 for both the red solid and blue dashed curves , which is almost the same as the ideal gas value . the critical temperature of the strongly interacting gas for the onset of pair condensation is then @xmath210 , in very good agreement with the values @xmath211 and @xmath212 that we obtain from the two fits to the @xmath4 measurements . this substantiates the conjecture that the change in the power law behavior observed at @xmath8 in our experiments corresponds to the superfluid transition . in our previous study of the heat capacity , we determined an empirical temperature @xmath213 as a function of the total energy of the gas @xcite . the gas was initially cooled close to the ground state and a known energy was added by a release and recapture method . then a thomas - fermi profile for an ideal fermi gas was fit to the low temperature cloud profiles to determine the fermi radius . holding the fermi radius constant , the best fit to the cloud profiles at higher temperatures determined the effective reduced temperature , which is denoted @xmath214 . the @xmath215 data @xcite was observed to scale as @xmath216 for @xmath217 , while below @xmath218 , the energy was found to scale as @xmath219 . the transition point occurs at an energy @xmath220 , which is close to the value @xmath221 obtained from power - law fit to the @xmath4 data for the fit with a heat capacity jump . assuming that @xmath222 corresponds to the superfluid - normal fluid transition , we can determine the corresponding value of @xmath12 for the strongly interacting gas . to calibrate the empirical temperature we start with @xmath215 . then , as discussed in [ sec : energyvstemp ] , eq . [ eq : evsslowengtwopower ] determines @xmath19 and hence @xmath223from the fits to our @xmath4 data . hence @xmath224 $ ] , where @xmath225 is the reduced temperature of the strongly interacting gas and @xmath226 is the reduced energy . for simplicity , we give the analytic results obtained using the @xmath141 fit to the @xmath4 data , @xmath227 fig . [ fig : tempsivni ] shows the full calibration ( green dashed curve ) . for comparison , the calibration obtained from the pseudogap theory of the cloud profiles gave @xmath228 for @xmath229 , and @xmath230 above @xmath218 . for @xmath222 , we obtain from eq . [ eq : tildetcalib ] @xmath231 ( see fig . [ fig : tempsivni ] ) , in good agreement with the value obtained for the onset of pair condensation and with the values @xmath232 and @xmath233 determined from the fits to the @xmath4 data . measurement of the ground state energy of a unitary fermi gas provides a stringent test of competing many - body theoretical predictions and is therefore of great interest . for a unitary fermi gas of uniform density in a 50 - 50 mixture of two spin states , the ground state energy per particle can be written as @xmath234 where @xmath235 is the local fermi energy corresponding to the density @xmath192 . the ground state energy of the unitary fermi gas differs by a universal factor @xmath236 from that of an ideal fermi gas at the same density . the precise value of @xmath237 has been of particular interest in the context of neutron matter @xcite , and can be measured in unitary fermi gas experiments @xcite . the sound speed at temperatures near the ground state determines @xmath16 according to eq . [ eq : soundres ] . we have made precision measurements of the sound speed in a trapped fermi gas at the feshbach resonance @xcite . at 834 g , we vary the density by a factor of 30 to demonstrate universal scaling and obtain the value @xmath238 . using eq . [ eq : soundres ] then yields @xmath239 . note that the reference fermi velocity @xmath240 depends on the fermi energy of an ideal gas at the trap center and hence on both the trap frequencies and atom number ( as @xmath241 ) , which are carefully measured to minimize systematic errors @xcite . while the energy of the gas as measured from the mean square cloud size was close to the ground state value , the precise temperature of the gas was not determined . the universal parameter @xmath16 also can be determined by measuring the ground state energy @xmath18 of a harmonically trapped unitary fermi gas , which is given by @xmath242 our @xmath4 data enables a new determination of @xmath18 by extrapolating the measured energy @xmath4 to @xmath9 . as pointed out by hui et al . @xcite , this method avoids a systematic error arising when the finite temperature is not determined in the measurements . from both of our fit functions below @xmath7 , we obtain @xmath243 . [ eq : hobeta ] yields @xmath244 . this result is slightly more negative than that obtained in the sound speed experiments , which is reasonable since the sound speed measurements are done at finite temperature . both results are in very good agreement . one possible systematic error in these measurements arises from the determination of the atom number . the measurements of @xmath16 from the sound speed and from the energy - entropy measurements were done in different laboratories . the close agreement is gratifying , considering that the imaging systems that determine the atom number employed @xmath245-polarized light for the sound speed experiments , while the entropy - energy measurements used @xmath246-polarized light , for which the resonant optical cross section is a factor of two smaller than for @xmath245 polarization . to examine the systematic error arising from the atom number determination , we employ a third method to measure @xmath16 based on the measured ratio of the cloud size at 840 g and at 1200 g , which is number independent . the ratio of the ground state mean square sizes for the weakly and strongly interacting gases is predicted to be @xmath247 note that we obtain @xmath248 from a mean field calculation @xcite , in agreement with that obtained using a many - body calculation @xcite . our measurements for the ground state mean square size at 1200 g are accomplished by fitting a sommerfeld expansion of the axial density for an ideal fermi gas to the cloud profile @xcite . the fit determines the fermi radius @xmath249 and reduced temperature @xmath133 , yielding @xmath250 for @xmath14 , close to the predicted value of @xmath251 . the ground state energy @xmath252 at 840 g from the entropy - energy experiments determines the ground state mean square size as @xmath253 . hence , @xmath254 . the corresponding @xmath255 from eq . [ eq : groundstateratio ] . since the mean square sizes are determined from the images and the ratio @xmath256 is number independent , this result shows that the systematic error arising from the number measurement is within the quoted error estimate . we also can determine @xmath16 by directly extrapolating to zero entropy the ratio of the axial mean square size of the weakly interacting fermi gas at 1200 g to that of strongly interacting gas at 840 g. when this is done , we obtain @xmath257 , in very good agreement with the estimates based on the sound speed and ground state energy . finally , we can estimate the correction to the ground state energy arising from the finite scattering length at 840 g , @xmath258 . for the trap conditions in the @xmath4 measurements , @xmath259 , where @xmath260 is the wavevector for an ideal fermi gas at the trap center . to estimate the true unitary ground state energy at @xmath261 , we first determine the leading order @xmath262 correction to the trapped atom density , where @xmath263 is the local fermi wavevector corresponding to the density @xmath192 . the local chemical potential is estimated from ref . @xcite . using the notation of eq . [ eq : reducedtruepotential ] and a harmonic approximation , the corrected density yields @xmath264 $ ] , where @xmath265 is the fermi radius for the unitary gas . according to the virial theorem ( see eq . [ eq : energymeas ] ) , the mean square size and energy of the unitary gas are corrected by the same factor . the unitary ground state energy is then @xmath266 for @xmath267 , we obtain @xmath268 and the value of @xmath269 obtained directly from @xmath270 is shifted to @xmath271 . we also obtain the corrected value of @xmath272 in eq . [ eq : groundstateratio ] and @xmath273 . table [ tbl : beta ] compares the values of @xmath16 obtained in our experiments to several recent predictions . note that the table does not include the finite @xmath274 correction for the @xmath4 measurement at 840 g described above . ' '' '' & @xmath16 + ' '' '' @xmath4 experiment & -0.59(2 ) + ' '' '' sound velocity experiment & -0.565(15 ) + ' '' '' cloud size ratio experiment & -0.61(2 ) + ' '' '' ref . @xcite & -0.58(1 ) + ' '' '' ref . @xcite & @xmath275 + ' '' '' ref . @xcite & -0.599 + ' '' '' ref . @xcite & -0.60(1 ) + ' '' '' ref . @xcite & -0.646(4 ) + using the @xmath4 data for the strongly interacting fermi gas and the temperature determined from the two power - law fits , we estimate several universal functions . first , we determine the dependence of the energy on temperature @xmath19 and the corresponding heat capacity , @xmath20 . then we find the global chemical potential of the trapped gas as a function of the energy @xmath21 . the energy is readily determined as a function of temperature using eq . [ eq : evsslowengtwopower ] for the case where there is a heat capacity jump and @xmath141 , @xmath276;\,\,\,0\leq t\leq t_c\nonumber\\ e_>(t)&=&e_c+{\frac{s_ct_c}{d}}\left[\left(\frac{t}{t_c}\right)^{\frac{d}{d-1}}-1\right ] ; \,\,\,t\geq t_c , \label{eq : energyvstemp}\end{aligned}\ ] ] where the energy ( temperature ) is given in units of @xmath46 ( @xmath47 ) and the critical energy @xmath15 is @xmath277 with @xmath18 the ground state energy . for the case with @xmath146 , where the heat capacity is continuous , we determine the ordered pairs @xmath278 $ ] as a function of @xmath2 and plot @xmath19 . fig . [ fig : energyvstemp ] shows the results using the best fits for both cases . of particular interest is the low temperature power law . for @xmath141 , we obtain @xmath279 and @xmath280 . since @xmath147 is near @xmath281 , the energy relative to the ground state scales approximately as @xmath282 . this is consistent with sound modes dominating the low energy excitations . however , one would expect instead that the free fermions on the edges of the trapped cloud would make an important contribution to the low energy excitations @xcite . over an extended range of @xmath201 , the net entropy arising from the bose and fermi excitations has been predicted to scale as @xmath283 , yielding an energy scaling @xcite as @xmath284 . in this case , one would expect that @xmath285 , i.e. , @xmath286 in eq . [ eq : energyvstemp ] , so that @xmath287 . hence , the low energy power law exponents for the entropy should be between @xmath281 and @xmath288 , which is barely distinguishable for our data . the heat capacity at constant trap depth @xmath289 is readily obtained from eq . [ eq : energyvstemp ] ( where there is a heat capacity jump , since we have constrained @xmath141 in eq . [ eq : evsslowengtwopower ] ) . for this parameterization , @xmath290 where @xmath6 and @xmath8 are given in units of @xmath47 , and @xmath7 is given in units of @xmath124 . for the fit with a continuous heat capacity , we use @xmath195 to find @xmath291 , and plot the ordered pairs @xmath292 $ ] . the heat capacity curves for both cases are shown in fig . [ fig : cvst ] . according to eq . [ eq : heatcapacity ] , a jump in heat capacity occurs at @xmath7 : @xmath293 and @xmath294 differ when the power law exponents @xmath147 and @xmath148 are different . this is a consequence of the simple two power - law structure assumed for the fit function @xmath4 given by eq . [ eq : evsslowengtwopower ] for @xmath141 , and can not be taken as proof of a true heat capacity jump . at present , the precise nature of the behavior near the critical temperature can not be determined from our data , and it remains an open question whether the data exhibits a heat capacity jump or a continuous heat capacity . the global chemical potential @xmath295 is readily determined from the fits to the @xmath4 data for a strongly interacting fermi gas , which obeys universal thermodynamics . the local energy density generally takes the form @xmath296 , where @xmath297 is the local internal energy , which includes the kinetic energy and the interaction energy . here , @xmath192 is the local density , @xmath95 is the local chemical potential , @xmath298 is the pressure and @xmath299 is the total entropy per unit volume . the local chemical potential can be written as @xmath300 , where @xmath24 is the trap potential . in the universal regime , where the local pressure depends only on the local density and temperature , we have @xmath301 , as noted by ho @xcite . hence , @xmath302 . integrating both sides over the trap volume and using @xmath303 , where @xmath1 and @xmath304 ) are the total energy and average potential energy per particle , respectively , we obtain @xmath305 where @xmath2 is the entropy per particle . for simplicity , we assume harmonic confinement and use the virial theorem result , @xmath306 , from eq . [ eq : energymeas ] , which holds in the universal regime . then , eq . [ eq : globalchempot1 ] yields the global chemical potential of a harmonically trapped fermi gas in the universal regime , @xmath307 by using the fit to the measured entropy - energy data to obtain the temperature @xmath5 from eq . [ eq : evsslowengtwopower ] , the global chemical potential of a trapped unitary fermi gas can be calculated from eq . [ eq : globalchempot ] . for @xmath141 , where the heat capacity has a jump , the simple power law fits above and below @xmath15 each yield a different linear dependence of @xmath295 on @xmath1 , @xmath308 where @xmath309 . we plot the chemical potential in fig . [ fig : chemenergy ] . the data points are obtained using eq . [ eq : globalchempot ] with the measured energy @xmath1 and entropy @xmath2 and the temperature determined from the fit to the @xmath4 data , using @xmath141 in eq . [ eq : evsslowengtwopower ] , i.e. , with a heat capacity jump . the solid red curve is given by eq . [ eq : globalchempotvse ] . we note that the low temperature data points in @xmath4 are best fit with the power law @xmath279 , which is close to @xmath281 . according to eq . [ eq : globalchempotvse ] , this produces a slope near zero for @xmath310 . since the power - law fit above @xmath15 gives @xmath311 , the slope according to eq . [ eq : globalchempotvse ] changes from nearly zero for @xmath310 to negative for @xmath312 . note that from eq . [ eq : globalchempot ] , we obtain the slope @xmath313 since the entropy @xmath2 is continuous , we see that a jump in the heat capacity produces a corresponding jump in the slope of @xmath295 versus @xmath1 . for comparison , fig . [ fig : chemenergy ] also shows the chemical potential obtained for @xmath146 in eq . [ eq : evsslowengtwopower ] , where the heat capacity is continuous ( blue dashed curve ) . we have studied the thermodynamic properties of a strongly interacting fermi gas by measuring both the energy and the entropy . the model - independent data obtained in both the superfluid and the normal fluid regimes do not employ any specific theoretical calibrations , and therefore can be used as a benchmark to test the predictions from many - body theories and simulations . parameterizing the energy - entropy data determines the temperature of the strongly interacting fermi gas and also yields estimates of the critical parameters . we use the measured data to calibrate two different temperature scales that were employed in observations of the onset of pair condensation and in heat capacity studies . these calibrations yield critical temperatures in good agreement with the results estimated from our energy - entropy data . our data does not determine whether the heat capacity exhibits a jump or is continuous at the critical temperature . however , for a finite system with nonuniform density , the latter is most likely . considering that there is huge interest in determining the detailed behavior of the superfluid transition in a strongly interacting fermi gas @xcite , more precise determinations of the critical temperature , the heat capacity , and the chemical potential near the critical point , as well as the high temperature behavior and the approach to the ideal gas limit , will be important topics for future research . this research has been supported by the physics divisions of the army research office and the national science foundation , the physics for exploration program of the national aeronautics and space administration , and the chemical sciences , geosciences and biosciences division of the office of basic energy sciences , office of science , u. s. department of energy . we thank willie ong for a careful reading of the manuscript .

the diffuse x - ray emission from the thin disk surrounding the galactic mid - plane ( the so - called galactic ridge ) was measured with _ rxte _ pca in order to determine the spatial extent , spectral nature , and origin of the emission . spatial examination of the diffuse emission in the central of the plane in galactic longitude reveals the presence of two components : a thin disk of full width centered roughly on the galactic mid - plane , and a broad component which can be approximated as a gaussian distribution with fwhm of about . assuming an average distance of 16 kpc to the edge of the galaxy , a scale height of about pc and 500 pc is derived for the thin and broad disk components , respectively . spectral examination of the emission clearly reveals the presence of a hard power law tail above 10 kev and an emission line from he - like iron , indicating both thermal and possibly non - thermal origins for the diffuse emission . the averaged spectrum from the ridge in the kev band can be modelled with a raymond - smith plasma component of temperature kev and a power law component of photon index . based on this finding , we argue that the temperature of the hot phase of the interstellar medium ( ism ) is less than the previously reported values of kev . motivated by the similarities between the characteristics of the thermal component of the galactic ridge emission in our model and the thermal emission from supernova remnants ( snrs ) , we discuss the origin of the thermal emission in terms of a population of snrs residing in the galactic disk . we find that a sn explosion rate of less than 5 per century is adequate to power the thermal emission from the ridge . the origin of the emission in the hard x - ray band modelled by a power law remains uncertain . possible contributions from non - thermal bremsstrahlung of cosmic ray electrons and protons , inverse compton scattering of energetic electrons from ambient microwave , infrared , and optical photons , non - thermal emission from snrs , and emission from discrete x - ray sources are discussed . we speculate that bremsstrahlung of accelerated electrons and protons in snr sites can play a significant role in producing the hard tail of the spectrum . moreover , their collisional losses can play a major role in the ionization of the ism . accepted for publication in the astrophysical journal

astro-ph9804012

a galaxy can be well described by an ecosystem . there is an intimate relationship , much like a symbiosis , between the discrete components of the galaxy such as stars , and its interstellar medium ( ism ) . the ism provides the foundation for the birth of a new generation of stars , while at the same time it is enriched by the remains of the older generations and their byproducts during their life cycle . hence , it is natural to expect that the understanding of formation and evolution of galaxies is closely related to the understanding of their ism . the ism of the milky way has been found to have many components . magnetic fields and cosmic ray gas compose the relativistic fluid , while the gaseous phase consists of both ionized and neutral components . the hot ionized medium observable in uv and x - ray has a temperature above @xmath8 k and is composed of hot coronal gas heated by supernova shocks . a good portion of the energy of the ism resides in this component . the warm ionized medium ( e.g. hii , planetary nebulae ) is visible in @xmath9 , uv , and optical , and has a temperature as high as @xmath10 k. the neutral atomic gas appears to have both cold ( @xmath11 k ; e.g. hi clouds ) and warm ( 100 k @xmath12 8000 k ) components . molecular clouds compose the self - gravitating gaseous phase . finally the ism ( with the exception of its hot phase ) is filled with dust , particles typically a few tenth of microns in size and visible through their infrared radiation . since the peak emission of each component arises in different wavebands , multi - wavelength observations from radio to @xmath13-rays are needed to probe the ism . diffuse x - ray emission from our galaxy is a powerful diagnostic of the hot phase of the ism . it is our purpose here to use this tool to probe the processes that contribute to its energetics and dynamics . the first detection of the x - ray emission from the galactic disk was achieved by the pioneering rocket experiment of bleach et al . they detected excess emission associated with a narrow disk component of angular size @xmath14 . since then , x - ray emission from the galactic plane and in particular the ridge ( the narrow region centered on the galactic mid - plane extending approximately to @xmath15 in longitude and @xmath16 in latitude ) has been measured in the past with several satellites ( e.g. _ heao-1 _ [ 2 - 50 kev ] : worrall et . 1982 ; _ exosat _ [ 2 - 6 kev ] : warwick et al . 1985 ; _ tenma _ [ 2 - 11 kev ] : koyama et al . 1986 ; _ ginga _ [ 2 - 16 kev ] : yamasaki et al . 1997 ; _ asca _ [ 0.5 - 10 kev ] : kaneda et al . 1997 ) . the presence of the @xmath17 kev iron line in the spectrum discovered with _ tenma _ has motivated the idea that part of the emission below 10 kev is due to a hot optically thin plasma of temperature @xmath7 kev . because of its high spectral resolution , _ asca _ has provided the most accurate measurement of the spectrum of the emission to date . the presence of mg , si , and fe k - lines in the spectrum obtained by _ asca _ suggests that at least part of the emission is of thermal origin . close examination of _ asca _ data has also revealed that unresolved , discrete sources are not responsible for the bulk of the emission ( yamauchi et al . 1996 ; kaneda 1997 ) . the most recent investigation of the diffuse emission from the scutum arm region with _ asca _ ( kaneda et al . 1997 ) has concluded that the emission below 10 kev has both soft ( @xmath18 kev ) and hard ( @xmath19 kev ) thermal components . if indeed the super hot gas ( @xmath20 kev ) exists in an extended form in the ism , it is not clear how to explain its confinement to the galactic disk since its temperature exceeds the gravitational potential of the disk by at least an order of magnitude ( townes 1989 ) . unfortunately , the two - temperature model does not produce a good fit to the data above 10 kev indicating the presence of additional component(s ) at higher energies . indeed , a hard power law tail has been detected in the hard x - ray / soft @xmath13-ray band from observations of the ridge with _ ginga _ and the balloon experiment _ welcome-1 _ ( yamasaki et al . 1997 ) , and _ osse _ ( skibo et al . 1997 ) . in this paper , we present the results from _ rxte _ measurement of the diffuse x - ray emission from the galactic ridge in the 3 - 35 kev band . observations in the hard x - ray / soft @xmath13-ray band have usually been complicated by the presence of numerous variable discrete sources , and the fact that the instruments generally have large fields of view and no imaging capabilities , or have imaging capability but no diffuse emission sensitivity . the combination of these factors makes the separation of emission between diffuse and compact sources a difficult task . the advantage of _ rxte _ over previous missions is its small field of view ( @xmath21 fwhm ) combined with its wide energy bandpass ( 2 - 60 kev for the pca ) , allowing for the subtraction of the contribution of discrete sources from the diffuse emission spectrum in the hard x - ray band . in addition to reporting on the detection of a hard power law tail in the _ rxte _ data , we also offer an alternative interpretation for the origin of the emission below 10 kev ( i.e. instead of a super hot plasma of temperature @xmath7 kev ) . in agreement with previous studies , our results indicate that the emission is most likely of diffuse origin as opposed to the superposition of discrete sources . however , we present a model in which the x - ray emission is the superposition of both thermal ( modelled by a raymond - smith plasma ) and possibly non - thermal ( modelled by a power law ) components . we discuss the origin of the thermal component in terms of a population of snrs residing in the disk . the origin of the power law component remains uncertain . by comparing the spectrum of the diffuse emission in hard x - rays ( _ rxte _ ) and soft @xmath13-rays ( _ osse _ ) , we find indications that the emission in the two bands are related . we discuss its origin in terms of both discrete hard x - ray sources and radiation mechanisms such as non - thermal bremsstrahlung from cosmic ray electrons and protons , and inverse compton scattering . the plan of this paper is as follows . in 2 , we describe the observations . in 3 , we present the results of our spatial and spectral analysis of the data . 4 is devoted to the discussion of the results , and their implications for the origin of the emission . finally in 5 , we present our conclusions . the galactic ridge observations were performed in september of 1996 , with the pca instrument onboard the _ rxte _ satellite . the pca ( jahoda et al . 1996 ) has a total collecting area of 6500 @xmath22 , an energy range of @xmath23 kev , and energy resolution of @xmath24 at 6 kev . the collimator field of view is approximately circular ( @xmath25 diameter ) with fwhm of @xmath21 . the observations consist of 14 long scans ( from @xmath26 to @xmath27 in galactic longitude ) parallel to the galactic plane up to @xmath28 and separated by @xmath29 in galactic latitude . the galactic mid - plane ( i.e. @xmath30 ) was scanned twice . fourteen shorter scans at higher latitudes parallel to the plane ( @xmath31 ; @xmath32 and @xmath33 ) were also performed . the speed of scans ranged from @xmath34 per minute for the shorter scans to @xmath35 per minute for the longer ones . we constructed maps of the galactic ridge x - ray emission from these scans in 3 energy bands . the results are shown in figure 1 . after subtraction of the instrument background ( described in 3.2 ) , the net counts were binned into @xmath36 pixels for the long scans ( @xmath37 ) in each pixel . to smooth the images , we have interpolated them using idl ( interactive data language ) . since the pixel size for the long and short scans were different , each image was broken into 3 pieces : one consisting of long scans , and two others consisting of short scans at larger positive and negative latitudes . each piece was then interpolated separately . some of the scans at larger @xmath38 were occulted by the earth . therefore , we have only shown regions that were devoid of data gaps and could be smoothly interpolated . in order to emphasize the galactic ridge diffuse emission in the images , we have used a color table that saturates the emission from discrete sources . furthermore , we did not attempt to de - convolve the detector s response function with the light curves . as a result , discrete sources appear as brown extended regions . for the first time , it is clear from these maps that the longitude extent of the diffuse emission in the hard x - ray band ( above 10 kev ) is similar to that in the softer 2 - 10 kev band . to measure the spectrum of the diffuse x - ray emission from the galactic ridge , we must first identify and exclude the contribution of compact sources within the paths of the scans . this task was done by an automatic routine which identifies sources with a minimum detection of @xmath39 . the location of each source was calculated and compared with a catalogue of known x - ray sources . the time interval in which each source was within the field of view of pca was then excluded . finally , the remaining good time intervals were inspected manually for detection of possible new sources . a detection of at least 10 @xmath40 ( @xmath41 mcrab ) above background was interpreted as a discrete x - ray source and the time interval corresponding to the excess counts above background was excluded . from hereafter , these sources are excluded in the analysis reported in this paper . table 1 lists all the identified sources and their positions which lied in the path of the scans . the faintest detected source in the scans has an intensity of @xmath42 mcrab . we used the pca background estimator program _ pcabackest _ ( version 1.4h ) provided by the _ rxte _ gof ( guest observer facility ) to estimate the background . the total background consists of the instrument background and the cosmic x - ray background ( cxb ) . variations in the instrument background are modelled as two components . one component is related to changes in the anti - coincidence rate , while the other is related to the activation by the saa ( southern atlantic anomaly ) passages . our galactic ridge scans were performed during non - saa orbits and therefore the activation induced background was estimated to be zero by _ pcabackest_. for modeling the latitude distribution of the diffuse emission ( 3.3 ) , we subtracted the instrument background only . for spectral analysis of the diffuse emission ( 3.4 ) , we have subtracted both the instrument and the cosmic x - ray background . since galactic absorption distorts the cxb spectrum at lower energies , this will result in an underestimate of the galactic ridge flux at energies below @xmath43 kev . we describe how we correct the spectrum for this effect in 3.4.2 . before performing a spectral analysis of the diffuse emission , it is crucial to have some understanding of its spatial distribution in the galactic plane . determining the scale height of the emission is particularly important for at least two reasons . firstly , it provides important clues for the origin of the emission . for example , the scale height of the emission from inverse compton scattering is expected to be larger than that from bremsstrahlung . this is due to the homogeneous nature of the cosmic microwave background radiation and the large scale height of the optical and infrared photons and cosmic ray electrons . secondly , it reveals the multi - component nature of the emission . for example , it can be determined whether the emission is confined to a narrow disk centered on the galactic mid - plane or whether a broad halo around the disk also contributes to the diffuse emission . in order to quantify the latitude distribution of the diffuse emission shown in figure 1 , we convolve the detector s response function with a given spatial model and perform a least squared fit to the data . in figure 2a , we have plotted the averaged diffuse emission count rate in the 5 - 8 kev and 8 - 35 kev energy bands vs. galactic latitude . the count rate is obtained by averaging the counts at a fixed latitude over the central @xmath0 of the galactic plane ( @xmath44 ) and then normalizing for exposure time . instrumental background is subtracted . we chose the 5 - 8 kev energy band since above 5 kev the effects of galactic absorption is small , and also this band implicitly maps the 6.7 kev iron line . the 8 - 35 kev band was chosen to map the distribution of the hard x - rays . we found that the best fit is given by a 2-component model consisting of thin and broad distributions for both energy bands . we modelled the thin component with a uniform thin disk , and the broad component with a disk of gaussian distribution as a function of galactic latitude . the contribution from the thin component is proportional to the distance that our line of sight traverses through the thin disk . its value drops rapidly as @xmath38 increases , and reaches a minimum at @xmath45 . the contribution from the gaussian distribution is integrated along the line of sight to the edge of the galaxy . moreover , we added a constant component to the model to account for the cosmic x - ray background . to determine the parameters of the thin and broad components , we constrained the count rate at large galactic latitudes to that of the cosmic x - ray background estimated from _ pcabackest _ and searched the parameter space for the best fit . for the @xmath46 kev energy band , the best fit parameters are found to be @xmath47 for the full width of the thin component , and @xmath48 for the fwhm of the broad component at the far edge of the galaxy ( @xmath49 , @xmath50 ) . the best fit is found when the thin component is centered at mid - galactic plane , while the broad component is centered at @xmath51 . in figure 2b , we have deconvolved the detector s response function and plotted the actual contribution of each component from the model in the @xmath46 kev band . for the @xmath52 kev band , the best fit parameters are @xmath53 and @xmath54 for the full width of the thin component and fwhm of the broad component at the far edge of the galaxy , respectively ( @xmath55 , @xmath50 ) . the best fit is obtained when both components are centered at @xmath56 . error bars indicate @xmath57 confidence limits . ( since the fits are poor , the confidence limits are derived by setting @xmath58 . ) we notice that a survey of the latitude distribution of the emission in the @xmath59 kev with _ osse _ yields a broad distribution that is approximated with a gaussian distribution of @xmath60 fwhm with no significant evidence of energy dependence ( purcell et al . 1996 ) . however , in our survey , the detection of the thin disk component is statistically significant . if we fit the emission in the @xmath52 kev band with a single gaussian component , the @xmath61 of the fit will increase from @xmath62 ( for the 2-component model ) to @xmath63 . it is likely that the detection of the thin component may not be possible with _ osse _ due to its large field of view . however , we can not determine the functional form of the thin component other than deriving limits on its width , due to the fact that the upper limit is smaller than the field of view of the pca . as for the broad component , we point out that it is most likely related to the one reported by iwan et al . ( 1982 ) from the measurements taken with the _ heao _ a-2 instrument . the scale height of the diffuse emission from that study was determined to be greater than @xmath64 kpc . from the reduced @xmath61 reported above , it is clear that the fits are poor . this is partly the result of the asymmetry of the overall profile of the diffuse emission with respect to the galactic mid - plane . from figure 2a , it also appears that at some latitudes the count rates substantially deviate from that of the model . we have inspected the source excluded data , but they do not appear to be contaminated with any sources above our detection level . hence , we conclude that additional spatial components at certain latitudes corresponding to localized features of the diffuse emission are needed to produce a better fit . however , the derived parameters provide a good global description of the spatial distribution of the diffuse emission in the galactic plane . to determine the spectral variation of the diffuse emission from the galactic ridge as a function of galactic latitude , we performed separate spectral analysis on the data from the central ridge ( r1 ) , the northern ridge ( r2 ) , and the southern ridge ( r3 ) . the coordinates of each region are given in table 2 . we will discuss the results of spectral fitting for the central ridge first . spectral analysis for regions r2 and r3 will be presented in 3.4.3 . since galactic disk absorption distorts the spectrum of both the galactic ridge emission and the cosmic x - ray background ( cxb ) at lower energies , we analyse the data above 10 kev first , where absorption is negligible for the expected galactic column densities . this strategy will allow us to constrain the component(s ) that dominate the spectrum in the hard x - ray band . we will then use this derived information to model the spectrum at lower energies . the presence of a hard power law tail above 10 kev has been reported from combined observations with _ ginga _ and _ welcome-1 _ ( yamasaki et al . 1997 ) and _ osse _ ( skibo et al . 1997 ) . here , we search for this component by fitting the _ rxte _ data in the @xmath65 kev energy range with a single power law component . the best fit yields a photon index of @xmath66 with a normalization of @xmath67 at 1 kev ( @xmath68 ) . the surface brightness for the @xmath65 kev band is @xmath69 . figure 3 shows the unfolded spectrum . a thermal bremsstrahlung for the data above 10 kev requires a temperature of @xmath70 kev . however , we prefer the power law model since it is not clear how the hot plasma can be bound to the galactic disk ( the gravitational potential of the disk is @xmath71 kev ) . furthermore , the presence of the power law at soft @xmath13-ray energies has been established with _ osse_. it is likely that the emission in the hard x - ray and soft @xmath13-ray both modelled by a power law are related . we will discuss this further in 3.6 . let us now extend the data analysis to lower energies . below 10 kev , the presence of a @xmath17 kev emission line from he - like iron was discovered with _ tenma _ ( koyama et al . further extensive study of the ridge spectrum with the high resolution capability of _ asca _ has revealed the presence of lower energy emission lines from mg , si , and s ( kaneda et al . these findings have motivated the idea that part of the emission has thermal origin . furthermore , we use the results of the previous section where the presence of a power law tail in the spectrum above 10 kev has been established . hence , we model the data with both raymond - smith ( rs ) plasma and power law components . notice from figure 2 that the emission in the central ridge ( r1 ) arises from both the thin disk and the broad components . however , since the spectral decomposition of the broad and thin components is complicated and not well constrained with current statistics , we model the averaged effect of the two components . before modeling the data , we should first consider the effect of galactic disk absorption on the final spectrum . since we have subtracted the high latitude cosmic x - ray background from our data , the galactic ridge flux at lower energies may have been underestimated . this is because for expected galactic column densities , absorption reduces the cxb flux mostly below @xmath72 kev , but will leave the spectrum above 10 kev unchanged . therefore , in order to compensate for this effect , we subtract the high latitude cxb model from our spectral model but add the absorbed cxb model ( had it been absorbed by the galactic disk ) . the spectrum of the cxb below 15 kev was modelled from _ rxte _ high latitude observations . the spectrum was found to have a power law shape of photon index @xmath73 with normalization @xmath74 at 1 kev corresponding to a surface brightness of @xmath75 in the @xmath76 kev . notice that since _ rxte _ is not designed to separate instrument background from the cosmic x - ray background , some residual counts inherent to the detectors background may exist in the cxb model at higher energies , but its effect on our modeling is negligible . in effect , we add the term @xmath77 to our model where @xmath78 is the cxb model and @xmath79 and @xmath80 are the galactic column density and the photoelectric cross section as a function of energy , respectively . however , the precise value of the galactic hydrogen column density as a function of position and its averaged effect over @xmath81 in galactic longitude ( the span of the central ridge studied here ) are not well known . therefore , while we freeze the parameters of the cxb model in our analysis , we fit for the galactic column density along with the rest of the model parameters . this correction is applied to all models we present hereafter . the best fit parameters for the model described above are presented in table 3 . the data and unfolded spectrum are shown in figure 4 . from these results , on average , the emission can be explained by a raymond - smith plasma of temperature @xmath82 kev and a power law component with photon index @xmath83 @xmath6 which dominates above 10 kev . in the @xmath76 kev band , 54% of the flux is from the thermal raymond - smith plasma component . above 10 kev , the power law component dominates by providing 95% of the flux . confidence contours for @xmath84 are shown in figure 5 . the average column density derived from this model is well within the expected galactic values . it may be argued that the column density @xmath79 absorbing the cxb component may be larger than that absorbing the rs plasma and power law components since in the case of cxb the entire material between us and the edge of the galaxy acts as the absorber ( i.e. longer path length ) . had we frozen the value of the column density absorbing the cxb component at an upper limit of @xmath85 , the key fit parameters would change to @xmath86 , @xmath87 , and @xmath88 kev ( @xmath89 ) . these are all within the error bars of the values presented in table 3 . we point out that these results do not indicate that the spectrum measured with _ rxte _ is inconsistent with that from _ asca_. we modeled the _ data from region r1 in the @xmath76 kev with the same components with which _ asca _ data from the scutum arm region was modeled ( kaneda et al . 1997 ) , namely 2 thermal components of temperature @xmath90 and @xmath20 kev , respectively . the fit was acceptable with @xmath91 . however , when we extend this model to higher energies , the fit is poor ( @xmath92 ) . this is mainly due to the observed count rate being greater than that of the model at higher energies , and can be easily seen from figure 6 . in the case of _ rxte _ analysis , the additional data at higher energies allows us to constrain the component that dominates in the hard x - ray band . this in turn , affects the results of spectral modeling at lower energies . finally , we present a slight variation of our model in which the power law is flattening below some energy between @xmath93 kev . the motivation for this model is two - fold . first , unusually large power is required to be injected into the galaxy in order to explain the presence of the power law tail via non - thermal bremsstrahlung process . we will discuss the physical reasons for this in 4.2 . second , there is evidence for the flattening of the slope at lower energies ( see 3.6 where we discuss simultaneous fitting of the spectrum with that from _ osse _ ) . but for now , we will only present the model . the attenuation of the power law can be mimicked by multiplying the power law function by an exponentially absorbing function @xmath94 . table 4 lists the best fit parameters for this model . figure 7 shows the model and unfolded spectrum . approximately 58% of the flux in the @xmath76 kev is due to the thermal component in this model . let us now examine the spectrum of the diffuse emission from the ridge at larger galactic latitudes . from figure 2 , it can be seen that the broad spatial component dominates the emission at larger latitudes . we used the same model discussed in the previous section to fit the data for galactic regions r2 and r3 ( table 3 ) . the results of the fits are summarized in tables 5 and 6 , respectively . since , the broad component is expected to be less absorbed than the thin disk component , it is reasonable that the average values of the hydrogen column density for these regions have decreased compared to that of the central ridge . the key fit parameters ( namely @xmath95 of the rs component and the photon index of the power law ) have remained within error bars of those derived for the central ridge indicating similar physical origins at higher latitudes . the average flux in the @xmath4 kev band from both regions is about 82% of that from the central ridge after removing the effect of absorption . let us now estimate the x - ray luminosity of the diffuse emission for the galactic disk in the @xmath96 kev and @xmath97 kev bands . in reality , this is a complex problem since it depends on the geometry and spatial distribution of the emitting and absorbing regions . a rigorous treatment of this problem is outside of the scope of this paper . however , we proceed to estimate the luminosity by making a few simple assumptions . from our spatial modeling of the latitude distribution of the emission ( 3.3 ) , we found that 2 spatial components ( i.e. a thin disk and a broad component ) contribute to the emission . since the flux at @xmath98 is dominated by the broad component ( as seen from figure 2 ) , we will first estimate the average volume emissivity of the broad component from spectral models of regions r2 and r3 . from the models presented in tables 5 and 6 , we find an average flux of @xmath99 and @xmath100 in the @xmath96 kev and @xmath97 kev bands , respectively , after removing the effect of absorption . dividing the flux by the _ effective _ solid angle of _ rxte _ ( @xmath101 sr ) yields an average surface brightness @xmath102 of @xmath103 . for a spherical geometry with uniform emissivity , the average volume emissivity along the line of sight can be estimated to be @xmath104 where @xmath105 is the depth of the emitting region . this is a good approximation for the broad component . we assume that the distance to the galactic center and the radius of the galactic disk are 8 and 10 kpc , respectively . choosing an average line of sight of length 16 kpc , we find the average volume emissivity @xmath106 to be @xmath107 . the fwhm of the broad component is about @xmath108 corresponding to a disk of height @xmath42 kpc . this yields @xmath109 for the the x - ray luminosity of the broad disk . from our spatial model presented in 3.3 , the ratio of the luminosity of the broad to the thin disk is about 14 . the total luminosity of the disk ( thin plus broad ) in this model is then estimated to be @xmath110 in the @xmath96 kev band . similarly , we derive a luminosity of @xmath111 in the @xmath97 kev . for comparison purposes , we should mention that koyama et al . ( 1986 ) have reported a luminosity of @xmath112 from observations with _ tenma _ ( @xmath113 kev ) . worrall et al . ( 1982 ) have reported a luminosity of @xmath114 in the @xmath96 kev band with heao-1 , while warwick et al . ( 1985 ) have reported a luminosity of @xmath115 with _ exosat _ in the @xmath116 kev band ( notice that this band does not include the iron line emission ) . from the latest _ asca _ survey , kaneda et al . ( 1997 ) report a luminosity of @xmath117 for the hard component in the @xmath118 kev band where the hard component is modelled by a bremsstrahlung of temperature 7 kev . the effect of absorption has been taken into account in all the previously reported estimates . in the previous sections , we established the presence of a hard power law tail above 10 kev . recent diffuse emission measurement of the galactic plane at longitude @xmath119 with _ osse _ in the @xmath120 kev band also reveals that the spectrum can be fitted with a power law model ( @xmath121 ; skibo et al . [ 1997 ] ) . if the spectrum extends smoothly from the hard x - ray to the soft @xmath13-ray regime , it means that the same physical processes are dominating the emission in a wide bandpass . unfortunately _ rxte _ data at galactic longitude of @xmath119 are not available yet . however , assuming that the spectral shape of the diffuse emission is the same as a function of galactic longitude for @xmath122 ( purcell et al . 1996 ) , and that only the normalization varies , we attempt a simultaneous spectral fit to the spectrum of the diffuse emission from both _ rxte _ and _ osse_. we use the _ rxte _ data from the central ridge ( r1 ) above 10 kev and simultaneously fit it with the data obtained with _ osse _ at galactic longitude of @xmath119 . the results of the fit appear in figure 8 . the fit yields a power law photon index of @xmath123 which is within the errors bars of the values derived from either _ rxte _ above 10 kev ( @xmath66 ) or _ osse _ ( @xmath121 ) alone . the normalization for the _ rxte _ and _ osse _ data are @xmath124 and @xmath125 @xmath126 at 1 kev , respectively . the fit is acceptable with a total @xmath61 of @xmath127 for 118 degrees of freedom . in order to compare the normalizations of the power law component derived from _ rxte _ and _ osse _ , we must calculate the flux per steradian ( @xmath128 ) for each instrument . however , for extended emission , the measured flux depends on the field of view of each instrument and the distribution function of the diffuse emission . hence , an _ effective _ solid angle ( i.e. the convolution of the distribution of the emission with the detector s response function ) should be calculated for each instrument . in the case of _ osse _ ( fov of @xmath129 fwhm ) , convolving the soft @xmath13-ray galactic diffuse emission ( assuming a gaussian distribution with @xmath130 fwhm [ purcell et al . 1996 ] ) , and _ osse _ s triangular response function yields an _ effective _ solid angle of @xmath131 sr . similarly _ rxte _ s solid angle is @xmath132 sr . therefore , the normalization parameter for the power law component from _ osse _ should be multiplied by @xmath133 to be compared with the normalization derived from _ rxte_. this yields a normalization of @xmath134 at 1 kev for the power law component measured with _ osse _ , which is a factor of @xmath135 smaller than the normalization of the power law component from _ rxte _ data . this is expected since the intensity of the emission seen with _ rxte _ decreases away from the galactic center . if we assume that the physical origin of the power law is the same for the averaged central ridge and at galactic longitude @xmath119 , then the combined _ rxte / osse _ fit gives a narrower constraint on the power law slope . however , the slope at higher energies is noticeably steeper than that derived for the spectrum below 35 kev , perhaps hinting that the power law either flattens or slowly attenuates at lower energies . we first consider the contribution of unresolved discrete sources to the diffuse emission spectrum . the important question is whether there exist classes of unresolved discrete sources with appropriate spatial number densities and similar spectra to that of the diffuse emission that can comprise the bulk of the spectrum . in the hard x - ray band ( above 10 kev ) , the number of known galactic hard x - ray sources with similar spectra to that of the galactic ridge diffuse emission is small ( levine et al . black hole candidates in their soft state have spectra similar to that of the ridge , but this is usually seen at high luminosities when individual sources can be easily detected . neutron star binaries are known to have hard power law spectra with an exponential cut off around 20 kev . however , only a handful of them are known to be active at a time in the galactic disk . also , their luminosities are mostly in the range @xmath136 which can easily be detected from the diffuse emission . unresolved pulsars are also thought to contribute to the galactic ridge diffuse emission particularly in the @xmath13-ray band . a recent study , however , concludes that pulsar contribution to the diffuse galactic emission is significant only above 1 gev ( pohl et al . 1997 ) . in the soft @xmath13-ray band , sigma_/_osse _ observations of the galactic center region has been performed to distinguish the diffuse emission from that of compact sources ( purcell et al . comparison of the source excluded spectrum from the galactic center and that from regions which are known to have fewer sources ( i.e. @xmath137 and @xmath138 ) has revealed that the spectra are similar in both intensity and spectral index . if the emission is due to discrete sources , 10 x - ray sources of flux @xmath139 at 100 kev must be present in the field of view of _ osse _ to make up the spectrum . because of the lack of knowledge of such class of sources with a uniform space density in longitude , it is likely that the emission above 10 kev is of diffuse origin . below 10 kev , because of its better spatial resolution , _ asca _ is far better equipped than _ rxte _ to resolve individual faint sources . a recent fluctuation analysis of the surface brightness of the ridge emission in the scutum arm region ( @xmath140 ) with _ asca _ has concluded that if discrete sources are to make up the entire spectrum , their luminosity must be less than @xmath141 ( yamauchi et al . this conclusion narrows the field of candidate class of sources that can make up the diffuse emission spectrum of the galactic ridge . among possible class of sources , neutron stars ( with x - ray luminosities mostly in the range @xmath142 ) are unlikely due to their large luminosities . a likely class of sources that has been thought to account for the emission is the rs cvn type binaries , since they exhibit the iron k line in their emission spectra however , a study by ottmann & schmitt ( 1992 ) on the contribution of rs cvn systems to the galactic diffuse emission concludes that while they contribute about 6% to the total galactic background , their contribution to the galactic ridge iron line is too small ( @xmath143 ) . therefore , given the strength of the iron line in the spectrum of the diffuse emission , it seems that unresolved rs cvn sources do not contribute to the bulk of the emission . cataclysmic variables ( cvs ) are another candidate that may contribute to the galactic ridge diffuse emission . they have similar spectra to that of the ridge and their luminosities fall in the range @xmath144 . however , from the analysis of the _ asca _ gis pointings of the galactic ridge , kaneda ( 1997 ) has shown that in order for cvs to make up the diffuse emission , their space densities would have to be @xmath145 , a factor of 100 larger than the estimated density of cvs ( patterson 1984 ) . for these reasons , it is more likely that the emission is due to diffuse or extended sources . a more detailed review of the previous works on the contribution of the discrete sources to the ridge spectrum can be found in kaneda ( 1997 ) and yamasaki ( 1996 ) , and references therein . finally , we point out that with the upcoming launch of _ axaf _ , a more sensitive number - flux distribution of galactic sources and their exact contribution to the diffuse galactic background can be determined . based on our measurements presented in this paper , we conclude that the diffuse emission from the ridge in the broad x - ray and soft @xmath13-ray band has both thermal and possibly non - thermal origins . in what follows , we speculate on the origin of the emission , first focusing on the thermal component and then speculating on the origin of the power law tail . the thermal component can be modelled with a raymond - smith plasma of temperature @xmath5 kev ( tables 3 and 4 ) . however , we point out that the spectral resolution of _ asca _ gis or sis is far superior to that of _ rxte _ pca . therefore , in order to interpret the spectrum accurately , we need to carefully examine the spectral features of the diffuse emission below 10 kev using the _ asca _ data . for this reason , we refer to the latest investigation of the diffuse emission from the galactic ridge with _ asca _ ( kaneda 1997 ; kaneda et al . the spectral examination of the _ asca _ data has clearly verified the presence of lower energy emission lines from mg , si , and s. the spectrum has also been carefully examined for the presence of fe k lines . while the emission line from he - like iron was clearly detected , the presence of the emission lines from neutral and h - like iron was not firmly established . this is important since the equivalent width of the h - like iron line sets an upper limit for the plasma temperature . kaneda et al . ( 1997 ) give an upper limit of @xmath146 of that of the he - like iron for the normalization of the h - like iron line . this implies an upper limit of @xmath147 kev for the plasma temperature . this is in agreement with our results since the emission line from h - like fe is expected to be very weak from a @xmath148 kev plasma , supporting our finding that the thermal component of the emission arises from a gas of cooler temperature . therefore , we argue that unlike previous conclusions ( koyama et al . 1986 ; kaneda et al . 1997 ) , the temperature of the x - ray emitting gas need not be so high ( 5 - 15 kev ) in order to produce the observed spectrum . this interpretation lessens the difficulties toward explaining the origin and confinement of the super hot ( @xmath149 kev ) gas . due to the similarities between the characteristics of supernova remnants and the thermal nature of the diffuse emission spectrum presented here , we speculate that a collection of supernova remnants in the galactic disk , below current x - ray and/or radio detection limits , is responsible for the thermal origin of the diffuse emission . the idea was originally introduced by koyama , ikeuchi , & tomisaka ( 1986 , hereafter kit86 ) . however , in their calculation , they assumed that the temperature of the emitting gas is about @xmath20 kev . in addition to the fact that no snr with such high temperature has ever been detected , this assumption also requires a large sn explosion rate ( one per 10 years ) which exceeds the expected galactic rate of 2 - 3 per century ( tammann 1982 ) . below , we revisit their argument in order to derive the sn explosion rate capable of powering the thermal emission from the ridge using the adjusted temperature of @xmath148 kev for the emitting plasma . the first step is to estimate the criteria for which a given snr is below radio detection limit . due to absorption in the galactic disk , the minimum radio surface brightness for detecting a snr is @xmath150 . using the empirical relation between the surface brightness @xmath151 , and the diameter of the snr ( milne 1979 ) , and accounting for the ambient gas density as a function of scale height of the galactic plane ( tomisaka , habe , & ikeuchi 1980 ) , kit86 reduce the criteria to @xmath152 where @xmath153 , and @xmath154 where @xmath155 is the radius of the snr in the adiabatic stage described by the similarity solution ( sedov 1959 ; spitzer 1968 ) , and @xmath156 , @xmath157 and @xmath158 are the age ( years ) , total energy of the snr ( in units of @xmath159 erg ) , and the density of the ambient medium ( @xmath160 ) , respectively . substitution of equation ( 2 ) in equation ( 1 ) yields @xmath161 where @xmath162 . assuming that the thickness of the emitting region at the shock is @xmath163 , the x - ray luminosity of the blast wave shell can be calculated from @xmath164 where @xmath155 is described by equation ( 2 ) and @xmath165 is the total emissivity of the gas and depends on the electron density @xmath166 and the cooling function @xmath167 via the relation @xmath168 . for a strong shock @xmath169 . the cooling function @xmath170 is estimated to be ( raymond , cox , & smith 1976 ) @xmath171 for a gas of @xmath172 ( @xmath173 ) . in figure 9 , we have plotted the condition for the undetectability of a snr in the radio band ( eq . [ 1 ] ) , and the restriction on the temperature of the blast wave ( @xmath174 ) in the @xmath175 phase space where the temperature just behind the shock front can be calculated from @xmath176 and @xmath177 and @xmath178 are the mean mass of the gas particle and the boltzman constant , respectively . we have also plotted relevant constant luminosity lines . from the region of the allowed parameters in this plot , we conclude that the average luminosity and age of the contributing snrs is expected to be @xmath179 and @xmath180 years , respectively . the x - ray luminosity of the entire disk in the @xmath96 kev band from the _ thermal _ component ( which accounts for approximately 50% of the emission ) is estimated to be @xmath181 ( 3.4.2 ) . this implies that a sn explosion rate of @xmath182 per century is required to power the thermal emission from the entire disk . on the other hand , the luminosity of the thermal component of the thin disk is about @xmath183 . then a sn explosion rate of @xmath42 per century is sufficient to power the thermal emission from the thin disk . whether this scenario is viable depends on the number of snrs that are actually detected in the x - ray band in the galactic ridge . if the snrs have an average luminosity of @xmath184 , approximately 200 of them are needed to power the thermal emission from the ridge . so far , 215 snrs have been detected in the radio band ( green 1996 ) . based on the prevalence of snrs among a sample of unidentified galactic radio sources , helfand et al . ( 1989 ) have built a model which predicts the existence of up to @xmath185 snrs in the galactic disk in the radio band . according to their results , these remnants are within detection capability of current instruments but are not yet identified . one of the goals of the _ asca _ galactic plane survey project is to identify snrs in the galactic disk ( in the region @xmath186 and @xmath187 ) that have already been catalogued in other wavebands , and to discover new ones . because of galactic absorption , remnants with low surface brightness will be missed in the survey . however , snrs with fluxes greater than 10 jy at 1ghz will be detected with _ asca _ provided no bright sources are in the field of view . from the most recent _ asca _ survey covering the region @xmath188 and @xmath189 pointed at @xmath30 , 13 snr were detected ( yamauchi et al . 1998 and references therein ) , where 25 snrs in radio and 1 snr in x - ray ( discovered with rosat ) were already known to exist . based on this trend , it can be estimated that approximately half of the radio snrs have significant emission in x - ray . it may then be expected that out of the 215 snrs catalogued in radio , approximately 100 of them should be detected in x - ray . if , we take the prediction of helfand et al . ( 1989 ) at face value , this number increases to @xmath190 . ultimately , whether the galactic snr population can power the thermal emission from the ridge depends on the final number of detected snrs and their average luminosity when the _ asca _ survey is completed . we now turn to the discussion of the origin of the power law tail of the spectrum dominating the hard x - ray / soft @xmath13-ray band . recent analysis of diffuse emission from the galactic ridge with _ asca _ ( kaneda et al . 1997 ) indicates that the spectrum can be fitted with a two temperature non - equilibrium ionization plasma of @xmath191 and 7 , respectively . two factors lead us to believe that the reported high temperature component of 7 kev with _ asca _ is in fact a different interpretation of the power law component reported in this paper . firstly , _ asca _ s sensitivity above 10 kev decreases substantially and goes to zero around 12 kev . thus , distinguishing the power law from the hard thermal component in the _ asca _ data may not be possible . secondly , extension of the two - temperature model above 10 kev does not produce a good fit to the data at higher energies ( as shown in on the other hand , we have unambiguously shown here the presence of the hard power law tail in hard x - ray data . this leads us to conclude that the previously reported super hot ( @xmath192 kev ) component of the ism , is most likely due to non - thermal emission with a power law spectrum that dominates above 10 kev and extends to beyond 100 kev . below , we speculate on the origin of the power law tail . one scenario is to interpret the hard x - ray emission as the result of the superposition of non - thermal tails from snrs . previously non - thermal x - ray emission from sn1006 ( koyama et al . 1995 ) and rx j1713.7 - 3946 ( koyama et al . 1997 ) has been reported . non - thermal @xmath13-ray emission from cas a ( the et al . 1995 ) and a high energy tail in its x - ray spectrum which can be possibly attributed to non - thermal processes ( allen et al . 1997 ) has also been discovered . however , it is not clear if this scenario can account for the bulk of the emission . the x - ray luminosity of the non - thermal emission discovered is about an order of magnitude smaller than the luminosity of the thermal component . in the case of cas a for example ( g. e. allen , private communication ) , the luminosity in the 2 - 10 kev band is @xmath193 , while the luminosity in the 10 - 60 kev is about @xmath194 . from our estimates , the luminosity of the ridge is approximately @xmath195 and @xmath196 in the 2 - 10 kev and 10 - 60 kev bands , respectively . the contribution of the snrs to the thermal component of the spectrum was previously addressed in this section . for the power law tail ( interpreted as non - thermal component ) , this suggests that approximately 750 snrs with non - thermal luminosities of @xmath194 are needed to make up the high energy tail of the spectrum . thus far , substantial number of remnants that exhibit non - thermal emission has not been discovered . therefore , the viability of this scenario awaits further observations . x - rays of 10 - 100 kev can also be produced via inverse compton scattering of energetic electrons on the ambient microwave background , optical , and infrared photons . a power law scattered photon spectrum with photon index @xmath197 is expected if the electron spectrum has energy index @xmath198 . however , the exact contribution from this process is model dependent due to the uncertainty in the interstellar radiation field and the electron spectrum . direct measurement of the electron spectrum is reliable only at gev energies because of the effects of solar modulation at lower energies . irrespective of existing models for the compton scattering contribution and their extrapolation to lower energies , skibo , ramaty , & purcell ( 1996 ) have argued that this mechanism does not produce the bulk of the emission below 100 kev . since compton scattering of microwave background and star light photons require electron energies of 100 mev-10 gev , the electron spectrum must turn up somewhere in that energy range . however , electrons at these energies produce the galactic synchrotron radio emission at 100 mhz frequencies . but since the observed radio spectrum breaks at @xmath199 mhz in the opposite sense to that of the @xmath13-ray emission , they argue that an inverse compton origin is unlikely . a likely possibility for the origin of the power law tail is non - thermal bremsstrahlung . however , bremsstrahlung of sub - mev electrons is highly inefficient due to the fact that ionization and coulomb collision losses exceed bremsstrahlung radiation . for example , for a 100 kev electron , the radiation yield ( i.e. the ratio of energy loss from radiation to that from collision ) is @xmath200 ( berger & seltzer 1964 ) . for a 10 kev electron , the yield drops to @xmath201 . from our estimate , the luminosity of the ridge in the @xmath97 kev band is @xmath202 . an order of magnitude estimate implies that in order to account for the power law tail via electron bremsstrahlung , a power of @xmath203 is required . skibo et al . ( 1996 ) have calculated that a total power input up to @xmath204 is required if the tail extends down to 10 kev . the total injected power into the galaxy via supernova explosion is only @xmath205 assuming @xmath206 is released in a sn explosion every 30 years . several attempts have been made to explain the source of this large power . skibo et al . ( 1996 ) have suggested that the power is derived from the release of the gravitational potential upon the passage of ism through galactic spiral arm compressions . schlickeiser ( 1997 ) explains the implied presence of large population of sub - mev electrons with the existence of interstellar in - situ reacceleration of cosmic ray particles by the ambient interstellar plasma turbulence . based on the empirical results of modelling the data in the wide bandpass of 3 - 500 kev , we suggest that the power law tail should gradually attenuate below some energy in the 10 - 100 kev range ( table 4 ) . this speculation is motivated by comparing the spectral fits of the data above and below 10 kev . above 10 kev , we found that the spectrum can be modelled with a power law of photon index @xmath207 up to 500 kev . on the other hand , when modelling the data in the @xmath4 kev band , the power law slope is @xmath208 . this interpretation lessens the large power requirement for the origin of the power law component via bremsstrahlung and results in the required injection power to be consistent with that expected from sn explosion in the galaxy . unfortunately , the electron spectrum can not be measured due to solar modulation at lower energies . however , our interpretation of the data in this way makes a prediction for the shape of the electron spectrum at lower energies . a detailed calculation of the electron spectrum in this scenario is out of the scope of this paper . in addition to electron bremsstrahlung , we suggest the possibility that _ inverse _ ( or proton ) bremsstrahlung may also contribute to the diffuse emission from the galactic ridge . in the inverse process , the rapidly moving proton radiates instead of the electron and the center of momentum of the proton - electron system is approximately that of the moving proton . boldt & serlemitsos ( 1969 ) have shown that significant radiation is associated with the collisions of suprathermal protons with ambient electrons ( free and bound ) . for the non - relativistic case ( @xmath209 ) , the spectral intensity from the bremsstrahlung of an electron with kinetic energy @xmath210 is the same as that from a suprathermal proton of kinetic energy @xmath211 , where @xmath212 and @xmath213 are the rest mass of the electron and proton , respectively . for example , an electron with kinetic energy of 10 kev produces the same spectral intensity as that of a proton with kinetic energy of @xmath214 mev . the scenario is particularly attractive in the context of particle acceleration and reacceleration in supernova shocks via the diffusive fermi process . for many years , the perception has persisted that supernova remnants are a major source of galactic cosmic rays ( e.g. lagage & cesarsky 1983 ) . recently much work has been done regarding the production of @xmath13-rays in the snrs . modelling of @xmath13-ray production via interaction of the energetic particles with the remnants ambient environment through neutral pion decay , bremsstrahlung , and inverse compton scattering is reviewed by baring ( 1997 ) . one expects that if both electrons and protons are accelerated with equal velocities in the explosion process , both should contribute to the bremsstrahlung emission . the ratio of the bremsstrahlung losses to the collisional losses is roughly the same for both electrons and protons since it is only dependent on the velocity of the radiating particles . however , a simple calculation shows that not only the accelerated protons lifetime is longer than that of electrons , but also they travel farther than electrons from their birthplace . to illustrate this , let us consider bremsstrahlung radiation from an electron of 25 kev . its spectral intensity is similar to that from a proton of 50 mev . however , the range @xmath215 of the 25 kev electron in hydrogen is @xmath216 as opposed to @xmath217 for the 50 mev protons ( berger & seltzer 1964 ; barkas & berger 1964 ) . this translates to distances of approximately 100 pc and @xmath218 mpc for the electrons and protons , respectively . the radial distance @xmath219 from the birthplace of the energetic particles which the particles will travel before they lose all their energy depends on the random walk step length @xmath220 via the relation @xmath221 . while the value of @xmath220 is not well known , it is expected to be much larger than the gyro radii of the particles for the expected galactic magnetic fields but somewhat smaller than the height of the galactic disk . a good estimate is that it is on the order of the coherence length of the magnetic fields . as a quantitative value for this length , we use the study of kim , kronberg , & lancecker ( 1988 ) where they probed the structure of magnetic fields in and around the supernova remnant oa 184 ( g166.2 + 2.5 ) using background radio source rotation measures . based on their study , they find an interstellar magnetic irregularity scale of order 100 pc . using this value , we find the total radial distance @xmath219 from the birthplace of the particles to be approximately 100 pc and @xmath222 kpc for the electrons and protons , respectively . in any case , it is clear that the radial distance travelled by the protons is much larger than that by electrons . likewise , the energy loss lifetime of protons ( @xmath223 ) is much larger than that of the electrons ( by a factor of @xmath224 ) since it is proportional to the mass of the particles . from these estimates , one expects that the proton bremsstrahlung should play a significant role if bremsstrahlung is responsible for the presence of the hard power law tail in the spectrum . this is due to the longer range and energy loss lifetime of the protons compared to electrons . if indeed the birthplace of the energetic electrons and protons are the sites of supernova explosions in the galaxy , electron bremsstrahlung will produce localized hard x - ray emission ( i.e. hot spots ) . on the other hand , proton bremsstrahlung will cause the radiation to be smoothly distributed over the galaxy . this interpretation implies that snrs play a dominant role in the energetics of ism via both thermal and non - thermal emission processes . finally , we calculate the hydrogen ionization rate for the ism in this model . if non - thermal bremsstrahlung dominates above 10 kev , then from our model , the average volume emissivity in the @xmath97 kev for the ridge is @xmath225 . the average radiation yield in the @xmath97 kev band is @xmath226 . this yields a collisional loss rate of @xmath227 . for an average energy loss of 36 ev ( @xmath228 erg ) per hydrogen atom ionized in the ism ( dalgarno & griffing 1958 ) , we find an average ionization rate of @xmath229 for the entire ridge where @xmath230 is the average hydrogen density . previous values derived from observations have been within @xmath231 ( dalgarno & mccray 1972 ; reynolds et al . 1973 from @xmath232 observations ) . our estimate is consistent with these values for an average @xmath233 in the ism . independent of @xmath230 , from the intensity of galactic @xmath234 background at high galactic latitudes , reynolds ( 1984 ) has derived @xmath235 hydrogen ionizations ( with a `` best fit '' value of @xmath236 ) per @xmath237 column perpendicular to the galactic disk for the region within @xmath148 kpc of the sun . our results imply @xmath238 hydrogen ionizations averaged for the ridge . since in our model , the thickness of the broad disk is @xmath42 kpc , our results imply hydrogen ionizations of @xmath239 of the galactic disk which is larger than the @xmath234 estimate by more than a factor of 2 . one reason could be that our estimate is derived by averaging over the central @xmath0 of the plane in longitude where the hard x - ray emission is more intense , while the rate from @xmath234 observation is for regions within @xmath148 kpc of the sun . in any case , this indicates that the collisional losses associated with the bremsstrahlung emission from energetic particles to produce the hard x - ray tail are more than adequate to account for the observed hydrogen ionization rate in the ism . our results are summarized as follows : ( 1 ) from the _ rxte _ survey of the galactic plane , we find that the diffuse emission from the galactic ridge has two spatial components : a thin disk of width @xmath240 , and a broad component with a functional form that can be approximated by a gaussian distribution of about @xmath108 fwhm . assuming an average distance of 16 kpc to the edge of galaxy , this translates to a height of 70 pc and 500 pc for the thin and broad disk , respectively . ( 2 ) a hard power law tail is clearly detected in the spectrum and dominates above 10 kev , implying that the emission in hard x - ray possibly has non - thermal origin . on the other hand , the detection of the emission line from he - like iron in the spectrum ( and also lower energy lines from mg , si , s in the spectrum from _ asca _ ) motivates the idea that part of the emission below 10 kev has thermal origin . ( 3 ) the spectrum in the @xmath4 kev band can be well modelled with a raymond - smith plasma component of @xmath148 kev plus a power law component of photon index @xmath6 . the spectrum above 10 kev simultaneously fitted with _ osse _ s spectrum can be fitted with a power law of photon index @xmath123 . the change in the power law slope at lower energies hints at the possibility that the power law either flattens or gradually attenuates below some energy between 10 - 100 kev . ( 5 ) the characteristics of the thermal component of the diffuse emission resembles that of the snrs . from this interpretation , we calculate that a sn explosion rate of less than 5 per century is adequate to power the thermal emission from the ridge . ( 6 ) the origin of the emission in the hard x - ray band modelled by a power law is uncertain . while unresolved discrete sources are expected to contribute , the bulk of the emission is expected to be of diffuse origin . one possibility is the non - thermal bremsstrahlung of electrons and protons which may have been accelerated in the snr sites . based on the empirical modeling of the data in a wide energy band , we speculate that the power law either flattens or gradually attenuates at lower energies . this lessens the large power injection required to explain the hard power law tail via either electron or proton bremsstrahlung . it will also make it consistent with the power injected to the galaxy via supernova explosion . ( 7 ) the hydrogen ionization rate implied from our model is @xmath229 averaged over the central ridge . generally , values of @xmath231 are expected in the ism . this is an indication that the collisional losses associated with the bremsstrahlung radiation of energetic particles that produce the hard power law tail are sufficient to explain the ionization rates observed in the ism . finally , we expect that simultaneous _ rxte_/_osse _ observations of the diffuse emission from the galactic ridge will allow the exclusion of hard discrete sources from the spectrum and will tightly constrain the hard tail slope that dominates in the hard x - ray / soft @xmath13-ray band . this will enhance our understanding of the origin of the power law tail and the energetics of ism . we are grateful to elihu a. boldt for many valuable discussions , in particular for bringing the process of proton bremsstrahlung to our attention and his contribution to the relevant arguments presented in this paper . we also thank him for a careful reading of the manuscript . we acknowledge g. e. allen , e. boldt , k. ebisawa , s. hunter , k. jahoda , g. m. madejski , r. m. mushotzky , and r. petre for useful discussions . we are grateful to j. g. skibo for providing the diffuse emission spectrum measured with _ osse_. the help of pat tyler and seth digel in producing plate 1 is greatly appreciated . we thank the referee , k. koyama , for many insightful comments that led to the improvement of this paper . acknowledges support from the national academy of sciences and the national research council through a research associateship at nasa s goddard space flight center . green , d. a. 1996 , a catalogue of galactic supernova remnants ( 1996 august version) , mullard radio astronomical observatory , cambridge , uk ( available on the www at http://www.mrao.cam.ac.uk/survey/snr/ ) . j1744 - 28 & 0.044 & 0.301 grs1739 - 278 & 0.660 & 1.167 sl1735 - 269 & 0.785 & 2.400 k1731 - 260 & 1.074 & 3.655 4u1744 - 26 & 2.294 & 0.793 gx5 - 1 & 5.079 & -1.018 4u1745 - 203 & 7.725 & 3.795 gx9 + 1 & 9.072 & 1.154 gx13 + 1 & 13.518 & 0.108 gx17 + 2 & 16.432 & 1.276 4u1812 - 12 & 18.033 & 2.362 g21.5 - 0.9 & 21.493 & -0.887 gr11(1 ) & 22.001 & -0.002 exo1846 - 031 & 29.954 & -0.917 g1843 + 009 & 33.108 & 1.746 gps1858 + 015 & 35.416 & -1.648 xte j1856 + 053 & 38.258 & 1.266 exo1902 + 054 & 39.33 & -0.390 ss433 & 39.695 & -2.247 4u1909 + 07 & 42.089 & -0.940 w49b & 43.275 & -0.189 a1907 + 09 & 43.743 & 0.477 grs1915 + 105 & 45.366 & -0.219 msh15 - 52 & 320.321 & -1.161 circinusx-1 & 322.117 & 0.038 4u1553 - 542 & 327.945 & -0.857 2e1614 - 505 & 332.421 & -0.422 1608 - 522 & 330.926 & -0.850 4u1624 - 49 & 334.914 & -0.263 gx340 + 0 & 339.587 & -0.079 2s1702 - 429 & 343.887 & -1.314 oao1653 - 40 & 344.369 & 0.326 nsco94 & 344.982 & 2.456 4u1708 - 408 & 346.327 & -0.928 gps1709 - 396 & 347.300 & -0.333 4u1700 - 37 & 347.755 & 2.173 gps1713 - 388 & 348.394 & -0.543 ctb37a / b & 348.598 & 0.232 gx349 + 2 & 349.105 & 2.750 exo1722 - 363 & 351.474 & -0.548 a1744 - 361 & 354.121 & -4.192 mxb1728 - 34 & 354.305 & -0.151 x1730 - 333 & 354.843 & -0.159 terzan2 & 356.320 & 2.298 sl1746 - 331 & 356.816 & -2.997 h1741 - 322 & 357.125 & -1.607 r1747 - 313 & 358.556 & -2.168 2s1742 - 294 & 359.559 & -0.390 a1742 - 289 & 359.930 & -0.043 absorption & @xmath79 ( @xmath247 ) & @xmath248 raymond - smith & @xmath95 ( kev ) & @xmath249 & abundance ( solar ) & @xmath250 & normalization & @xmath251 power law & photon index & @xmath208 & normalization & @xmath252 & @xmath253 & absorption & @xmath79 ( @xmath247 ) & @xmath254 raymond - smith & @xmath95 ( kev ) & @xmath255 & abundance ( solar ) & @xmath250 & normalization & @xmath256 power law & photon index & @xmath257 & @xmath258 & @xmath259 & normalization & @xmath260 & @xmath261 & absorption & @xmath79 ( @xmath247 ) & @xmath262 raymond - smith & @xmath95 ( kev ) & @xmath263 & abundance ( solar ) & @xmath250 & normalization & @xmath264 power law & photon index & @xmath265 & normalization & @xmath266 & @xmath267 & absorption & @xmath79 ( @xmath247 ) & @xmath268 raymond - smith & @xmath95 ( kev ) & @xmath269 & abundance ( solar ) & @xmath250 & normalization & @xmath270 power law & photon index & @xmath271 & normalization & @xmath272 & @xmath273 &

absorption and emission lines originating from the c - h stretching manifold of gas phase ch were discovered in the high resolution (=25,000 ) infrared l band spectrum along the line of sight toward ngc 7538 : irs9 . these observations provide a diagnostic of the complex dynamics and chemistry in a massive star forming region . the line shapes resemble p cygni profiles with the absorption and emission components shifted by with respect to the systemic velocity . similar velocity components were observed in co at 4.7 , but in contrast to ch , the co shows deep absorption due to a high velocity outflow as well as absorption at the systemic velocity due to the cold outer envelope . it is concluded that the gas phase ch abundance varies by an order of magnitude in this line of sight : it is low in the envelope and the outflow ( x[ch] ) , and at least a factor of 10 larger in the central core . the discovery of solid ch in independent ground and space based data sets shows that methane is nearly entirely frozen onto grains in the envelope . it thus appears that ch is formed by grain surface reactions , evaporates into the gas phase in the warm inner regions of protostellar cores and is efficiently destroyed in shocks related to outflows .

astro-ph0407270

[ sec : intro ] among the suite of molecules observed in interstellar and circum - protostellar media , methane is relatively poorly studied . ch@xmath1 has no permanent dipole moment , and therefore can not be observed by pure rotational transitions at radio wavelengths . its strongest fundamental ( @xmath8 ) ro - vibrational transitions , i.e. the @xmath9 c - h stretching and the @xmath10 c - h bending modes , occur at 3.32 and 7.67 @xmath5 , respectively . strong telluric absorption at these wavelengths hinders interstellar ch@xmath1 studies . gas and solid phase ch@xmath1 have been positively identified at 7.67 @xmath5 in three sight - lines : the massive protostars ngc 7388 : irs9 , w 33a , and gl 7009 s @xcite . ch@xmath1 was detected only in the solid state in low resolution _ infrared space observatory ( iso ) _ observations at 7.67 @xmath5 toward @xmath11 low mass protostars @xcite , the galactic center @xcite , and an external galactic nucleus @xcite . a recent low resolution _ spitzer space telescope _ spectrum shows absorption by solid ch@xmath1 toward the young , low mass protostar iras 08242 + 5100 @xcite . observations of ch@xmath1 provide insight into the basic principles of astrochemical networks . in the presence of atomic c and h , ch@xmath1 is rapidly formed on cold grains , just as h@xmath12o is formed from atomic o and h ( e.g. @xcite ) . perhaps surprisingly , the observed gas and solid state ch@xmath1 abundances are low , not more than a few percent with respect to co and h@xmath12o @xcite . this points to relatively low atomic c abundances at the time of ch@xmath1 formation , when most c is already locked up in co. in this same scenario the high ch@xmath13oh abundances in several lines of sight may be explained by hydrogenation of abundantly present co @xcite , and large co@xmath12 abundances by oxidation of co. the dominance of a co- and o - based , rather than c - based , chemistry is stressed by the profiles of solid co@xmath12 absorption bands which show an intimate mixture of co@xmath12 , ch@xmath13oh , and h@xmath12o in the ices @xcite . nevertheless , even at relatively low abundances methane is proposed to be the starting point of a rich chemistry leading to the complex organic molecules observed in dense clouds like the tmc-1 ridge @xcite . to further test the role of ch@xmath1 in interstellar chemistry , basic diagnostics , such as gas phase abundances , gas / solid phase abundance ratios , gas phase temperatures and abundance gradients need to be measured in more lines of sight . in this work we report the discovery of absorption and emission features arising from the 3.32 @xmath5 c - h stretching mode of gas and solid phase ch@xmath1 . a previous attempt to detect ch@xmath1 at this wavelength was unsuccessful and indicated an abundance of @xmath14 with respect to h in the orion bn object @xcite . we study the line of sight toward the massive protostar ngc 7538 : irs9 , because its high radial velocity of @xmath15 @xmath4(e.g . @xcite ) facilitates the separation of telluric and interstellar lines . ngc 7538 : irs9 is a well studied massive protostar , known for its large ice and gas columns @xcite and a prominent molecular outflow ( mitchell , maillard , & hasegawa 1991 ) . the r(0 ) and r(2 ) lines of the bending mode of gas phase ch@xmath1 and solid state ch@xmath1 were detected in this source by @xcite . unambiguous confirmations of these detections were obtained with space based observations @xcite . here we independently confirm the presence of ch@xmath1 by detecting many lines from the 3.32 @xmath5 stretching mode as well as the ice band . the high spectral resolution of these observations permits a line profile analysis in which we determine gas phase conditions in the various environments along the line sight , such as the envelope , the hot core , and the outflow . this , in turn , elucidates the formation and destruction pathways of ch@xmath1 in these different environments . the keck / nirspec and iso / sws observations are discussed in [ sec : obs ] , and presented in [ sec : feat ] . a profile analysis of the gas phase absorption and emission ch@xmath1 lines , in conjunction with previously reported high resolution co observations , is presented in [ sec : ch4 ] and [ sec : co ] . the ice band profile is discussed and compared with the c - h bending mode in [ sec : ice ] . the implications of these results on the formation and evolution history of the ch@xmath1 molecule in protostellar environments are discussed in [ sec : disc1 ] . the origin of the low velocity , warm gas , responsible for the p cygni - like line profiles , is further constrained in [ sec : disc2 ] . we conclude with future prospects in [ sec : concl ] . [ f : obs ] [ sec : obs ] [ sec : obsns ] the high resolution l band spectrum of the massive protostar ngc 7538 : irs9was obtained with the nirspec spectrometer @xcite at the keck ii telescope atop mauna kea . observations were made on five different nights with clear and dry weather conditions : ut 2002 december 17 and 18 , 2003 july 10 and 11 , and 2003 november 2 . nirspec was used in the echelle mode with the 0.43@xmath16 slit , providing a resolving power of @xmath17 ( @xmath1812 @xmath4 ) . the 2.94.1 @xmath5 region was almost fully covered in five echelle settings , each covering five nyquist sampled wavelength orders of @xmath19 @xmath5 on the 1024@xmath20 pixel array . two small regions centered on 2.97 and 3.12 @xmath5 were not covered . the on - source integration time per setting was 20 minutes . several ch@xmath1 ro - vibrational lines as well as the solid ch@xmath1 band at 3.32 @xmath5were observed multiple times to verify their reality . at @xmath21 @xmath4 for ngc 7538 : irs9 , the doppler velocity with respect to earth was @xmath2253.5 , @xmath2282.4 , and @xmath2264.0 @xmath4 for the three observing runs . this proved to be sufficient to separate interstellar and telluric ch@xmath1 lines and guarantees their reality . the spectra presented in this paper are an average of all the data . the data were reduced in a standard way , using idl routines ( boogert , blake , & tielens 2002 ) . atmospheric absorption features were divided out using the standard star hr 8585 ( a1v ) , which is bright ( @xmath23=3.78 ) and reasonably close to ngc 7538 : irs9 ( an airmass of 1.20 - 1.30 versus 1.34 for ngc 7538 : irs9 ) . the spectral shape and hydrogen absorption features in the standard star were divided out with a kurucz model atmosphere . the hydrogen absorption profiles in the kurucz model are accurate to 1 - 2% , and residuals can be seen near strong lines . this does not affect the region of main interest for this paper ( 3.153.65 @xmath5 ) , except perhaps near 3.297 @xmath5 due to the hi pf @xmath24 line . overall , a good telluric correction was achieved , resulting in signal - to - noise values of @xmath18110 in the region of ch@xmath1 at 3.32 @xmath5 , @xmath25 at the bottom of the h@xmath12o ice band at 3.0 @xmath5 and @xmath26 at wavelengths @xmath27 @xmath5 . strong atmospheric lines , i.e. with less than 50% of the maximum transmission in each setting , leave residuals and were removed from the final spectrum . the spectra were wavelength calibrated on the atmospheric emission lines , using an atmospheric model spectrum ( p. van der valk and p. roelfsema , priv . all orders were subsequently combined by applying relative multiplication factors . for the analysis of the emission lines ( [ sec : ch4 ] ) , the spectrum is flux calibrated by scaling to the continuum brightness of the data presented in @xcite . finally , in conjunction with the ch@xmath1 features , we analyze keck / nirspec m band co lines in ngc 7538 : irs9 , previously published in @xcite . [ sec : obssws ] the 3.26 - 3.38 @xmath5 spectral region of ngc 7538 : irs9 was observed multiple times in the high resolution r=1,500 aot06 mode of the short wavelength spectrometer on board the infrared space observatory ( iso / sws ; @xcite ; @xcite ) . from the iso archive we have selected the longest integrations in this wavelength range , corresponding to tdt numbers 39002336 ( ut 11 dec 1996 ) , 75101049 ( ut 05 dec 1997 ) , and 85200455 ( ut 16 mar 1998 ) . all observations were part of the ice_band / dwhittet guaranteed time program , and have not been previously published . the combined on - target integration time in the 3.26 - 3.38 @xmath5 region amounts to 5219 seconds . the individual data sets were processed with sws interactive analysis and calibration files version 10.1 in april 2003 at sron / groningen . the three aar products were combined and the detector scans were aligned to the mean using second order polynomials . then , in this highly oversampled spectrum , data points deviating more than 3@xmath28 from the mean per resolution element were removed . finally , the data points were averaged and re - sampled to an r=1,500 nyquist sampled grid . the resulting s / n is 100 , but significantly poorer at the outer 0.015 @xmath5 edges . [ sec : res ] [ sec : feat ] [ f : gas ] the l band spectrum of ngc 7538 : irs9 is dominated by the deep absorption band of the o - h stretching mode of h@xmath12o ice @xcite . its peak optical depth at 3.0 @xmath5 is @xmath29 . the most prominent substructure in the long wavelength wing of h@xmath12o is the well known , yet unidentified , 3.47 @xmath5 absorption band @xcite . we applied a smooth spline continuum to extract this band on an optical depth scale ( fig . [ f : obs ] ) . the results were verified by applying low order polynomials to continuum regions described in previous studies ( e.g. brooke , sellgren , & geballe 1999 ) . the absorption shape is in good agreement with the low resolution spectrum of @xcite , as is the presence of a sub - band at 3.53 @xmath5 , attributed to the c - h stretching mode of solid ch@xmath13oh . a previously undetected absorption band is apparent by a downturn in the continuum slope at 3.32 @xmath5 . although it is in a spectral region badly affected by the telluric ch@xmath1 q - branch , its presence is confirmed in keck / nirspec observations on different dates . in addition , a feature at the same wavelength is found in the space based iso / sws spectrum ( fig . [ f : obs ] ) . we thus conclude that this feature is real and is due to the stretching mode of solid ch@xmath1 ( [ sec : ice ] ) . not immediately obvious in fig . [ f : obs ] but clearly detected upon close inspection ( fig . [ f : gas ] ) are p , q , and r - branch lines of gas phase ch@xmath1 . it is the first time that the 3.32 @xmath5 ice and gas features are detected in a circum - protostellar environment . a number of additional features , not further discussed in this paper , are visible in the high resolution keck / nirspec spectrum . although they are reproduced in observations at different dates , they are entirely new and need to be confirmed in a larger sample of sources , or , as for solid ch@xmath1 , in independent data . a weak feature at 3.36 @xmath5 coincides with the c - h stretch modes of ethanol @xcite and ethane ( boudin , schutte , & greenberg 1998 ) . interestingly , another band of ethane at 3.47 @xmath5 seen in laboratory ices appears to be present in ngc 7538 : irs9 as well . finally , weak absorptions at 3.47 and 3.41 @xmath5 are also close to the c - h stretch mode of single bonded carbon in ch and ch@xmath12 groups in aliphatic hydrocarbons , normally seen in diffuse lines of sight only ( e.g. @xcite ) . [ sec : gas ] [ sec : ch4 ] ro - vibrational absorption and emission lines of gaseous ch@xmath1 are clearly detected toward ngc 7538 : irs9 ( fig . [ f : gas ] ) . although the telluric ch@xmath1 lines are deep , the large blue - shift of ngc 7538 : irs9 with respect to the earth ( [ sec : obsns ] ) separates many interstellar lines well . the lines appear mostly in absorption up to lower energy levels with @xmath30 , and entirely in emission above @xmath31 . intermediate @xmath32 levels have p cygni - like line profiles , i.e. red - shifted emission and blue - shifted absorption . such line profiles are usually associated with expanding envelopes ( see [ sec : disc2 ] , however ) . note that for @xmath33 multiple emission and absorption peaks appear for each @xmath32 level . these line splittings are intrinsic to the ch@xmath1 molecule and result from rotation - vibration interactions with the spin manifolds of the four equivalent h nuclei . thus e , f- , and a type multiplets occur with spin statistical weights of @xmath34=2 , 3 , and 5 respectively . lcccc [ t : ch4em ] + r(15 ) & 3.158123.15864 & 11 & 84 & 5.2 ( 1.1 ) + & 3.157673.15792 & 3 & 13 & @xmath353.3 + r(14 ) & 3.166773.16720 & 13 & 73 & 3.8 ( 0.7 ) + & 3.166263.16665 & 5 & 26 & 1.5 ( 0.4 ) + r(13 ) & 3.175453.17595 & 13 & 81 & 9.0 ( 0.8 ) + & 3.174983.17534 & 4 & 16 & 4.3 ( 0.8 ) + r(12 ) & 3.184353.18485 & 14 & 74 & 13.6 ( 3.0 ) + & 3.183903.18428 & 4 & 26 & 4.8 ( 0.5 ) + r(11 ) & 3.192953.19325 & 2 & 21 & 3.8 ( 1.5 ) + r(10 ) & 3.202113.20253 & 3 & 34 & 3.5 ( 1.0 ) + r(9 ) & 3.211783.21229 & 6 & 78 & 10.7 ( 3.0 ) + & 3.211463.21179 & 2 & 22 & 3.0 ( 0.8 ) + r(8 ) & 3.220873.22118 & 3 & 47 & 4.8 ( 0.8 ) + r(6 ) & 3.240143.24080 & 6 & 100 & 25.1 ( 1.5 ) + r(5 ) & 3.250193.25064 & 4 & 100 & 12.5 ( 1.4 ) + r(4 ) & 3.260173.26054 & 4 & 100 & 7.5 ( 1.0 ) + r(3 ) & 3.270353.27063 & 3 & 100 & 6.6 ( 0.5 ) + r(1 ) & 3.291053.29127 & 1 & 100 & 1.5 ( 0.5 ) + r(0 ) & 3.301653.30187 & 1 & 100 & 1.5 ( 0.6 ) + p(2 ) & 3.334353.33463 & 2 & 100 & 1.8 ( 0.4 ) + p(3 ) & 3.345733.34600 & 3 & 45 & 2.1 ( 0.5 ) + p(4 ) & 3.357083.35736 & 1 & 23 & 1.9 ( 0.6 ) + & 3.356793.35703 & 3 & 77 & 1.9 ( 0.9 ) + p(5 ) & 3.368603.36901 & 2 & 45 & @xmath364.8 + & 3.368143.36860 & 2 & 55 & 4.5 ( 0.6 ) + p(6 ) & 3.379673.38022 & 3 & 48 & 7.8 ( 0.8 ) + p(7 ) & 3.391823.39230 & 2 & 42 & @xmath37 + & 3.391383.39176 & 2 & 32 & 3.5 ( 0.4 ) + p(8 ) & 3.403793.40411 & 1 & 14 & @xmath361.6 + & 3.403183.40370 & 3 & 48 & 6.8 ( 0.4 ) + p(9 ) & 3.415773.41613 & 2 & 18 & @xmath38 + & 3.415093.41560 & 2 & 22 & 4.1 ( 0.4 ) + p(10 ) & 3.427193.42772 & 3 & 35 & 4.2 ( 0.7 ) + p(11 ) & 3.441283.44172 & 3 & 30 & 3.2 ( 0.4 ) + & 3.440723.44109 & 1 & 11 & 1.6 ( 0.2 ) + & 3.440093.44057 & 3 & 37 & 3.9 ( 0.2 ) + & 3.439443.43982 & 2 & 22 & 2.0 ( 0.2 ) + p(12 ) & 3.453803.45441 & 4 & 43 & 4.7 ( 0.3 ) + & 3.453133.45347 & 2 & 13 & 1.4 ( 0.3 ) + & 3.452523.45296 & 2 & 16 & 1.1 ( 0.4 ) + & 3.451753.45208 & 3 & 27 & 1.9 ( 0.3 ) + p(13 ) & 3.466643.46711 & 3 & 23 & 1.1 ( 0.4 ) + & 3.465663.46629 & 3 & 31 & 0.8 ( 0.3 ) + & 3.465013.46548 & 3 & 28 & 1.8 ( 0.3 ) + & 3.464063.46457 & 2 & 17 & 0.8 ( 0.3 ) + p(14 ) & 3.479523.48004 & 3 & 22 & 1.9 ( 0.4 ) + & 3.478143.47899 & 4 & 35 & 0.6 ( 0.3 ) + & 3.477573.47796 & 2 & 16 & @xmath39 + & 3.476593.47707 & 3 & 27 & 1.7 ( 0.4 ) + + + + + the observed lines are analyzed using the 2001 update of the hitran 2000 database @xcite . all hitran transitions contributing to each observed interstellar line are identified and the number of transitions responsible for at least 90% of the line strength is indicated in column 3 of tables 1 and [ t : ch4ab ] . for example , at the resolution of our observations , the p(11 ) transition is split in 4 separate emission lines , which are composed of some 9 strong transitions . the contribution of each observed line to the total emission or absorption per @xmath32 level is indicated in column 4 of tables 1 and [ t : ch4ab ] and is calculated in the optically thin limit as @xmath40 where @xmath41 is the product of the rotation ( 2@xmath32 + 1 ) and nuclear spin ( @xmath34 ) statistical weights and @xmath42 is the corresponding einstein coefficient for transition @xmath43 at rotational level @xmath32 . the summation over @xmath43 is for all transitions in the wavelength range of an observed emission line ( column 2 in tables 1 and [ t : ch4ab ] ) while the summation over @xmath44 is for all transitions at rotational level @xmath32 . for several @xmath32 levels only part of the emission or absorption is observed as some components lie in poor atmospheric regions . for example , only half the r(8 ) line flux is seen and for the q - branch the clearly detected absorption component represents only a few percent of the total q - branch absorption . finally , the observed emission line fluxes @xmath45 listed in table 1 are corrected for the underlying h@xmath12o absorption using a wide coverage iso / sws spectrum @xcite . the resulting continuum flux of 1.5 jy is corrected for foreground extinction using the galactic extinction law @xcite and the @xmath46 derived in [ sec : disc2 ] . thus , the observed line fluxes are scaled to an unextincted continuum flux of 23.7 jy . emission lines toward ngc 7538 : irs9 . p - branch lines are represented by filled circles and r - branch lines by open circles . the solid line represents a least - squares best fit to the data corresponding to a rotational temperature of 600 k at a column density of 5.8@xmath47 @xmath48 in a solid angle of 4.3@xmath49 sterad . the dashed lines represent extreme ( 3@xmath28 ) physical parameters that still fit the data : @xmath50500 k and @xmath51(ch@xmath1)=20@xmath47 @xmath48 , and @xmath50700 k and @xmath51(ch@xmath1)=2.6@xmath47 @xmath48 . ] [ f : rotch4 ] in order to derive basic physical parameters , we construct a rotation diagram for the ch@xmath1 emission lines . the boltzmann equation of a single transition @xmath43 at upper rotational level @xmath32 is @xmath52 where @xmath53 and @xmath54 are the column density and the energy of the upper level of a single multiplet component respectively . assuming that collisional excitation is applicable , @xmath54 includes a vibrational @xmath55 excitation energy of 4348 k @xcite . @xmath56 and @xmath57 are the total ch@xmath1 column density and rotational temperature . throughout this paper we take the ro - vibrational partition function @xmath58 from the hitran database , which is applicable at @xmath59 k @xcite . at lower temperatures we calculate @xmath58 from lines in the hitran database . in the optically thin limit , the flux @xmath60 of a partial @xmath32 transition is obtained by summation over all multiplet lines @xmath61 and thus the observed @xmath60 relates to eq . [ e : bol ] as @xmath62 [ f : velo ] [ t : ch4ab ] lccccc @xmath63 & @xmath64@xmath65 & lines@xmath66 & @xmath67@xmath68 & eq . width@xmath69 + & @xmath5 & & % & @xmath70 @xmath71 + r(4 ) & 3.260033.26031 & 3 & 77 & 10.0 ( 1.4 ) + r(3 ) & 3.270193.27051 & 3 & 100 & 12 ( 1 ) + r(1 ) & 3.290923.29121 & 1 & 100 & 12.7 ( 0.4 ) + r(0 ) & 3.301513.30179 & 1 & 100 & 17.6 ( 1.0 ) + q & 3.312343.31296 & 4 & 2 & 35 ( 6 ) + p(2 ) & 3.334233.33453 & 2 & 100 & 6.5 ( 0.8 ) + p(3 ) & 3.345303.34584 & 3 & 100 & 12.3 ( 0.7 ) + p(4 ) & 3.356603.35694 & 3 & 61 & 5.9 ( 0.8 ) + + + + + here we make use of the fact that the upper energy level @xmath72 and transition frequency @xmath73 are nearly the same for all multiplet lines ( to within 0.1% ) . @xmath74 is the size of the emitting region , for which we assume the value of 4.3@xmath49 sterad derived from co emission lines , two orders of magnitude smaller than the solid angle corresponding to the 0.42@xmath75 wide slit ( [ sec : disc2 ] ) . the rotation diagram constructed from the observed ch@xmath1 emission lines indicates a rotational temperature of 600@xmath76 k ( fig . [ f : rotch4 ] ) . although @xmath57 is well constrained , the derived column density of @xmath77@xmath48 is accurate to within a factor of @xmath183 only ( 3@xmath28 errors ) . to further illustrate the goodness of fit , to investigate possible optical depth effects and to model the complex blend of ch@xmath1 absorption and emission lines , we have also constructed single temperature lte models . the level populations are calculated according to eq . [ e : bol ] . emission line fluxes and absorption equivalent widths at any optical depth are calculated following @xcite , and @xcite respectively , applying the voigt function for the line profiles . the observed line profiles are fitted by adding a single temperature red - shifted emission model to a single temperature , blue - shifted , absorption model . we find that a blue - shift of 6 @xmath4 and a red - shift of 8 @xmath4 are needed , and that both components are unresolved at the spectral resolution of 12 @xmath4 . the validity of the temperature and column density derived from the emission line rotation diagram is confirmed by the relatively low model line optical depth ; for example , at a doppler parameter of @xmath78=1 @xmath4 , @xmath79 @xmath48 , and @xmath80=600 k , the peak optical depth is 0.33 ( in the r(8 ) line ) . the absorption column is well constrained to @xmath51(ch@xmath1)=3@xmath81 @xmath48 at @xmath82=55@xmath8315 k for intrinsic line widths of @xmath78=0.53 @xmath4 ( table [ t : col ] ) . the narrowest lines are optically thick ( @xmath84=5 at @xmath78=0.5 @xmath4 ) , while they are optically thin for @xmath78@xmath85 @xmath4 . we take @xmath78=3 @xmath4 as a preferred value throughout this paper , because the larger line widths are usually preferred toward ngc 7538 : irs9 @xcite , but values of @xmath78@xmath86 @xmath4 are inconsistent with the unresolved ch@xmath1 lines observed . the physical parameters derived from the ch@xmath1 stretching mode are in good agreement with the parameters from the bending mode at 7.67 @xmath5 in lower resolution spectra @xcite . finally , absorption or emission lines of the @xmath87ch@xmath1 and ch@xmath13d isotopes have not been detected . study of the @xmath87ch@xmath1 lines is complicated , because the isotope wavelength shift is comparable to the separation of @xmath32 levels and thus the @xmath87ch@xmath1 and @xmath88ch@xmath1 lines are blended . the column density upper limits are a factor of 37 smaller compared to the ch@xmath1 column density . at @xmath88c/@xmath87c and h / d isotope ratios of at least 50 in dense clouds @xcite , these upper limits are not significant . [ sec : co ] [ t : co ] lcc @xmath89 & + & + & @xmath90 & @xmath91 + r(18 ) & 2 ( 1 ) & + r(17 ) & 5 ( 2 ) & + r(16 ) & 3 ( 1.1 ) & + r(15 ) & 6.5 ( 2 ) & + r(13 ) & 10 ( 2 ) & + r(12 ) & 16 ( 3 ) & + r(11 ) & 12 ( 4 ) & @xmath354.5 + r(10 ) & 16 ( 3 ) & @xmath354.5 + r(9 ) & 18 ( 5 ) & 5 ( 1 ) + r(7 ) & 30 ( 8) & 2.2 ( 1.0 ) + r(6 ) & 30 ( 2 ) & 3.9 ( 1.2 ) + r(5 ) & 41 ( 3 ) & 4.7 ( 1.2 ) + r(4 ) & 71 ( 4 ) & 5.6 ( 1.0 ) + r(3 ) & 71 ( 4 ) & 5.2 ( 0.8 ) + r(2 ) & & 13.7 ( 1 ) + r(0 ) & 55 ( 3 ) & 10 ( 2 ) + p(1 ) & 55 ( 2 ) & 6.4 ( 1.5 ) + p(2 ) & 56 ( 3 ) & + p(3 ) & 74 ( 2 ) & + p(4 ) & 27 ( 2 ) & + p(18 ) & @xmath3515 & + p(19 ) & 3.4 ( 0.4 ) & + p(21 ) & 2.5 ( 0.8 ) & + p(22 ) & @xmath92 & + + + ( filled circles ) and @xmath91 ( open circles ) absorption lines toward ngc 7538 : irs9 . the @xmath91 column densities have been scaled up by a factor of 6.9 to match the abundance of interstellar @xmath90 @xcite . the good agreement of these isotopes in this plot implies that the @xmath90 absorption lines are optically thin . the solid line represents a two component gas with rotational temperatures of 20 and 270 k both at a @xmath90 column density of 4.65 @xmath47 @xmath48 . note that the 20 k gas absorbs at the systemic velocity , while the warm component absorbs at much higher velocity ( fig . [ f : velo ] ) . ] [ f : rotco ] at first glance , the line profiles of the fundamental ro - vibrational co transitions at 4.7 @xmath5 , as published in @xcite and @xcite , look similar to those of ch@xmath1 . they show p cygni profiles as well , in @xmath93 , @xmath90 , and perhaps even the @xmath91 isotope , with the strength of emission versus absorption increasing for higher @xmath32 levels . upon closer inspection , however , the absorption profiles of co and ch@xmath1 differ . the low-@xmath32 isotopic co lines peak at 0 @xmath4 , compared to 6 @xmath4 for ch@xmath1 ( fig . [ f : velo]a ) . co lines ( @xmath94 k ) , on the other hand , are dominated by absorption at blue - shifted velocities @xmath95 @xmath4 , which is absent in ch@xmath1 ( fig . [ f : velo]b ) . this provides clues to the formation and evolution history of ch@xmath1 ( [ sec : disc1 ] ) , and in this context it is relevant to quantify the physical conditions of the absorbing and emitting co gas as well ( table [ t : col ] ) . we analyzed the r=25,000 keck / nirspec spectrum of @xcite and arrive at different conclusions compared to mitchell et al . ( 1990 , 1991 ) . first , and most importantly , in the keck / nirspec data blue - shifted absorption at 16 @xmath4 is seen up to @xmath32-levels as high as 20 , but only up to @xmath30 in @xcite . in the rotation diagram constructed from the integrated @xmath90 absorption equivalent widths , this high velocity gas is responsible for the linear part above energies of 100 k ( fig . [ f : rotco ] ; table [ t : co ] ) . from this a column density of @xmath51(@xmath93)=3.5@xmath96 @xmath48 at a temperature of @xmath97 k ( 3@xmath28 ) is derived . the column density is comparable to that found by @xcite , but @xmath57 is a factor of 10 higher in the keck / nirspec data . a second difference is that @xmath91 lines are detected in the high s / n keck / nirspec data . the @xmath90/@xmath91 absorption line ratios of corresponding @xmath32 levels are equal to the interstellar isotope ratio of 6.9 at large galactocentric radii @xcite within a 20% error margin ( see the @xmath91 data points in the @xmath90 rotation diagram ; fig . [ f : rotco ] ) . an important consequence is that reliable co column densities can be derived , and thus reliable ch@xmath1 abundances in the various velocity ranges ( 4.1 ) . the steep part of the @xmath90 rotation diagram , corresponding to gas at velocities of 8 to + 11 @xmath4 , indicates @xmath5020@xmath98 k and a column density of @xmath51(@xmath93)=3.5@xmath96 @xmath48 , which is a factor of 4 less compared to that found by @xcite . this is due to our factor of 2 smaller measured equivalent widths and , presumably , lower assumed line opacity . for the emission lines , the observed @xmath93/@xmath90 peak strength ratio is @xmath99 at intermediate upper @xmath32 levels of 7 - 15 . using the standard radiative transfer equation , the peak optical depth of the @xmath93 emission lines , @xmath100 can be derived from @xmath101 at a @xmath93/@xmath90 isotope ratio of 80 @xcite , it follows that @xmath102 . such an optical depth corresponds to a column density of @xmath51(@xmath93)=@xmath103 @xmath48 at an intrinsic line width of @xmath78=3 @xmath4 . we further discuss the origin of the emitting gas in [ sec : disc2 ] . [ sec : ice ] solid ch@xmath1 has been detected by its c - h bending mode at 7.67 @xmath5 in iso / sws observations of ngc 7538 : irs9 @xcite . the peak position and width of the interstellar band are well fitted with an amorphous , low temperature ( 10 k ) laboratory ice in which ch@xmath1 is diluted in an h@xmath12o matrix . this same laboratory ice gives an excellent fit to the new 3.32 @xmath5 absorption feature detected in the keck / nirspec and iso / sws spectra , not only in peak position and width , but also to the peak optical depth of 0.1 . we thus attribute this interstellar band in its entirety to the c - h stretching mode of solid ch@xmath1 in an h@xmath12o rich ice ( fig . [ f : ice ] ) . a mismatch with the observed iso / sws spectrum on the low wavelength wing is attributed to emission by the strong q - branch of gaseous ch@xmath1 ( fig . [ f : ice ] ) . indeed , a better match is obtained after subtracting the model constructed from the observed p and r - branch lines from the iso / sws spectrum ( 3.2.1 ) . in contrast , contamination of gas phase emission to the 7.67 @xmath5 ice band is small because the continuum emission at this wavelength is more than an order of magnitude brighter compared to the continuum at 3.32 @xmath5 @xcite . indeed , a weak inflection at 7.66 @xmath5 is due to q - branch _ absorption_. simultaneous fits of the c - h stretch and bend modes highlight the good fit of ch@xmath1:h@xmath12o ices ( fig . [ f : ice ] ) . pure ch@xmath1 ices and mixtures with co , otherwise highly abundant in this line of sight @xcite , can be excluded at a high confidence level . note that the c - h stretching mode of ch@xmath1 better discriminates between polar and apolar ices than does the bending mode . mixtures with other high - dipole molecules , such as nh@xmath13 and ch@xmath13oh , do not provide equally good fits compared to h@xmath12o , but at present we can not exclude combinations of these polar species . observed with iso / sws ( thick histogram ) and keck / nirspec ( thin histogram ) , compared to a number of laboratory ice simulations in panel ( a ) . at the very top the effect of removing the q - branch emission of gas phase ch@xmath1 is shown . the ch@xmath1 bending mode in panel ( b ) is fitted with the same laboratory ice , without scaling the peak optical depth . best fits are obtained with h@xmath12o rich ices . note the poor fit of co rich ices to the ch@xmath1 stretching mode . ] [ f : ice ] [ sec : disc ] [ t : col ] lllccccl species & location@xmath65 & phase@xmath66 & @xmath104&@xmath105 & @xmath106 & @xmath107@xmath108 & notes + & & & @xmath4 & k & @xmath109 @xmath48 & 10@xmath110 & + ch@xmath1 & outflow & gas / abs . & @xmath2216 & 270 & @xmath111 & @xmath112 & assumed @xmath80 + ch@xmath1 & core & gas / abs . & @xmath226 & 55@xmath8315 & [email protected] & @xmath113 & @xmath78=0.53 @xmath4 + ch@xmath1 & envelope & gas / abs . & 0 & 20 & @xmath111 & @xmath114 & assumed @xmath80 + ch@xmath1 & core & gas / em . & + 8 & 600@xmath83100 & 0.58@xmath115 & 1.9@xmath116 & @xmath117 ster@xmath118 + ch@xmath119 & envelope & ice / abs . & & & 1.3@xmath120 & 2.1@xmath121 & + co & outflow & gas / abs . & @xmath2216 & 270@xmath83100 & 35@xmath8310 & 100 & opt . thin @xmath90 + co & core & gas / abs . & @xmath226 & 55 & @xmath122 & 100 & assumed @xmath80 + co & envelope & gas / abs . & 0 & 20@xmath837 & 35@xmath8310 & 100 & opt . thin @xmath90 + co & core & gas / em . & + 8 & 500@xmath83150 & 30@xmath8310@xmath123 & 100 & @xmath117 ster@xmath118 + co@xmath124 & envelope & ice / abs . & & & 18@xmath832 & 36@xmath125 & + [ sec : disc1 ] the detections of both the 3.32 @xmath5 stretch and the 7.67 @xmath5 bend modes firmly establish the presence of both gaseous and solid ch@xmath1 in the interstellar medium . the gas+solid state abundance averaged along the line of sight of the massive protostar ngc 7538 : irs9 is x[ch@xmath1]=@xmath126 , and the absorption gas / solid state abundance ratio is 0.23 . a combined high resolution co / ch@xmath1 line profile analysis shows that the abundance as well as the gas / solid state abundance ratio of ch@xmath1 vary _ locally _ by an order of magnitude , however ( table [ t : col ] ) . an ice+gas abundance of a few @xmath127 is found in the outer envelope ( systemic velocity ) , the inner envelope ( 6 @xmath4 ) and the compact emission component ( + 8 @xmath4 ) . in the outer envelope , at 20 k the coldest component along the line of sight , at least 90% of the ch@xmath1 is frozen on grains . within the envelope a segregation is present as well ; ch@xmath1 is intimately mixed with h@xmath12o and ch@xmath13oh rather than with co. in the warm , low velocity inner regions of ngc 7538 : irs9 , ch@xmath1 is mostly present in the gas phase . these enhanced abundances strongly contrast to the absence of ch@xmath1 in the massive , high velocity 16 @xmath4 , warm ( @xmath128 k ) outflow seen in co ( x[ch@xmath1]@xmath7 ) . these new results provide clues to the formation and evolutionary history of ch@xmath1 , refining previous scenarios @xcite . in time dependent pure gas phase models , such as the ` standard ' dense cloud model of @xcite and the umist model of @xcite , gas phase ch@xmath1 abundances of a few 10@xmath110 are achieved after 10@xmath129 years . although this is comparable to the observed abundances , there are strong arguments against a gas phase origin : the low gas / solid state ratio and the intimate mixture between ch@xmath1 and h@xmath12o in the ices in the envelope . for comparison , co , a typical gas phase product , has an order of magnitude larger gas / solid phase ratio compared to ch@xmath1 , even in the cold gas phase . h@xmath12o on the other hand is formed by hydrogenation of oxygen on the grains at low temperatures ( e.g. @xcite ) , and a low gas / solid state ratio is observed @xcite . it is thus likely that ch@xmath1 was formed on grain surfaces by hydrogenation of atomic c , at the same time as h@xmath12o was formed from atomic o. the low observed ch@xmath1 abundance points to a formation time when most of the c was already locked up in co @xcite . the gas phase ch@xmath1 abundance in the 6 and + 8 @xmath4 components is comparable to the solid state abundance in the envelope . the ch@xmath1 may thus have been released from the grains in the inner regions near the protostar . because ch@xmath1 is diluted in an h@xmath12o rich ice ( fig . [ f : ice ] ) , its sublimation temperature is close to that of h@xmath12o , 90 k @xcite . the temperature of the + 8 @xmath4emission component is indeed well above 90 k. the gas at 6 @xmath4 , however , has a temperature of 55@xmath8315 k only . while in one massive protostar , w 33a , ch@xmath1 gas at @xmath130 k was detected @xcite , a lower temperature of 50 k was also measured in another line of sight , gl 7009s @xcite . we speculate that the icy mantles did not thermally evaporate as a result of protostellar radiative heating , but by a mild shock that has briefly heated the grains above 90 k. subsequently the gas has cooled radiatively or adiabatically as the inner ngc 7538 : irs9 core expands outwards with a ( current ) velocity of 6 @xmath4 . shocks may have had a profound impact on the higher velocity gas . ch@xmath1 is absent at velocities of 16 @xmath4 and above , where warm co ( @xmath131 k ) is very abundant . after release from the grains , ch@xmath1 must have efficiently chemically reacted . the observed co velocities up to 35 @xmath4 point to the presence of non - destructive c - type shocks . these shocks have been well studied in the context of molecular chemistry . at initial densities of 10@xmath129 @xmath132 , shocks with velocities of 1535 @xmath4 have peak temperatures of 500 - 2500 k ( bergin , neufeld , & melnick 1998 ) . at such high temperatures , and at the low expected shock pressures bar assuming a factor of 2 density enhancement at a maximum temperature of 2500 k for a 35 @xmath4 shock @xcite ] , ch@xmath1 is rapidly driven into co @xcite . the occurrence of a c shock therefore explains the absence of ch@xmath1 in the outflow of ngc 7538 : irs9 . the importance of shock chemistry in high mass cores was deduced from low gas / solid co@xmath12 abundance ratios as well @xcite , and by the abundance ratios of s bearing species @xcite . one distinct characteristic of shock chemistry is the formation of h@xmath12o at temperatures above 250 k ( e.g. @xcite ) . h@xmath12o columns comparable to co are expected , i.e. 2.8@xmath133 @xmath48 for the ngc 7538 : irs9 outflow ( table [ t : col ] ) . this is a factor of 3.5 larger than the upper limit to the gas phase h@xmath12o abundance derived from iso / sws 6 @xmath5 spectra @xcite . high resolution 6 @xmath5spectra are required to further investigate the presence of h@xmath12o in the line of sight of ngc 7538 : irs9 , the effect of blended emission and absorption components and the relation to shocks and outflows . [ sec : disc2 ] the elevated temperature of the low velocity , 6 - 8 @xmath4 , absorption and emission components seen both in co and ch@xmath1 locates this gas in the inner regions of the protostellar core . the present observations further constrain the origin of this material . the area of the radiating physical body can be calculated by matching the @xmath93 column density derived from the isotope ratio ( [ sec : co ] ) to the emission line fluxes . the measured @xmath93line fluxes need to be corrected for continuum absorption by foreground material . we derive a 4.7 @xmath5 continuum extinction magnitude of @xmath134 using the interstellar extinction law of @xcite and an optical extinction @xmath46 of 60@xmath8320 . the latter is derived from the @xmath46/@xmath135 relation of @xcite and the average line of sight hydrogen column density from this work ( table [ t : col ] ) and that from the silicate absorption band @xcite . the extinction corrected line fluxes are thus matched at a solid angle of 4.3@xmath49 sterad . at the distance to ngc 7538 : irs9 of 2.8 kpc @xcite this corresponds to a radius of 70 au . the similarity of co and ch@xmath1 velocities and emission line widths ( fig . [ f : velo]c ) indicates that both molecules are present in this same volume . the co and ch@xmath1 gas detected on these scales may well be related to the dense material that was inferred on scales of @xmath136 au from unresolved 3 millimeter continuum interferometer images @xcite . this intense emission is unrelated to the envelope , and instead may originate from a dense shell or circumstellar disk . the p cygni - like line profiles in both co and ch@xmath1 may suggest that an expanding envelope is the preferred geometry . in these line profiles the absorption component is red - shifted by the shell expansion velocity , and the emission component is centered at the systemic velocity and is broadened by twice the expansion velocity . @xcite argue that the co emission component is blue - shifted due to the effect of deep absorption by foreground material at the systemic velocity . this hypothesis can be tested by the profiles of the highest @xmath32 transitions , which should not have foreground contributions . we find that even the lines from the highest rotational levels of co are blue shifted by + 8 @xmath4(fig . [ f : velo]c ) . furthermore , the ch@xmath1 emission lines peak at this same velocity and are also unresolved . perhaps the ngc 7538 : irs9 core moves within its envelope by 8 @xmath4 and the observed peak of the emission line is the center of the p cygni profile . in this scenario the shell expansion velocity is 14 @xmath4 and an emission line width of 28 @xmath4 is expected . such broad lines are inconsistent with the observations as well . we conclude that the line profiles are not true p cygni type profiles and thus are inconsistent with a spherically symmetric expanding envelope . it is unclear what the relation is between the 55 k absorbing gas seen at @xmath226 @xmath4 and the 500 k emitting gas at + 8 @xmath4 . more detailed dynamical models of circumstellar disks , accretion flows and outflows are required . the possibility of photon excitation needs to be considered as well . as noted in @xcite the similarity of co absorption and emission line profiles between a number of massive protostars suggests that these lines trace a fundamental property of their inner cores . this is strengthened by the similarity of the ch@xmath1 line profiles . [ sec : concl ] using high resolution l and m band spectroscopy we have shown that the gas phase ch@xmath1 abundance varies considerably in the line of sight of the massive protostar ngc 7538 : irs9 . this is a result of specific molecule formation and destruction processes : ch@xmath1 is formed on grain surfaces and subsequently evaporated from the grains in the inner regions of the protostar . both evaporation by photon heating and by shocks likely have played roles . ch@xmath1 efficiently ` burns ' to co in the highest velocity shocked regions . the gas phase ch@xmath1/co abundance ratio is thus a sensitive tracer of shock chemistry . high spectral resolution observations of other species , in particular h@xmath12o at 6 @xmath5 , are required to further test the importance and nature of shock chemistry in high mass star forming regions . the _ stratospheric observatory for infrared astronomy ( sofia ) _ will offer this opportunity in the near future @xcite . observations of the 3.32 @xmath5 ch@xmath1 stretch mode in more lines of sight require high radial velocities ( @xmath137 @xmath4 ) or temperatures that populate the highest @xmath32 levels to avoid saturated telluric lines . in this respect , the 7.67 @xmath5 bending mode may be more suited . future observations of ch@xmath1 toward low mass protostars are particularly exciting , as the ch@xmath1/co ratio is a sensitive tracer of protoplanets . in the relatively low pressure solar nebula low ch@xmath1/co abundances occur . the higher pressure in protoplanetary sub - nebulae , however , favors higher ch@xmath1/co ratios @xcite . variations of the ch@xmath1 abundance between solar system comets possibly trace such physical condition gradients in the early protoplanetary disk @xcite . the research of a.c.a.b . and g.a.b . is supported by the spitzer legacy science program and by the owens valley radio observatory through nsf grant ast-0228955 . 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 thank jan cami for providing a copy of iso / sws ia , fred lahuis for an independent reduction , tim brooke for suggesting the possible ethane identification , and yuk yung for help with the wavelength calibration . suggestions by an anonymous referee have led to an improved presentation of the data . alexander , r. d. , casali , m. m. , andr ' e , p. , persi , p. , & eiroa , c. 2003 , , 401 , 613 allamandola , l.j , sandford , s.a . , tielens , a.g.g.m . , & herbst , t.m . 1992 , , 399 , 134 bergin , e. a. , neufeld , d. a. , & melnick , g. j. 1998 , , 499 , 777 bohlin , r. c. , savage , b. d. , & drake , j. f. 1978 , , 224 , 132 boogert , a. c. a. , et al . 1996 , , 315 , l377 boogert , a. c. a. , schutte , w. a. , helmich , f. p. , tielens , a. g. g. m. , & wooden , d. h. 1997 , , 317 , 929 boogert , a. c. a. , helmich , f. p. , van dishoeck , e. f. , schutte , w. a. , tielens , a. g. g. m. , & whittet , d. c. b. 1998 , , 336 , 352 boogert , a.c.a . , ehrenfreund p. , gerakines , p.a . , tielens , a.g.g.m . , whittet , d.c.b . , et al . 2000 , , 353 , 349 boogert , a. c. a. , blake , g. a. , & tielens , a. g. g. m. 2002 , , 577 , 271 boogert , a. c. a. , et al . 2004 , , 154 , in press boonman , a. m. s. & van dishoeck , e. f. 2003 , , 403 , 1003 boudin , n. , schutte , w. a. , & greenberg , j. m. 1998 , , 331 , 749 brooke , t. y. , sellgren , k. , & smith , r. g. 1996 , , 459 , 209 brooke , t. y. , sellgren , k. , & geballe , t. r. 1999 , , 517 , 883 brown , p. d. , charnley , s. b. , & millar , t. j. 1988 , , 231 , 409 brown , l. r. , et al . 2003 , journal of quantitative spectroscopy and radiative transfer , 82 , 219 ceccarelli , c. , loinard , l. , castets , a. , tielens , a. g. g. m. , caux , e. , lefloch , b. , & vastel , c. 2001 , , 372 , 998 charnley , s. b. & kaufman , m. j. 2000 , , 529 , l111 chiar , j. e. , tielens , a. g. g. m. , whittet , d. c. b. , schutte , w. a. , boogert , a. c. a. , lutz , d. , van dishoeck , e. f. , & bernstein , m. p. 2000 , , 537 , 749 crampton , d. , georgelin , y. m. , & georgelin , y. p. 1978 , , 66 , 1 dartois , e. , dhendecourt , l. , boulanger , f. , jourdain de muizon , m. , breitfellner , m. , puget , j .- l . , & habing , h. j. 1998 , , 331 , 651 dartois , e. , schutte , w. , geballe , t. r. , demyk , k. , ehrenfreund , p. , & dhendecourt , l. 1999 , , 342 , l32 dartois , e. , dhendecourt , l. , thi , w. , pontoppidan , k. m. , & van dishoeck , e. f. 2002 , , 394 , 1057 de graauw , t. , et al . 1996 , , 315 , l49 draine , b. t. 2003 , , 41 , 241 fischer , j. , gamache , r. r. , goldman , a. , rothman , l. s. , & perrin , a. 2003 , journal of quantitative spectroscopy and radiative transfer , 82 , 401 genzel , r. 1992 , in the galactic interstellar medium , ed . d. pfenniger & p. bartholdi ( berlin : springer ) , 275 gerakines , p.a . , whittet , d.c.b . , ehrenfreund , p. , boogert , a.c.a . , tielens , a.g.g.m . , et al . 1999 , , 522 , 357 gibb , e. l. , mumma , m. j. , dello russo , n. , disanti , m. a. , & magee - sauer , k. 2003 , icarus , 165 , 391 g " urtler , j. , klaas , u. , henning , t. , ' abrah ' am , p. , lemke , d. , schreyer , k. , & lehmann , k. 2002 , , 390 , 1075 hatchell , j. & viti , s. 2002 , , 381 , l33 hiraoka , k. , miyagoshi , t. , takayama , t. , yamamoto , k. , & kihara , y. 1998 , , 498 , 710 jones , a. p. & williams , d. a. 1984 , , 209 , 955 kaufman , m. j. & neufeld , d. a. 1996 , , 456 , 250 kessler , m. f. , et al . 1996 , , 315 , l27 knacke , r. f. , noll , k. s. , geballe , t. r. , & tokunaga , a. t. 1985 , , 298 , l67 lacy , j. h. , carr , j. s. , evans , n. j. , baas , f. , achtermann , j. m. , & arens , j. f. 1991 , , 376 , 556 lacy , j. h. , knacke , r. , geballe , t. r. , & tokunaga , a. t. 1994 , , 428 , l69 lacy , j. h. , richter , m. j. , greathouse , t. k. , jaffe , d. t. , & zhu , q. 2002 , , 114 , 153 lee , h .- h . , bettens , r. p. a. , & herbst , e. 1996 , , 119 , 111 lide , d. r. , 1994 , crc handbook of chemistry and physics ( 75th ed . ; boca raton : crc press ) markwick , a. j. , millar , t. j. , & charnley , s. b. 2000 , , 535 , 256 mclean , i.s . , becklin , e.e . , bendiksen , o. , brims , g. , & canfield , j. 1998 , proc . spie , 3354 , 566 millar , t. j. , farquhar , p. r. a. , & willacy , k. 1997 , , 121 , 139 mitchell , g.f . , maillard , j .- , allen , m. , beer , r. , & belcourt , k. 1990 , , 363 , 554 mitchell , g. f. , maillard , j .- p . , & hasegawa , t. i. 1991 , , 371 , 342 noriega - crespo , a. , et al . 2004 , , 154 , in press pendleton , y. j. , sandford , s. a. , allamandola , l. j. , tielens , a. g. g. m. , & sellgren , k. 1994 , , 437 , 683 pontoppidan , k. m. , dartois , e. , van dishoeck , e. f. , thi , w .- f . , & dhendecourt , l. 2003 , , 404 , l17 prinn , r. g. , & fegley , b. 1989 , in : `` origin and evolution of planetary and satellite atmospheres '' , eds . s. k. atreya , j. b. pollack , m. s. matthews , the university of arizona press , p. 78 rothman , l. s. , et al . 2003 , journal of quantitative spectroscopy and radiative transfer , 82 , 5 sandford , s.a . , & allamandola , l.j . 1988 , icarus , 76 , 201 sandford , s. a. , allamandola , l. j. , tielens , a. g. g. m. , sellgren , k. , tapia , m. , & pendleton , y. 1991 , , 371 , 607 schutte , w. a. , gerakines , p. a. , geballe , t. r. , van dishoeck , e. f. , & greenberg , j. m. 1996 , , 309 , 633 spitzer , l. 1978 , physical processes in the interstellar medium ( new york : wiley ) spoon , h. w. w. , keane , j. v. , tielens , a. g. g. m. , lutz , d. , & moorwood , a. f. m. 2001 , , 365 , l353 tielens , a.g.g.m . , tokunaga , a.t . , geballe , t.r . , & baas , f. 1991 , , 381 , 181 tielens , a. g. g. m. & whittet , d. c. b. 1996 , iau symp . 178 : molecules in astrophysics : probes & processes , 178 , 45 van der tak , f.f.s . , van dishoeck , e.f . , evans , n.j . , & blake , g.a . 2000 , , 537 , 283 white , g. j. , et al . 2000 , , 364 , 741 whittet , d.c.b . , schutte , w.a . , tielens , a.g.g.m . , boogert , a.c.a . , de graauw , t. , et al . 1996 , , 315 , 357 willner , s. p. , et al . 1982 , , 253 , 174 wilson , t.l . , & rood , r.t . 1994 , , 32 , 191

we present a new numerical method of special relativistic resistive magnetohydrodynamics with scalar resistivity that can treat a range of phenomena , from nonrelativistic to relativistic ( shock , contact discontinuity , and alfvn wave ) . the present scheme calculates the numerical flux of fluid by using an approximate riemann solver , and electromagnetic field by using the method of characteristics . since this scheme uses appropriate characteristic velocities , it is capable of accurately solving problems that can not be approximated as ideal magnetohydrodynamics and whose characteristic velocity is much lower than light velocity . the numerical results show that our scheme can solve the above problems as well as nearly ideal mhd problems . our new scheme is particularly well suited to systems with initially weak magnetic field , and mixed phenomena of relativistic and non - relativistic velocity ; for example , mri in accretion disk , and super alfvnic turbulence .

1105.5683

the magnetohydrodynamics ( mhd ) approximation has some interesting properties , for example , the flux freezing and magnetic pressure ; the former can be used for the collimation of the jet , and the latter for the acceleration of the plasma . thus , the magnetic field is considered an essential ingredient for many astrophysical phenomena . in particular , many observations indicate that most of the high energy phenomena in astrophysics are related to the strongly magnetized relativistic plasma around some compact objects , for example , agn @xcite , relativistic jet @xcite , pulsar wind @xcite , gamma - ray bursts @xcite , and so on . since it is extremely difficult to solve the relativistic mhd ( rmhd ) equations analytically , the theoretical investigations in fully nonlinear regimes are mainly based on the numerical simulations @xcite . most of these studies approximates the plasma as the ideal rmhd fluid . one reason for this is that the ideal rmhd is an excellent approximation of high energy phenomena for ordinary parameters . however , when one considers extreme phenomena , such as the neutron star mergers , or the central engines of grb , the electrical conductivity can be small , and highly resistive regions may appear . in addition , when one considers the magnetic reconnection , the resistivity plays an essential role in this phenomenon . magnetic reconnection is one of the most important phenomena , since it is highly dynamic , and it changes magnetic field energy into fluid energy @xcite . though numerical results of ideal rmhd exhibit magnetic reconnection , this originates in the purely numerical resistivity , and this is unphysical . for this reason , using resistive rmhd is important for the understanding of reconnection and related phenomena . in order to consider ohmic dissipation , one only has to take into account an additional term @xmath0 in the induction equation of non - relativistic mhd . however , similar to other non - relativistic dissipation , this induction equation is parabolic and it is well - known that this equation is acausal . as a result , if one takes into account ohmic dissipation in a relativistic mhd in a similar way , the equation inevitably includes unphysical exponential growing modes , and unstable for small perturbations similar to other dissipation @xcite . this unphysical divergence results from the fact that one neglects the time derivative of the electric field in the induction equation with ohmic dissipation . for this reason , when one takes into account the ohmic dissipation , one has to consider the time evolution of the electric field , that is , one has to deal with the relativistic electromagnetic hydrodynamic equation . this equation is a telegrapher equation , and satisfies the causality . in this paper , we present a new numerical scheme for the resistive rmhd . there are several examples of pioneering work for resistive rmhd , for example , komissarov ( 2007 , hereafter k07 ) proposed numerical method that solves hyperbolic fluxes by using the harten - lax - van leer ( hll ) prescription , and damping of the electric field by ohmic dissipation that is very stiff by using strang - splitting techniques ; palenzuela et al . ( 2009 , hereafter p09 ) proposed a numerical method that solves hyperbolic fluxes by local lax - friedrichs approximate riemann solver , and the stiff part by using implicit - explicit ( imex ) runge kutta methods . however , these methods use light velocity as the characteristic velocity , and their numerical solutions are diffusive when one considers problems whose characteristic velocity is much lower than light velocity . this indicates that their numerical solutions are diffusive in many important high plasma @xmath1 dynamics , and also their solutions become highly diffusive when the characteristic velocity of phenomena is much lower than light velocity . in particular , when one solves the dynamics of the accretion disk around a black hole with a relativistic jet , one has to use relativistic resistive mhd code that can solve both highly relativistic and non - relativistic dynamics with resistivity for the following three reasons : ( 1 ) the saturation of the magnetorotational instability ( mri ) depends on the resistivity ; ( 2 ) the dynamics of an accretion disk are not ordinarily relativistic , especially , the dynamics of the mri is sub - alfvnic ; ( 3 ) the dynamics of the jet are highly relativistic . for these reasons , previous schemes are diffusive in such phenomena , and we need more accurate numerical schemes . we are developing a new numerical scheme capable of accurately solving problems whose characteristic velocity is quite different from light velocity . in this scheme , we obtain numerical flux of fluid by using sound velocity as the characteristic velocity , and numerical flux of electromagnetic field by using appropriate characteristic velocities of rmhd . this enables us to obtain accurate numerical results when we consider problems whose characteristic velocity is much lower than light velocity . in addition , p09 pointed out that the strang - splitting method used in the komissarov method is unstable when applied to discontinuous flows with large conductivities . however , we find that this problem is not related to the strang - splitting method , but the evolution of electric field @xmath2 during the primitive recovery , that is introduced in the method by p09 . by considering this procedure , we can apply the strang - splitting method to discontinuous flows with large conductivities . this paper is organized as follows . in section [ sec : sec2 ] , the basic equations of resistive rmhd are presented . in section [ sec : sec3 ] , we present the numerical method . results of numerical test problems previously presented are shown in section [ sec : sec4 ] . in section [ sec : sec5 ] , we present results of numerical test problems that can not be solved accurately by previous codes . throughout this paper , we use the units @xmath3 in cartesian coordinates , the minkowski metric tensor @xmath4 is given by @xmath5 variables indicated by greek letters take values from @xmath6 to @xmath7 , and those indicated by roman letters take values from @xmath8 to @xmath7 . the covariant maxwell equations can be written as @xmath9 where @xmath10 is the maxwell tensor , @xmath11 the faraday tensor , and @xmath12 the four - vector of electric current . if we consider highly ionized plasma , the electric and magnetic susceptibilities can be neglected . then , one has @xmath13 where @xmath14 is the levi - civita alternating tensor of space - time , and @xmath15 is the four - dimensional levi - civita symbol . we introduce a future - directed unit timelike vector @xmath16 normal to a spacelike hypersurface @xmath17 . using @xmath16 , we can decompose the maxwell tensor into following forms : @xmath18 similarly , the current four - vector @xmath12 can be decomposed into : @xmath19 where @xmath20 is the charge density observed in the rest frame of @xmath16 , and @xmath21 the conduction current satisfying @xmath22 . in the following , we consider only minkowski spacetime , so @xmath23 . by using the decomposition of the maxwell tensor ( [ dec_m ] ) and the current four - vector ( [ dec_i ] ) , the maxwell equations can be split into the familiar set @xmath24 from maxwell equations , we can derive the electric charge conservation law @xmath25 the relativistic hydrodynamic equations can be obtained from the conservation of mass , momentum , and energy @xmath26 where @xmath27 is the mass density current and @xmath28 the energy - momentum tensor defined respectively as @xmath29 where @xmath30 here @xmath31 is the specific enthalpy , @xmath32 is the proper rest mass density , @xmath33 is the thermodynamic pressure , and @xmath34 is the specific internal energy . the evolution equation of a relativistic resistive mhd is @xmath35 where @xmath36 , @xmath37 , @xmath38 is the density , momentum density , total energy density . in the laboratory frame , @xmath36 , @xmath39 , @xmath38 are given by @xmath40 where @xmath41 is the fluid three - velocity , @xmath42 is the lorentz factor , and numerical fluxes are @xmath43 this is the most common form of perfect fluid equations for the numerical hydrodynamics . the system of eqs . ( [ maxwell_1 ] ) - ( [ maxwell_4 ] ) , ( [ fluid ] ) is closed by means of ohm s law . although there are various forms of ohm s law , we consider only the simplest kind of relativistic ohm s law that accounts only for the plasma resistivity , and that assumes that it is isotropic similar to previous studies k07 and p09 . in the covariant form , it is given by @xmath44 where @xmath45 is the conductivity , @xmath46 is the resistivity , and @xmath47 is the electric charge density as measured in the fluid frame . as the maxwell equations and fluid equations , we can decompose eq . ( [ ohm_cov ] ) into 3 + 1 form , and then the space component of eq . ( [ ohm_cov ] ) is given by @xmath48 + q \mathbf{v } \label{ohm_lab } , \ ] ] in the fluid rest frame , eq . ( [ ohm_lab ] ) becomes @xmath49 the ideal mhd limit of ohm s law can be obtained in the limit of infinite conductivity ( @xmath50 ) . in this limit , eq . ( [ ohm_lab ] ) reduces to @xmath51 splitting this equation into the components that are normal and parallel to the velocity vector , it becomes @xmath52 from these equations , we can obtain the usual result @xmath53 in this section , we present our new numerical scheme for the resistive rmhd . since the pioneering studies of resistive rmhd k07 and p09 use light velocity as the characteristic velocity , their solution becomes highly diffusive when characteristic velocity is much lower than light velocity . in our new scheme , we obtain numerical flux of fluid by using sound velocity as the characteristic velocity , and numerical flux of the electromagnetic field by using alfvn velocity as the characteristic velocity . this enables us to obtain accurate numerical results even when characteristic velocity is much lower than light velocity . in the following sections , we consider the one - dimensional case . the extension to the multi - dimensional scheme using the constrained transport method @xcite . will be shown in our next paper . the relativistic resistive mhd is hyperbolic - relaxation equations . in previous work k07 and p09 , they assume that characteristic velocity is the speed of light . thus , their schemes are highly diffusive when the characteristic velocity is lower than light velocity . for this reason , we apply the strang splitting method @xcite and solve the basic equations by using each appropriate characteristic velocity . first , we split fluid equations eq . ( [ fluid ] ) as follows : @xmath54 where @xmath55 \eta^{ix } \label{flux_em_m } , \\ f_{e , em}^x & = & ( \mathbf{e \times b})^x \label{flux_em_e } . \end{aligned}\ ] ] the flux of the fluid component @xmath56 can be calculated by using the riemann solver ; the flux of the electromagnetic component @xmath57 can be calculated by the method of characteristics . next , we consider the maxwell equations eqs . ( [ maxwell_1 ] ) - ( [ maxwell_4 ] ) . ( [ maxwell_1 ] ) and ( [ maxwell_2 ] ) are not evolution equations but constraint equations , and we treat them separately from evolution equations . the evolution equations of @xmath58 and @xmath59 are eqs . ( [ maxwell_3 ] ) and ( [ maxwell_4 ] ) . by using ohm s law eq . ( [ ohm_lab ] ) , eq . ( [ maxwell_3 ] ) reduces to @xmath60 - q \mathbf{v } \label{maxwell_3_2 } .\ ] ] the source term of this equation includes evolving variables @xmath58 , so this equation is a hyperbolic equation with stiff relaxation terms that requires special care to capture the dynamics in a stable and accurate manner . thus , we split the charge current @xmath61 into two parts similar to k07 @xmath62 \label{current_stiff } .\end{aligned}\ ] ] then , we split eq . ( [ maxwell_3_2 ] ) into two parts @xmath63 eq . ( [ maxwell_3_3 ] ) is non - stiff equations , and eq . ( [ maxwell_3_4 ] ) is stiff equations . as a result , the evolution part of the maxwell equations can be rewritten as @xmath64 in component form , eqs . ( [ maxwell_4_2 ] ) and ( [ maxwell_3_3_2 ] ) reduce to @xmath65 we solve eqs . ( [ maxwell_4_2_y ] ) , ( [ maxwell_4_2_z ] ) , ( [ maxwell_3_3_2_y ] ) , and ( [ maxwell_3_3_2_z ] ) using method of characteristics ( moc ) , which will be shown in sec . [ sec : sec3.2 ] . eq . ( [ maxwell_3_3_2_x ] ) is solved using the runge - kutta method . the numerical scheme for the stiff equation eq . ( [ maxwell_3_4_2 ] ) will be shown in sec . [ sec : sec3.3 ] . the method of characteristics can be used to solve the initial value problems of advective and hyperbolic equations . as is well known , the maxwell equations are hyperbolic , so we can solve the maxwell equations accurately by using this method . the maxwell equations for the transverse fields are eqs . ( [ maxwell_4_2_y ] ) , ( [ maxwell_4_2_z ] ) , ( [ maxwell_3_3_2_y ] ) , and ( [ maxwell_3_3_2_z ] ) . by adding and subtracting these equations , for @xmath66 , @xmath67 , and @xmath68 , we obtain @xmath69 ^{\pm}f = - \frac{1}{2 } j^y \label{charact } , \\ & & ^{\pm}f \equiv \frac{1}{2 } ( e^y \pm b^z ) , \end{aligned}\ ] ] where @xmath70 is the characteristic velocity , and this is equal to the speed of light in ordinal maxwell equations . the transverse fields are recovered from @xmath71 by @xmath72 the left - hand side of eq . ( [ charact ] ) is the total derivative @xmath73 for an observer moving at velocity @xmath74 . let us consider conservative discretizations of eqs . ( [ maxwell_4_2_z ] ) and ( [ maxwell_3_3_2_y ] ) : @xmath75 \label{maxwell_4_2_2_z } , \\ \bar{e}_i^{y , n+1 } & = & \bar{e}_{i}^{y , n } - \frac{\delta t^n}{\delta x_i } \left [ ( b^z_{i+1/2})^{n+1/2 } - ( b^z_{i-1/2})^{n+1/2 } \right ] \nonumber \\ & - & q^{n+1/2}_i v_i^{y , n+1/2 } \delta t^n \label{maxwell_3_3_3_y } , \end{aligned}\ ] ] where superscript @xmath76 means the time - step , and subscript @xmath77 means the coordinate of cell center . using eqs . ( [ ch1 ] ) and ( [ ch2 ] ) , we can obtain the numerical flux of eqs . ( [ maxwell_4_2_2_z ] ) and ( [ maxwell_3_3_3_y ] ) . ( see fig . [ the same procedure can be done for time advance of @xmath78 , @xmath79 . the characteristic velocity of the maxwell equations in vacuum is light velocity . however , since we consider the electromagnetic hydrodynamics equations , appropriate characteristic velocity has to be used for them . also , because we consider resistive systems , the characteristic velocity varies with the conductivity @xmath80 and the scale of wave modes . for example , as shown in appendix . the transverse electromagnetic hydrodynamic waves propagate with the light velocity when @xmath81 is large , where @xmath82 is the wave number , and they propagate with the alfvn velocity when @xmath81 is smaller than a critical value depending on @xmath83 , and @xmath84 . because of the finite resistivity , the frequency of the transverse waves has an imaginary part @xmath85 ( damping rate ) , which is a increasing function of @xmath81 . in this scheme , we use the light velocity as the characteristic velocity when @xmath80 is smaller than the critical value ; when @xmath80 is larger than the critical value , we use appropriate magnetohydrodynamic characteristic velocities . the critical value is determined so that the transverse waves whose phase velocities are light velocity are dissipated within the numerical integration timestep @xmath86 . a detailed procedure to judge whether we use the light velocity or magnetohydrodynamic characteristic velocities is given in appendix . in addition to the numerical flux of the maxwell equations , the characteristic velocity is also required to construct the maxwell stress terms and the poynting flux term . when the characteristic velocity obtained from the analysis of the transverse waves is the light velocity , we use the light velocity as the characteristic velocity for them ; when the transverse wave characteristic velocity is the alfvn velocity , we use the characteristic velocities for them as follows . note that if the following characteristic velocities are not used , numerical integration becomes unstable , and unstable numerical oscillation occurs . then , the necessary procedures are as follows : 1 . for the numerical flux of maxwell equation eqs . ( [ maxwell_4_2_2_z ] ) and ( [ maxwell_3_3_3_y ] ) , we use the alfvn wave velocity in laboratory frame because the information of transverse electromagnetic fields are transmitted by the alfvn wave limit , the alfvn velocity in laboratory frame @xmath87 becomes @xmath88 . we find that when this @xmath88 is also small , numerical oscillations occur and numerical integration becomes occasionally unstable . we can prevent this purious oscillations , if we use the characteristic velocity as @xmath89 when @xmath90 . in this case , our scheme for solving the maxwell equations becomes equivalent to the hlle scheme . note that introduction of this modification does not change any results presented in this paper . ] . in relativistic mhd , the alfvn velocity in laboratory frame @xmath87 can be obtained by solving @xcite @xmath91 where @xmath92 , @xmath93 , @xmath94 , and @xmath95 is the covariant magnetic field defined as @xmath96 .\ ] ] 2 . for the maxwell tension terms @xmath97 in eq . ( [ flux_em_m ] ) , we use the alfvn wave velocity in fluid comoving frame @xmath98 because magnetic tension force is originated by the alfvn wave . in relativistic mhd , the alfvn velocity in fluid comoving frame is given by @xmath99 3 . for the poynting flux @xmath100 of energy equation eq . ( [ flux_em_e ] ) and the maxwell pressure terms @xmath101 in eq . ( [ flux_em_m ] ) , we use the fast magnetosonic wave velocity in laboratory frame because the magnetic pressure originates in the magnetosonic wave . in relativistic mhd , fast magnetosonic wave velocity in the laboratory frame @xmath102 can be obtained by solving @xmath103 \label{charact_fm } , \end{aligned}\ ] ] eq . ( [ charact_fm ] ) is a quartic equation , and in an ordinary one has to use the newton - raphson method or the quartic formula for obtaining solutions . however , since our scheme splits the fluid part and the electromagnetic part , the sound velocity @xmath104 can be set equal to zero . then , the characteristic equation eq . ( [ charact_fm ] ) reduces to @xmath105 by using the quadratic formula , one can obtain solutions of above equation : @xmath106 to sum up , we only have to substitute the appropriate characteristic velocities @xmath107 , and @xmath102 into @xmath70 in eq . ( [ charact ] ) , and calculate the electromagnetic field @xmath108 at half time step . then , the numerical fluxes of electromagnetic hydrodynamics equations are given by @xmath109 \eta^{ix } , \\ f_{e , em}^x & = & ( \mathbf{e}_{fm } \times \mathbf{b}_{fm})^x , \end{aligned}\ ] ] where @xmath110 means that they are calculated by using the alfvn velocity in comoving frame , and @xmath111 by using the fast magnetosonic wave velocity in laboratory frame . for the numerical flux of the maxwell equation , one has to use the alfvn velocity in laboratory frame @xmath87 for the calculation . as explained sec . [ sec : sec3.1 ] , eq . ( [ maxwell_3_4 ] ) contains stiff terms . following the previous work k07 , we split the equation into components normal and parallel to the velocity vector . @xmath112 = 0 , \\ \partial_t \mathbf{e}_{\perp } & + & \sigma \gamma \left [ \mathbf{e}_{\perp } + \mathbf{v \times b } ) \right ] = 0 , \end{aligned}\ ] ] since we use the strang splitting method , the right - hand side of the above equations can be considered constant other than the electric field @xmath58 . as a result , these equations can be solved analytically @xmath113 \label{exact1 } , \\ \mathbf{e}_{\perp } & = & \mathbf{e}^*_{\perp } + ( \mathbf{e}^0_{\perp } - \mathbf{e}^*_{\perp } ) \exp \left [ - \sigma \gamma t \right ] \label{exact2 } , \end{aligned}\ ] ] where @xmath114 and suffix 0 indicates the initial component . if we use the explicit integrator , the stiff equation has to be solved in very small time steps @xmath86 . however , since eqs . ( [ exact1 ] ) and ( [ exact2 ] ) are formal solutions , we can avoid the stability constraints of the time step . in the context of ambipolar diffusion in partially ionized plasma , a similar numerical technique using the piecewise formal solution of stiff part is known to be useful scheme @xcite . it is well known that eqs . ( [ maxwell_1 ] ) and ( [ maxwell_2 ] ) are constraints on the cauchy surface . though maxwell equations ensure that these constraints are preserved at all times , straightforward numerical integration of maxwell equations does not preserve these properties because of the accumulated numerical error . this causes corruption of numerical results , and results in a crash in the end . for this reason , there are a number of numerical techniques for avoiding this problem . we have implemented hyperbolic divergence cleaning for the electric field . the main idea of the hyperbolic divergence cleaning is that one defines new variable @xmath115 as the deviation from constraint equations , and arranges a system of equations to decay or carry the deviation @xmath115 out of the computational domain by high speed waves . for the magnetic field , if one sets @xmath116 constant , the constraint equation can be satisfied in one - dimensional case . in the multi - dimensional case , we can implement constrained transport method @xcite . the detailed implementation will be presented in our next paper . for hyperbolic divergence cleaning , we modify eqs . ( [ maxwell_1 ] ) and ( [ maxwell_3_3_2_x ] ) @xmath117 where @xmath115 is a new dynamic variable and @xmath118 a positive constant . clearly , when we set @xmath119 , we can recover standard maxwell equation eq . ( [ maxwell_1 ] ) . from these equations , we can obtain the telegrapher equation for @xmath115 @xmath120 thus , @xmath115 propagates at the speed of light and decays exponentially over a timescale @xmath121 . similar to eq . ( [ maxwell_3_4 ] ) , eq . ( [ dce ] ) contains stiff source terms . thus , we split the equation into a stiff part and non - stiff part @xmath122 the analytical solution of eq . ( [ dce_stiff ] ) is @xmath123\ ] ] where @xmath124 is the initial value of @xmath115 . in order to compute numerical flux ( [ flux_fluid_d ] ) , ( [ flux_fluid_m ] ) , ( [ flux_fluid_e ] ) , ( [ flux_em_m ] ) , and ( [ flux_em_e ] ) , the primitive variables @xmath125 have to be recovered from the conserved variables @xmath126 . in conserved variables , @xmath58 and @xmath59 can be obtained by evolving the maxwell equations . however , as pointed out by p09 , it is more stable to perform evolution of stiff part eqs . ( [ exact1 ] ) and ( [ exact2 ] ) during this primitive recovery process when @xmath80 is large , i.e. , ideal mhd approximation is valid . this is because when we consider mhd approximation , the electric field @xmath58 is equal to @xmath127 ; however , in general , primitive recovered @xmath58 does not satisfy this relation . in what follows we explain the primitive recovery procedure following p09 . 1 . set an initial guess for the velocity by using previous time step value then , evolve electric field @xmath58 using eqs . ( [ exact1 ] ) and ( [ exact2 ] ) . 2 . subtract poynting flux and electromagnetic energy density from conserved variables , and new variables can be defined as follows : @xmath128 then , variables @xmath129 are the ideal relativistic fluid conserved variables , and can be recovered by using the ordinary procedures . 3 . replace the initial guess for the velocity with the obtained velocity @xmath130 , and repeat the steps 1 - 3 until the primitive variables converge . in this section , we provide the detailed numerical algorithm . in cartesian coordinates , the relativistic resistive mhd equations written in conservative fashion are @xmath131 where @xmath132 \eta^{ix } , \\ f_{e , em}^x & = & ( \mathbf{e \times b})^x . \end{aligned}\ ] ] the electric field @xmath58 and magnetic field @xmath59 are evolved by the maxwell equations . if the ohmic dissipation is considered , the maxwell equations have stiff and non - stiff part . the non - stiff part is @xmath133 @xmath134 where @xmath135 + q v^x$ ] . the maxwell equations are consistent with the equation of charge conservation . however , numerical errors in general destroy the conservation law in a way similar to the constraint equations . thus , the above equation contains the equation of charge conservation . as explained in sec . [ sec : sec3.3 ] and [ sec : sec3.4 ] , the stiff - part is evolved by using the formal solution @xmath113 \label{exact_1.2 } , \\ \mathbf{e}_{\perp } & = & \mathbf{e}^*_{\perp } + ( \mathbf{e}^0_{\perp } - \mathbf{e}^*_{\perp } ) \exp \left [ - \sigma \gamma t \right ] \label{exact_2.2 } , \\ \psi & = & \psi_0 \exp[- \kappa t ] \label{exact_3.2 } .\end{aligned}\ ] ] using the above system equations , the second - order numerical algorithm is given as follows . 1 . advance the stiff - part equations over @xmath136 by using the formal solutions eqs . ( [ exact_1.2 ] ) - ( [ exact_3.2 ] ) . 2 . advance the non - stiff part of maxwell equations eqs . ( [ 4_maxwell ] ) and ( [ maxwell_stiff ] ) over @xmath137 by using method of characteristics as explained in sec . [ sec : sec3.2 ] , and calculate numerical flux @xmath138 ( [ flux_em_m ] ) and ( [ flux_em_e ] ) . on the other hand , numerical flux @xmath139 ( [ flux_fluid_d ] ) - ( [ flux_fluid_e ] ) can be calculated by using approximate riemann solver @xcite . in this paper , we use the hllc solver . 3 . advance conserved variables d , @xmath39 , e over the half time - step @xmath137 by using eqs . ( [ rrmhd ] ) - ( [ flux_em_e ] ) . then , calculate primitive variables of half time step @xmath140 by primitive recovery explained in sec . [ sec : sec3.5 ] . in our scheme , electric field @xmath58 has to be evolved @xmath136 by using formal solution ( [ exact_1.2 ] ) and ( [ exact_2.2 ] ) during primitive recovery . primitive variables obtained through this procedure are used for the calculation of the numerical flux at @xmath141 . 4 . again , advance initial stiff variables over @xmath137 by using the formal solutions eqs . ( [ exact_1.2 ] ) - ( [ exact_3.2 ] ) . calculate temporal second - order numerical flux ( [ flux_fluid_d ] ) - ( [ flux_em_e ] ) by using primitive variables obtained through the procedure 3 . then , advance conserved variables @xmath142 over @xmath86 by eq . ( [ rrmhd ] ) , and electric field @xmath58 and magnetic field @xmath59 by the maxwell equations of stiff - part eq . ( [ maxwell_stiff ] ) . 6 . calculate primitive variables by a primitive recovery process . during this process , the electric field @xmath58 is advanced over @xmath137 by using formal solutions ( [ exact_1.2 ] ) and ( [ exact_2.2 ] ) . for the spatial second - order , we use the muscl scheme by van leer explained in appendix . note that if we evolve electric field @xmath2 in integration of stiff equations or primitive recovery procedure , we have to evolve other primitive variables . this is because conserved variables are not changed during those procedures , and this means that the change of electric field @xmath2 affects all other primitive variables . in this section , several one - dimensional test simulations given in previous studies k07 and p09 are presented . for the numerical flux of fluid , we use the hllc solver @xcite . we use an ideal equation of state @xmath143 with @xmath144 , and courant number , @xmath145 . this test consists of the propagation of a large amplitude circularly - polarized alfvn waves along a uniform background field @xmath146 . the analytical exact solution of this problem is given by del zanna et al . ( 2007 ) @xcite ; and this problem is used as the ideal - mhd limit test problem by p09 . we use the same condition as p09 . @xmath147 , \sin[k ( x - v_a t ) ] \right ) , \\ ( v^y , v^z ) & = & - \frac{v_a}{b_0 } ( b^y , b^z ) , \end{aligned}\ ] ] where @xmath148 , @xmath149 , @xmath82 is the wave number , and @xmath150 is the amplitude of the wave . the special relativistic alfvn speed @xmath151 is given by @xmath152 for the initial data parameters , we have used @xmath153 , and @xmath154 . using these parameters , the alfvn velocity is @xmath155 . for the boundary condition , the periodic one is used . in addition , we use a high uniform conductivity @xmath156 following p09 , since this is the exact solution of ideal relativistic mhd . [ fig1 ] is results of our new code at @xmath157 ( one alfvn crossing time ) for three different resolution cases with @xmath158 . the computational domain is @xmath159 $ ] . this result indicates that our new code reproduces ideal relativistic mhd solutions when the conductivity @xmath80 is high . in these test problems , we can not achieve full second - order accuracy . the left - hand side of fig . [ convergence ] is the @xmath160 norm errors of the tangential magnetic field @xmath161 of this test problem . this figure shows that our numerical result is nearly @xmath162-order convergence . we estimate this is because our scheme uses many operator splittings , and the time accuracy of our scheme worsens . note that this problem is one of the most difficult to solve in relativistic resistive mhd , since this is the limit of large conductivity @xmath80 . the right - hand side of fig . [ convergence ] is the @xmath160 norm errors of the @xmath161 of the next test problem . since the conductivity @xmath80 is moderate value in that test problem , our new scheme achieves second - order convergence . this problem is used as the test problem of highly resistive cases in k07 and p09 . in this test , it is assumed that the magnetic pressure is much smaller than gas pressure , so that the background fluid is not influenced by the evolution of the magnetic field . we assume the magnetic field has only tangential component @xmath163 , and @xmath164 changes its sign within this current sheet . since we are interested only in the evolution of the magnetic field , the background fluid is set initially in equilibrium , @xmath165 . in addition , we assume that the conductivity @xmath80 is high , and the diffusion timescale is much longer than the light propagating timescale . although the resistive relativistic mhd equation is hyperbolic , this assumption allows us to neglect the displacement currents at least in the rest frame . as the result , the evolution equation is reduced to @xmath166 this equation has exact solution @xmath167 where erf is the error function . following k07 and p09 , we set the initial condition at @xmath168 with @xmath169 , @xmath170 , @xmath171 , and @xmath172 . the computational domain is @xmath173 $ ] , and the number of grid points is @xmath174 . fig . [ fig2 ] is the numerical result at @xmath175 . this figure shows that our scheme can solve a highly resistive problem accurately . the convergence rate is consistent with the second - order spatial and temporal discretization . in order to confirm the capability of our method for the relativistic resistive mhd , we perform the test calculation of the propagation of alfvnic transverse waves with ohmic dissipation , and compare the results with the exact dispersion relation eq . ( [ eq6 ] ) obtained in appendix . as explained in appendix , the resistive relativistic magnetohydrodynamic equation contains transverse wave modes that become the light wave in large @xmath81 region , and become the alfvn wave in small @xmath81 region . to demonstrate the propagation of transverse waves , we set the initial condition by eigenfunctions of the mode obtained from eqs . ( [ eqd1 ] ) - ( [ eqd4 ] ) @xmath176 where @xmath177 and @xmath178 , and @xmath179 is the solution of the dispersion relation eq . ( [ eq6 ] ) . we set the same parameters in appendix.@xmath180 since the enthalpy includes the information of the equation of state , one can take any value of @xmath181 . in this calculation , we set @xmath144 and @xmath182 . the computational domain covers the region @xmath183 $ ] where the periodic boundary condition is imposed , and the number of grid points is @xmath174 . the propagation speed of the numerical waves can be determined by tracing the position where @xmath184 is maximum . we measure the propagation speed and evaluate re@xmath185 $ ] based on the time when the maximum of @xmath67 reaches @xmath186 again , i.e. one - wave crossing period . the damping rate im@xmath185 $ ] is measured by using @xmath187 t]$ ] where @xmath188 is the maximum of @xmath67 after the one - wave crossing time . in figs . [ fig_d ] , we plot the real and imaginary part of @xmath189 against @xmath81 . the solid line is the exact dispersion relation obtained in appendix . we have performed the calculation in the cases of @xmath190 . these figures show that our new numerical code can reproduce the propagation of alfvnic transverse waves accurately for any value of the conductivity @xmath80 . for the shock tube test problem , we consider the simple mhd version of the brio and wu test as p09 . the initial left and right states are given by @xmath191 all the other fields are set to @xmath6 . [ fig3 ] is the numerical results at @xmath192 that change grid points @xmath193 . the computational domain covers the region @xmath194 $ ] . we also plot an ideal rmhd solution by the solid line computed by a publicly available code developed by giacomazzo and rezzolla @xcite . the conductivity is uniform with @xmath156 . the solution of this riemann problem contains a rarefaction moving to the left , a shock moving to the right , and a tangential discontinuity between them . [ fig3 ] shows that our numerical solution of the resistive mhd can reproduce the profile of an ideal mhd shock tube problem using high conductivity @xmath80 . in addition , our numerical solution captures contact discontinuity as sharp as p09 . [ fig4 ] is the numerical results of the same problem that changes the conductivity @xmath195 . we also plot the ideal rmhd solution by the solid line . the number of grid points is @xmath196 . this result shows that our numerical solution reproduces nearly the same results as p09 . p09 reports that strang s splitting method becomes unstable for moderately high values of the conductivity for this shock tube problem , and one has to use the implicit method . however , this is not related to whether one uses strang s splitting or implicit method , but to the revision of the electric field during the iteration of the primitive recovery ( h. r. takahashi 2010 , private communication ) . our scheme uses strang s splitting , but can solve this shock tube problem stably even when @xmath197 , if we revise the electric field during the primitive recovery as explained in sec . [ sec : sec3.5 ] . the previous studies k07 and p09 use light velocity for the characteristic velocity . thus , their numerical solutions become highly diffusive when one considers problems whose sound velocity or alfvn velocity is much lower than light velocity . in this section , we perform test problems in such cases , and compare the results of the hll code with that of our code . in this section , we compute a high plasma @xmath1 shock tube problem , and compare results of our code with those of the hll code . the initial left and right states are given by @xmath198 all the other fields are set to @xmath6 . [ fig5 ] are the numerical results of our code and the hll one , being compared with ideal solutions at @xmath199 . the number of grid points is @xmath196 . these figures show that hll solver becomes more diffusive than our code . in addition , figs . [ fig5 ] show that the density profile of the shock heated region somewhat overshoots that of the ideal solution , and tangential magnetic field @xmath79 slightly undershoots that of the ideal solution . these results show that when the plasma @xmath1 is high , the hll solver becomes highly diffusive and does not reproduce the correct value of the shock heated region . in contrast , our numerical results reproduce ideal solutions very well even for high @xmath1 problems . in this section , we calculate the propagation of a contact discontinuity , and study the accuracy of capturing the contact discontinuity for various advection velocities . when one uses the hll code by komissarov , the numerical results can be expected to be diffusive for the case of very slow advection velocity , since the hll code uses light velocity for the characteristic velocity . in contrast , our new code uses sound velocity for the fluid characteristic velocity , and the numerical results will be more accurate for any advection velocity . we consider the propagation of contact discontinuity of magnetohydrodynamics . the initial condition is @xmath200 all the other fields are set to @xmath6 . since we want to consider the ideal fluid case , we consider high conductivity @xmath156 . we use an equation of state with @xmath201 , and the computational domain covers the region @xmath183 $ ] with @xmath202 grid points . the cfl number is @xmath203 , and the integration is carried out until @xmath204 fluid crossing time . for the boundary condition , the periodic one is used . for the advection velocity , we use the following velocities : @xmath205 on the left of figs . [ fig9_new ] are the numerical results of the density profile calculated by using our new code , and on the left of figs . [ fig9_kom ] are the numerical results of the density profile calculated by using the hll code , the solid lines are the ideal solution . this figure shows that the numerical results of density profiles by hll code of @xmath206 is nearly equal to that of our new code . however , the numerical results by hll code become more diffusive than by our code as the advection velocity becomes small ; in contrast , the accuracy of the numerical results by our code is nearly independent of the advection velocity . the right hand side of figs . [ fig9_new ] are the numerical results of tangential magnetic field @xmath79 calculated by using our new code , and the right hand side of figs . [ fig9_kom ] are the numerical results of tangential magnetic field @xmath79 calculated by using the hll code . similar to the density profile , the numerical results by using the hll code become more diffusive than by using our code as the advection velocity becomes small ; in contrast , the accuracy of the numerical results by our code is nearly independent of the advection velocity . in conclusion , the hll code is not capable of accurately solving problems whose advection velocity is smaller than light velocity , since the hll code uses light velocity for the characteristic velocity . the diffusive result of hll code is always problematic for any discontinuity when the propagation velocity is much smaller than light velocity . in contrast , since our code uses appropriate characteristic velocities , the numerical dissipation does not depend on the characteristic velocity . for this reason , our new code can solve any advection velocity problems accurately , especially for the problems including discontinuities . in this section , we consider the propagation of small amplitude alfvn waves in high @xmath1 plasma . the integration is performed for different resolutions , and we compare the numerical results of komissarov s hll code and our code . for the application to the numerical simulation of mri , the integration is performed for a small number of grid points : @xmath207 for one wavelength of the alfvn wave ; this corresponds to the number of grid points for resolving the wavelength of maximum growth rate of mri . for the initial condition , we consider @xmath208 in this case , the alfvn velocity @xmath151 and plasma beta @xmath1 are given by @xmath209 where the alfvn velocity and the plasma beta is defined as @xmath210 since the initial magnetic field is very weak for most of the mri phenomenon , a weak magnetic field is considered . in order to consider the ideal fluid case , we set a high conductivity @xmath156 . we use an equation of state with @xmath144 . the computational domain covers the region @xmath183 $ ] . the cfl number is @xmath211 , and the integration is carried out until @xmath8 alfvn wave crossing time . for the boundary condition , the periodic one is used . the numerical results of our code and hll are presented in figs . [ fig10 ] . although the amplitude of both results falls because of the numerical diffusion , it can be seen that hll results are more diffusive than our numerical results when the number of grid points is @xmath212 . when the number of grid points is @xmath213 , the numerical result of hll code is a little more accurate than that of our code . this is because our new scheme uses an operator split for the accuracy , and the convergence rate is a little less than second order in time . however , from a practical point of view , it is impossible to cost 64 grid points for the wavelength of maximum growth rate of mri in many cases , and still our new code can integrate the growth of magnetic field by mri more accurately . in conclusion , when one considers the high @xmath1 plasma , our code is more accurate than the hll code because our code uses sound velocity and alfvn velocity as the characteristic velocity . in particular , the above results show that our new method is useful for the application to the phenomena including mri . this instability occurs in the system whose angular momentum changes as @xmath214 , and the amplitude of the perturbative alfvn waves grows exponentially in over the duration of nearly one kepler rotation . since the above condition is satisfied in most of the differential rotating systems in gravity , mri is one of the most important astrophysical phenomena . in order to reproduce this instability numerically , one has to resolve the wavelength of maximum growth rate . however , this is difficult for most problems , since this wavelength is proportional to the initial weak magnetic field . for this reason , in order to reproduce mri numerically , one has to use numerical schemes that can integrate small amplitude alfvn waves accurately by smaller number of grid points . then , the results of test problems in this section show that our new numerical scheme can deal such problems more accurately than previous codes . in these three test problems , we consider extremely high density cases in order to distinguish differences easily . however , this can always happen when the magnetic field is weak . as a result , if one considers problems including an initially weak magnetic field like magnetic rotational instability ( mri ) in the accretion disk , our code can produce more accurate results . in this paper , we have presented a new numerical scheme of resistive rmhd for one - dimensional case which can solve matter dominated problems more accurately than the existing numerical method . since this new scheme uses different characteristic velocity for obtaining the numerical flux of fluid and electromagnetic field , one can solve accurately and stably problems whose characteristic velocity is much lower than that of light . when one considers relativistic problems , one has to solve stiff equations for electric fields . in general , it is difficult to deal with stiff equations , and special methods have been presented ; for example , k07 uses the strang s splitting method , and p09 use the implicit method . p09 report that strang s splitting method is incapable of solving problems that include discontinuity , such as shock . however , we find that this is not related to the method for the stiff equations , and one can solve problems including shock if one evolves the electric field during the primitive recovery ; we use strang s splitting method , and the solver is well behaved for shock tube problems . the results of other test problems show that our new scheme is capable of accurately solving both highly resistive problems and nearly ideal mhd ones . in addition , it has been shown that our code can solve low characteristic velocity problems more accurately than the hll code . the problems of high density and high plasma @xmath1 appear when one considers magnetorotational instability ( mri ) in the accretion disk with a relativistic jet , for example . in this case , one has to use relativistic resistive mhd code that can solve both highly relativistic and non - relativistic dynamics with resistivity for the following three reasons : ( 1 ) the saturation of the magnetorotational instability ( mri ) depends on the resistivity ; ( 2 ) the dynamics of an accretion disk are not ordinarily relativistic , especially , the dynamics of the mri is sub - alfvnic ; ( 3 ) the dynamics of the jet are highly relativistic . our new scheme can solve such problems accurately even when the initial magnetic field is very weak . we present multi - dimensional extension of our scheme in our next paper . we would like to thank bruno giacomazzo for providing the code computing the exact solution of the riemann problem in ideal mhd . numerical computations were in part carried out on cray xt4 at center for computational astrophysics , cfca , of national astronomical observatory of japan . this work is supported by grant - in - aids from the ministry of education , culture , sports , science , and technology ( mext ) of japan , no . 22@xmath2153369 ( t. i. ) . as explained in sec . [ sec : sec3.2 ] , we solve the evolution of electric and magnetic field by the method of characteristics . for the characteristic velocity , we use the appropriate mhd characteristic velocity when @xmath80 is large , that is , ideal mhd approximation is valid . however , when the conductivity @xmath80 is not so large , we have to replace the characteristic velocity with the speed of light . in this section , we discuss when to switch the characteristic velocity from appropriate mhd characteristic velocity to light velocity . in the following , we calculate the linear perturbation of the relativistic electromagnetic equation in order to obtain characteristic velocity . the relativistic electromagnetic fluid equations are given by @xmath216 + q \mathbf{v } \label{eq4 } , \\ q & = & \nabla \cdot \mathbf{e } , \qquad \nabla \cdot \mathbf{b } = 0 \label{eq5 } .\end{aligned}\ ] ] to obtain the dispersion relation , we start by expanding physical variables around an unperturbed state in the following frame : * the fluid is at rest : @xmath217 * the x - coordinate is parallel to the @xmath218 : @xmath219 * the magnetic field is in the x - direction : @xmath220 * charge neutrality : @xmath221 since we only want to judge when to switch characteristic velocity , we consider propagation of the transverse waves along the magnetic field . when one uses this procedure during the numerical simulation , one only has to calculate @xmath222 of the simulation data , and substitute its square root into the above @xmath116 . this is because @xmath222 is a scalar , and becomes the square of the magnetic field in the fluid comoving frame because of the assumption of the charge neutrality . since the magnetic field appears only in the form of @xmath223 in the following procedure , one can neglect the sign of magnetic field . in the following , we consider only the characteristic velocity of transverse waves . in the above condition , the alfvn mode is included in the z - component of the velocity @xmath224 and magnetic field @xmath225 , and decouples from other variables . for this reason , we consider only variables related to @xmath224 and @xmath225 . we replace the current vector in eq . ( [ eq1 ] ) with eq . ( [ eq3 ] ) . then , the perturbed equations are @xmath226 from these equations , the following dispersion relation is obtained : @xmath227 \omega^2 - i \sigma ( b^2 + \rho h ) k^2 \omega + \sigma^2 ( b^x)^2 k^2 = 0 , \label{eq6}\ ] ] eq . ( [ eq6 ] ) is the biquadratic equation with respect to @xmath179 , and has the formula of radicals . however , the analytical formula is very complex and hard to analyze , and is not suitable for obtaining the characteristic velocity . as shown below , the transverse wave becomes an alfvn wave in the long wavelength regime , and the light wave in the short wavelength regime shown in fig . [ figa1 ] . fig . [ figa1 ] shows that the damping rate is a monotonically increasing function of @xmath82 . for this reason , we establish the following criterion for the characteristic velocity : when all light modes damp during one time step @xmath228 we use appropriate mhd characteristic velocity for the method of characteristics ; when some light modes do not damp during one time step @xmath228 , we use light velocity for the method of characteristics . we will discuss this method in detail in the following . first , we substitute @xmath229 into eq . ( [ eq6 ] ) , and divide the dispersion relation into a real part and imaginary part . then , the real part is @xmath230 \omega_r^2 \nonumber \\ & + & \rho h \omega_i^4 + \sigma ( b^2 + 2 \rho h ) \omega_i^3 \nonumber \\ & + & [ \rho h k^2 + \sigma^2 ( b^2 + \rho h ) ] \omega_i^2 + \sigma ( b^2 + \rho h ) k^2 \omega_i + ( b^x)^2 \sigma^2 k^2 = 0 , \label{eq7}\end{aligned}\ ] ] and the imaginary part is @xmath231 \omega_r^3 - [ \rho h ( \sigma + 2 \omega_i ) \{k^2 + 2 \omega_i ( \sigma + \omega_i ) \ } + b^2 \sigma \{k^2 + \omega_i ( 2 \sigma + 3 \omega_i ) \ } ] \omega_r = 0 , \label{eq8}\ ] ] eq . ( [ eq8 ] ) implies that the solution for @xmath232 is @xmath6 and a conjugate complex numbers . the solutions of @xmath233 are pure decaying modes , so the other modes are the desired propagating ones that become light velocity in the limit of small @xmath80 and alfvn velocity in the limit of large @xmath80 . figs . [ figa1 ] are the dispersion relation for the propagation modes of the following parameters : @xmath234 these figures show that this mode becomes light in the limit of small @xmath80 and alfvn wave in the limit of large @xmath80 . although the form of @xmath232 does not become as fig . [ figa1 ] for some parameters , this mode always becomes light waves in the limit of small @xmath80 . from eq . ( [ eq8 ] ) , this desired mode can be obtained as follows : @xmath235 + b^2 \sigma [ k^2 + \omega_i ( 2 \sigma + 3 \omega_i ) ] } { b^2 \sigma + 2 \rho h ( \sigma + 2 \omega_i ) } . \label{eq9}\ ] ] we substitute this @xmath236 into eq . ( [ eq7 ] ) , and we obtain @xmath237 where @xmath238 this equation should include the propagation modes . ( [ eq10 ] ) is the biquadratic equation with respect to @xmath82 , but includes unknown quantity @xmath85 . [ figa1 ] show that the propagation mode becomes light waves in the short wavelength region , and damping rate @xmath239 is a monotonically increasing function of @xmath82 . note that what we want to know is whether the undamped shortest wavelength mode is light waves or alfvn waves , and we do not necessarily have to solve the biquadratic equation directly . for this reason , we substitute @xmath240 into @xmath85 of eq . ( [ eq10 ] ) , and solve it with respect to @xmath241 : @xmath242 @xmath243 \nonumber \\ & - & b^2 \rho h \sigma ( \sigma + 2 \omega_i ) [ -4 ( b^x)^2 \sigma^2 + \rho h ( 5 \sigma^2 + 18 \sigma \omega_i + 16 \omega_i^2 ) ] \\ \beta_2 & = & -8 \rho^2 h^2 \omega_i ( \sigma + 2 \omega_i)^3 [ b^2 \sigma + 2 \rho h ( \sigma + \omega_i ) ] [ b^2 \sigma + \rho h ( \sigma + 2 \omega_i ) ] ^3 \nonumber \\ & + & [ -b^6 \sigma^3 ( \sigma + 2 \omega_i ) + 2 \rho^2 h^2 ( \sigma + 2 \omega_i)^2 \{2 ( b^x)^2 \sigma^2 - \rho h ( \sigma + 2 \omega_i)^2\ } \nonumber \\ & + & b^4 \sigma^2 \{(b^x)^2 \sigma^2 - 2 \rho h ( 2 \sigma^2 + 7 \sigma \omega_i + 6 \omega_i^2)\ } \nonumber \\ & + & b^2 \rho h \sigma ( \sigma + 2 \omega_i ) \{4 ( b^x)^2 \sigma^2 - \rho h ( 5 \sigma^2 + 18 \sigma \omega_i + 16 \omega_i^2)\}]^2 , \\ \beta_3 & = & \rho^2 h^2 ( \sigma + 2 \omega_i ) [ b^2 \sigma + \rho h ( \sigma + 2 \omega_i ) ] , \\ \omega_i & = & - \frac{2 \pi}{\delta t}\end{aligned}\ ] ] where eq . ( [ eq10 ] ) has two solutions of @xmath241 , and we adopt the larger one since @xmath82 is a real number . substituting above @xmath241 into eq . ( [ eq9 ] ) , one can obtain the desired characteristic velocity . since what we need is appropriate mhd characteristic velocity , obtained velocity can not be used as the characteristic velocity . however , if obtained velocity is not alfvn velocity , it shows that we should use light velocity for the characteristic velocity . this method requires some further explanation . first , note that the above method needs @xmath223 , @xmath116 , and @xmath228 in the comoving frame , and one should transform numerical data from lab frame to comoving frame . second , numerical experiments indicate that @xmath232 becomes @xmath6 for some range of @xmath82 for some parameter , and @xmath241 of eq . ( [ eq11 ] ) becomes negative . in this case , we use speed of light as the characteristic velocity . finally , fig . [ figa1 ] implies that @xmath244 has some maximum value . this can be proved as follows . first , dividing eq . ( [ eq6 ] ) by @xmath245 , one obtains @xmath246 \bar{\omega}^2 - i ( b^2 + \rho h ) \bar{k}^2 \bar{\omega } + ( b^x)^2 \bar{k}^2 = 0 , \label{eq12}\ ] ] where @xmath247 , and @xmath248 . [ figa1 ] implies that the maximum value of @xmath244 is obtained in the limit of large @xmath249 , and the propagation mode is light wave in this limit . for this reason , we substitute @xmath250 into eq . ( [ eq12 ] ) . then it reduces to @xmath251 \bar{k}^2 \nonumber \\ & + & i [ - b^2 \bar{\omega_i } ' ( - 2 + 3 \bar{\omega_i } ' ) + 2 \rho \bar{\omega_i } ' ( 1 - 3 \bar{\omega_i } ' + 2 \bar{\omega_i}'^2 ) ] \bar{k } \nonumber \\ & - & b^2 ( - 1 + \bar{\omega_i } ' ) \bar{\omega_i}'^2 + \rho ( - 1 + \bar{\omega_i}')^2 \bar{\omega_i}'^2 = 0 . \end{aligned}\ ] ] since this is in the limit of large @xmath249 , what we have to consider is only the highest degree of @xmath249 . then , we set its coefficient equal to @xmath6 , and it reduces to @xmath252 this shows that @xmath244 becomes @xmath253 in the limit of large @xmath81 , and we use the speed of light as the characteristic velocity when @xmath254 is less than @xmath255 . for the second - order scheme , one has to compute the cell boundary numerical flux using riemann solver or method of characteristics with left and right states obtained by using muscl . in this section , we explain muscl of van leer @xcite . since we need a second - order scheme , the left and right states of primitive variables @xmath256 are @xmath257 where @xmath258 follows from a predictor step @xmath259 , \ ] ] where @xmath260 is the conserved variables . in the above equation , @xmath261 can be computed from eqs . ( [ eq01 ] ) and ( [ eq02 ] ) by replacing @xmath258 with @xmath262 . when one uses muscl , the @xmath263 in eqs . ( [ eq01 ] ) and ( [ eq02 ] ) are computed as follows : @xmath264 where @xmath265 aloy , m. a. , ibez , j. m. , mart , j. m. , mller , e. 1999 , , 122 , 151 anile , a. m. 1990 , relativistic fluids and magneto - fluids , by a. m. anile , pp . isbn 0521304067 . cambridge , uk : cambridge university press , february 1990 . , antonucci , r. 1993 , , 31 , 473 banyuls , f. , font , j. a. , ibanez , j. m. a. , marti , j. m. a. , & miralles , j. a. 1997 , , 476 , 221 blandford , r. d. , & konigl , a. 1979 , , 232 , 34 camus , n. f. , komissarov , s. s. , bucciantini , n. , & hughes , p. a. 2009 , , 400 , 1241 del zanna , l. , & bucciantini , n. 2002 , , 390 , 1177 del zanna , l. , zanotti , o. , bucciantini , n. , & londrillo , p. , 473 ( 2007 ) 11 . evans , c. r. , & hawley , j. f. 1988 , , 332 , 659 font , j. a. , miller , m. , suen , w .- m . , & tobias , m. 2000 , , 61 , 044011 giacomazzo , b. , & rezzolla , l. 2006 , journal of fluid mechanics , 562 , 223 hiscock , w. a. , & lindblom , l. 1983 , annals of physics , 151 , 466 hiscock , w. a. , & lindblom , l. 1985 , , 31 , 725 inoue , t. , inutsuka , s. , & koyama , h. 2007 , , 658 , l99 inoue , t. , & inutsuka , s. 2008 , , 687 , 303 inoue , t. , & inutsuka , s. 2009 , , 704 , 161 inoue , t. , asano , k. , & ioka , k. 2010 , arxiv:1011.6350 komissarov , s. s. 2007 , , 382 , 995 ( k07 ) palenzuela , c. , lehner , l. , reula , o. , & rezzolla , l. 2009 , , 394 , 1727 ( p09 ) li , c. , tang , y. h. , dai , y. , fang , c. , & vial , j .- c . 2007 , , 472 , 283 marti , j. m. , & mller , e. 1994 , journal of fluid mechanics , 258 , 317 mart , j. 1996 , journal of computational physics , 123 , 1 mart , j. m. , mller , e. 2003 , living reviews in relativity , 6 , 7 mckinney , j. c. , & gammie , c. f. 2004 , , 611 , 977 mignone , a. , & bodo , g. 2005 , , 364 , 126 mignone , a. , plewa , t. , & bodo , g. 2005 , , 160 , 199 mirabel , i. f. , & rodrguez , l. f. 1999 , , 37 , 409 piran , t. 2004 , reviews of modern physics , 76 , 1143 pons , j. a. , ma mart , j. , mller , e. 2000 , journal of fluid mechanics , 422 , 125 rees , m. j. , & gunn , j. e. 1974 , , 167 , 1 stone , j. m. , & norman , m. l. 1992 , , 80 , 791 strang , g. 1968 , siam journal on numerical analysis , 5 , 506 urry , c. m. , & padovani , p. 1995 , , 107 , 803 van leer , b. 1979 , journal of computational physics , 32 , 101 woosley , s. e. 1993 , , 405 , 273 zenitani , s. , hesse , m. , & klimas , a. 2009 , , 696 , 1385 zweibel , e. g. , & yamada , m. 2009 , , 47 , 291

in the context of warped extra - dimensional models with all fields propagating in the bulk , we address the phenomenology of a bulk scalar higgs boson , and calculate its production cross section at the lhc as well as its tree - level effects on mediating flavor changing neutral currents . we perform the calculations based on two different approaches . first , we compute our predictions analytically by considering all the degrees of freedom emerging from the dimensional reduction ( the infinite tower of kaluza klein modes ( kk ) ) . in the second approach , we perform our calculations numerically by considering only the effects caused by the first few kk modes , present in the 4-dimensional effective theory . in the case of a higgs leaking far from the brane , both approaches give the same predictions as the effects of the heavier kk modes decouple . however , as the higgs boson is pushed towards the tev brane , the two approaches seem to be equivalent only when one includes heavier and heavier degrees of freedom ( which do not seem to decouple ) . to reconcile these results it is necessary to introduce a type of higher derivative operator which essentially encodes the effects of integrating out the heavy kk modes and dresses the brane higgs so that it looks just like a bulk higgs . +

1301.7692

warped extra dimensional models have become very popular because they are able to address simultaneously two intriguing issues within the standard model ( sm ) : the hierarchy problem and the mass / flavor problem . they were originally introduced to treat the first issue @xcite in a setup where the sm fields were all localized at one boundary of the extra dimension . later it was realized that by allowing fields to propagate into the bulk , different geographical localization of fields along the extra dimension could help explain the observed masses and flavor mixing among quarks and leptons @xcite . flavor bounds and precision electroweak tests put pressure on the mass scale of new physics in these models @xcite , but extending the gauge groups and/or matter content ( e.g. @xcite ) or by slightly modifying the spacetime warping of the metric ( e.g @xcite ) , it is possible to keep the new physics scale at the tev level at the reach of the large hadron collider ( lhc ) . electroweak symmetry breaking can still happen via a standard higgs mechanism in these scenarios ( although it can also be implemented as as pseudo - nambu - goldstone boson ( pngb ) @xcite or described within the effective theory formalism @xcite ) . as the lhc announced the discovery of a light higgs - like particle of a mass around 125 gev @xcite , it becomes crucial to have a detailed prediction of the properties of the physical higgs particle in these models . the higgs boson itself must be located near the tev boundary of the extra dimension in order to solve the hierarchy problem , and so typically it is assumed to be exactly localized on that boundary ( brane higgs scenario ) . nevertheless , it is possible that it leaks out into the bulk ( bulk higgs scenario ) , and in doing so indirectly alleviate some of the bounds plaguing these models @xcite . the calculation of the production cross section of the brane higgs in these scenarios has been addressed before @xcite but we will pay close attention to the more recent works of @xcite . the towers of fermion kaluza - klein ( kk ) modes will affect significantly the sm prediction and in @xcite it was found that the higgs boson production rate can receive important corrections , either enhancing or suppressing the standard model prediction . the suppression or enhancement depends on the model parameters considered , in particular on the phases appearing in the different yukawa - type operators present in the 5d action . previously , the analysis of @xcite , in which only the first few modes were considered , gave no contribution to the rate from the towers of kk fermions . finally , the analysis of @xcite seems to indicate that with just a few kk modes a substantial effect is obtained , but of opposite sign as the one predicted from summing the infinite tower @xcite . in this work we consider the effects of allowing the higgs boson to propagate in the bulk , with its profile more or less localized towards the ir brane depending on the value of the mass parameter @xmath0 , related to the bulk mass of the 5d higgs field . to keep matters as simple as possible we will set up a model containing a single family of up - type 5d fermions along with a bulk higgs scalar . generalization to a more realistic scenario is straight forward but we prefer to stay as transparent as possible due to the many subtleties involved in the calculation . we first compute the contribution of the complete tower of kk fermions to the higgs production cross section as well as to the tree - level shift happening between the light fermion mass and its yukawa coupling ( leading to flavor violating couplings when considering three fermion families ) . these calculations , as outlined in @xcite , are analytically straightforward and allow us to obtain simple and compact results . we then repeat the same analysis numerically from the point of view of an effective theory in which only the first few kk fermions contribute . we show that for a bulk higgs with a thickness of the order of inverse tev scale , the results obtained are the same as the results obtained by summing the complete kk tower ( i.e. heavier modes decouple ) . moreover , these results are consistent with the predictions obtained in @xcite for the specific case of a brane localized higgs . the two aproaches outlined seem to give different predictions as the bulk higgs is continuously pushed towards the brane . it turns out that in order to maintain the consistency of both approaches we need to include in the analysis the effects of a special type of higher order operators . after these effects are included , we will come back and address in the discussion section the differences among the existing calculations in the literature and stress the importance of including the mentioned higher order operators in the analysis . our paper is organized as follows . in sec . [ sec : setup ] we summarize the simple 5d warped space model used in the calculation . in sec . [ sec : infinite ] we present analytical results for the higgs flavor - changing effects ( [ subsec : infinitefcnc ] ) and production ( [ subsec : infiniteprod ] ) , using the full tower of kk fermions . we use numerical methods to calculate the effects of including just a few kk modes in sec . [ sec : individual ] , both for flavor - changing neutral currents effects ( [ subsec : individualfcnc ] ) and higgs boson production ( [ subsec : individualprod ] ) . we include the effect of the higher order operator in sec . [ sec : ramanop ] and discuss the misalignment between the higgs boson profile and its vacuum expectation value ( vev ) in sec . [ sec : vevmisalignment ] . we discuss the significance of our results , compare them to previous analyses and conclude in sec . [ sec : conclusion ] . we leave some of the details for the appendices [ apen : i ] and [ apen : ii ] . we consider the simplest 5d warped extension of the sm , in which we keep the sm local gauge groups and just extend the space - time by one warped extra dimension . the spacetime metric is the usual randall - sundrum form @xcite : ds^2 = ( _ dx^dx^-dz^2 ) , [ rs ] with the uv ( ir ) branes localized at @xmath1 ( @xmath2 ) . we denote the @xmath3 doublets by @xmath4 and the @xmath3 singlets by @xmath5 where @xmath6 are flavor indices and @xmath7 represents the 4d spacetime coordinates while @xmath8 represents the extra dimension coordinate . the fermions are expected to propagate in the bulk @xcite . the up - sector fermion action that we consider is therefore & & s_fermion = d^4x dz , with @xmath9 being the covariant derivative , and we have added a yukawa interaction with a higgs field @xmath10 which in principle can be either brane or bulk localized . from the 5d fermion mass terms one defines dimensionless parameters @xmath11 , @xmath12 which are _ a priori _ quantities of @xmath13 . the coefficients @xmath14 have inverse energy units ( @xmath15 ) since yukawa couplings in 5d are higher dimensional operators . after separating 5d fields into left and right chiralities we impose a mixed ansatz for separation of variables q_l(x , z ) & = & q_l^0(z ) q_l^0 ( x ) + q_l^1(z ) _ l^1(x ) + ... , [ sep1 ] + q_r(x , z ) & = & q^0_r(z ) u^0_r(x ) + q_r^1(z ) _ r^1 ( x ) + ... , [ sep2 ] + u_l(x , z ) & = & u^0_l(z ) q_l^0(x ) + u_l^1(z ) _ l^1 ( x)+ ... , [ sep3 ] + u_r(x , z ) & = & u^0_r ( z ) u_r^0(x ) + u_r^1(z ) ^1_r(x ) + ... , [ sep4 ] where @xmath16 and @xmath17 are the sm fermions and @xmath18 are the heavier kk modes . in order to obtain a chiral spectrum , we choose boundary conditions for the fermion wavefunctions @xmath19 so that before electroweak symmetry breaking only @xmath20 and @xmath21 will be massless ( zero modes ) with wavefunctions : @xmath22 where we have defined @xmath23 and the hierarchically small parameter @xmath24 . thus , if we choose @xmath25 , the zero mode wavefunctions are localized towards the uv brane ; if @xmath26 , they are localized towards the ir brane . in order to implement minimally the higgs sector out of a 5d scalar we use the following action @xcite @xmath27 - v_{uv}(h)\delta(z - r ) - v_{ir}(h)\delta ( z - r'),\ ] ] where @xmath28 is the 5d mass for the higgs boson . the boundary potentials @xmath29 and @xmath30 yield boundary conditions that can accommodate electroweak symmetry breaking , so that one obtains a higgs vev with a non - trivial profile along the extra - dimension . around that vev , one should then add perturbations and obtain the spectrum of physical modes , i.e. a sm - like higgs boson and a tower of kk higgs fields . the expansion should look like h(x , z)= v_(z ) + h_(z ) h(x ) + .... , and we can choose the boundary conditions such that the profile of the higgs vev @xmath31 takes the simple form @xmath32 where @xmath33 and @xmath34 where @xmath35 is the sm higgs boson vev . one should note that the wave function @xmath36 of the light physical higgs ( lightest kk higgs field ) will have the form h_(z)= ( 1+o ( ) ) , [ hvsv ] so that for a light enough higgs boson mass both profiles @xmath36 and @xmath31 are aligned ( i.e. proportional to each other ) . the previous bulk higgs sector is capable of reproducing the brane higgs limit , since the wavefunction of the light higgs ( and its vev ) both depend exponentially on the parameter @xmath0 . as this parameter is increased , the wavefunctions are pushed more and more towards the ir brane mimicking a perfectly localized higgs sector . parameter and so these fields will decouple from the theory for very large @xmath0 . ] indeed , the wave function of the higgs can act as a brane localizer since _ h^2(z ) = _ v^2(z ) = ( z - r ) , where the dirac delta function is defined as the limit of a sequence of functions with increasing value of @xmath0 . one can easily prove that for any wavefunction @xmath37 ( or a product of wavefunctions ) we have _ _ r^r^+h^2(z ) f(z ) dz = f(r ) . there is however an issue about localizing the whole higgs sector towards the brane since we just showed that only quadratic higgs operators will `` become '' brane localizers . when a 5d action operator contains more than two ( or less than two ) higgs fields , the ( successful ) localization of such operators is not guaranteed . in fact in order to ensure that the 5d bulk higgs scenario correctly tends smoothly to a fully localized higgs sector , one should implement a prescription enforcing a precise @xmath0 dependence on the coefficients of all operators containing higgs fields . more precisely , the coefficient @xmath38 of an operator containing @xmath39 higgs fields ( before electroweak symmetry breaking ) should behave as y^n()= y^n_1^ , where @xmath40 . this is the only way to ensure that we can have _ _ r^r^+ y^n ( ) h^n(z ) f(z ) dz = _ _ r^r^+ y_1^n ^h^n(z ) f(z ) dz= y^n_1 f(r ) , or in other words _ y^n ( ) h^n(z ) = y^n_1 ( z - r ) . in particular for 5d yukawa type couplings this prescription implies that the 5d yukawa coupling will have to carry a @xmath41 dependence in order to ensure that the brane limit yukawa coupling is non - vanishing @xcite ( see also @xcite ) . but it also means that any other 5d action operator containing a single higgs field would need to carry the same @xmath41 dependence . on the other hand , 5d action operators containing 3 higgs fields ( like the operator @xmath42 ) would have a diverging limit for @xmath0 large unless its action coefficient @xmath43 is itself suppressed by @xmath44 . the previous prescription makes it technically possible to define a localized higgs sector from a 5d bulk higgs field , but it certainly seems quite contrived to appropriately fix all operator coefficients such that they all can give non - zero and finite contributions when the higgs is localized . a brane localized higgs sector could seem `` un - generic '' or `` un - natural '' if it is to be seen as a limiting case of a bulk higgs . more details about the complete prescription for operators containing higgs fields are presented in appendix [ apen : ii ] . for completeness and consistency , we present first a result previously obtained in @xcite , namely the computation of the shift between the light sm fermion mass term and its yukawa coupling with the higgs field ( leading to higgs mediated fcnc when more than one fermion family is considered ) . we then calculate the coupling between the physical higgs and two gluons for the 5d bulk higgs case . we can follow two routes to obtain our predictions . the computation of the flavor violating couplings of the higgs scalar with fermions will be obtained in an approach based on considering _ first _ electroweak symmetry breaking and _ then _ solving the 5d equations of motion for the fermions ( i.e. the effect of the higgs vev is directly taken into account in the equations of motion and during the dimensional reduction procedure ) . the alternative ( and equivalent ) approach would be to consider _ first _ the dimensional reduction ( i.e. obtain the 4d effective theory in the gauge basis ) , and _ then _ consider the electroweak symmetry breaking in the presence of the infinite tower of kk fermions . after performing the diagonalization of the infinite fermion mass matrix ( as well as canonical normalization of the fermion kinetic terms ) we should recover the same results . we use the first approach in the first subsection , and the second approach in the computation of the higgs coupling to gluons and also in the following sections where we will truncate the infinite mass matrix in order to consider only the effect of the first few kk modes . after imposing electroweak symmetry breaking in the higgs sector , the four profiles @xmath45 and @xmath46 introduced in eqs . ( [ sep1 ] ) to ( [ sep4 ] ) must obey the coupled equations coming from the equations of motion : & & -m_u q_l - q_r + c_q+2z q_r + ( rz)v_(z ) y_u u_r=0 , [ qr ] + & & -m^*_u q_r + q_l + c_q-2z q_l + ( rz)v_(z ) y_u u_l=0 , [ ql ] + & & -m_u u_l - u_r + c_u+2z u_r + ( rz)v_(z ) y^*_u q_r=0 , [ dr ] + & & -m^*_u u_r + u_l + c_u-2z u_l + ( rz ) v_(z ) y^*_u q_l=0 , [ dl ] [ eqmot ] where the @xmath47 denotes derivative with respect to the extra coordinate @xmath8 and @xmath48 is 5d yukawa coupling . it is simple to deduce from these equations an exact expression for the mass eigenvalue @xmath49 in terms of the fermion profiles @xcite m_u = r^4^r_r dz ( ( |u_l|^2+|q_r|^2 ) + rv_(z)z^5(y_u u_rq^*_l - y^*_uq_ru^*_l ) ) , and compare it to the expression of the fermion yukawa coupling , i.e y^u_4 = r^5^r_r dz h_(z)z^5(y_u u_rq_l^*+ y^*_dq_ru_l^ * ) , where @xmath36 is the profile of the physical higgs field . with these two expressions we compute the shift ( or misalignment ) between the fermion mass @xmath49 and the yukawa coupling @xmath50 as ^u = m_u - v_4 y^u_4 , which becomes simply ^u&=&r^4^r_r dz ( ( |u_l|^2+|q_r|^2 ) -2y^*_u r v_(z)z^5 q_ru_l^ * ) . [ delta ] in order to proceed further , a perturbative approach is used , such that we assume that @xmath51 where @xmath35 is the sm higgs vev . knowing the analytical form of the vev profile @xmath31 and using the @xmath52 small parameter it is possible to solve perturbatively the system of coupled equations ( [ qr ] ) to ( [ dl ] ) to any order in @xmath52 ( see @xcite for details ) . the result for the shift in the top quark yukawa coupling is @xmath53 , \end{aligned}\ ] ] where we have only included the contribution from the third term in eq . ( [ delta ] ) , as the other terms are subdominant for light quarks , although not necesarily for the top quark . for clarity we omit their analytical expression here , but the complete analytical result can be found in @xcite and in the appendix [ apen : i ] of this work . the shift in the yukawa coupling has some dependence on the higgs localization parameter @xmath0 and it is shown in figure [ fig : topshift ] as the `` infinite sum '' result , as the procedure we followed is equivalent to diagonalizing the infinite fermion mass matrix in the gauge eigenbasis . . each line represents an effective theory containing the given amount of kk fermions . the lower line ( blue ) represents the contribution from the infinite tower of kk modes . apart from the direct phenomenological impact of this result , this term also affects the @xmath54 coupling , as discussed in the text . the dimensionless 5d yukawa couplings are fixed at @xmath55 and the kk scale is set at @xmath56 gev ( the overall effect scales as @xmath57).,height=302 ] in this section we follow the approach of working with the infinite fermion kk modes with wavefunctions in the gauge basis . this is not the physical basis after electroweak symmetry breaking since yukawa couplings will introduce off - diagonal terms in the infinite fermion mass matrix , which should be properly diagonalized in order to obtain the physical basis . to the higgs - gluon - gluon coupling . in the sm , the dominant contribution is through the top quark due to its large yukawa coupling with the higgs boson . in rs the heavier kk fermions contribute to the coupling with potentially large effects , either suppressing or enhancing the sm coupling , depending on the phases present in the different yukawa - type operators present in the 5d action , and on the localization of the higgs ( see text for details).,height=151 ] since the higgs field is not charged under qcd the main contribution to its coupling to gluons comes from a top quark loop , as shown in figure [ fig : toploop ] ; if the model contains many heavy quarks the resulting cross section for the process is @xmath58 is @xcite ^sm_ggh = _ s m_h^2 576|_qy_qm_q a_1/2(_q)|^2(s - m_h^2)[hggsm ] , with @xmath59 , @xmath60 being the @xmath61 invariant mass squared and @xmath62 representing the physical fermions with physical yukawa couplings @xmath63 and masses @xmath64 . the form factor is given by a_1/2 ( ) = 32[+(-1)f()]^-2[ff ] , with f ( ) = [ ] ^2 + -14 ^ 2 > 1[ftau ] . here we want to figure out the contribution to the @xmath54 coupling coming from 5d quark doublets and a singlets , i.e. containing the sm quarks ( which includes doublets and singlets @xmath65 ) , along with the associated towers of vector - like kk fermions , @xmath66 . the relevant quantity to calculate is c_hgg=_qy_qm_q a_1/2(_q ) , where @xmath67 is the physical yukawa coupling of the physical dirac fermion @xmath62 and @xmath64 is its mass . as stated before , it will prove useful to work in the gauge basis , and so we represent the yukawa couplings between the kk fermions @xmath68 and @xmath69 in the gauge basis as @xmath70 . its values will be obtained by performing the overlap integral of the higgs profile and the corresponding bulk fermionic wave functions , i.e. y_q_lu_r^u = y^u_r^rdz(rz)^5 q_l^u(i)(z)u_r^(k)(z)[yu ] , where we have assumed that the nontrivial higgs vev and the physical higgs profile are perfectly aligned in section [ sec : vevmisalignment ] . ] . the yukawa couplings between different chirality kk fermions and also between zero modes and heavy kk fermions are obtained and written in a similar way so that we can write the infinite dimensional fermion mass matrix as [ decom ] y^u_q_lu_r v_4 & 0 & y^u_q_lu_r v_4 + y^u_q_lu_r v_4 & m_q & y^u_q_lu_r v_4 + 0 & y^u*_u_lq_r v_4 & m_u u_r^(0 ) + q_r^u(k ) + u_r^(l ) [ infinitem ] , where @xmath71 and @xmath72 are the kk mass matrices for the corresponding fermion fields in the gauge basis , and we have suppressed fermion family indices to simplify notation . from eqs . ( [ ff ] ) and ( [ ftau ] ) , we notice that in eq . ( [ hggsm ] ) the form factors , @xmath73 for light fermions , and @xmath74 for the much heavier kk modes and the top quark . therefore , separating the contribution of the light fermions from the heavy ones we write c_hgg=_lighty_q m_qa_1/2(_q ) + _ heavyy_qm_q , where in the first ( second ) term the sum is only over light ( heavy ) fermion generations . noting that _ _ lighty_qm_q = ( ^-1 ) , where @xmath75 is the fermion mass matrix given in ( [ decom ] ) , while @xmath76 is the yukawa matrix , we have c_hgg=(^-1 ) + _ lighty_q m_q(a_1/2(_q ) - 1)[sumym ] . we also note that @xmath77 and since the trace is invariant under unitary transformations , we can compute it in the gauge basis ( so we can use the fermion mass matrix in that basis ) . up to first order in @xmath78 one finds ( ^-1 ) = ( ) -v_4_i , j2m_q_im_u_j(y^u_q_l_iu_r_jy^u*_u_l_jq_r_i - y^u_q_lu_r_jy^u*_u_l_jq_r_iy^u_q_l_iu_r y^u_q_lu_r)[trace ] . noting that the sm masses and yukawa couplings are also modified ( shifted ) as @xcite [ ym ] y_qm_q| _ ( 1 + 2v_4 ^ 2 y^u_q_lu_r_i , jy^u_q_lu_r_jy^u*_u_l_jq_r_iy^u_q_l_iu_r m_q_im_u_j ) , [ shiftsum ] we can write the total @xmath54 coupling as c_hgg=-2v_4_i , jy^u_q_l_iu_r_jy^u*_u_l_jq_r_i m_q_im_u_j + y_qm_q|_a_1/2(_q_light ) [ c ] . where we have used equations ( [ sumym ] ) , ( [ trace ] ) and ( [ ym ] ) . as we mentioned before , the form factor is negligible for the light fermion generations . therefore neglecting the last term above , and using ( [ yu ] ) we have [ hggyu ] c_hgg=-2v_4y^uy^u*r_i , jdzdz(rz)^5(rz)^5 q^(i)_l(z)q^(i)_r(z ) m_q_iu^(j)_r(z)u^(j)_l(z)m_u_j h_(z)h(z ) , where the 5d bulk physical higgs profiles can be normalized as @xcite h_(z ) = r(zr)^2 + , with @xmath79 being the warp factor . the sums in eq . ( [ c ] ) are given by @xcite _ i=1^ q^(i)_l(z)q^(i)_r(z)m_q_i = -z^2+c_qz^2-c_q r^4 , [ sumdbrane ] and _ j=1^ u^(j)_r(z)u^(j)_l(z)m_u_j = z^2+c_uz^2-c_u r^4[sumubrane ] . substitution of these sums and of the higgs profile in eq . ( [ hggyu ] ) and assuming . for any physically acceptable model @xmath80 . ] @xmath81 will finally give the total higgs coupling for the light fermions which is given in appendix i. if we assume that @xmath82 and @xmath83 , which is the case for light fermions ( up - like fermion ) , the expression for @xmath84 can be simplified as c^up_hggy^uy^u*r^22(1+)(2++c_q - c_u)14 + 2[4 + 2beta ] . in the case of the top quark , we have to add the contribution due to the last term in eq . ( [ c ] ) , since @xmath85 . following the notation in @xcite , we write the additional contribution as eq . ( [ ym ] ) multiplied by the form factor , @xmath86 and last term , is the result of kinetic term corrections due to the shift in yukawa couplings , which are also not negligible for the heavy fermions . the shift is given by _ 2^top&=&r^4^r_r dz ( ( |u_l|^2+|q_r|^2 ) ) . [ delta ] for a complete discussion on this , we refer the reader to @xcite . so finally for ir localized fermions with @xmath87 and @xmath88 ( top - like ) we have @xmath89 . \end{aligned}\ ] ] following our ansatz for localizing the higgs sector , and in order to compare with previous brane higgs results , we need to replace the @xmath90 yukawa couplings with the dimensionless and @xmath0-independent couplings = ( 2-c_q+c_u+)y^5d . the results obtained in this section , of the contribution of a 5d top - like quark and a 5d up - like quark to the @xmath54 coupling are shown in both panels of figure [ fig : cggh ] as the infinite sum " result . coming from the kk partners of the `` up '' quark ( left panel ) and from the full top quark sector ( right panel ) as a function of the bulk higgs localization parameter @xmath0 . each line represents the numerical result obtained in an effective theory containing the amount of kk fermions indicated . the upper line ( blue ) represents the contribution of the infinite tower of kk modes ( computed in the text analytically ) . the dimensionless 5d yukawas are fixed at @xmath55 and the kk scale is set at @xmath56 gev ( the overall effect scales as @xmath57).,title="fig:",width=336 ] coming from the kk partners of the `` up '' quark ( left panel ) and from the full top quark sector ( right panel ) as a function of the bulk higgs localization parameter @xmath0 . each line represents the numerical result obtained in an effective theory containing the amount of kk fermions indicated . the upper line ( blue ) represents the contribution of the infinite tower of kk modes ( computed in the text analytically ) . the dimensionless 5d yukawas are fixed at @xmath55 and the kk scale is set at @xmath56 gev ( the overall effect scales as @xmath57).,title="fig:",width=336 ] in this section we take a different approach and compute the effects on higgs phenomenology ( fcnc and production cross section ) due to only the first few kk fermions in the model . that is , we consider a 4-dimensional effective theory which contains the sm matter content , augmented by a few levels of kk fields . this procedure is better fitted within the framework we work in ( low cut - off effective theories ) , the drawback being that it is not possible to obtain general analytical predictions in a close form . our strategy will be to assign some generic values to the parameters of the model and perform the computations numerically . in particular we will fix the bulk mass parameters of the 5d fermions @xmath62 and @xmath91 to be @xmath92 and @xmath93 ( for an up - type quark ) and @xmath94 and @xmath95 ( for a top quark ) . the value of the dimensionless 5d yukawa coupling will be taken to be @xmath96 . in order to evaluate the shift in the yukawa coupling of the sm fermion ( the zero mode ) due to the presence of a finite number of kk fermions , we can simply use eq . ( [ shiftsum ] ) , with the understanding that now the sum is finite , and so we shall sum up to the maximum number of kk modes chosen . we are interested in computing the top quark yukawa shift as it is the most interesting for direct phenomenology , and also because it will also enter in the calculation of the @xmath54 coupling . we perform the sum numerically and stop the summation at different maximum numbers of kk fermions . the results are shown in figure [ fig : topshift ] in which we focus on the variation of the yukawa coupling shift with respect to the bulk higgs localization parameter @xmath0 and we compare these to the results obtained in the previous section for the infinite kk degrees of freedom . the main observation is that for small @xmath0 , the finite sums are in good agreement with the infinite sum result . on the other hand for large values of @xmath0 the yukawa shift obtained from the finite sums becomes more and more irrelevant and is clearly at odds with the infinite sum prediction . to evaluate the contribution to the @xmath54 coupling coming from the individual kk fermion modes we proceed as in the previous subsection . we now use eq . ( [ hggyu ] ) , and sum up to the maximum number of kk modes desired . we perform the sum numerically and show the results in figure [ fig : cggh ] . again we are interested in the variation of the couplings with @xmath0 and compare them to the result for the @xmath84 obtained by calculating the infinite sum , as shown in the previous section . the two panels of the figure show the contribution to the @xmath54 coupling coming from a 5d up - like quark ( left panel ) and the contribution coming from a 5d top - like quark ( right panel ) for @xmath0 values up to 100 . we can see how the sums over different maximum number of kk modes converge to the infinite sum limit as we vary @xmath0 . the approximation obtained by considering just a few kk modes is much better for low values of @xmath0 . for example , from the left panel of figure [ fig : cggh ] , for @xmath97 , 8 kk modes saturate some 90% of the infinite sum , while for @xmath98 , 8 kk modes saturate some 60% of the infinite sum . for @xmath99 ( corresponding to a higgs highly localized towards the brane ) , 8 kk modes represent only some 10% of the total kk contribution . this dependence on @xmath0 is in agreement with the results found in @xcite , in which a brane localized higgs was considered ( i.e. @xmath100 ) and the first few kk fermions considered were found to give a negligible contribution to the @xmath54 coupling . + we conclude that in all the previous calculations ( the top quark yukawa shift and the contributions to the @xmath54 coupling coming from up - type and top - like 5d quarks ) we have observed the same feature , namely that in the case of a bulk higgs ( small @xmath0 ) , the effect of the heavier kk modes decouples ( i.e. performing the infinite sum is equivalent to sum only over the first few kk modes ) . on the other hand , when @xmath0 is very large , the heavier degrees of freedom do not seem to decouple hinting towards some type of uv sensitivity of the brane higgs case . this is not that surprising since the thickness of a higgs being crushed against the brane is becoming smaller and smaller , and the scale associated with the higgs localization eventually becomes much larger than the cut - off of the scenario . we will now see how adding a type of higher derivative operators will be sufficient to make the finite sums consistent with the infinite sum results obtained earlier . we have just seen how the results obtained in the previous section ( [ sec : individual ] ) , where we sum over a few kk modes agree with the complete kk tower summation of section [ sec : infinite ] only in the case of a bulk higgs boson . when the higgs is on the brane , or very much pushed towards the brane , the results for the two approaches do not seem to agree ( see figure [ fig : cggh ] when @xmath101 ) . we will reconcile the two methods by including , in the effective theory calculation , the contribution of higher derivative operators . in particular we consider the effect of the following operator in the action with a dimensionful coupling constant @xmath102 ( flavor indices are suppressed ) , s d^4x dz . the operator is of yukawa - type as it couples two fermions with the higgs , but it involves derivatives of fields . the coupling @xmath102 should be in units of @xmath103 , the cut - off of the theory , and so obviously this operator is cut - off suppressed ( we note that the standard 5d yukawa @xmath48 coupling is also dimensionfull and cut - off suppressed , but by two units less than @xmath102 ) . since @xmath104 and @xmath105 satisfy dirichlet boundary conditions on the ir brane , their derivatives along the extra dimension can be large after electroweak symmetry breaking and so we focus on the operator s d^4x dz ( rz)^3 , [ r ] which includes only the @xmath106 chirality fermion components @xmath104 and @xmath105 as it could lead to potentially large effects . as explained in the previous sections we can proceed in two ways in order to compute the effects of this operator . we could study the effect of the operator into the 5d equations of motion _ after _ electroweak symmetry breaking ( esb ) and calculate its effects from these . alternatively , we could solve the equations of motion and perform the dimensional reduction _ before _ esb , and then consider the effects produced by the operator by working in this gauge eigenbasis . both methods should be equivalent , but we will follow the second one . in this approach , we obtain the effective 4d theory and since it is non - renormalizable , we cut - off its spectrum at the cut - off scale thus effectively we only allow a few physical kk modes into the calculation . the effects from higher modes are integrated out and encoded in all higher order operators of the theory with their effects under control by the cut - off suppression . in the case of the @xmath102 operator the potentially large derivatives of @xmath104 and @xmath105 can offset the cut - off suppression and so we should keep this operator in the calculations . in the approach in which the kk modes are in the gauge basis , the @xmath102 operator will affect the fermion mass matrix from eq . ( [ infinitem ] ) , and in particular it will contribute to the @xmath107 terms . ( relative to the standard model ) as a function of the higgs localization parameter @xmath0 when considering _ only _ a five - dimensional up - type quark , and computed with the higher derivative term discussed in the text in addition to the standard 5d yukawa coupling term . since both contributions have independent phases we add and subtract their generic size to obtain the shaded region of possible values . these results are calculated by using only the first 3 kk modes ( i.e. considering an effective theory with a cut - off of the order the the fourth kk mass ) . the dimensionless 5d yukawas are fixed at @xmath55 and the kk scale is set at @xmath56 gev ( the overall effect scales as @xmath57).,height=302 ] @xmath108{ramanshifttop.pdf } & \includegraphics[width=3.5in , height=7cm]{ramanhggtop.pdf } \end{array}$ ] its effects can therefore be tracked into the effects of these _ wrong _ chirality terms , as was already noted in the appendix of @xcite . we can thus formally treat the situation as before , where a truncated version of the infinite mass matrix of eq . ( [ infinitem ] ) is considered ( with just a few kk levels ) , but now we redefine the terms @xmath107 to include the contributions from @xmath102 as y_u_lq_r^u = _ r^rdz(rz)^5 ( y^u u_l(z)q_r(z ) + y_r^u z^2r^2 _ zu_l(z)_zq_r(z ) ) . it is now easy to compute numerically the new effects since from here we just have to repeat the previous procedure . the results are shown in figures [ rl ] and [ rh ] . in both figures we show the individual contributions coming from the normal yukawa coupling @xmath109 , from the new @xmath110 coupling , as well as the combined effect . this combined effect is represented by the shaded region , the reason being that the two types of couplings @xmath109 and @xmath110 have independent phases and so can add up constructively or destructively , or in between . in figure [ rl ] we focus on the contribution to @xmath54 due to an up - like 5d quark . in figure [ rh ] we show the predictions for the both shift in the sm top quark yukawa coupling as well as the prediction for the contribution to the @xmath54 coupling coming from a 5d top - like quark . as we can see , the shift in the top quark yukawa coupling can be quite large , and for low values of the higgs localization parameter @xmath0 the shift obtained always results in a suppression in the yukawa coupling . for large values of @xmath0 the shift can be in either direction ( suppression or enhancement ) . in the case of the @xmath54 coupling , we see that the contribution represents an enhancement with respect to the sm prediction for small values of @xmath0 , and again for large values of @xmath0 the @xmath84 coupling can be either enhanced or suppressed depending on the relative phases between @xmath102 and @xmath109 . for this comparison we have taken the absolute value of both couplings @xmath102 and @xmath109 to be the same , i.e. @xmath55 , in appropriate units of the cutoff . the main feature to remember is that the effects of the higher derivative operator @xmath102 are subdominant for small @xmath0 but become dominant for large @xmath0 . the large contribution obtained at large @xmath0 is precisely what makes these new predictions consistent with the results obtained with the original infinite sum , and so the higher derivative operators that we have considered here somewhat encode the uv sensitivity found in the previous section . in this section we present a discussion on how to treat the case where the higgs profile is different form its vev profile . this is equivalent to consider the mixing effects between the massless zero mode higgs boson , and the heavy kk higgs modes and its effects on the higgs observables computed in this paper . we follow closely an argument by azatov @xcite and for simplicity we will discuss a simple situation in which the 4d effective theory contains only two new heavy vector - like fermions , @xmath62 and @xmath91 , doublet and singlet of @xmath3 respectively . this is the situation one would have when the kk fermion towers are truncated after the first kk excitation . let s first define our notation for the following quantities y^_ijdz()^5 _ i _ j + x^_ijdz()^5 _ i _ j h_(z ) , where @xmath35 is the sm higgs vev and @xmath111 are the 5d profiles of the higgs vev and the higgs physical field , which are generically different . that is , after ewsb , the higgs field is expanded around the nontrivial vev @xmath31 as h(x , z)=v_(z ) + h(x ) h_(z ) + ... . in the case of the bulk higgs sector considered here , both profiles @xmath111 are almost the same , ( see eq . ( [ hvsv ] ) ) , the order of the misalignment between them being controlled by powers of @xmath112 ( a small quantity ) . we consider all the possible couplings between the higgs and the fermions of the effective theory which after ewsb can be written as the matrix @xmath113 as (|q_l,|q_l|u_l ) ( y^_q_lu_r v_4 + x^_q_lu_r h(x ) & 0 & y^_q_lu_r v_4 + x^_q_lu_rh(x ) + y^_q_lu_r v_4 + x^_q_lu_rh(x ) & m_q & y^_q_lu_r v_4 + x^_q_lu_rh(x ) + 0 & y^_u_lq_r v_4 + x^_u_lq_rh(x ) & m_u ) (u_r + q_r + u_r ) , the coupling between the physical higgs and the two gluons is controlled by the physical yukawa couplings @xmath114 and masses @xmath115 of the heavier physical fermions running in the loop ( top quark and kk modes ) , i.e. _ heavy = tr(*y*_phys * m*_phys^-1)-_light , where @xmath116 is the physical yukawa coupling and @xmath117 is the physical fermion mass matrix of the setup . because the trace is invariant under unitary transformations , we can rotate to the gauge basis and write tr(*y*_phys*m*_phys^-1 ) = tr(*y*_gauge*m*_gauge^-1 ) , and note that we can now relate this to the matrix @xmath113 as tr(*y*_gauge*m*_gauge^-1 ) = _ h(det m(v_4,h))|_h=0 . the procedure is the same as was followed in section [ sec : infinite ] , i.e. we compute the the determinant by expanding in powers of @xmath118 and after combining everything we obtain _ heavy&=&_hdet m(v_4,h)- + & = & v_4(-- ) . this result is the equivalent to eq . ( [ c ] ) with the effect of the misalignment between @xmath31 and @xmath36 . one sees that the difference lies in the substitution of one of the @xmath119 terms by an @xmath120 term , and so the correction to the result of eq . ( [ c ] ) is = v_4(-- ) , which is controlled by ( x^_u_lq_r - y^_u_lq_r)= dz()^5 u_l(z)q_r(z ) ( h_(z)- ) , and ( x^_q_lu_r - y^_q_lu_r)= dz()^5 q_l(z)u_r(z ) ( h_(z)- ) , and since the misalignment between @xmath31 and @xmath36 can be computed perturbatively @xcite as h_(z)-= ( + + o(m_h^4r^4 ) ) , we obtain ( x^_u_lq_r - y^_u_lq_r ) = m_h^2 r^2 ( + ) + o(m_h^4r^4 ) . in other words , the effect of considering the misalignment between @xmath31 and @xmath36 is to add a correction with the same structure as the result of eq . ( [ c ] ) , but with a suppression of @xmath121 , i.e. the correction is at most @xmath122 , and becomes much smaller for increasing values of @xmath0 . of the integrals @xmath123 and @xmath124 is quite mild and so , in terms of order of magnitude , we have @xmath125 . ] in this work we have presented the results for the predictions of higgs phenomenology in a toy - model rs setup in which the higgs field is allowed to propagate in the bulk and with a single 5d fermion field . our results can be extended to three families to include full flavor effects , but the generic predictions that we would obtain are expected to be basically the same as the ones presented in @xcite . that is , that in the context of flavor anarchy , where the action parameters are all of the same order but with more or less random values and phases ( with the constraint of obtaining correct sm predictions ) the couplings of the higgs with fermions and gluons and photons can receive important corrections , either enhancing or suppressing the sm predictions . however , the two references mentioned present calculations performed by including the effect of all the kk fermions , technically assuming an infinite cut - off for the model ( where a brane higgs is considered ) . in general , all these scenarios break down at a low cut - off , becoming strongly coupled for both gauge and yukawa interactions . the implicit assumption made in @xcite was that the effects of the heavier modes should decouple quickly , at least for the case of a bulk higgs field . the main motivation to perform the calculations by considering the full infinite fermion kk tower , as well as pushing the higgs into the brane was mainly of technical nature . indeed both the flavor structure of the higgs yukawa couplings as well as the coupling to gluons and photons can be computed analytically with those ingredients . in @xcite , the authors checked analytically that the corrections to the higgs yukawa couplings were actually of the same order for a bulk higgs and a brane higgs . however it was pointed out in @xcite that in the brane higgs case , the effects of the heavier kk fermion modes do not decouple and that they all contribute evenly in the computation of the higgs couplings in the model . on the other hand , we showed in sections [ sec : infinite ] and [ sec : individual ] of this paper that the heavier kk modes in the case of a bulk higgs do decouple very quickly , so that the analytical result obtained by using the infinite kk tower approaches with great precision the numerical result obtained by considering an effective theory with only a few kk fermion modes . moreover , when considering the effective theory with only a few kk modes , one should include in the action all possible operators and in particular the higher derivative ones introduced in section [ sec : ramanop ] . these effects were omitted in @xcite , and as we showed in this work , the importance of these operators increases as the higgs is more and more localized towards the brane . in @xcite , the authors considered an rs setup with a highly localized higgs and the presence of only a few kk fermions and studied the effects on the higgs couplings to gluons and photons , among other observables . in the limit of the sm gauge group ( they did consider an extended gauge group ) they found no significant deviations from the sm predictions . indeed this result is consistent with our findings of section [ sec : individual ] ( no higher derivative operators invoked yet ) , since as it can be seen on figures [ rh ] and [ rl ] , the shift in higgs yukawa couplings and the new effects to higgs - gluon - gluon coupling vanish in the limit of highly localized higgs ( large @xmath0 parameter ) . on the other hand , in @xcite it is claimed that large effects should be present in the case of a brane higgs and with only a few kk modes present in the effective theory ( and no higher derivative operators ) , a result inconsistent with both our findings and those found in @xcite . we can trace the origin of the disagreement in their calculation of the higgs yukawa couplings . those are computed by using the full 5d equations of motion , which as we have said earlier is equivalent to considering the complete tower of kk modes . then , using these couplings , they calculate the @xmath54 radiative coupling but now including only a finite amount of kk fermions . this treatment leads to a highly suppressed top quark yukawa coupling ( due to effects from the infinite kk tower ) and a vanishing contribution to @xmath54 from the loops of kk fermions considered ( one would need the whole tower to obtain a finite effect ) . their end result is a suppressed top quark yukawa coupling and a suppressed @xmath54 coupling ( due to the smaller top quark yukawa ) , predictions which are at odds with the findings of @xcite and of this paper . the procedure of @xcite seems inconsistent because essentially the authors use infinite kk degrees of freedom in one part of their calculation ( the sm quark yukawa couplings computation via equations of motion ) but then they truncate the kk degrees of freedom in order to compute the @xmath54 coupling . in any case , had they included the higher derivative operators introduced in this paper , their results would have changed dramatically since then , the effect to @xmath54 coming from the top quark yukawa loop would remain basically the same , but the effects due to loops of a few kk fermions would dominate the overall effect ( and thus the result would start to become consistent with the findings of @xcite ) . also , the predictions of @xcite should change if one considers the effects of the higher derivative operators introduced in section [ sec : ramanop ] . in that situation , the higgs couplings can receive large corrections , and can be of any sign ( suppression or enhancement ) due to the different phases present in the couplings @xmath119 and @xmath102 . in fact we have found here that for a higgs field in the bulk , our results are more predictive than for a brane higgs field , because the effect of the higher derivativer operators is subdominant for a bulk higgs field . the effects from only the 5d yukawa operators are aligned @xcite , and thus all the kk quarks add up in phase . in that situation we can have definite predictions for the effects caused by a single family of fermions , i.e. it will produce a suppression in the light quark yukawa coupling and an enhancement in higgs boson production ( as well as suppression in the higgs to photons coupling ) @xcite , with the caveat of taking the dimensionless couplings of both yukawa terms and higher derivative operators to be the same ( consistent with the usual assumption that all 5d coefficients have to be of the same order ) . taking into account the three fermion families in conjunction with a bulk higgs field might weaken this prediction due to complicated flavor mixings and structure , but still one should be able to draw a correlation between yukawa couplings and higgs production ( and @xmath126 ) for the case of a bulk higgs field . the parameter space of the bulk higgs scenario can therefore be under a tighter pressure as more and more precise experimental measurements in higgs observables at the lhc become available . in particular if the predicted and correlated deviations of higgs couplings is not clearly observed this should put bounds on the kk scale of the bulk higgs scenario . the situation for a higgs on the brane is different . the higher order derivative terms are now important . each kk tower of light quarks and the top will contribute to the @xmath54 coupling , but their effect depends on arbitrary relative phases ( between @xmath102 and @xmath127 ) , and so one can not make a firm statement about the magnitude and phase of the overall contribution : it can be a suppression or an enhancement , or in between . finally we comment again on the apparent problem of a highly localized higgs scenario ( brane higgs ) in which predictions made from a truncated fermion kk tower are very different from predictions made from an infinite fermion kk tower . this apparent uv - sensitivity can actually be lifted by considering the higher derivative operators described here ( first introduced in @xcite ) . when these are included , the predictions made with a finite kk fermion tower become consistent with the original predictions obtained with an infinite fermion tower . a more esthetic problem with the brane higgs scenario remains , since the definition of the higgs operators seems highly unnatural , if one understands a brane higgs field as a limit of a bulk higgs field . all operators involving higgs fields will have to have a precise and definite dependence on @xmath0 ( a large number ) , which seems quite contrived , specially in a framewrok in which no big numerical hierarchies should arise from fundamental 5d coefficients . in any case , with the ansatz outlined in the text and reviewed in appendix [ apen : ii ] , one can still work consistently with a brane higgs field as a limit case of a bulk higgs field . would like to thank kaustubh agashe for many discussions and comments and specially alex azatov and lijun zhu for their invaluable help and collaboration in the early stages of this work . this work is supported in part by nserc under grant number sap105354 . from equation ( [ ym ] ) the shift defined as @xmath128 , can be also derived from = v_4y^u_qu _ i , jy^u_qu_jy^u*_u_jq_iy^u_q_ium_q_im_u_j . therefore simply replacing the second yukawa coupling with the yukawa coupling of the higher derivative operator , @xmath129 will give [ yyy ] = v_4y^u_qu _ i , jy^u_qu_jy^ru*_u_jq_iy^u_q_ium_q_im_u_j . here , we present explicit analytic expressions for the @xmath54 production and also the yukawa coupling - mass shifts by performing the infinite sums over the kk modes . we also include the result given in reference @xcite for the shift due to the usual yukawa term , @xmath130 , for completeness . ) , and our results match the one given in the text of @xcite . note however that there a few typos in the eq . a1 of their appendix . ] to summarize , we have @xmath131+{y_t^{rs}\over m_t}a_{1/2}(\tau_{t})+{\delta_2^t\over m_t{v_4}}\ , , \end{aligned}\ ] ] for the @xmath54 production and @xmath132 \ , , \end{aligned}\ ] ] for the shifted yukawa coupling . also , there is a misalignment due to the kinetic term @xcite , which as discussed in the text , is only important for the case of the third generation quarks . we do not repeat that result here . for the higher derivative term the shift is : @xmath133.\nonumber\end{aligned}\ ] ] we summarize the matching prescription for operators containing higgs field for the case where the higgs boson is localized on the brane . as explained in section [ sec : setup ] , these prescriptions insures that the 5d bulk higgs scenario transitions smoothly to a brane - localized higgs case . the brane prescription for the higgs associates a delta function to the higgs normalization integral @xmath134 ^ 2=1\,.\ ] ] as the @xmath135 , rather than @xmath10 field , is associated with a @xmath136 function , one must include a @xmath0 dependence to the bulk higgs fields to be able to match operators , in the limit @xmath137 to the brane ones . the conversion is : @xmath138 for matching brane to bulk in the appropriate limit . for the case of brane higgs , the contributions for a finite number of modes for @xmath139 give exactly @xmath140 ( because of boundary values on the brane ) . this confirms the work of @xcite . however , we must add higher order operators @xmath102 , which give a significant result ( converging to a constant for @xmath141 and anything beyond ) . the result obtained by summing over a finite number of modes in the brane on the @xmath102 contribution must be compared with the result in the paper by @xcite for the infinite sum of @xmath139 on the brane . for bulk higgs , the shift contribution from a finite number of modes on the @xmath139 contribution is no longer @xmath140 . however , adding to this the @xmath102 contribution , we notice that the @xmath102 contribution for bulk higgs is much smaller ( two orders of magnitude ) than the corresponding one in the brane . this is a clear indication that higher order corrections are much more important for the brane higgs case than for the bulk . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) ; l. randall and r. sundrum , phys . rev . lett . * 83 * , 4690 ( 1999 ) . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . b * 473 * , 43 ( 2000 ) ; a. pomarol , phys . b * 486 * , 153 ( 2000 ) ; s. chang , j. hisano , h. nakano , n. okada and m. yamaguchi , phys . d * 62 * , 084025 ( 2000 ) . k. agashe _ et al_. , phys . d * 71 * ( 2005 ) 016002 ; d * 75 * ( 2007 ) 015002 ; d * 74 * ( 2006 ) 053011 ; phys . * 93 * ( 2004 ) 201804 ; s. j. huber and q. shafi , phys . b * 512 * ( 2001 ) 365 ; b * 544 * ( 2002 ) 295 ; b * 583 * ( 2004 ) 293 ; t. appelquist _ et al_. , phys . d * 65 * ( 2002 ) 105019 ; t. gherghetta , phys . * 92 * ( 2004 ) 161601 ; g. moreau _ et al_. , jhep * 0601 * ( 2006 ) 048 ; 0603 ( 2006 ) 090 ; eur . j. c * 40 * ( 2005 ) 539 ; s. chang _ et al_. , phys . d * 73 * ( 2006 ) 033002 . c. csaki , a. falkowski and a. weiler , jhep * 0809 * , 008 ( 2008 ) ; [ arxiv:0804.1954 [ hep - ph ] ] . a. l. fitzpatrick , g. perez and l. randall , arxiv:0710.1869 [ hep - ph ] . m. blanke , a. j. buras , b. duling , s. gori and a. weiler , jhep * 0903 * , 001 ( 2009 ) [ arxiv:0809.1073 [ hep - ph ] ] ; m. blanke , a. j. buras , b. duling , k. gemmler and s. gori , jhep * 0903 * , 108 ( 2009 ) [ arxiv:0812.3803 [ hep - ph ] ] ; m. e. albrecht , m. blanke , a. j. buras , b. duling and k. gemmler , jhep * 0909 * , 064 ( 2009 ) [ arxiv:0903.2415 [ hep - ph ] ] ; m. bauer , s. casagrande , l. grunder , u. haisch and m. neubert , phys . d * 79 * , 076001 ( 2009 ) [ arxiv:0811.3678 [ hep - ph ] ] ; m. bauer , s. casagrande , u. haisch and m. neubert , arxiv:0912.1625v1 [ hep - ph ] . g. cacciapaglia , c. csaki , j. galloway , g. marandella , j. terning and a. weiler , jhep * 0804 * , 006 ( 2008 ) ; a. l. fitzpatrick , l. randall and g. perez , phys . rev . lett . * 100 * , 171604 ( 2008 ) ; j. santiago , jhep * 0812 * , 046 ( 2008 ) ; c. csaki , g. perez , z. surujon and a. weiler , phys . d * 81 * , 075025 ( 2010 ) ; m. c. chen , k. t. mahanthappa and f. yu , phys . d * 81 * , 036004 ( 2010 ) ; m. c. chen and h. b. yu , phys . b * 672 * , 253 ( 2009 ) ; g. perez and l. randall , jhep * 0901 * , 077 ( 2009 ) ; c. csaki , c. delaunay , c. grojean and y. grossman , jhep * 0810 * , 055 ( 2008 ) [ arxiv:0806.0356 [ hep - ph ] ] ; f. del aguila , a. carmona and j. santiago , arxiv:1001.5151 [ hep - ph ] . c. csaki , a. falkowski and a. weiler , phys . d * 80 * , 016001 ( 2009 ) [ arxiv:0806.3757 [ hep - ph ] ] . cabrer , g. von gersdorff , and m. quiros , new j. phys . * 12 * ( 2010 ) 075012 ; j.a . cabrer , g. von gersdorff , and m. quiros , phys . d * 84 * ( 2011 ) 035024 ; j.a . cabrer , g. von gersdorff , and m. quiros , jhep * 1105 * ( 2011 ) 083 ; a. carmona , e. ponton and j. santiago , jhep * 1110 * ( 2011 ) 137 ; p. r. archer , j.s . huber and s.jager , jhep bf 1112(2011 ) , 101 . r. contino , y. nomura and a. pomarol , nucl . b * 671 * , 148 ( 2003 ) , k. agashe , r. contino and a. pomarol , nucl . phys . b * 719 * , 165 ( 2005 ) . g. f. giudice , c. grojean , a. pomarol and r. rattazzi , jhep * 0706 * , 045 ( 2007 ) [ arxiv : hep - ph/0703164 ] . b. lillie , jhep * 0602 * , 019 ( 2006 ) [ arxiv : hep - ph/0505074 ] . a. falkowski , phys . d * 77 * , 055018 ( 2008 ) [ arxiv:0711.0828 [ hep - ph ] ] . a. djouadi and g. moreau , phys . b * 660 * , 67 ( 2008 ) [ arxiv:0707.3800 [ hep - ph ] ] . s. casagrande , f. goertz , u. haisch , m. neubert , and t. pfoh , jhep * 0810 * ( 2008 ) 094 ; jhep * 1009 * ( 2010 ) 017 , m. carena , s. casagrande , f. goertz , u. haisch and m. neubert , jhep * 1208 * , 156 ( 2012 ) [ arxiv:1204.0008 [ hep - ph ] ] .

recent studies have established that , in addition to the well - known kicked harper model ( khm ) , an on - resonance double kicked rotor model ( ordkr ) also has hofstadter s butterfly floquet spectrum , with strong resemblance to the standard hofstadter s spectrum that is a paradigm in studies of the integer quantum hall effect . earlier it was shown that the quasi - energy spectra of these two dynamical models ( i ) can exactly overlap with each other if an effective planck constant takes irrational multiples of and ( ii ) will be different if the same parameter takes rational multiples of . this work makes some detailed comparisons between these two models , with an effective planck constant given by , where and are coprime and odd integers . it is found that the ordkr spectrum ( with two periodic kicking sequences having the same kick strength ) has one flat band and non - flat bands whose largest width decays in power law as , where is a kicking strength parameter . the existence of a flat band is strictly proven and the power law scaling , numerically checked for a number of cases , is also analytically proven for a three - band case . by contrast , the khm does not have any flat band and its band width scales linearly with . this is shown to result in dramatic differences in dynamical behavior , such as transient ( but extremely long ) dynamical localization in ordkr , which is absent in khm . finally , we show that despite these differences , there exist simple extensions of khm and ordkr ( upon introducing an additional periodic phase parameter ) such that the resulting extended khm and ordkr are actually topologically equivalent , i.e. , they yield exactly the same floquet - band chern numbers and display topological phase transitions at the same kick strengths . a theoretical derivation of this topological equivalence is provided . these results are also of interest to our current understanding of quantum - classical correspondence considering that khm and ordkr have exactly the same classical limit after a simple canonical transformation .

1306.6128

as one important paradigm in the studies of quantum chaos and quantum - classical correspondence , the kicked rotor ( kr ) model @xcite has received tremendous theoretical and experimental interest in the last three decades @xcite . for some experimental activities on kr within the last three years , we would like to mention those listed in ref . a one - dimensional kr is described by the hamiltonian @xmath7 in terms of dimensionless variables , where @xmath8 and @xmath9 are conjugate ( angular ) momentum and angle variables , @xmath6 and @xmath10 are the kick strength and the period of the @xmath11-kicks . the dynamical evolution of the system for a period from time @xmath12 to @xmath13 can be expressed as a quantum map , which is given by the following unitary floquet operator @xmath14 for our considerations below , we confine ourselves to a rotor hilbert space defined by the periodic boundary condition in @xmath9 , with @xmath15 . the hilbert space can then be represented by the eigenfunctions @xmath16 of @xmath8 , with @xmath17 , @xmath18 , @xmath19 being an integer , and @xmath20 being a dimensionless effective planck constant . through extensive numerical simulations and mathematical analysis , it is now well known that in general the kr dynamics can be classified into two categories @xcite . for an irrational ( hence generic ) value of @xmath21 the system can diffuse in ( angular ) momentum space only for a short time due to `` dynamical localization '' , regardless of the kick strength . this hints at a discrete spectrum of @xmath22 and is closely related to anderson localization @xcite . on the other hand , for @xmath21 being a rational multiple of @xmath0 ( except for odd multiples of @xmath0 ) , @xmath22 has continuous bands : a time - evolving state would keep spreading out in ( angular ) momentum space ballistically . this category of dynamics was termed as `` quantum resonance '' @xcite . another important quantum chaos model is the kicked harper model ( khm ) @xcite , originally introduced in ref . @xcite as an approximation of the problem of kicked charges in a magnetic field . remarkably , the khm and even a whole class of its generalized versions were shown to be equivalent to the problem of a charge kicked periodically in the presence of a magnetic field @xcite . the associated khm quantum map for each period is given by @xmath23 with @xmath24 being an additional system parameter . throughout we assume the khm is also treated in the same hilbert space as the kr and is quantized on a rotor hilbert space . the dynamics of khm differs from that of kr as described above in several aspects . for example , for all irrational values of @xmath25 , the system in general tends to delocalize ( localize ) in ( angular ) momentum space for @xmath26 ( @xmath27 ) @xcite . of particular interest is the symmetric case of @xmath28 , for which the quasi - energy spectrum of @xmath29 is fractal - like in general . scanning the spectrum collectively for fixed @xmath30 versus a varying @xmath20 forms a pattern that resembles the hofstadter s butterfly spectrum @xcite , a paradigm in studies of the integer quantum hall effect . the associated dynamics is extended in general and may be connected with the fractal dimensions of the floquet spectrum . given the above - mentioned differences between kr and khm , the work of ref . @xcite by two of the authors emerged somewhat unexpectedly . there it was shown that a variant of kr also has hofstadter s butterfly spectrum . in particular , motivated by the double - kicked rotor model studied both experimentally and theoretically in ref . @xcite , which is a special case of multiple kr s " first introduced in ref . @xcite , ref . @xcite studied a double - kicked model under a quantum - resonance condition . for a total period of @xmath31 ( @xmath32 ) , a double kicked rotor model is associated with two periodic @xmath11-kicks of strengths @xmath6 and @xmath24 , separated by a time interval set to be unity , yielding the following floquet operator @xmath33 in ref . @xcite , @xmath31 is chosen to satisfy the quantum resonance condition @xmath34 . then @xmath35 due to the discreteness of the momentum eigenvalues . this leads us to an on - resonance double kicked rotor model ( ordkr ) , whose floquet operator is given by @xcite : @xmath36 note that we have deliberately used symbols @xmath6 and @xmath24 in both @xmath29 and @xmath37 because in this paper , parameter @xmath6 or parameter @xmath24 from both models will always be assigned the same value . experimental realization of such an ordkr propagator in atom optics is possible by loading a bose - einstein - condensate ( bec ) in a kicking optical lattice , with the initial quasi - momentum spread of the bec negligibly small as compared with the recoil momentum of the optical lattice @xcite . interestingly , for @xmath20 being an irrational multiple of @xmath0 , the ordkr and the khm share the same quasi - energy spectrum @xcite . our main plan for this paper is to make some detailed comparisons between khm and ordkr as two closely related dynamical models , both possessing hofstadter s butterfly spectrum . our motivations are as follows . first of all , in refs . @xcite , it was shown that @xmath37 and @xmath29 have different spectra if @xmath20 is a rational multiple of @xmath0 . on the other hand , as @xmath25 approaches an arbitrary irrational number , the spectral difference between @xmath37 and @xmath29 , which is characterized by a hausdorff metric in ref . @xcite , was shown to approach zero . it is therefore highly worthwhile looking into the actual spectral differences for rational values of @xmath25 , because , up to a classical canonical transformation , ordkr and khm have exactly the same classical limit @xcite ( obtained by letting @xmath20 approach zero while fixing @xmath38 and @xmath39 ) . indeed , given their equivalence in the classical limit , the spectral differences we analyze constitute beautiful examples to illustrate how quantization of classically equivalent systems may lead to remarkable system - specific consequences . second , by working on the details we hope to find some clues as to why the dynamics of ordkr can be so different from that of khm . we indeed succeed in doing this , finding that even on a qualitative level , the floquet bands of ordkr behave much differently from that of khm , for @xmath40 , with @xmath2 and @xmath3 being coprime and both odd . in particular , we shall prove the existence of a flat floquet band @xcite for ordkr with @xmath28 , which may be of interest to current studies of strongly correlated condensed - matter systems with an almost flat energy band @xcite . the existence of a flat floquet band has been shown elsewhere to be important in explaining the intriguing exponential quantum spreading dynamics in ordkr @xcite . third , motivated by recent interests in topological characterization of periodically driven systems @xcite and given the interesting relationship of the two models described previously , we ask whether , after all , ordkr and khm have any interesting topological connections . based on our numerical and analytical studies , the answer is yes and we shall claim that ordkr and khm are topologically equivalent in the sense that their extended floquet bands ( obtained upon introducing a phase shift parameter defined in sec . iii ) always have the same band chern numbers . this paper is organized in the following order . in sec . ii we present detailed results regarding a spectral comparison between khm and ordkr , for @xmath28 , and @xmath40 with @xmath2 and @xmath3 being coprime and odd integers . numerical findings will be described first , followed by analytical considerations when possible ( e.g. , band width scaling for a three - band case and the general proof of a flat band for ordkr ) . the implications of peculiar spectral properties of ordkr for its dynamics are also discussed via some numerical studies . in sec . iii we study the khm and ordkr by extending them to accommodate a new periodic parameter and demonstrating the topological equivalence of the resulting extended models . section iv concludes this paper . as far as numerics are concerned , the spectrum of the unitary operators can be obtained in a straightforward manner . for completeness we describe some details here . the key step is to take advantage of the periodic property of @xmath29 or @xmath37 in the ( angular ) momentum space , which arises naturally for @xmath20 being a rational multiple of @xmath0 . denote @xmath41 to refer to either @xmath29 or @xmath37 . letting @xmath42 , one easily finds @xmath43 for @xmath40 . this indicates a unit cell in ( angular ) momentum space , with a size of @xmath3 . the spectrum is then equivalent to that of a reduced @xmath44 matrix @xmath45 , whose elements are given by @xmath46_{j , k}=\sum\limits_{l } e^{il\varphi}u_{j , k+ln}$ ] , with @xmath47 being the bloch phase in momentum space and @xmath48 running over all integers . as off - diagonal elements of @xmath49 decay exponentially , the summation in @xmath50 can be truncated safely at certain large enough value of @xmath51 ( in our analytical studies below , we do not do such truncations ) . numerical results are then checked by further increasing the truncation radius . once @xmath52 is numerically obtained , the standard diagonalization algorithm for a unitary matrix can be exploited to obtain @xmath3 values of quasi - energy @xmath53 . by varying @xmath54 in @xmath55 we have @xmath3 floquet bands . -1.0 cm -0.5 cm in fig . [ squid ] we show our obtained quasi - energy values of @xmath37 and @xmath29 as a function of the kick strength @xmath6 . though for each fixed value of @xmath6 , we only show the quasi - energy values for a limited number of bloch phase choices , the locations of the bands , the band width , and a few avoided band crossings can already be seen clearly for not too large values of @xmath28 . in particular , at @xmath56 , nice floquet bands can be identified clearly for both ordkr and khm , though for very large values of @xmath6 the merging of the bands does occur . spectral differences between @xmath37 and @xmath29 in the shown example are also obvious . based on the results shown in fig . [ squid ] , we have carried out extensive numerical investigations for other cases with @xmath40 , with @xmath2 and @xmath3 coprime and both odd . some of the main features are presented and commented on below . first , the band structure of @xmath37 is symmetric with respect to the zero quasi - energy axis , which is however not the case for @xmath29 . this interesting symmetry is absent in both @xmath29 and @xmath22 . we shall prove this property below . second , consistent with the above - mentioned symmetry , @xmath37 is seen to have a flat band with @xmath57 . by flat band , we mean that this quasi - energy value is independent of the bloch phase @xmath54 . so the overall picture is that the @xmath3 bands can be classified into @xmath58 pairs , with each pair having opposite quasi - energy values , plus a flat band in the middle . again , this is not the case for @xmath29 . the existence of a flat floquet band was previously observed in studies of the quantum antiresonance phenomenon in kicked systems @xcite . however , unless in the case of @xmath59 ( @xmath2 odd ) that also corresponds to a quantum antiresonance condition , here the flat band of @xmath37 coexists with other nonflat bands . this coexistence of a flat band with nonflat bands constitutes an interesting feature . as a side note , ref . @xcite suggested that for a kr defined in this paper under the quantum resonance condition of any order ( i.e. , @xmath60 , with @xmath2 and @xmath3 arbitrary coprime integers ) , none of the floquet bands of @xmath22 is flat . so the existence of one single flat band of @xmath37 is also remarkable as compared with @xmath22 . -0.90 cm -0.5 cm third , the largest bandwidth of the other @xmath4 non - flat floquet bands of ordkr scales with @xmath6 as @xmath5 , in the limit of @xmath61 . in sharp contrast , the bandwidths of khm scale with @xmath6 linearly . representative numerical results are shown in fig . [ bw ] , where the bandwidth of the widest band is plotted against small values of @xmath6 , for @xmath40 , with @xmath62 , @xmath63 . the power law decay of the ordkr bandwidth in the form of @xmath5 can be clearly identified , whereas the bandwidth of khm remains a linear function of @xmath6 , irrespective of the value of @xmath3 . this being the case , in the small @xmath6 regime ( @xmath64 ) , the maximum bandwidths of ordkr is @xmath65 times narrower than that of khm . flat bands in solid - state systems are of vast interest in condensed matter physics because they offer new opportunities for understanding strongly correlated systems without landau levels . for this reason the existence of a flat band in a periodically driven system can be useful , too . to further understand the flat band of ordkr , we present a theoretical proof in this subsection . in doing so we shall also prove the band symmetry noted above . we shall also discuss how an eigenstate on a flat band , which is infinitely degenerate , may be numerically found . for @xmath40 with @xmath2 and @xmath3 being coprime integers , the spectrum becomes that of a reduced @xmath44 floquet matrix with elements @xmath66_{n , m}=\sum_{l=-\infty}^{\infty } { \left\langle n\right|}{\hat u}_{\text{ordkr}}{\left| m+l\times n\right\rangle } e^{il\varphi}$ ] . after performing some necessary integrals and using the fact that both @xmath2 and @xmath3 are odd , one can express @xmath67_{n , m}$ ] as a summation of finite terms ( see appendix for details ) . in the following discussions regarding the existence of a flat band and the band inversion symmetry , we shall restrict ourselves to the cases of @xmath28 ( note however , in the next section , the notation introduced here will be extended to the cases with @xmath68 ) . we first introduce diagonal unitary matrices @xmath69 , @xmath70 , @xmath71 and unitary matrix @xmath72 , with matrix elements @xmath73 , @xmath74 , @xmath75 and @xmath76 , where indices @xmath19 and @xmath77 take values @xmath78 . note that in obtaining our expression for @xmath79 , we made use of the fact that @xmath80 . we then have the following compact form for the reduced floquet matrix @xmath81 to prove that there is a flat band for ordkr , we show that @xmath82 has an eigenvalue equal to one , regardless of the value of @xmath54 . consider then a matrix @xmath83 transformed from @xmath82 by a unitary operation @xmath84 , which takes the form @xmath85 the eigenvalue equation of @xmath83 may be rewritten as @xmath86 where @xmath87 , @xmath88 denotes an eigenvector , and @xmath89 is an eigenvalue of @xmath90 . detailed calculations show that @xmath91 is a symmetric matrix ( see appendix for details ) and since @xmath92 is a diagonal matrix , @xmath93 must be an antisymmetric matrix of odd dimension . it immediately follows @xmath94 . thus , regardless of the bloch phase @xmath54 , @xmath95 is a permissible solution to eq . ( [ neweigeneq ] ) . we have thus shown that @xmath90 always has a unity eigenvalue or zero quasi - energy for @xmath40 . this is nothing but the existence of a flat floquet band . our considerations above also lead us to a proof of the band inversion symmetry of ordkr for odd @xmath2 and @xmath3 . specifically , because @xmath96 , we see that if @xmath97 , then @xmath98 as well . that is , both @xmath89 and @xmath99 are solutions to the eigenvalue equation of eq . ( [ neweigeneq ] ) . as such , if we have a quasi - energy @xmath100 , we must have @xmath101 in the spectrum . this completes our proof of the inversion symmetry of the ordkr . a flat band is infinitely degenerate as states on the band can still have a continuous bloch phase @xmath54 . due to such an independence upon the bloch phase , the band dispersion relation directly yields a zero group velocity in the ( angular ) momentum space , thus indicating a zero mobility in the ( angular ) momentum space . further , the infinite degeneracy allows us to construct a flat - band eigenstate that is localized in the ( angular ) momentum space ( though the floquet operator itself is periodic in momentum with a period @xmath102 ) . it is interesting to outline a simple approach to the construction of flat - band states . it is found that , highly localized flat - band states can be obtained by directly truncating the full floquet matrix @xmath103 to a small size , such that there is only one eigenstate whose eigenvalue is real and still equals to unity ( thus not affected by the truncation ) . other localized states on the flat band can be obtained by shifting it by a multiple of n sites , or by superimposing these states localized at different locations . figure [ eigenstate ] depicts one computational example of a flat - band eigenstate strongly localized in the ( angular ) momentum space . we have checked that if we use a flat - band state we constructed as the initial state for time evolution , then indeed this state does not evolve with iterations of our ordkr quantum map . this situation is more subtle than the quantum antiresonance phenomenon @xcite : for ordkr with multiple bands , only special states prepared on the single flat band can remain localized , whereas in the case of quantum antiresonance an arbitrary state should remain localized . -0.5 cm for @xmath40 with @xmath2 and @xmath3 being coprime integers , the reduced @xmath44 floquet matrices @xmath82 and @xmath104 ( see our general expressions in the appendix ) can be obtained analytically . to further understand and confirm the bandwidth scaling of ordkr and khm , we have also carried out analytical studies for a three - band case , with @xmath28 and @xmath105 . for ordkr , the three eigenvalues are found to be @xmath106 and @xmath107 , where @xmath108 $ ] . one finally finds @xmath109\}$ ] where @xmath110 . for @xmath111 it can be shown that the edges of the band correspond to @xmath112 and @xmath113 . the bandwidth can thus be determined to be @xmath114/3\right\ } -\arccos\left[\frac{1}{3}+\frac{2}{3}\cos\left(\frac{3k}{2\hbar}\right)\right]$ ] . taylor expanding this expression for the bandwidth , we find the first nonzero term to be @xmath115 , a clear power - law scaling of @xmath116 . for khm , the eigenvalues can be deduced from the equation @xmath117 . the resulting explicit expression of the eigenvalue equation is @xmath118 where @xmath119 . note that all eigenvalues are in the form of @xmath120 , since the reduced floquet matrix is always unitary . the three eigenvalues are found to be @xmath121 , @xmath122 and @xmath123 where @xmath124 . for @xmath111 , the edges of the band correspond to @xmath125 and @xmath126 . the band width can thus be determined to be @xmath127 , @xmath128 and @xmath129 . taylor expanding the expressions of eigevalues for @xmath130 and keeping the lowest order in @xmath6 , we have @xmath131 , @xmath132 and @xmath133 , a clear linear scaling of k. -0.0 cm the very fast decay of the floquet bandwidth of ordkr suggests that in a considerable range of @xmath6 the bandwidths will be very narrow . in other words , for a small @xmath6 , all the floquet bandwidths would be effectively zero for a reasonably long time scale . therefore , when it comes to the dynamical evolution of the system , effectively the system will not feel its continuous floquet spectrum and hence displays localization behavior , for a time scale inversely proportional to the bandwidths . we call this the time scale of transient dynamical localization and denote it by @xmath134 . we then have @xmath135 . the overall expectation is the following : within @xmath134 , ordkr displays localization in the ( angular ) momentum space , but afterwards it begins to show ballistic behavior in the ( angular ) momentum space . because of the power law scaling , the intriguing time scale @xmath134 can be very sensitive to a change in the kick strength @xmath6 . our numerical calculations indeed confirm this . figure [ p2t](a ) shows an example of the dynamics of the kinetic energy of ordkr , starting from an initial state with zero momentum . in all three of the shown cases , the kinetic energy is seen to freeze over a time scale before it starts to increase ballistically . the time scale of the freezing stage is shown to increase rapidly as we decrease the value of @xmath6 . as a comparison , figure [ p2t](b ) shows the parallel dynamics of khm , for the same three values of @xmath6 . there it is seen that the transient stage of localization is only weakly dependent upon @xmath6 , which is again consistent with the linear @xmath6-dependence of the bandwidth of khm . quantitatively , the transient localization time scale @xmath134 is numerically determined from the duration of kinetic energy freezing . the @xmath134 thus obtained numerically and shown in fig . [ p2t](c ) indeed satisfies the scaling @xmath135 for ordkr , which is in sharp contrast to the @xmath136 scaling for khm . the results here can also be understood as a quantitative explanation of our earlier finding of transient dynamical localization in ref . @xcite . for future experiments , the observation of the aforementioned scaling of @xmath134 versus @xmath6 may serve as the first piece of evidence of a successful realization of an ordkr . in this section , we devote ourselves to a detailed comparison of the floquet band topologies of ordkr and khm . we first describe our motivation and introduce new notation . next , we report numerical findings of the floquet band topological numbers of both models . finally , an exact analytical proof of the topological equivalence between ordkr and khm is presented . one early study @xcite suggested that topological properties of the floquet bands of khm may be connected with the regular - to - chaos transition in the classical limit . because ordkr and khm share the same classical limit ( up to a canonical transformation ) , we suspect that there should be some similarity in their floquet band topologies . our second motivation for a topological study is related to an earlier finding that , when @xmath25 is a _ rational _ number , the spectral union of @xmath137 ( variant of ordkr defined below ) over all @xmath138 is the same as that of @xmath139 ( variant khm defined below ) over all @xmath138 @xcite . this previous mathematical result further suggests a possible topological connection between the two models . interestingly , as we explore this possible topological connection , we are able to see a connection between khm propagator and ordkr propagator for each individual value of @xmath138 along with an individual value of the bloch phase , thus going beyond ref . @xcite that considered a unification of all values of @xmath138 and the bloch phase . further , as we shall see below , the connection is established by a mapping in the parameter space , which can not be achieved by a unitary transformation between the two propagators . next , we introduce necessary notation for our discussion of band topology . to characterize the band topology for both ordkr and khm , we introduce an additional periodic phase parameter @xmath140 to the ordkr and khm maps , namely , @xmath141 for @xmath40 , both operators are periodic in ( angular ) momentum space with period @xmath102 . hence , their eigenvalues are @xmath0-periodic in the bloch phase @xmath54 and also in @xmath138 , giving rise to @xmath3 _ extended _ floquet bands which disperse as a function of @xmath54 and @xmath138 . these 2-dimensional bands may be topologically characterized by chern numbers , denoted @xmath142 for the @xmath77th band . in what follows , we denote @xmath143 as an ( generalized ) eigenstate of either @xmath137 or @xmath139 , in the @xmath77th band , with an eigenvalue @xmath144 $ ] . such a generalized eigenstate lives on the entire ( angular ) momentum space . we then denote @xmath145 as the reduced @xmath44 floquet matrix constructed from either @xmath137 or @xmath139 using the method described at the beginning of section ii . we next define the state @xmath146 , which is @xmath143 projected onto @xmath3 sites of one unit cell in the ( angular ) momentum space , i.e. , @xmath147 . we further assume that @xmath146 is normalized over one unit cell consisting of @xmath3 sites . using the above notation , the berry curvature of the @xmath77th band is then defined as @xcite @xmath148 where we have suppressed the explicit dependences on @xmath54 and @xmath138 for brevity . from the berry curvature we obtain the chern number @xmath142 , @xmath149 we have conducted extensive numerical evaluations of the floquet band chern numbers associated with both @xmath137 and @xmath139 . we find that for the same @xmath6 and @xmath24 respectively in both models , the chern numbers are always equal . for example , for @xmath105 and @xmath28 , fig . [ cherntable_keql ] represents the floquet band chern numbers for _ both _ models versus a varying @xmath6 . the chern numbers obtained for @xmath137 are identical with those for @xmath139 . here , we adopt the convention that the band with largest absolute value of chern number is always represented by the line in the middle . vertical lines represent collisions between quasi - energy bands , during which chern number transitions can take place . note that in some cases band 1 and band 3 can collide directly with each other through the boundary of the quasienergy brillouin zone . it is also important to stress that the chern numbers of ordkr match those of khm for all @xmath6 values , despite their jumps at various topological phase transition points . we are thus clearly witnessing , albeit numerically , a remarkable topological equivalence between ordkr and khm ! @xmath150{cherntable_dkrm_keql_3bands.pdf}\\\ \end{array}$ ] some insight into this observed topological equivalence may be obtained by comparing the quasienergy dispersions of the two models . in fig . [ keqlplots ] , we present the floquet band structure for both ordkr and khm , in the case of @xmath151 . interestingly , the ordkr band profile appears to be the same as that of khm , up to some translation along the @xmath54 and @xmath138 axes , followed by a rotation of the spectrum about the quasi - energy axis . this observation is consistent with our proof of topological equivalence in the next section . @xmath152{dkrm_h2pio3_k1k2eq3h_band1.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_kleq3h_band1.png}\\ \includegraphics[width=80 mm , height=!]{dkrm_h2pio3_k1k2eq3h_band2.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_kleq3h_band2.png}\\ \includegraphics[width=80 mm , height=!]{dkrm_h2pio3_k1k2eq3h_band3.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_kleq3h_band3.png}\\ \end{array}$ ] we have numerically observed that the topological equivalence also occurs for @xmath68 . as one example of this , fig . [ cherntable_keqml ] depicts a zoo of chern numbers for ordkr and khm , with @xmath105 , @xmath153 fixed but @xmath6 varying . we again see the same equivalence of chern numbers across a few topological phase transition points . in addition , we found computationally that the chern numbers are invariant upon an exchange between @xmath24 and @xmath6 . this was found to hold true also in other cases with more bands . @xmath154{cherntable_dkrm_k_gt_l_3bands.pdf}\\ \end{array}$ ] we have also plotted the floquet band structure for a @xmath155 case in fig . [ k_gt_lplots ] . here we consider the case of @xmath156 , @xmath157 . it is seen that the band profiles of ordkr and khm are once again similar and appear to be related by a rotation and translation . @xmath152{dkrm_h2pio3_k2eq3h_k1eqh_band1.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_keq3h_leqh_band1.png}\\ \includegraphics[width=80 mm , height=!]{dkrm_h2pio3_k2eq3h_k1eqh_band2.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_keq3h_leqh_band2.png}\\ \includegraphics[width=80 mm , height=!]{dkrm_h2pio3_k2eq3h_k1eqh_band3.png } & \includegraphics[width=80 mm , height=!]{khm_h2pio3_keq3h_leqh_band3.png}\\ \end{array}$ ] to strictly confirm our claim of topological equivalence , we present an analytical proof in this subsection . the proof proceeds as follows . we first show that the reduced ordkr floquet matrix and the reduced khm floquet matrix are equivalent up to a series of unitary transformations and a mapping between their parameter values . we then show that these matrices obtained under the unitary transformations and mapping of parameters still correspond to the same chern numbers as the original reduced matrices . these steps constitute a proof of topological equivalence . we consider cases with @xmath40 , with @xmath2 and @xmath3 co - prime and both odd . in these cases , the reduced floquet matrices of @xmath137 and @xmath139 ( see the appendix for details ) can be written compactly as a product of @xmath158 unitary matrices @xmath159 where @xmath160 , @xmath161 , @xmath162 , @xmath163 are diagonal unitary matrices , with matrix elements @xmath164 , @xmath165 , @xmath166 , @xmath167 , where the index @xmath77 takes values @xmath78 . @xmath79 and @xmath168 are defined as they were in section ii . we begin the proof by applying a unitary transformation given by @xmath169 to the @xmath170 matrix to obtain @xmath171 . writing @xmath172 as the exponential of a matrix , we obtain @xmath173d_{2l}\\ & = \exp\left[-i\frac{k}{2\hbar}\left(e^{-i\frac{\varphi}{n}}c+ e^{i\frac{\varphi}{n}}c^\dagger\right)\right ] \begin{pmatrix } \ddots & & \\ & e^{-i\frac{l}{\hbar}\cos(2\pi\frac{m}{n}n-\alpha)}&\\ & & \ddots\\ \end{pmatrix } , \end{split } \label{unitarykhm}\ ] ] where @xmath174 in the following steps , we will apply a series of unitary transformations to the reduced matrix @xmath175 and show that the result is equivalent to the above unitarily transformed version of @xmath176 provided a condition between @xmath54 and @xmath138 in the two models is obeyed . applying a transformation given by @xmath84 to @xmath177 , we obtain @xmath178 , which we simplify as follows . @xmath179d_1f^\dagger d_{1l}\\ & = \exp\left[i\frac{k}{2\hbar}\left(e^{-i\frac{\varphi}{n}}fd_1^\dagger cd_1f^\dagger + e^{i\frac{\varphi}{n}}fd_1^\dagger c^\dagger d_1f^\dagger\right)\right]d_{1l}. \end{split}\ ] ] denoting @xmath180 , @xmath181d_{1l}$ ] . the explicit expression for @xmath182 is @xmath183 next , we introduce the @xmath184 permutation matrix @xmath185 which is made up entirely of zeroes except that in the @xmath186-th row , the @xmath187-th column equals 1 , with @xmath188 . here , @xmath186 and @xmath187 take values @xmath189 . note that @xmath185 is unitary and that the set of @xmath187 values will include all of the @xmath3 values @xmath190 . we apply the unitary transformation @xmath185 to @xmath191 and obtain @xmath192d'_{1l}$ ] , where @xmath193 . @xmath194 is a diagonal unitary matrix with diagonal elements @xmath195 . the effect of the permutation matrix on @xmath182 is as follows . @xmath196 we can see that the structure of the above matrix is very similar to @xmath197 and would be made identical with it if we were to replace all the nonzero elements with 1 . this is achieved by a transformation via the diagonal unitary matrix @xmath198 which has diagonal elements @xmath199}$ ] . it can be shown that @xmath200 . denoting @xmath201 and using that @xmath202 and @xmath203 commute due to their both being diagonal , we obtain @xmath204d'_{1l}\\ & = \exp\left[-i\frac{k}{2\hbar}\left(e^{-i\frac{\varphi+n\pi}{n}}c + e^{i\frac{\varphi+n\pi}{n}}c^\dagger \right)\right ] \begin{pmatrix } \ddots & & \\ & e^{-i\frac{l}{\hbar}\cos(2\pi\frac{m}{n}j+\frac{\varphi}{n}-\alpha)}&\\ & & \ddots\\ \end{pmatrix}. \end{split } \label{unitaryordkr}\ ] ] from eq . ( [ unitarykhm ] ) and ( [ unitaryordkr ] ) , we observe that @xmath205 and @xmath206 are identical , provided that @xmath207 and @xmath208 . summarizing what we have found so far , we have learned that if we unitarily transform from @xmath209 to @xmath210 , where @xmath211 , and unitarily transform from @xmath212 to @xmath213 , where @xmath214 , we find that the two unitarily transformed matrices are identical up to some mapping between @xmath215 and @xmath216 . figure [ spectrun_compare ] represents one example of the quasi - energy band plot for both ordkr and khm . referring to panel ( b ) and panel ( c ) , we thus directly see that provided that @xmath207 and @xmath208 , the extended floquet band structure for ordkr and khm are the same ( though the boundaries on the @xmath217 plane are different ) . @xmath218{khm_ordkr_middle_band_k_3.pdf}\\ \end{array}$ ] recapping our proof so far , with the mapping @xmath219 and @xmath220 , we have the following @xmath221 where @xmath222 , and the definitions of the matrices @xmath223 , @xmath72 , @xmath162 , @xmath198 and @xmath185 all are previously given . for example , @xmath224 , and @xmath225 . let @xmath226 be the @xmath77th eigenstate of @xmath209 and @xmath227 be the @xmath77th eigenstate of @xmath212 . equation ( [ eqn : equvilance ] ) then leads to @xmath228 because scanning all the values of @xmath216 will scan all the values of @xmath215 , it is obvious now that the union of the spectrum of @xmath209 ( after considering all values of @xmath229 and @xmath230 ) should be the same as the union of the spectrum of @xmath212 ( after considering all values of @xmath54 and @xmath138 ) , thus directly confirming an early proof in ref . we stress , however , that the one - to - one correspondence between @xmath177 and @xmath231 is a new result that we did not find previously . finally , we show that @xmath205 and @xmath232 have the same set of chern numbers as their respective original matrices @xmath212 and @xmath170 . to do this , we make use of the line integral version of the chern number of the @xmath77th band given by @xmath233 where @xmath234 and the line integral is around the perimeter of the brillouin zone @xmath235 \times ( 0,2 \pi]$ ] in @xmath236 parameter space . here @xmath237 again refers to the @xmath77th band eigenstate of either @xmath238 or @xmath239 at the point @xmath240 . the eigenstates of @xmath241 and @xmath242 , denoted @xmath243 , are related to the original eigenstates by @xmath244 respectively . we may substitute this into eq . ( [ lineint ] ) and obtain an expression for @xmath142 in terms of @xmath243 . because the transformations @xmath245 depend on @xmath54 but not on @xmath138 , it can be shown , by making use of the fact that the line integrals along @xmath246 and @xmath247 are in opposite directions , that the resulting expression for @xmath142 reduces back to that of the form of eq . ( [ lineint ] ) , except with the transformed eigenstates taking the place of the original ones . this proves that the chern numbers of the unitarily transformed reduced matrices match those of the original ones . next , we note that when we impose @xmath248 and @xmath249 , working out the line integral in eq . ( [ lineint ] ) for @xmath242 over a typical square perimeter space in @xmath216 space with corners @xmath250 is equivalent to working out the line integral for @xmath206 over some parallelogram in @xmath215 space with corners @xmath251 . to complete the proof of topological equivalence , we need only show that the aforementioned line integral in @xmath215 for @xmath252 gives a result equal to that when we calculate the line integral around the perimeter of the usual @xmath253\times(0,2\pi]$ ] brillouin zone . however , this can be easily shown to be the case by converting the line integral around the parallelogram into a surface integral using stokes theorem . we then obtain a surface integral of the form of eq . ( [ cherneq ] ) enclosing the area of the parallelogram . because the berry curvature as seen in eq . ( [ bcurvature ] ) is exactly @xmath0-periodic along both @xmath54 and @xmath138 , it is trivial to see that we can map the area of the parallelogram back onto that of the original @xmath235 \times ( 0,2 \pi]$ ] brillouin zone , without any difference in the result of the integral . in other words , the chern numbers of @xmath254 and @xmath242 are always identical . putting this together with the result of the previous paragraph , we may conclude that the chern numbers of the original matrices @xmath239 and @xmath238 are indeed the same . this completes our proof of topological equivalence . in this work , we have mainly focused on two topics : the spectral difference between ordkr and khm ( comparing quantum maps @xmath37 and @xmath29 ) and their topological equivalence upon introducing an additional periodic phase parameter @xmath138 ( comparing quantum maps @xmath137 and @xmath139 ) . one important spectral difference we have found is the existence of a flat band for @xmath37 under the condition @xmath28 , but not for @xmath29 . to our knowledge , this is the first example of a periodically driven model that has a mixture of flat band and non - flat bands . states launched from a flat band will be strictly localized , and this feature might be useful for benchmarking experimental errors in any future realizations of ordkr . the coexistence of a flat band with non - flat bands may also open up new applications of @xmath11-kicked systems . we have also shown that for small kick strength @xmath28 , the band width of the non - flat bands of @xmath37 scales with @xmath6 in a power law with a high exponent @xmath255 , indicating that for sufficiently small kick strength , all floquet bands will be effectively flat for a long time scale . the dynamical consequence is a transient dynamical localization in ordkr ( absent in khm ) for a long time scale . the topological equivalence between @xmath137 and @xmath139 makes our ordkr - khm comparison even more interesting . that is , for a fixed @xmath138 , ordkr and khm have many different features . but topologically speaking , upon introducing one extra parameter @xmath138 we have a topological equivalence between an extended ordkr model , previously proposed in studies of quantum ratchet acceleration without using a bichromatic lattice @xcite , with a simple extension of the standard khm . to have a pair of models that are topologically equivalent should be a useful contribution to the general understanding of the topological properties of periodically driven systems @xcite . j.w . and j.g . acknowledge helpful discussions with prof . italo guarneri , who also confirmed our flat - band result by showing us an alternative proof by him . d.h . thanks adam zaman chaudhry for helpful discussions . j.w . received support from nnsf ( grants no.11275159 and no . 10925525 ) and srfdp ( grant no . 20100121110021 ) of china , and j.g . is supported by arf tier i , moe of singapore ( grant no . r-144 - 000 - 276 - 112 ) . . dedicates this work to his late beloved wife huairui zhang . for @xmath40 with @xmath2 and @xmath3 being coprime and odd integers , reduced @xmath44 floquet matrix is given by @xmath256_{n , m}=\sum_{l=-\infty}^{\infty}{\left\langle n\right|}\hat u { \left| m+l\times n\right\rangle}e^{il\varphi}$ ] . the floquet operator of ordkr is @xmath257 the reduced @xmath44 floquet matrix is thus @xmath258_{n , m}&=\sum_{l=-\infty}^{\infty } { \left\langle n\right|}\hat u_{\text{ordkr}}{\left| m+l\times n\right\rangle } e^{il\varphi}\\ & = \sum_{l=-\infty}^{\infty}\sum_{l'=-\infty}^{\infty}\sum_{m'=0}^{n-1 } { \left\langle n\right|}e^{i\frac{p^2}{2\hbar}}e^{-i\frac{k}{\hbar}\cos(q)}{\left| m'+l'\times n\right\rangle}\\ & \qquad\times { \left\langle m'+l'\times n\right|}e^{-i\frac{p^2}{2\hbar } } e^{-i\frac{l}{\hbar}\cos(q)}{\left| m+l\times n\right\rangle } e^{il\varphi}\\ & = \sum_{m'=0}^{n-1}\frac{1}{2\pi}e^{i\frac{\hbar}{2}n^2}\int_0^{2\pi } d\theta_2 e^{-i\frac{k}{\hbar}\cos(\theta_2)}e^{i\theta_2(m'-n)}\sum_{l'=-\infty}^{\infty}e^{i\theta_2l'n}\\ & \qquad\times\frac{1}{2\pi}e^{-i\frac{\hbar}{2}(m'+l'n)^2}\int_0^{2\pi } d\theta_1 e^{-i\frac{l}{\hbar}\cos(\theta_1)}e^{i \theta_1(m - m')}\sum_{l=-\infty}^{\infty}e^{i \theta_1(l - l')n}e^{il\varphi}\\ & = \sum_{m'=0}^{n-1}\frac{1}{2\pi } e^{i\frac{\hbar}{2}n^2}\int_0^{2\pi } d\theta_2 e^{i\frac{k}{\hbar}\cos(\theta_2+\pi)}e^{i(\theta_2+\pi)(m'-n)}e^{i\pi(m'-n)}\sum_{l'=-\infty}^{\infty}e^{i ( \theta_2+\pi ) l'n } \\ & \qquad\times\frac{1}{2\pi } e^{-i\frac{\hbar}{2}m'^2}\int_0^{2\pi } d\theta_1 e^{-i\frac{l}{\hbar}\cos(\theta_1)}e^{i\theta_1(m - m')}e^{-i \theta_1 l'n}\sum_{l=-\infty}^{\infty}e^{i \theta_1 ln } e^{il\varphi}. \end{split}\end{aligned}\ ] ] to simplify , we make use of the poisson summation formula @xmath259 and obtain @xmath258_{n , m } & = e^{i\frac{\hbar}{2}n^2}\sum_{m'=0}^{n-1}e^{-i\frac{\hbar}{2}m'^2}e^{i \pi(m'-n ) } \frac{1}{n}\sum_{j_2=0}^{n-1}e^ { i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j_2-\frac{\varphi}{n } ) } e^{i ( \frac{2\pi}{n}j_2-\frac{\varphi}{n})(m'-n)}\\ & \qquad\times\frac{1}{n}\sum_{j_1=0}^{n-1}e^ { -i \frac{l}{\hbar } \cos(\frac{2\pi}{n}j_1-\frac{\varphi}{n})}e^{i ( \frac{2\pi}{n}j_1-\frac{\varphi}{n})(m - m')}\\ & = \frac{1}{n^2}\sum_{j_2=0}^{n-1}\sum_{m'=0}^{n-1}\sum_{j_1=0}^{n-1 } e^{i n\frac{\varphi}{n}}e^{-i m\frac{\varphi}{n}}\\ & \qquad\times e^{-i\frac{2\pi-\hbar}{2}n^2}e^{-i \frac{2\pi}{n}n j_2 } e^ { i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j_2-\frac{\varphi}{n})}e^{i \frac{2\pi}{n}j_2 m'}\\ & \qquad\times e^{i\frac{2\pi-\hbar}{2}m'^2}e^{-i \frac{2\pi}{n}m ' j_1 } e^ { -i \frac{l}{\hbar } \cos(\frac{2\pi}{n}j_1-\frac{\varphi}{n})}e^{i \frac{2\pi}{n}j_1 m}. \end{split}\end{aligned}\ ] ] for the sake of illustration , we write the reduced floquet matrix as a product of unitary matrices @xmath260 if we introduce an additional periodic phase parameter @xmath140 to the ordkr map , the floquet operator becomes @xmath261 the corresponding reduced floquet matrix is @xmath262_{n , m } & = \frac{1}{n^2}\sum_{j_2=0}^{n-1}\sum_{m'=0}^{n-1}\sum_{j_1=0}^{n-1}e^{i n\frac{\varphi}{n}}e^{-i m\frac{\varphi}{n}}\\ & \qquad\times e^{-i\frac{2\pi-\hbar}{2}n^2}e^{-i \frac{2\pi}{n}n j_2}e^ { i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j_2-\frac{\varphi}{n})}e^{i \frac{2\pi}{n}j_2 m'}\\ & \qquad\times e^{i\frac{2\pi-\hbar}{2}m'^2}e^{-i \frac{2\pi}{n}m ' j_1}e^ { -i \frac{l}{\hbar } \cos(\frac{2\pi}{n}j_1-\frac{\varphi}{n}+\alpha)}e^{i \frac{2\pi}{n}j_1 m}. \end{split}\end{aligned}\ ] ] written again as a product of unitary matrices , @xmath263 the floquet operator of khm is @xmath264 with reduced @xmath44 floquet matrix @xmath265_{n , m}&=\sum_{l=-\infty}^{\infty}\langle n|\hat u_{\text{khm}}|m+l\times n\rangle e^{il\varphi}\\ & = \frac{1}{2\pi}e^{-i\frac{l}{\hbar}\cos(n\hbar)}\int_0^{2\pi } d\theta e^{-i\frac{k}{\hbar } \cos(\theta)}e^{i \theta(m - n)}\sum_{l=-\infty}^{\infty}e^{i \theta ln } e^{il\varphi}\\ & = e^{-i\frac{l}{\hbar}\cos(n\hbar)}\frac{1}{n}\sum_{j=0}^{n-1}e^{-i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j-\frac{\varphi}{n})}e^{i ( \frac{2\pi}{n}j-\frac{\varphi}{n})(m - n)}\\ & = \frac{1}{n}\sum_{j=0}^{n-1}e^{i n\frac{\varphi}{n}}e^{-i\frac{l}{\hbar}\cos(n\hbar)}e^{-i \frac{2\pi}{n}n j}e^ { i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j-\frac{\varphi}{n})}e^{i \frac{2\pi}{n}jm}e^{-i m\frac{\varphi}{n}}. \end{split}\end{aligned}\ ] ] for the sake of illustration , we write the reduced floquet matrix as a product of unitary matrices . @xmath266 if we introduce an additional periodic phase parameter @xmath140 to the khm map , the floquet operator becomes @xmath267 the corresponding reduced floquet matrix is @xmath268_{n , m } = \frac{1}{n}\sum_{j=0}^{n-1}e^{i n\frac{\varphi}{n}}e^{-i\frac{l}{\hbar}\cos(n\hbar-\alpha)}e^{-i \frac{2\pi}{n}n j}e^ { i\frac{k}{\hbar } \cos(\frac{2\pi}{n}j-\frac{\varphi}{n})}e^{i \frac{2\pi}{n}jm}e^{-i m\frac{\varphi}{n}}.\end{aligned}\ ] ] written as a product of unitary matrices , @xmath269 @xmath70 is a diagonal unitary matrix and @xmath72 is a unitary matrix . the corresponding matrix elements are @xmath270_{n , n}=e^{i\frac{2\pi-\hbar}{2}n^2}$ ] , @xmath76 , where @xmath271 , and indices @xmath19 and @xmath77 take values @xmath78 . we also assume @xmath272 is an integer ranging from 1 to @xmath3 , and @xmath273 is an integer ranging from 1 to @xmath274 , with @xmath275 . from @xmath87 and using the fact that @xmath276 is an odd number , we have @xmath277}\\ = & \frac{1}{n}\sum_{k=1}^{n}(-1)^ke^{i\frac{2\pi}{n}\left[\frac{m}{2}k^2+(m - n)k\right]}\\ = & \frac{1}{n}\sum_{\tilde k=1}^{q}e^{i\frac{2\pi}{n}\left[\frac{m}{2}(2\tilde k)^2+(m - n)(2\tilde k)\right]}\\ & \quad-\frac{1}{n}\sum_{\tilde k=1}^{q}e^{i\frac{2\pi}{n}\left[\frac{m}{2}(n-2\tilde k)^2+(m - n)(n-2\tilde k)\right]}\\ & \quad-\frac{1}{n}e^{i\frac{2\pi}{n}\left[\frac{m}{2}n^2+(m - n)n\right]}\\ = & \frac{1}{n}+ \frac{1}{n}\sum_{\tilde k=1}^{q}e^{i\frac{2\pi}{n}\left[2m\tilde k^2 + 2(m - n)\tilde k\right]}\\ & \quad-\frac{1}{n}\sum_{\tilde k=1}^{q}e^{i\frac{2\pi}{n}\left[2m\tilde k^2+\frac{m}{2}n^2 - 2mn\tilde k+(m - n)n-2(m - n)\tilde k\right]}\\ = & \frac{1}{n}+\frac{1}{n}\sum_{\tilde k=1}^{q}\left[e^{i\frac{2\pi}{n}\left(2m\tilde k^2 + 2(m - n)\tilde k\right)}+e^{i\frac{2\pi}{n}\left(2m\tilde k^2 - 2(m - n)\tilde k\right)}\right]\\ = & \frac{1}{n}+\frac{2}{n}\sum_{\tilde k=1}^{q}e^{i4\pi\frac{m}{n}\tilde k^2}\cos{\left[4\pi\frac{\tilde k}{n}(m - n)\right]}. \end{split}\ ] ] it is now seen that @xmath91 is a symmetric matrix , i.e. , @xmath278 . g. casati , b.v . chirikov , f.m . izraelev , and j. ford , in _ stochastic behavior in classical and quantum hamitonian systems _ , edited by g. casati and j.ford , lecture notes in physics , vol . 93 ( springer , berlin , 1979 ) . f. m. izrailev , phys . rep . * 196 * , 299 ( 1990 ) . g. lemari , h. lignier , d. delande , p. szriftgiser , and j. c. garreau , phys . 105 , 090601 ( 2010 ) ; i. talukdar , r. shrestha , and g. s. summy , phys . rev . lett . 105 , 054103 ( 2010 ) ; a. ullah and m. d. hoogerland , phys . e 83 , 046218 ( 2011 ) ; m. lopez , j. f. clement , p. szriftgiser , j. c. garreau , and d. delande , phys . 108 , 095701 ( 2012 ) ; m. sadgrove , t. schell , k. nakagawa , and s. wimberger , , 013631 ( 2013 ) ; b. gadway , j. reeves , l. krinner , and d. schneble , , 190401 ( 2013 ) . s. fishman , d.r . grempel , and r.e . prange , phys . * 49 * , 509 ( 1982 ) . f. m. izrailev and d.l . shepelyansky , theor . mat . phys . * 43 * , 553 ( 1980 ) . p. leboeuf , j. kurchan , m. feingold , and d. p. arovas , , 3076 ( 1990 ) . t. geisel , r. ketzmerick , and g. petschel , , 3635 ( 1991 ) ; r. lima and d. shepelyansky , _ ibid _ * 67 * , 1377 ( 1991 ) . r. artuso , f. borgonovi , i. guarneri , and g. casati , , 3302 ( 1992 ) ; t. prosen , i. i. satija , n. shah , , 066601 ( 2001 ) ; i. i. satija , , 056213 ( 2005 ) . g. m. zaslavskii , m. yu . zakharov , r. z. sagdeev , d. a. usikov and a. a. chernikov , sov . jetp * 64 * , 294 ( 1986 ) . i. dana , phys . lett . a * 197 * , 413 ( 1995 ) . d. r. hofstadter , , 2239 ( 1976 ) . j. wang and j. b. gong , , 031405(r ) ( 2008 ) . h. jones , m. m. stocklin , g. hur , and t. s. monteiro , , 223002 ( 2004 ) . i. dana , e. eisenberg , and n. shnerb , , 5948 ( 1996 ) . j. b. gong and j. wang , phys . e * 76 * , 036217 ( 2007 ) . to our knowledge , the smallest quasi - momentum spread in kicked - rotor experiments so far was achieved in the experiment reported in c. ryu , m. .f . andersen , a. vaziri , m. b. darcy , j. m. grossman , k. helmerson , and w. d. phillips , , 160403 ( 2006 ) . w. lawton , a. s. mouritzen , j. wang , and j. b. gong , j. math . * 50 * , 032103 ( 2009 ) ; note that the key result of this work is eq . ( 42 ) , which is clear only after referring to the proposition 29 detailed on page 23 . equation ( 42 ) itself is about the unitary equivalance between two extended versions of ordkr and khm after considering a union of their respective spectrum . section iii of the present work makes one step further by exposing a unitary equivalence under a certain parameter mapping without considering a spectral union . i. guarneri , _ a note about double kicked rotors _ ( unpublished ) , 2008 . j. wang , a. s. mouritzen , and j. b. gong , j. mod . opt.*56 * , 722 ( 2009 ) . i. dana , , 1609 ( 1994 ) ; i. dana and d. l. dorofeev , , 046205 ( 2005 ) ; i. dana , e. eisenberg , and n. shnerb , , 686 ( 1995 ) ; i. dana and d. l. dorofeev , , 026206 ( 2006 ) . j. wang , i. guarneri , g. casati , and j. b. gong , , 234104 ( 2011 ) . h. l. wang , j. wang , i. guarneri , g. casati , and j. b. gong , , 052919 ( 2013 ) . see also arxiv : 1308.3527 . d. y. h. ho and j. b. gong , , 010601 ( 2012 ) . note that in this reference , the eigenstates used to compute chern numbers differ from those used here by a unitary transformation @xmath279 $ ] that has no effect on the value of the chern numbers .

we show that when supersymmetry is broken at the tev scale by strong dynamics , the higgs sector of the mssm can be drastically modified . this arises from possible sizeable mixings of the higgs with the resonances of the strong sector . in particular the mass of the lightest higgs boson can be significantly above the mssm bound ( gev ) . furthermore only one higgs doublet is strictly necessary , because the yukawa couplings can have a very different structure compared to the mssm . using the ads / cft correspondence electroweak precision observables can be calculated and shown to be below experimental bounds . the most natural way to generate sparticle masses is through mixing with the composite states . this causes the gauginos and higgsinos to easily obtain dirac masses around 200 gev , while scalar masses can be generated either from extra-terms or also through mixing with the strongly - coupled states . finally one of the most interesting predictions of these scenarios is the sizeable decay width of the higgs boson into a very light gravitino ev ) and a higgsino . -1.0 cm -0.5 cm -0.5 cm july 2011 + + + _ _ + _ _ nsf - kitp-11 - 137 + su - itp-11/38

1107.4697

one of the most remarkable predictions of the minimal supersymmetric standard model ( mssm ) arises in the higgs sector . supersymmetry requires the presence of two higgs doublets @xmath3 with a fully fixed yukawa structure . only one higgs @xmath4 can couple to the up - quark sector , while the other higgs @xmath5 couples to the down - quark and lepton sectors . the higgs potential quartic couplings are also determined in terms of the gauge couplings . when supersymmetry is softly broken some of these predictions are modified by quantum effects , but nevertheless the higgs quartic and yukawa couplings can be predicted as a function of the soft masses . this leads to some general predictions , the most important being the presence of a light higgs boson with mass @xmath6 gev . in addition there are relations between the heavy higgs masses such as @xmath7 . these predictions occur for the most popular models of supersymmetry - breaking , such as in gauge - mediated supersymmetry breaking ( gmsb ) models . the purpose of this paper is to show that the higgs sector can be drastically modified if supersymmetry is broken dynamically by a strong sector at low energies @xmath8 tev . this can occur if the higgs responsible for electroweak symmetry breaking has sizeable couplings to this sector , which then leads to mixings with composite states . in particular , we will show that the mass of the lightest higgs boson can not only be heavier than the tree - level value , @xmath9 ( without large radiative corrections as in usual mssm models ) , but can also be significantly enhanced well beyond @xmath0 gev . furthermore the yukawa couplings can fully arise from supersymmetry - breaking effects , implying that it is possible to have a supersymmetric standard model with only one higgs doublet . to accomplish this , there are several obvious generic problems that need to be addressed . first , a strong sector with a partly - composite higgs generically leads to corrections to standard model ( sm ) observables that have been very well measured at collider experiments . a second problem is related with the absence of large flavor changing neutral currents ( fcnc ) . finally , we must generate sizeable gaugino and squark / slepton masses . however in contrast to gmsb models , the problem of generating sufficiently large higgsino masses ( the @xmath10-problem ) is easily solved in our setup because the higgsino can have sizeable mixing with the composite states . we will make the simplifying assumption that there is a unique strong sector responsible for supersymmetry breaking . as in gmsb models , this sector will contain fields charged under the sm gauge group . due to our present lack of complete understanding of strong interactions , we will not consider any specific model of dynamical supersymmetry breaking . instead we will make use of the ads / cft correspondence in order to calculate observables and obtain predictions . generically this correspondence allows strongly - coupled gauge theories in the large-@xmath11 and large t hooft coupling limit to be described by gravitational theories in five dimensions @xcite . we will also consider the possibility of having the strong sector being responsible not only for supersymmetry breaking , but also for electroweak symmetry breaking . in this case the mssm higgs obtains a vacuum expectation value ( vev ) from a tadpole , as in bosonic technicolor models @xcite . a generic prediction of our scenario is that by mixing with the strong sector the lightest higgs boson can become much heavier than 130 gev without causing any conflict with electroweak precision tests ( ewpt ) . since the scale of supersymmetry breaking is low ( @xmath8 tev ) , novel ways are needed to generate sparticle masses consistent with collider bounds . for example , the problem of having sufficiently heavy gaugino masses can be remedied by marrying the gaugino with composite states thereby generating a dirac gaugino . similarly the squark and slepton masses will depend on how the mssm matter sector couples to the strong sector . if there is no direct coupling , one is forced to introduce an extra @xmath12 to mediate the breaking of supersymmetry to squarks and sleptons via flavor - independent @xmath1-terms . alternatively , the matter fields can mix with composite states and generate sizeable soft masses . in this case , yukawa couplings can be generated from the strong sector without requiring a second higgs doublet , but flavor symmetries are needed to avoid fcnc . our novel scenario leads to extraordinary mssm physics . for example , the lightest higgs boson may be observed through the golden @xmath13 decay channel , and can also have sizeable decays to the gravitino and higgsino whenever the @xmath14 channel is kinematically closed . signals of compositeness in the quark sector could also be visible at the lhc , and a tev - scale mass @xmath15 gauge boson , associated with the new @xmath12 gauge symmetry , could also be easily detected through its decay to leptons . the scenarios presented here share some similarities with models in a warped extra dimension with tev - scale supersymmetry breaking @xcite . also recently , detailed strong sectors with similar properties has been proposed in ref . @xcite using seiberg dualities . for previous general studies see ref . @xcite . in addition , while this article was being completed , we became aware of analogous phenomenological studies in ref.@xcite , that partially overlap with our present work . we go beyond the analysis in ref.@xcite by making use of the ads / cft correspondence to make quantitative predictions for large @xmath11 gauge theories . the strong sector will be broadly characterized by two parameters , the scale @xmath16 tev associated with the mass gap ( or , equivalently , the mass of the first resonance ) , and the number of colors " @xmath11 . we will assume that supersymmetry is completely broken in the strong sector with supersymmetry - breaking mass - splittings of order @xmath17 . the vacuum energy is therefore estimated to be @xmath18 ( assuming strong sector fields transform in the fundamental representation of the gauge group ) which implies that @xmath19-terms are generically of order f~^2 , [ fterm ] and similarly for @xmath1-terms . in certain cases it will be useful to parametrize , as usual , the breaking of supersymmetry by a dimensionless spurion superfield ^2 . [ spurion ] the sm fields and superpartners , denoted generically as @xmath20 , will be assumed to couple to the strong sector linearly : _ int = g_i_i o__i , [ mixing ] where @xmath21 is the coupling to an operator @xmath22 of the strong sector . in particular , for the mssm vector supermultiplet @xmath23 an interaction of the type eq . ( [ mixing ] ) can arise , like in gmsb models , by assuming that the strong sector contains fields charged under the sm gauge groups . in this case we have the gauge interaction @xmath24 , where the superfield @xmath25 contains the conserved current operator of the strong sector ( @xmath26 ) and @xmath21 is the gauge coupling of the different sm gauge groups . for the mssm chiral multiplets the interaction eq . ( [ mixing ] ) can arise from a superpotential term @xmath27 , where the same notation will be used for general superfields and component fields . notice that , apart from couplings like eq . ( [ mixing ] ) , supersymmetry and gauge invariance allows for extra ( non - linear ) interactions between the mssm and the strong sector . if the strong sector is approximately conformal at energies above @xmath17 , the operators @xmath22 can be organized according to their conformal dimension @xmath28 $ ] . the couplings @xmath21 will then have a dimension @xmath29 where @xmath30 + { \rm dim}[\phi_i]-4 $ ] . the renormalization group ( rg ) equation of the dimensionless coupling , @xmath31 is then given by = _ ig_i+_ig^3_i+ , [ rgeqn ] where the second term in ( [ rgeqn ] ) originates from the wave - function renormalization of @xmath20 in the large-@xmath11 limit , with @xmath32 a coefficient of order one satisfying @xmath33 . we will require the value of the coupling at the mass gap scale , @xmath34 , where it is useful to define the quantity _ i~ , [ epsilon ] up to factors of order one . this ratio gives the coupling strength of the mssm fields to the strong sector , normalized with respect to the strong sector couplings @xmath35 @xcite . equivalently , @xmath36 parametrizes the mixing between the mssm fields and the composite states ( resonances ) associated with the operators @xmath22 . for @xmath37 the mixing is maximal and the mssm particle behaves as a resonance of the strong sector . we are interested in the case in which @xmath38 such that the coupling @xmath39 is relevant ( @xmath40 ) or marginal ( @xmath41 ) . for @xmath42 , we see from eq . ( [ rgeqn ] ) that @xmath39 grows towards low energies , before reaching the fixed - point value g_i()= . [ irrcoup ] in this case we obtain _ [ releg ] when @xmath41 , as for the mssm gauge couplings , the solution of eq . ( [ rgeqn ] ) gives = + _ i . [ rung ] for @xmath43 , we have @xmath44 and then @xmath45 . in particular , for the gauge couplings we have @xmath46 that will be useful later . also we set @xmath47 by a redefinition of @xmath11 . as mentioned in the introduction the higgs sector can either contain one or two higgs doublets . let us begin by considering the case of two higgs doublets with opposite hypercharge , @xmath5 and @xmath4 , and study their potential . the potential consists of a supersymmetric part containing not only the usual @xmath1-term , but also possible @xmath19-terms arising from @xmath48 . in addition there are supersymmetry - breaking masses arising from @xmath49 @xmath50 , and quartic couplings from @xmath51 with @xmath52 defined in eq . ( [ spurion ] ) . for the neutral @xmath53-even components , @xmath54 and @xmath55 , we then have @xmath56 where @xmath57 with @xmath9 the @xmath58-boson mass , @xmath59 gev , and the coefficients @xmath60 and @xmath61 are proportional to the degree of supersymmetry breaking in the strong sector which we assume to be of order one . the parametric dependence of the terms in eq . ( [ parest ] ) can be understood as follows . at large @xmath11 the strong sector can be described by a theory of resonances with masses @xmath62 and quartic couplings @xmath63 . assuming a mixing between these resonances and the higgs of order @xmath64 , we obtain a higgs mass - squared of order @xmath65 and quartic couplings for the higgs of order @xmath66 . this is indeed supported by the ads / cft correspondence ( see appendix ) , where we find , for example , that the ratio between the higgs quartic coupling and the first scalar resonance coupling is given by @xmath67 . supersymmetric contributions from @xmath68 , if induced from the strong sector , are of order @xmath69 . compared to the supersymmetry breaking contributions they can be neglected , and we will assume for simplicity @xmath70 . we will be interested in the large @xmath71 limit because this gives the largest possible mass for the lightest mssm higgs boson . this limit can be naturally achieved by taking @xmath72 . indeed , in this case @xmath5 couples stronger than @xmath4 to the supersymmetry - breaking sector , leading to the hierarchy , @xmath73 , and @xmath74 . in this limit we can integrate @xmath54 out , and obtain the effective potential for @xmath55 @xmath75 where the coefficients @xmath76 and @xmath77 have been redefined to absorb order - one corrections . the potential eq . ( [ pot2 ] ) is obviously equivalent to the potential of a one - higgs doublet model with @xmath55 playing the role of the sm higgs . for @xmath78 we have electroweak symmetry breaking with @xmath79 and the higgs mass given by @xmath80 to have a sizeable effect on the higgs mass we must consider values of @xmath81 such that corrections to @xmath82 become sizeable . this is shown in fig . [ mhplot ] where we plot the tree - level higgs mass for @xmath83 and @xmath84 and @xmath85 . in particular , we find that for @xmath86 the tree - level higgs mass can be as large as 190 gev . for smaller values of @xmath87 the higgs mass , as expected , decreases as depicted in fig . [ mhplot ] . for @xmath84 ( upper lines ) and @xmath88 ( lower lines ) . we have taken @xmath83 and the solid ( dashed ) line is for tan@xmath89 ( tan@xmath90 ) . ] however , the stronger the higgs couples to the supersymmetry - breaking sector , the larger the corrections to the electroweak observables . these corrections arise from the strong sector that , coupled to the higgs , can generate large effects on the sm gauge boson self - energies . the main contributions are parametrized by the @xmath91 and @xmath92 parameters defined in ref . we will follow the notation in ref . @xcite where these parameters are denoted by @xmath93 and @xmath94 . these contributions can be parametrically written as @xmath95 where @xmath96 are parameters of order one . in order to make precise predictions we will take the values of @xmath96 as predicted by the ads / cft correspondence . in the regime @xmath97 , we have @xcite : @xmath98 where we identify @xmath99 this formula changes to @xmath100 . ] . we will take @xmath101 and @xmath47 as mentioned earlier . the contributions to @xmath93 and @xmath94 are shown in fig . [ stch ] for @xmath83 . the solid red line corresponds to @xmath102 tev for different values of @xmath103 . the origin of the @xmath93-@xmath94 plane corresponds to the sm with a reference higgs mass of @xmath104 gev . since our sm - like higgs @xmath55 has a mass different from @xmath105 , we have included this effect in @xmath93 and @xmath94 . in particular , the mass of @xmath55 is computed , as a function of @xmath103 , from eq . ( [ mhiggs ] ) with @xmath84 . in fig . [ stch ] we see that , as expected , increasing @xmath103 couples @xmath4 stronger to the supersymmetry - breaking sector , causing the contribution to @xmath93 and @xmath94 to increase . nevertheless , it is interesting to note that both contributions are positive , allowing them to remain inside the 99@xmath106 cl ellipse for values as large as @xmath86 . for these values of @xmath103 the higgs mass ( fig . [ mhplot ] ) receives important contributions . therefore a higgs mass well above the tree - level mssm value , @xmath9 is possible without affecting the ewpt of the sm . as we increase @xmath17 , the contributions to @xmath93 and @xmath94 becomes smaller , as is clear from the expressions ( [ stest0 ] ) and ( [ stest ] ) . this can allow for a higgs with a larger value of @xmath103 and therefore a larger mass . for @xmath107 tev ( @xmath108 tev ) the dashed blue ( dotted green ) line in fig . [ stch ] shows the contribution to @xmath93 and @xmath94 for several values of @xmath103 . for @xmath109 the higgs mass can obtain a very large value @xmath110 gev . and @xmath94 for @xmath83 and different values of the degree of compositeness @xmath103 . the solid red line corresponds to @xmath102 tev , the dashed blue line is for @xmath107 tev , and the dotted green line is for @xmath111 tev . the values of @xmath103 increase as we move out from the ellipses along the lines . the first dot ( inside the inner ellipse ) corresponds to @xmath112 , and each successive dot represents an increase of @xmath113 . ] ) , as a function of @xmath103 . the solid red line is for @xmath102 tev , the dashed blue line is for @xmath107 tev , and the dotted green line is for @xmath111 tev . we have taken @xmath83 and @xmath84 ( lower lines ) and @xmath114 ( upper lines ) . ] although increasing the values of @xmath17 and/or @xmath103 can lead , as we have seen , to large values of the higgs mass while still maintaining consistency with ewpt , it is clear that a large @xmath17 and @xmath103 leads to a large @xmath115 , unless we tune @xmath116 . this tuning , which is reminiscent of the hierarchy problem , can only be technically natural if also @xmath117 , since this would correspond to the supersymmetric limit @xmath118 . therefore , for a given value of @xmath17 and @xmath103 we can estimate the fine tuning of the model by calculating @xmath119 . using @xmath120 this ratio can be written as = . [ tun ] in fig . [ tplot ] we plot eq . ( [ tun ] ) as a function of @xmath103 . for @xmath102 tev we see that no tuning is really needed , while for @xmath107 tev the tuning is around 10@xmath106 . notice that the tuning is very sensitive to the coefficient @xmath77 and is reduced when this coefficient is smaller than one , as can be appreciated in fig . [ tplot ] . of course , smaller values of @xmath77 lead to smaller higgs masses as shown in fig . [ mhplot ] . it is also interesting to remark that the tuning does not increase as the higgs becomes more strongly coupled to the supersymmetry - breaking sector . we can then envisage a scenario with @xmath16 4 tev and @xmath121 , certainly with some degree of tuning ( but not higher compared to other mssm scenarios such as gmsb ) , in which the higgs is quite heavy , @xmath122 tev , and therefore difficult to be seen at the lhc . the elementary higgs scenario presented in the previous section can also be described using an alternative ( dual ) description . the fields @xmath3 can be thought of as arising from the strong sector with the identification @xmath123 - 1 $ ] . taking @xmath124 then corresponds to the limit in which these operators have dimension close to one . in this limit @xmath3 become free fields and decouple from the strong sector . this is an equivalent description of the elementary higgs scenario and gives rise to the same physics . from this perspective there is no elementary higgs and electroweak symmetry breaking would be triggered by the strong sector , just like in conformal technicolor theories @xcite . these two descriptions have been identified in ref . @xcite using the ads / cft correspondence . let us now assume that the strong sector not only breaks supersymmetry but also electroweak symmetry @xcite by a condensate parametrized by a doublet with hypercharge @xmath125 : @xmath126 where @xmath127 are pauli matrices . the vev of @xmath128 can be thought of as arising from a generic potential induced by the strong dynamics , like in qcd or technicolor models . the natural scale for @xmath129 is @xmath130 . the @xmath128 scalar could be identified with the @xmath5 field of the mssm . in this case its fermionic partner , a composite state , is responsible for canceling the anomalies of the @xmath4 higgsino . this case then corresponds to a particular case of the previous model with @xmath131 , implying that @xmath5 behaves as a composite field of the strong sector . contrary to the previous section , however , we must emphasize that here we are assuming that the vev of @xmath5 is generated independently from the vev of @xmath4 . the higgs potential for @xmath4 is given by @xmath132 where @xmath115 and @xmath133 are given by eq . ( [ parest ] ) ( with @xmath70 ) and we have neglected terms with higher powers of @xmath128 since they are irrelevant in the analysis . the novelty of this scenario is that @xmath4 gets a vev induced by the tadpole arising from the last term of eq . ( [ pot2s ] ) that is of order @xmath134 . the electroweak symmetry is then broken by the strong sector and the elementary higgs @xmath4 vev , @xmath135 , giving rise to @xmath136 where @xmath87 can be identified with @xmath137 . in this case the higgs mass is given by @xmath138 when the second term in eq . ( [ mh2 ] ) is positive , it gives an additional contribution to the higgs mass compared to eq . ( [ mhiggs ] ) , causing the higgs to be heavier than in the ordinary mssm . since in this case large values of @xmath103 are no longer needed to obtain sizeable contributions to the higgs mass , we will assume @xmath139 . the main contributions to @xmath94 and @xmath93 are no longer arising from the vev of @xmath4 but instead from electroweak symmetry breaking in the strong sector , @xmath128 . these contributions are therefore proportional to @xmath129 and given by @xmath140 where @xmath96 are parameters of order one . the predictions from ads / cft are given in eq . ( [ csct ] ) with the replacement @xmath141 - 1 $ ] , where @xmath142 is the operator of the strong sector that breaks electroweak symmetry . assuming dim@xmath143=2 $ ] and @xmath144 , we have @xmath145 if @xmath146 , the contributions ( [ sttcest1 ] ) and ( [ sttcest2 ] ) to @xmath94 and @xmath93 are the same as those of technicolor models without a custodial symmetry , which are clearly ruled out mostly due to the @xmath94 parameter . this means that we either need @xmath147 or @xmath148 . the first option can be naturally implemented since parametrically , neglecting the @xmath1-term , we have @xmath149 . however , this implies that @xmath150 tev which becomes too small to generate proper masses for the superpartners . therefore we will need to also rely on having @xmath148 ; this will not represent a severe fine tuning of the parameters of the model since supersymmetry could mildly protect @xmath129 to make it slightly smaller than its nda value . in fig . [ sttc ] we show the contributions to @xmath94 and @xmath93 using eq . ( [ csct2 ] ) , @xmath83 , and treat @xmath129 as a free parameter . as in fig . [ stch ] , the one - loop effects on @xmath94 and @xmath93 from a higgs @xmath55 heavier than @xmath104 gev have been included . its mass is calculated using eq . ( [ mh2 ] ) with @xmath112 and @xmath151 . the solid red line is for @xmath102 tev and the corresponding dots are for @xmath152 gev , while the dashed blue line is for @xmath107 tev and the corresponding dots are for @xmath153 gev . we can see that the model is consistent with the @xmath154 constraints for values of @xmath129 only slightly smaller than its nda value @xmath155 . in these scenarios the mass of the lightest higgs @xmath55 is dominated , for @xmath156 tev , by the second term of eq . ( [ mh2 ] ) , that gives m_hm_22~_h_2~200 gev ( ) ( ) . [ masshtc ] the three scalar states in the @xmath128 field , @xmath157 , that corresponds to a neutral and a charged state , obtain a common mass given by m_^2= , where @xmath158 has been determined as a function of the other parameters by the minimization condition . we plot in fig . [ mh2hybrid ] the mass of @xmath55 and @xmath157 in the regions of @xmath129 allowed by ewpt , assuming @xmath112 and @xmath151 . the generalization of eq . ( [ pot2s ] ) for two elementary higgs is straightforward . the potential becomes @xmath159 where @xmath160 is the ordinary mssm @xmath1-term contribution . again , we have neglected terms that are smaller in the limit @xmath161 and @xmath147 . we will discuss the implications of this potential below . [ cols="^,^ " , ] if the strong sector discussed above is the only source of supersymmetry breaking , it will have to generate masses for all the mssm superpartners . however , since we are assuming a low supersymmetry - breaking scale @xmath8 tev , generating sufficiently heavy superpartner masses represents a challenge . let us begin by considering the gaugino sector . as in gmsb models , majorana gaugino masses can be generated if the @xmath162-symmetry is broken . we obtain @xmath163 where @xmath39 are the gauge couplings . however for @xmath16 tev and @xmath164 these masses seem to be too small to satisfy collider constraints , at least for gluino masses . therefore we are forced to assume that the gauginos obtain dirac masses . recently gmsb models with dirac gauginos have been studied in @xcite . in our setup dirac masses can be generated by mixing the mssm gauginos with composite fermions arising from the strong sector , just like in dual five - dimensional ( 5d ) models @xcite . in this case , the gaugino mass becomes @xmath165 where @xmath166 are coefficients of order one . in ads@xmath167 models ( see eq . ( [ gauginomass ] ) in the appendix ) we find that @xmath168 . since composite fermions marry the mssm gauginos to obtain dirac masses , we must avoid the presence of majorana masses for these composite fermions ( or at least these must be smaller than @xmath169 of eq . ( [ mdirac ] ) ) . this can be achieved by demanding that the strong sector be approximately invariant under some global symmetry under which these composite fermions are charged . this could be an @xmath162-symmetry under which the spurion eq . ( [ spurion ] ) has charge @xmath170 and the composite fermions ( in the adjoint representation of the sm gauge group ) have opposite @xmath162-charges compared to the mssm gauginos , @xmath171 . an interesting possibility is to have only dirac masses for the gluino and the wino , but not for the bino . the bino mass will then arise from @xmath162-symmetry breaking terms and could be much smaller . in the case that the @xmath162-symmetry is only broken by gravitational interactions , the bino can be lighter than the gravitino , a possibility that can be viable phenomenologically as soon as the mixing between the bino and higgsino is smaller than @xmath172 @xcite . similarly , higgsinos can also obtain dirac masses by mixing with composite states . in the case of only one higgs superfield @xmath4 , anomaly cancellation requires the presence of an extra fermion doublet that could marry the higgsino @xmath173 and give a mass of order @xmath174 in the case of two mssm higgs superfields @xmath3 , the two corresponding higgsinos can marry each other giving @xmath175 notice that this mass term breaks any @xmath162-symmetry since @xmath5 and @xmath4 must have opposite @xmath162-charges in order to allow the presence of a mixing term for the scalar components , @xmath176 . alternatively , we could preserve the @xmath162-symmetry by marrying the two higgsinos @xmath177 with two extra composite fermions and obtain masses of order @xmath178 . finally , in these scenarios the lightest supersymmetric particle is the gravitino ( with the exception of scenarios in which the bino gets its mass from gravitational interactions @xcite ) , with a mass given by m_3/2~~ 10 ^ -4 ev ( ) ^2 . [ gravitino ] the gravitino contains the goldstino , which arises from the strong sector and couples to the mssm fields with a coupling suppressed by @xmath179 . if the full model has an exact @xmath162-parity , then the goldstino is a stable particle . if the matter sector , quarks and leptons , are elementary fields not directly coupled to the strong sector , then the corresponding squarks and sleptons obtain their masses from gmsb contributions . as for gauginos , these contributions , of order @xmath180 , are too small if @xmath16 tev . we find only two ways to generate sizeable squark and slepton masses . the first option is to introduce an extra @xmath12 to mediate supersymmetry - breaking from the strong sector to the mssm . this extra gauge symmetry must be broken around the scale @xmath17 . assuming that the mssm matter fields and strong - sector fields are both charged under this @xmath12 , @xmath1-terms can mediate the supersymmetry - breaking from the strong sector to the mssm @xcite . this leads to soft masses @xmath181 where the @xmath12 charges , @xmath182 are assumed to be positive for all mssm fields and family - independent to avoid fcnc . if the @xmath12 is broken by the strong sector , the gauge boson obtains a mass @xmath183 that seems to be too small to avoid experimental constraints for @xmath16 tev . this then requires that the @xmath12 is broken at a slightly larger energy scale than @xmath17 by some field in the strong sector that does not simultaneously break supersymmetry . in ads@xmath167 models this larger scale could be @xmath184 , which is related to the 5d cut - off scale of the theory ( see appendix ) . with this assumption we obtain @xmath185 for @xmath186 , allowing the mass to be a few tev . therefore , combining this with the nda estimate of eq . ( [ fterm ] ) , the soft masses ( [ dsofteq ] ) from the @xmath1-terms become of order @xmath187 the soft masses can therefore be as large as the gaugino masses . an alternative way to generate squark and slepton soft masses is via bilinear mixings with the strong sector , for example , @xmath188 where @xmath189 are the quark superfields and @xmath190 is an operator of the strong sector . soft masses will then be generated at the one - loop level giving _ i^2~^2 . we need @xmath191 to generate masses above the experimental constraints for @xmath16 tev . the yukawa couplings @xmath192 ( @xmath193 ) of the higgs fields can arise , as in the mssm , from superpotential terms : d^2 ( y_u h_2 qu+y_d h_1 qd+y_e h_1 le ) . we are therefore forced to have a two higgs - doublet model . this case does not suffer from fcnc , since all the flavor structure can be encoded in the yukawa couplings , @xmath192 . for the case of a cft ( @xmath194 constant ) with a high cutoff @xmath195 , we can obtain in these scenarios an upper bound on @xmath196 . this is due to the wave - function renormalization of @xmath3 arising from the mixing eq . ( [ mixing ] ) . indeed , from the rg equation for the yukawas we obtain for @xmath197 : y_u , d ( ) y_u , d ( _ uv)()^-_h_2,1 . [ rgy ] therefore the larger @xmath198 , the larger @xmath199 , and the smaller the yukawa couplings at low energies . the requirement of generating sizeable yukawa couplings for the 3rd family quarks then places a bound on @xmath198 . demanding @xmath200 , eq . ( [ rgy ] ) gives _ h_2 ^ 2 , _ h_1 ^ 2 . for @xmath201 , we have @xmath202 and @xmath203 . these constraints however can be relaxed either by taking a smaller uv cutoff @xmath195 or by having a non - constant @xmath194 ( _ i.e. , _ a departure from a cft ) . a second possibility for the matter sector is , as in the case of the higgs , to couple it linearly to the strong sector following eq . ( [ mixing ] ) . this implies that matter superfields mix with resonances of the strong sector , and then sm fermions are a mixture of elementary and composite fields _ i.e. _ , they are partially composite . the mixing angle is of order @xmath36 ( @xmath204 ) , defined in eq . ( [ epsilon ] ) . in this case squarks and sleptons obtain masses of order @xmath205 where now @xmath39 is the coupling of the matter superfield @xmath20 to the strong sector . for values of @xmath206 these scalar masses are of the same order of magnitude as the gaugino masses . to avoid fcnc the strong sector must preserve a flavor symmetry , implying that @xmath39 are the same for all families . large values of @xmath39 make the squark and slepton masses heavier , but at the expense of increasing the degree of compositeness of the matter fields . this is bound by vertex corrections to the @xmath207 and @xmath58 couplings , @xmath208 that have been very well measured at lep . we find that = c_z , w _ h_2 ^ 2 , [ couplingratio ] where @xmath209 is a coefficient of order one ; its prediction from ads / cft is @xmath210 ( see appendix ) . the strongest bounds are on the @xmath207 and @xmath58 boson couplings to leptons and left - handed quarks that are measured at the few per mille level . for these fields , using the ads / cft prediction and the rough experimental bound @xmath211 , we obtain the constraint _ q , e , l0.3 ( ) ( ) . note that for the right - handed quarks which have a small @xmath58-boson coupling , these deviations are only measured at the 10@xmath106 level and therefore do not put strong constraints on @xmath212 . as in the case of fully elementary matter , the yukawa couplings can arise from the superpotential . however with partially composite matter yukawa couplings can be generated from the strong sector . in this case , since supersymmetry is broken in the strong sector , only one higgs doublet @xmath4 is required . in superfield notation the couplings can be written as d^2 y_u h_2 qu+ d^4 ^(y_d h_2^qd + y_e h_2^le ) , that can preserve an @xmath162-symmetry with @xmath213 for all matter fields and @xmath214 for the higgs . to avoid fcnc from four - fermion operators that , without any flavor symmetry , will only be suppressed by @xmath215 and will generically be too large , we must demand that all sources of flavor violation are proportional to the yukawa couplings @xmath192 , _ i.e. _ , the minimal flavor violation ( mfv ) hypothesis . in this case flavor symmetry breaking effects are parametrized by three spurion fields of the strong sector , @xmath216 , that are related to the sm yukawa couplings by y_u = y_u_h_2_q_u , [ rely ] and similarly for the other yukawa couplings . we can now estimate the magnitude of the fcnc operators . for example , at the loop - level the following operator is generated ( in superfield notation ) : d^4 ( q y_u y_u^q^)^2 . [ fcnc ] using the flavor constraints from ref . @xcite and eqs . ( [ rely ] ) and ( [ fcnc ] ) we find that 4 _ h_2 ^ 2 ^ 2_u 5 tev , which if satisfied implies that the up - sector must be quite composite , @xmath217 , for @xmath218 tev . this can have important implications in present searches in dijet mass and angular distributions at the lhc @xcite . the main phenomenological implications of models with supersymmetry broken at low energy occur in the higgs sector . as we have seen , these models can either have one or two higgs doublets . in both cases , the lightest scalar @xmath219 can be much heavier than 130 gev as shown in figs . [ mhplot ] and [ mh2hybrid ] . it is then possible to detect this scalar via the golden decay channel @xmath220 . the strong sector also gives corrections to the higgs couplings of order @xmath221 for the @xmath222 coupling , and @xmath223 for the higgs couplings to sm fermions when the yukawa terms are generated from the strong sector @xcite . these later can be as large as @xmath224 for @xmath83 , @xmath102 tev and @xmath225 . when a second higgs doublet @xmath5 is present , as in the ordinary mssm , there are four extra scalars , @xmath226 and @xmath227 . in the limit @xmath72 they have a common mass @xmath228 . the mass - splitting between @xmath229 and @xmath227 is given by m_h^^2=m_a^2+(2_5-_4 ) , where we have defined @xmath230 . these quartic couplings , which in the mssm are given by @xmath231 and @xmath232 , receive corrections of order @xmath233 , then lead to deviations from the mssm relation : m_h^^2-m_a^2= m_w^2+o ( _ h_2 ^ 2m_h^^2 ) . these deviations can be large enough to allow for the decay @xmath234 which could discriminate these models from ordinary mssm models . when the second higgs is identified with @xmath128 , which itself breaks electroweak symmetry , we find that in the limit @xmath147 only three scalars , @xmath157 , are lighter than @xmath17 with a mass shown in fig . [ mh2hybrid ] . if two higgs doublets @xmath3 are added to @xmath128 we then have a three higgs - doublet model with a potential given in eq . ( [ pot12s ] ) . in the limit @xmath147 and @xmath235 the light spectrum is similar to the mssm ( since the @xmath157 are heavier ) . however the potential eq . ( [ pot12s ] ) has two extra parameters @xmath236 as compared with the mssm . we can not then predict the neutral higgs masses as a function of @xmath237 and @xmath87 , but we can predict @xmath238 and the mixing @xmath239 which determines the mass - eigenstate higgses @xmath219 and @xmath240 as a function of @xmath54 and @xmath55 . in the first case we find the same mssm relation @xmath241 , while for the second case we obtain ^2(-)= . one of the most interesting phenomenological implications of models with low - scale supersymmetry breaking is the possible decay of the higgs to the goldstino and neutralino . for the lightest higgs this decay width is given by ( hg _ i)= ^4 , where @xmath242 is a mixing angle @xcite . this decay width is important whenever the lightest neutralino @xmath243 is mostly the higgsino partner of @xmath219 . in this case we have @xmath244 and br(@xmath245 ) can be sizeable if the mass of the higgsino is around its present lep bound @xmath246 gev . in fig . [ hgh ] we show this branching ratio for @xmath83 , @xmath102 tev and @xmath247 gev . for the other heavy higgs see ref . @xcite . in scenarios with light binos , which as explained earlier can be naturally implemented in these models , the higgs can also decay , via a bino - higgsino mixing , to two binos . light binos , @xmath248 , decay mostly to @xmath249 but this can be outside the detector if @xmath250 gev ( for @xmath251 ev ) , or they can be even stable if @xmath252 . as a consequence the decay to binos corresponds to an invisible partial width for the higgs . . see the text for details . ] if our model has a conserved @xmath162-parity , its phenomenology will be very similar to gmsb models . there are however a few important differences . first , the fact that the @xmath19-term can be so small , ~700 gev ( ) ^1/4 ( ) , [ ftermv ] implies that the next - to - lightest supersymmetric particle decays inside the detector with no sizeable vertex displacement . therefore , contrary to gmsb models , searches for superpartners must be performed in this case without requiring displaced vertices . another important phenomenological difference is that the gauginos are dirac states , instead of majorana . this affects gaugino cross - sections and decays as shown in ref.@xcite . nevertheless , we expect that bounds on sparticle masses from the lhc and the tevatron on gmsb models , that are quite stringent , will be similar in our scenarios . if @xmath162-parity is not conserved in our models then the bounds on the sparticles are relaxed . the best limits to date arise from d0 where tau sneutrinos as light as 100 gev are allowed @xcite . a generic prediction of models with low - scale supersymmetry breaking is a light gravitino , ( see eq . ( [ gravitino ] ) ) , that couples dominantly via the longitudinal component or goldstino @xmath253 . if all superpartners are too heavy to be detected at the lhc , goldstinos can still be produced with sm particles . these interactions arise from dimension-8 operators suppressed by @xmath254 . since @xmath253 is either stable or decays outside the detector ( if @xmath162-parity is not conserved ) , the signal at hadron colliders is missing transverse momentum together with either a monojet or photon . this arises from @xmath255 or @xmath256 with initial - state bremsstrahlung and thus provides a distinctive signal of our setup . the tevatron data leads to the most stringent bound to date , @xmath257 gev @xcite , while the lhc will be able to probe @xmath258 up to approximately 2 tev . on the other hand , the fact that the compositeness scale @xmath16 tev is low , allows for the possible detection of resonances of the strong sector at the lhc . we expect resonances with the same quantum numbers as the sm gauge bosons . this is similar to usual models in a warped extra dimension , except that in our scenario the fermions can all be elementary ( localized on the uv boundary ) and the higgs partly composite . this implies that the resonance coupling to elementary fermions is universal and approximately given by @xmath259 , where @xmath39 ( @xmath260 ) are the sm gauge couplings , whereas the coupling to two higgs or goldstones behaves like @xmath261 . in warped spaces we can calculate these couplings and obtain @xmath262 for fermions and @xmath263 ( in the limit @xmath264 ) for higgs / goldstones . finally , if supersymmetry is mediated to the matter sector by a @xmath1-term of a @xmath12 gauge group , we must have , as explained above , an extra gauge boson @xmath15 with a mass around the tev scale . this @xmath15 boson must universally decay to leptons , which can be easily detected at the lhc . we have shown that when supersymmetry is broken at a low scale ( @xmath265 tev ) , the higgs sector of the mssm can undergo drastic modification . the higgs can now mix with the resonances of the strong sector giving rise to supersymmetry - breaking contributions to the quartic coupling of the higgs potential . this can raise the higgs mass not only above the tree - level mssm value , @xmath9 ( without depending upon large radiative corrections as in ordinary mssm models ) , but also for values significantly larger than @xmath0 gev . in this hybrid scenario the mixing parameter interpolates between the two extremes corresponding to an elementary mssm higgs and a strongly - coupled higgs resonance , with the degree of compositeness of the higgs increasing as the higgs becomes heavier . in this way the higgs mass can range all the way up to the tev scale , extending the usual higgs mass range allowed in extensions of the mssm such as nmssm models . even though the scale of supersymmetry breaking is low , a sufficiently heavy superparticle spectrum can still be generated to avoid the ever - constraining collider bounds . for instance , gauginos and higgsinos can marry composite states of the strong sector , combining to form a dirac state for which collider constraints are much less stringent . on the other hand the squarks and sleptons obtain masses either by mixing with composite states of the strong sector ( assuming flavor symmetries to avoid fcnc ) or introducing an extra u(1)@xmath266 to mediate supersymmetry - breaking via flavor - independent @xmath1-terms . the direct coupling of the mssm higgs to a strong sector generically leads to deviations in electroweak observables which have been measured at the per - mille level . numerical techniques at strong coupling are limited , but fortunately the ads / cft correspondence can be used to calculate precisely these observables in a class of strongly - coupled gauge theories with large @xmath11 and large thooft coupling . over most of the range of the mixing parameter the contributions to the @xmath91 and @xmath92 parameters are found to be within the 99@xmath106 cl ellipse . the fact that a sizeable degree of compositeness in the mssm higgs sector is consistent with ewpt provides an interesting alternative to the usual mssm . the phenomenological implications of a heavy higgs induced by strong dynamics include the possibility of observing the higgs boson decay via the golden channels ( @xmath13 ) . in addition since a light gravitino is a generic prediction of low - scale supersymmetry breaking a heavy higgs can decay to a gravitino and higgsino with a sizeable branching fraction . for the other mssm higgses the strong - coupling corrections to the mssm relation @xmath267 can be large enough to allow for the decay @xmath268 providing a distinctive signal of our scenario . other exotic possibilities include an mssm higgs sector with only one higgs doublet if the matter sector mixes with the strong sector , obviating the need for a second higgs doublet . alternatively electroweak symmetry may actually be broken by the strong sector itself . in this case the vev of @xmath4 is induced by a tadpole and large mixing parameters are not necessary . this possibility may occur with two higgs doublets , effectively becoming a three higgs doublet model . clearly the mssm higgs sector can be drastically altered if it directly couples to the strong sector , providing a number of novel higgs signatures at the lhc . other collider signals can arise from monojets or single photons with missing transverse momentum associated with dimension-8 operators which couple the goldstino to sm fields ( assuming heavy superpartners ) . furthermore the observation of resonances of the strong sector with quantum numbers of the sm gauge bosons , dirac gaugino states or a tev - scale mass @xmath15 gauge boson associated with the @xmath12 gauge symmetry are also predictions of our scenario . the possibility that both supersymmetry and supersymmetry - breaking strong dynamics are at the tev scale is an intriguing option . it gives rise to a rich phenomenology that will be fully explored at the lhc . @xmath269 * note added : * on december 13 , 2011 the atlas and cms collaborations updated the higgs mass exclusion limits based on the full 2011 data . in particular , the cms collaboration now excludes a sm - like higgs boson at the 95@xmath106 cl in the mass range @xmath270 gev @xmath271 gev . the models presented here can easily satisfy this constraint . for example , the higgs boson can be lighter ( heavier ) than 127 ( 600 ) gev if the higgs has a small ( large ) amount of mixing with the strong sector . alternatively , the higgs bosons in our model could have escaped detection due to a reduction of the event rates . this is possible because in our model the higgs bosons can have couplings to @xmath13 and to the top quark which are smaller than in the sm , or the higgs boson can have an appreciable decay width into supersymmetric particles . in addition the atlas collaboration reported a tantalising excess near 125 gev . if we assume that this is due to a sm - like higgs boson , then in our setup this can be easily achieved by requiring , for example , @xmath272 with @xmath273 , @xmath274 , and @xmath275 . this includes radiative corrections from a stop mass of 500 gev that arises from choosing @xmath276 and @xmath277 . however there are @xmath278 lhc mass limits on the gluinos and squarks that need to be avoided , and the simplest possibility to satisfy these constraints is to not assume @xmath162-parity . the work of tg is supported by the australian research council . tg thanks the ifae , barcelona and the sitp at stanford for support and hospitality during the completion of this work . this material is based upon work supported in part by the national science foundation under grant no.1066293 and the hospitality of the aspen center for physics . the work of ap was partly supported by cicyt - feder - fpa2008 - 01430 , 2009sgr894 and icrea academia program . ap thanks the kitp at santa barbara for hospitality . this research was supported in part by the national science foundation under grant no . nsf phy05 - 51164 a quantitative analysis of the strong dynamics can be performed using the ads / cft correspondence where a strongly - coupled four - dimensional ( 4d ) gauge theory in the large @xmath11 and large t hooft coupling limit is holographically related to a gravity theory in a 5d warped space . for our purposes we will simply assume a slice of ads@xmath167 with constant radius of curvature @xmath279 and 5d cutoff scale @xmath280 @xcite . the 5th dimension is labelled by @xmath281 , with a uv ( ir ) brane located at the boundaries , @xmath282 . the warp factor between the two boundaries , @xmath283 is chosen to explain the hierarchy between @xmath284 tev and the planck scale . this corresponds to a 4d cft with a planck scale uv cutoff and whose conformal symmetry is spontaneously broken at @xmath62 . the bulk contains the sm gauge group with 5d gauge couplings @xmath285 @xmath286 . the 4d gauge couplings @xmath39 are related to the bulk couplings via the relation @xmath287 , where @xmath288 are boundary gauge couplings . by comparing this equation with eq . ( [ rung ] ) we can identify @xmath289 we can choose @xmath47 by a redefinition of @xmath11 . in this case , with @xmath290 and the condition @xmath291 leads to the bound @xmath164 . the mass of the first kaluza - klein ( kk ) gauge field can be identified with the scale @xmath292 and is related to the ir scale @xmath293 via @xmath294 furthermore , to implement supersymmetry breaking at the scale @xmath17 , we will assume that the 5d bulk and the uv - boundary are supersymmetric , with supersymmetry broken by spurion fields @xmath52 localized on the ir - boundary @xcite . we can use the 5d gravity description to compute the quartic coupling of an elementary higgs field coupled to the cft . in the limit that the 4d higgs vev @xmath295 , it is a good approximation to consider the higgs as the zero - mode of a bulk scalar field @xmath240 , with bulk mass @xmath296 and boundary mass @xmath297 ( in units of @xmath298 ) . a massless zero mode , @xmath299 is obtained by a supersymmetric relation , @xmath300 between the bulk and boundary masses @xcite . the dimension of the operator @xmath301 dual to the 5d bulk higgs is dim@xmath302 = 2+|\delta_h-2|$ ] . using this relation , we can compute the dimension , @xmath303 of the higgs coupling to the strong sector . in the case of interest @xmath304 , corresponding to @xmath305 < 3 $ ] ( or @xmath306 ) , we can identify _ h= . [ epsdefn ] furthermore with @xmath306 , the zero - mode behaviour ranges from @xmath307 ( flat profile ) to @xmath308 ( fully composite higgs ) , corresponding to @xmath309 . our goal is to calculate the higgs quartic coupling induced from the non - supersymmetric ir - boundary term @xmath310 ( we are neglecting the bulk @xmath1-term contribution ) . this leads to the zero - mode quartic coupling @xmath311 where using ( [ epsdefn ] ) we have @xmath312 . comparing the quartic coupling of the zero mode , @xmath313 and the first excited kk state , @xmath314 gives rise to the prediction ( _ h -1)^2 ^20.92 _ h^4 , [ ratioquartic ] where @xmath315 is the mass of the first scalar resonance that is approximately given by ( [ lamdefn ] ) . note also that with the above identifications the calculation of @xmath96 in ( [ stest0 ] ) and ( [ stest ] ) using ref . @xcite leads to the results of eq . ( [ csct ] ) . alternatively , instead of directly computing the properties of the elementary higgs field , we can invoke the complementary viewpoint considered in section [ alvi ] and treat the higgs field as arising from the cft . in this way the usual ads / cft dictionary can be used , where the higgs operator of the cft is modeled by a bulk higgs field @xmath240 . electroweak symmetry will then be broken by considering a generic potential on the ir boundary : @xmath316 . the setup is similar to the gaugephobic higgs @xcite . the 5d vev is given by @xmath317^{1/2 } v z_{ir}\left(\frac{z}{z_{ir}}\right)^{\delta_h}~ , \label{v5vev}\ ] ] where the parameter @xmath318 has been eliminated in favour of the 4d higgs vev @xmath319 and we have again neglected bulk @xmath1-term contributions . however to mimic the elementary higgs we are interested in the case when @xmath320 < 2 $ ] , where now dim@xmath302 = \delta_h$ ] . this is the alternative cft interpretation that arises when @xmath306 @xcite . the physical higgs mass , @xmath321 and couplings can be obtained by considering fluctuations about the 5d vev of eq . ( [ v5vev ] ) and deriving the effective 4d action . the quartic coupling is again found to be the same as eq . ( [ quartic0 ] ) except that now we have the identification _ h^2 = dim[o_h ] -1 . similarly the zero - mode quartic coupling agrees with the result eq . ( [ ratioquartic ] ) . in the limit @xmath322 the higgs mass is found to be @xmath323 consistent with the expression given in ( [ mhiggs ] ) . dirac gaugino masses are generated naturally in a warped extra dimension by imposing dirichlet boundary conditions on the ir brane ( assuming neumann boundary conditions on the uv brane ) . generalizing the procedure in ref . @xcite by including boundary terms we obtain @xmath324 where in the second expression we have used eq . ( [ nrelation ] ) and eq . ( [ lamdefn ] ) . compared to the expression eq . ( [ mdirac ] ) , ads / cft predicts @xmath325 . in the case that squark and slepton soft masses are generated from @xmath1-terms , we need an extra u(1)@xmath266 in the bulk . the simplest way to break the u(1)@xmath266 symmetry is to impose dirichlet boundary conditions on the ir brane ( assuming neumann boundary conditions on the uv brane ) . in this case the gauge boson mass is similar to that obtained for the gaugino eq . ( [ gauginomass ] ) . however , this value of the gauge boson mass is in conflict with experimental constraints unless @xmath326 tev . alternatively we can suppose that the gauge symmetry is broken at the ir cutoff scale @xmath327 so that the gauge boson mass @xmath328 . this scale can be related to @xmath292 by estimating when bulk perturbation theory breaks down @xmath329 . this leads to the condition @xmath330 where we have again used eq . ( [ nrelation ] ) and eq . ( [ lamdefn ] ) . since @xmath331 is larger , the experimental constraints on the @xmath12 gauge boson mass can now be avoided . if we allow fermions to propagate in the bulk then the exchange of kk states lead to vertex corrections for the gauge couplings . modifying the result in @xcite to include an arbitrary bulk higgs vev , we obtain the corrections to the @xmath207 and @xmath58 couplings arising from the kk gauge bosons ( neglecting hypercharge effects ) @xmath332 where @xmath333 is the 5d bulk mass of the fermion , @xmath334 , and we have assumed @xmath335 and @xmath336 . there are also corrections coming from the kk fermions that are more model dependent @xcite . comparing to the result ( [ couplingratio ] ) we find that @xmath337 . j. m. maldacena , adv . * 2 * ( 1998 ) 231 [ arxiv : hep - th/9711200 ] . s. samuel , nucl . b * 347 * ( 1990 ) 625 ; m. dine , a. kagan and s. samuel , phys . b * 243 * ( 1990 ) 250 . c. d. carone , j. erlich and j. a. tan , phys . d * 75 * ( 2007 ) 075005 [ arxiv : hep - ph/0612242 ] ; c. d. carone and r. primulando , phys . rev . d * 82 * ( 2010 ) 015003 [ arxiv:1003.4720 [ hep - ph ] ] . t. gherghetta and a. pomarol , nucl . b * 602 * ( 2001 ) 3 [ arxiv : hep - ph/0012378 ] . l. j. hall , y. nomura , t. okui and s. j. oliver , nucl . b * 677 * ( 2004 ) 87 [ arxiv : hep - th/0302192 ] . n. craig , d. stolarski , j. thaler , [ arxiv:1106.2164 [ hep - ph ] ] ; c. csaki , y. shirman , j. terning , [ arxiv:1106.3074 [ hep - ph ] ] . a. brignole , j. a. casas , j. r. espinosa , i. navarro , nucl . * b666 * ( 2003 ) 105 [ arxiv : hep - ph/0301121 ] ; m. carena , k. kong , e. ponton , j. zurita , phys . rev . * d81 * ( 2010 ) 015001 [ arxiv:0909.5434 [ hep - ph ] ] ; i. antoniadis , e. dudas , d. m. ghilencea and p. tziveloglou , nucl . b * 848 * ( 2011 ) 1 [ arxiv:1012.5310 [ hep - ph ] ] . m. ibe , a. rajaraman and z. surujon , arxiv:1012.5099 [ hep - ph ] . r. davies , j. march - russell and m. mccullough , jhep * 1104 * ( 2011 ) 108 [ arxiv:1103.1647 [ hep - ph ] ] . s. mukhopadhyay , b. mukhopadhyaya and s. sengupta , arxiv:1103.3678 [ hep - ph ] . a. azatov , j. galloway and m. a. luty , arxiv:1106.3346 [ hep - ph ] ; arxiv:1106.4815 [ hep - ph ] . g. t hooft , nucl . b * 72 * ( 1974 ) 461 . m. e. peskin and t. takeuchi , phys . * 65 * ( 1990 ) 964 . r. barbieri , a. pomarol , r. rattazzi and a. strumia , nucl . b * 703 * ( 2004 ) 127 [ arxiv : hep - ph/0405040 ] . m. round , phys . d * 82 * ( 2010 ) 053002 [ arxiv:1003.2933 [ hep - ph ] ] ; j. a. cabrer , g. von gersdorff and m. quiros , phys . b * 697 * ( 2011 ) 208 [ arxiv:1011.2205 [ hep - ph ] ] . m. a. luty and t. okui , jhep * 0609 * ( 2006 ) 070 [ arxiv : hep - ph/0409274 ] . r. contino and a. pomarol , jhep * 0411 * ( 2004 ) 058 [ arxiv : hep - th/0406257 ] . see , for example , k. benakli , arxiv:1106.1649 [ hep - ph ] . see also z. chacko , m. a. luty and e. ponton , jhep * 0007 * ( 2000 ) 036 [ arxiv : hep - ph/9909248 ] ; m. nardecchia , a. romanino and r. ziegler , jhep * 1003 * ( 2010 ) 024 [ arxiv:0912.5482 [ hep - ph ] ] . m. bona _ et al . _ [ utfit collaboration ] , jhep * 0803 * ( 2008 ) 049 [ arxiv:0707.0636 [ hep - ph ] ] . v. khachatryan _ et al . _ [ cms collaboration ] , phys . rev . * 106 * ( 2011 ) 201804 . [ arxiv:1102.2020 [ hep - ex ] ] ; g. aad _ et al . _ [ atlas collaboration ] , new j. phys . * 13 * ( 2011 ) 053044 [ arxiv:1103.3864 [ hep - ex ] ] ; g. f. giudice , c. grojean , a. pomarol and r. rattazzi , jhep * 0706 * ( 2007 ) 045 [ arxiv : hep - ph/0703164 ] . a. djouadi and m. drees , phys . b * 407 * ( 1997 ) 243 [ arxiv : hep - ph/9703452 ] . s. y. choi , m. drees , a. freitas and p. m. zerwas , phys . d * 78 * ( 2008 ) 095007 [ arxiv:0808.2410 [ hep - ph ] ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 105 * ( 2010 ) 191802 [ arxiv:1007.4835 [ hep - ex ] ] . a. brignole , f. feruglio , m. l. mangano and f. zwirner , nucl . b * 526 * ( 1998 ) 136 [ erratum - ibid . b * 582 * ( 2000 ) 759 ] [ arxiv : hep - ph/9801329 ] . l. randall and r. sundrum , phys . * 83 * ( 1999 ) 3370 [ arxiv : hep - ph/9905221 ] . t. gherghetta and a. pomarol , nucl . b * 586 * ( 2000 ) 141 [ arxiv : hep - ph/0003129 ] . g. cacciapaglia , c. csaki , g. marandella and j. terning , jhep * 0702 * ( 2007 ) 036 [ arxiv : hep - ph/0611358 ] . i. r. klebanov and e. witten , nucl . b * 556 * ( 1999 ) 89 [ arxiv : hep - th/9905104 ] . k. agashe , r. contino , l. da rold and a. pomarol , phys . b * 641 * ( 2006 ) 62 [ arxiv : hep - ph/0605341 ] .

a recently developed quantum monte carlo approach to the holstein model with one electron [ prb * 69 * , 024301 ( 2004 ) ] is extended to two and three dimensional lattices . a moderate sign problem occurs , which is found to diminish with increasing system size in all dimensions , and not to affect simulations significantly . we present an extensive study of the influence of temperature , system size , dimensionality and model parameters on the small - polaron cross over . results are extrapolated to remove the error due to the trotter discretization , which significantly improves the accuracy . comparison with existing work and other quantum monte carlo methods is made . the method can be extended to the many - electron case .

cond-mat0401634

in recent years , a lot of experimental evidence has been given for the importance of electron - phonon interaction in strongly correlated systems such as the cuprates@xcite or the manganites.@xcite although considerable theoretical progress has been made in understanding and describing many details of the physics of these compounds , the quantum nature of the phonons has often been neglected in actual calculations.@xcite however , a quantum - mechanical treatment of the lattice degrees of freedom has been shown to give significantly different results in certain parameter regimes.@xcite due to the complexity of the models for the abovementioned classes of materials , numerical methods have been used extensively . one very powerful approach is the quantum monte carlo ( qmc ) method , which allows for simulations on relatively large lattices and gives quasiexact results ( i.e. , exact apart from statistical errors which can , in principle , be made arbitrarily small ) also at finite temperature . the latter point represents a major advantage over , e.g. , the density matrix renormalization group ( dmrg ) or variational methods which are restricted to the calculation of ground - state properties since fascinating phenomena such as high - temperature superconductivity and colossal magnetoresistance can be investigated . exact diagonalization ( ed ) can be performed at finite temperatures ( see , e.g. , ref . ) , but for the case of electron - phonon systems such calculations would be restricted to very small systems and unrealistic parameters . consequently , all existing work based on ed is for zero temperature only . additionally , for coupled electron - phonon systems , the infinite - dimensional hilbert space associated with the boson degrees of freedom represents a substantial difficulty for ed and dmrg , in contrast to qmc . nevertheless , qmc methods are often limited by ( 1 ) the minus - sign problem , which restricts simulations to high temperatures and/or small systems , ( 2 ) the fact that the calculation of dynamical properties , such as the one - electron green function , requires analytic continuation to the real - time axis which is an ill - posed problem , and ( 3 ) by strong autocorrelations and large statistical errors . in a recent paper,@xcite from here on referred to as i , we have proposed a new qmc approach to the holstein model , which is based on the canonical lang - firsov transformation@xcite and a principal component representation of the phonon degrees of freedom . the resulting algorithm has been shown to completely avoid the problem of autocorrelations , strongly reduce statistical errors and to allow for very efficient simulations.@xcite the work here and in i is restricted to the case of a single electron which is often called the `` _ holstein polaron problem _ '' . obviously , in connection with materials such as the cuprates or manganites , we are interested in studying the many - electron system . however , the one - electron case represents an important first step towards an understanding of more general situations . moreover , due to the coupling to the phonons , even the model with one electron constitutes a complex many - body problem which has received a lot of attention in the past . an important feature of our approach is the fact that it can be generalized to the more realistic many - electron case encountered in transition metal oxides . in this paper , we extend the qmc method proposed in i to square ( cubic ) lattices in two ( three ) dimensions . although the one - dimensional case considered in i is of great theoretical interest , real strongly correlated materials usually require two or three dimensional models . consequently , before turning to the more challenging many - electron case , it is crucial to examine the influence of dimensionality on our algorithm . to this end , we investigate the dependence of the minus - sign problem , which has been found to appear in simulations of the lang - firsov transformed model , on the dimensionality d of the lattice , as well as on the parameters of the model . we then use our method to study in detail the well - known small - polaron cross over , which occurs with increasing electron - phonon coupling strength , and its dependence on phonon frequency , temperature , system size and dimensionality . we remove the trotter error , which results from the discretization of imaginary time , by extrapolating our results to the limit of continuous time , thereby significantly improving the accuracy of the data . the paper is organized as follows . in sec . [ sec : holstein ] we review the physics of the holstein model with one electron , and in sec . [ sec : qmc ] we briefly discuss the qmc method introduced in i. results are presented in sec . [ sec : results ] and sec . [ sec : summary ] contains our conclusions . the holstein hamiltonian@xcite with dimensionless phonon variables reads@xcite @xmath0 where we have introduced the abbreviations @xmath1 , @xmath2 and @xmath3 for the kinetic , potential and interaction terms , respectively . spin indices have been suppressed in the notation , owing to the spin symmetry of the one - electron problem . in eq . ( [ eq : holstein ] ) , @xmath4 ( @xmath5 ) creates ( annihilates ) a spinless electron at lattice site @xmath6 , @xmath7 and @xmath8 denote the displacement and momentum operator of a harmonic oscillator at site @xmath6 , and @xmath9 . the coupling term @xmath3 describes the local interaction of the single electron considered here with dispersionless einstein phonons . in the first term , the symbol @xmath10 denotes a summation over all hopping processes @xmath11 and @xmath12 between neighboring lattice sites @xmath13 . the parameters of the model are the hopping integral @xmath14 , the phonon energy @xmath15 ( @xmath16 ) , and the electron - phonon coupling constant @xmath17 . as in i , we define the dimensionless coupling constant @xmath18 , where @xmath19 is the bare bandwidth in d dimensions . we shall also use the dimensionless phonon frequency @xmath20 , also called the `` _ adiabatic ratio _ '' , and express all energies in units of @xmath14 . consequently , the model can be described by two independent parameters , @xmath21 and @xmath22 . while the bulk of results in this paper has been calculated assuming periodic boundary conditions in real space , we shall also discuss the effect of the boundary conditions on the sign problem in sec . [ sec : sign - problem ] . since the literature on the holstein polaron problem is vast , we restrict our discussion to work in more than one dimension . a brief review of the one - dimensional model can be found in i. moreover , we will focus on recent progress in the field , and on numerical methods . a comprehensive overview of earlier analytical work can be found , e.g. , in the books of alexandrov and mott@xcite and mahan.@xcite the holstein polaron in @xmath23 has been studied using a large variety of numerical techniques . in contrast to many perturbative approaches,@xcite the latter can also accurately describe the physically most interesting regime of small but finite phonon frequency ( @xmath24 ) , and intermediate electron - phonon coupling ( @xmath25 ) . much information about the holstein polaron has been obtained using qmc . de raedt and lagendijk@xcite applied feynman s path - integral technique to the lattice problem of eq . ( [ eq : holstein ] ) . in this approach , the phonon degrees of freedom are integrated out analytically , and one is left with a single - particle system with a retarded self - interaction , which can be simulated using the monte carlo method . the only approximation consists of discretizing the imaginary time using the suzuki - trotter decomposition.@xcite as the phonons have been completely eliminated from the problem , very efficient simulations on large lattices even in higher dimensions are possible . moreover , the method is not restricted to the holstein hamiltonian ( [ eq : holstein ] ) . it can be extended to include long - range electron - phonon coupling as well as dispersive phonons.@xcite kornilovitch@xcite later employed the same method to study the whole range of values of the adiabatic ratio @xmath21 , and extrapolated the results to remove the error due to the trotter approximation . he also pointed out the connection of the algorithm to world - line methods.@xcite extending this work , kornilovitch developed a qmc approach which is formulated in continuous imaginary time , and which is capable of directly measuring the polaron band dispersion @xmath26 and the density of states in one to three dimensions , by sampling world lines with open boundary conditions in imaginary time.@xcite although the method gives very accurate results and can also be applied to more general models with , e.g. , long - range interaction , it is limited to the regime of intermediate and strong electron - phonon coupling as well as @xmath27 by a minus - sign problem.@xcite consequently , it is not as universally applicable as the algorithm of de raedt and lagendijk.@xcite as the above discussion reveals , the qmc methods of refs . are very well suited to study the one - electron problem . moreover , de raedt and lagendijk also extended their approach to the bipolaron problem of two electrons with opposite spins.@xcite however , as pointed out by kornilovitch,@xcite these are all world - line methods . consequently , despite the possibility of integrating out the phonon degrees of freedom even in the many - electron case , they face a serious sign problem in more than one dimension , for two or more fermions of the same spin , and can therefore not be used to study many - particle systems . in the context of superconductivity , the holstein and the holstein - hubbard model with many electrons have been investigated@xcite using the grand - canonical determinant method of blankenbecler _ et al._@xcite ( see also i ) . however , as discussed in i , due to strong autocorrelations , these simulations were restricted to rather large phonon frequencies @xmath28 , while , e.g. , the cuprates and the manganites fall into the adiabatic regime @xmath29 . this is exactly the point where our new approach enters . as we shall see below , it completely avoids the problem of autocorrelations . apart from qmc , several other methods have been successfully applied to the holstein polaron . this includes ed in combination with a truncation of the phonon hilbert space,@xcite finite - cluster strong coupling perturbation theory ( scpt),@xcite cluster perturbation theory ( cpt),@xcite dmrg,@xcite and a variational diagonalization method.@xcite as pointed out before , standard ed is restricted to rather small numbers of lattice sites and/or phonon states , while dmrg@xcite and the variational method of ref . are applicable over a wider range of values of phonon frequency and electron - phonon coupling , and are much less influenced by finite - size effects . the same is true for scpt and cpt , which exactly diagonalize small clusters for which enough phonon states can be included in the calculation and extrapolate the results to the thermodynamic limit by treating the rest of the system as a perturbation.@xcite nevertheless , as mentioned before , all this work was limited to zero temperature . finally , the holstein polaron has been investigated recently using weak- and strong - coupling perturbation theory,@xcite and a variational wave function which is a superposition of bloch states for the weak- and the strong - coupling regime.@xcite from all this work , many properties of the holstein polaron are well understood . similar to one dimension,@xcite there is a cross over from a quasiparticle with slightly increased effective mass to a heavy small polaron as the electron - phonon coupling strength increases . as pointed out by fehske _ et al._@xcite and capone _ et al._,@xcite the two conditions for the existence of a small polaron are @xmath30 and @xmath31 . we will first discuss the weak - coupling state . concerning its nature in higher dimensions , there exist two different views . from calculations based on the adiabatic approximation , i.e. , taking the limit @xmath32 , one expects a qualitatively different behavior in one dimension compared to @xmath23 , also for @xmath33 . et al._@xcite distinguish between the adiabatic ( @xmath34 ) and the nonadiabatic ( @xmath35 ) regime . in the adiabatic case , and for @xmath23 , the electron is expected to remain quasifree with an almost unchanged effective mass and kinetic energy for @xmath36 , corresponding to a quasiparticle with infinite radius . this contrasts strongly with the one - dimensional case , in which the electron is always self - trapped by the surrounding lattice distortion and forms a polaron with finite radius for any @xmath37 . on the other hand , in the nonadiabatic regime @xmath35 , the behavior is very similar in all dimensions , and a very gradual decrease of the kinetic energy ( or increase in effective mass ) is observed as the coupling strength increases . due to the large energy of the phonon excitations , only the zero - phonon state contributes significantly . moreover , the phonons are fast and react almost instantly to the motion of the electron . consequently , a lattice distortion only persists in the immediate vicinity of the electron , and this rather small quasiparticle is sometimes called a `` _ nonadiabatic lang - firsov polaron_''.@xcite romero _ et al._@xcite take on a slightly different viewpoint . they argue that the large polaron state is essentially the same in any dimension , and that the only effect of increasing the dimension of the system is given by the observed sharpening of the transition to a small polaron in @xmath23 compared to 1d . for @xmath22 larger than a critical value @xmath38 , determined by the aforementioned conditions @xmath30 and @xmath31 , where the cross over occurs , both views agree on the existence of a a small - sized so - called `` _ holstein polaron_''.@xcite the latter is a heavy quasiparticle with a strongly reduced mobility . the cross over at @xmath38 is very sharp , especially for @xmath29 , but it does not represent a real phase transition.@xcite even though the electron is trapped in the potential well originating from the response of the lattice to its motion , the ground state is still bloch - like . for simplicity , in the remainder of this paper , we shall always refer to the weak - coupling state as a large polaron , either with finite or infinite radius , depending on which of the abovementioned viewpoints one holds . in our opinion , a definite decision about which of the two alternatives is correct can not be made using qmc , which is restricted to finite clusters and , more importantly , finite temperatures . as a consequence , there will always be a contribution from excited states , making it difficult to reveal the true nature of the ground state in the weak - coupling regime . since the extension of the qmc algorithm to higher dimensions is straight forward , here we shall merely give an overview of the method which has been discussed in detail in i. the cornerstone of the new approach is the canonical lang - firsov transformation,@xcite which separates the polaron effects , due to the electron - phonon interaction , from the zero - point fluctuations of the harmonic oscillators in eq . ( [ eq : holstein ] ) . the transformed model with one electron takes the form@xcite @xmath39 with @xmath2 as defined in eq . ( [ eq : holstein ] ) , and the polaron binding energy @xmath40 . the parameter @xmath41 , which corresponds to the displacement of the harmonic oscillator in the presence of an electron,@xcite is given by @xmath42 . the method employs a trotter decomposition of the imaginary time axis into @xmath43 intervals of size @xmath44 , where @xmath45 is the inverse temperature . the partition function , obtained by integrating out the phonon coordinates,@xcite is given by @xmath46 where @xmath47 denotes the @xmath48 dimensional integral over all phonon momenta @xmath49 , and @xmath50 is the linear size of the lattice in d dimensions . the bosonic weight @xmath51 is defined as @xmath52 with the bosonic action @xmath53 @xmath54 , and a tridiagonal @xmath55 matrix @xmath56 with nonzero elements @xmath57 and periodic boundary conditions @xmath58 . as discussed in i , the electronic weight @xmath59 is given by @xmath60 here @xmath61 is @xmath62 with the phonon operators @xmath8 replaced by the values @xmath63 on the @xmath64th trotter slice . the exponential of the hopping term can be written as @xmath65 where @xmath66 is the @xmath67 tight - binding hopping matrix for the lattice under consideration . in fact , this is the only nontrivial change compared to the one - dimensional case . as pointed out in i , for a single electron , the fermionic trace can easily be calculated from the matrix representation of @xmath68 as @xmath69 . due to the complex - valued hopping term , @xmath59 is not strictly positive , which gives rise to the minus - sign problem discussed below . we would like to mention that , in contrast to some determinant qmc methods,@xcite the @xmath43-fold matrix product involved in the calculation of the matrix @xmath68 is well conditioned also for large systems at low temperatures , so that a time - consuming numerical stabilization is not necessary . in i , we have introduced the so - called principal component representation for the phonon degrees of freedom . in terms of the latter , the bosonic weight takes the simple gaussian form @xmath70 with the _ principal components _ @xmath71 . it is strictly positive . finally , the efficiency of the qmc algorithm can be greatly improved by the use of a reweighting of the probability distribution . in our case , this amounts to transferring all the influence of the electronic degrees of freedom which are treated exactly to the observables calculated via @xmath72 with the expectation values with respect to the bosonic weight @xmath51 defined as @xmath73 as outlined in i , the expectation values defined in eq . ( [ eq : ob ] ) can be determined during the qmc simulation . the applicability of the reweighting method has been discussed in detail in i , and here we only mention that we have carefully checked that it also works reliably in higher dimensions . this could be expected , since the physics of the holstein polaron is rather similar in all dimensions ( see sec . [ sec : holstein ] ) . details about the calculation of observables within our formalism can be found in i. due to the analytic integration over the phonon coordinates @xmath74 used here , interesting observables such as the correlation functions @xmath75 are difficult to measure accurately . other quantities such as the quasiparticle weight , and the closely related effective mass,@xcite can be determined from the one - electron green function at long imaginary times,@xcite but results but would not be as accurate as existing work on the one - electron case considered here ( see , e.g. , refs . , and ) . therefore , we have restricted ourselves to the kinetic energy of the electron , which contains a lot of information about the small - polaron cross over . for the more demanding many - electron case , to which our method can be extended ( see sec . [ sec : summary ] ) , other methods produce far less reliable data . we are now in a position to discuss the actual qmc procedure which has been explained more thoroughly in i. from the above discussion and i , it is obvious that we only simulate the phonons , while the electronic degrees of freedom have been integrated out analytically . this is equivalent to the method proposed by blankenbecler _ et al._@xcite for the many - electron case . however , here the monte carlo sampling is independent of the fermionic weight @xmath59 given by eq . ( [ eq : wf ] ) , with the latter entering the calculations only via the reweighting procedure [ see eq . ( [ eq : reweighting ] ) ] . thereby , we avoid the problem of nonpositive weights during the qmc updates , since we do not use @xmath59 as a probability for accepting or rejecting new phonon configurations . nevertheless , the sign problem still manifests itself in terms of statistical errors , as can be seen from eq . ( [ eq : reweighting ] ) . similar to other occurrences of the minus - sign problem , e.g. , for the case of the hubbard model away from half filling,@xcite simulations would become very difficult if @xmath76 should tend to zero . owing to the gaussian form of the bosonic action , when written in terms of the principal components @xmath77 , the latter can be sampled exactly by drawing random numbers from a normal distribution . in contrast to usual markov chain monte carlo simulations , every new configuration is accepted , and no autocorrelations between successive values of the @xmath77 exist . this is in strong contrast to conventional qmc methods for the holstein model , also with many electrons ( see discussion in i ) . except for situations in which the phonons are integrated out analytically , simulations become extremely difficult at low temperatures and for small phonon frequencies due to strongly increasing autocorrelations . we regard the complete absence of such correlations as a major advantage of our method . after the @xmath77 have been updated for each site of the @xmath48 space - time lattice , a transformation back to the momenta @xmath49 is performed using the matrix @xmath78 . then , for each observable of interest , @xmath79 as well as the fermionic weight @xmath59 are calculated for the current phonon configuration , after which the next update can begin . at the end of the program , results for observables are obtained according to eq . ( [ eq : reweighting ] ) ( see also i ) . before we come to a discussion of the sign problem in the approach presented here , we would like to give a quick review of its occurrence in other qmc methods . the situation is best illustrated for the case of the world - line algorithm ( see , e.g. , ref . ) . the use of the latter to simulate systems of interacting fermions is restricted to one dimension by the fermi statistics of the electrons . this is a consequence of the negative matrix elements @xmath80 , which appear when two fermion world lines wind around each other one or more times as they traverse the space - time lattice . with other methods , despite the occurrence of the sign problem , simulations can still be carried out in many situations . since the qmc sampling requires nonnegative weights , one uses @xmath81 instead of @xmath80 . as a consequence , the sign of the fermionic weight , @xmath82 , is treated as part of the observables . the latter then have to be calculated via @xmath83 and it is obvious that simulations will become extremely difficult if the denominator in eq . ( [ eq : obs_sign ] ) tends to zero . in fact , if a sign problem is present , the number of measurements during a qmc run has to be increased by a factor @xmath84 to obtain results of the same accuracy.@xcite as discussed in ref . , using @xmath81 instead of @xmath80 corresponds to simulating an effective model of hard - core bosons . the average sign can be written as @xmath85 where @xmath86 and @xmath87 denote the free energy per site of the fermionic and bosonic model , respectively , and @xmath88 is the volume of the system.@xcite from eq . ( [ eq : sign_free_energy ] ) , it is obvious that @xmath89 decreases exponentially with increasing @xmath90 and @xmath88 . the auxiliary - field method@xcite for the hubbard model faces similar problems . here the weight of a configuration is given by the product of determinants for @xmath91 and @xmath92 electrons , respectively . the product is strictly positive only for half filling , whereas simulations for other particle densities become very demanding at low temperatures and/or for large systems . since at large @xmath93 the determinant is similar to the weight of world lines of hubbard - stratonovitch variables,@xcite this similarity of the dependence of @xmath89 on the parameters of the system is not surprising . finally , there is no sign problem in determinantal grand - canonical simulations of the holstein model at any filling , since the coupling of electrons and phonons is the same for both spin directions.@xcite the sign problem in the current approach clearly has a fundamentally different origin , since there is only a single electron in the system , so that no winding of world lines around each other can take place . in fact , the negative fermionic weights are a result of the complex phase factor in the transformed hopping term [ eq . ( [ eq : htrans ] ) ] . we will see below that , in contrast to other methods , the sign problem in our approach diminishes , i.e. , @xmath94 , with increasing system size , and a detailed investigation of its dependence on the parameters of the model and the choice of boundary conditions will be given in sec . [ sec : results ] . as discussed in i , the error due to the trotter decomposition is proportional to @xmath95 . we perform simulations at different values of @xmath96 , typically @xmath97 , @xmath98 and @xmath99 , and exploit the linear dependence of the results on @xmath95 to extrapolate to @xmath100 . this is a common procedure in the context of discrete - time qmc methods,@xcite and allows one to remove the trotter error if the values of @xmath96 are sufficiently small . moreover , a given accuracy can often be reached using larger values of @xmath96 , compared to calculations which do not use the abovementioned extrapolation , thereby saving computer time . we conclude this section by a comparison of our approach with other qmc methods for the holstein polaron . as mentioned in sec . [ sec : introduction ] , the methods of de raedt and lagendijk@xcite and kornilovitch@xcite are based on an analytic integration over the phonon degrees of freedom . this separation of electronic and bosonic degrees of freedom greatly reduces the statistical noise due to phonon fluctuations , which are induced by the electron - phonon interaction . the fluctuations increase noticeably with decreasing phonon frequency , decreasing temperature and increasing electron - phonon coupling . although the lang - firsov transformation used here performs a very similar task , namely , to separate polaron effects from the zero - point and thermal fluctuations of the free oscillators ( see i ) , the integral over the bosonic degrees of freedom is calculated with monte carlo , thereby leaving us with a residual influence of the phonons . in fact , the numerical effort for calculations with our approach is proportional to the @xmath101 , similar to the grand - canonical determinant method for the holstein model.@xcite this could be improved to a computer time @xmath102 by linearizing the exponential of the hopping term to a tridiagonal matrix , which has not been done in the work presented here . although our algorithm is not as efficient as the methods of refs . , in which the numerical effort is independent of @xmath50 , proportional to @xmath103 and depends only linearly on d , we will see in the following section that it is well suited for accurate simulations on large clusters in one and two dimensions , and on reasonably large clusters in 3d . moreover , as we are interested in developing a qmc method that can also be applied to the many - electron system in the adiabatic regime in order to study , e.g. , quantum phonon effects in the manganites ( see i)such a decrease in performance when compared to the world - line methods@xcite is acceptable , since the latter can not be applied to the many - electron case in more than one dimension . additionally , the advantage of the analytical integration over the phonons is biggest in the one - electron case , in which they represent the majority of the degrees of freedom in the path - integral representation of the partition function.@xcite this is no longer true at finite band filling , where the contributions of fermions and bosons are comparable . this section is divided into two parts . first , in sec . [ sec : sign - problem ] , we investigate in detail the dependence of the aforementioned sign problem on electron - phonon coupling , dimensionality , boundary conditions , system size , phonon frequency , and temperature . in sec . [ sec : polaron ] , we present results for the electronic kinetic energy to study the small - polaron cross over discussed in sec . [ sec : holstein ] . average sign of the fermionic weight @xmath59 as a function of electron - phonon coupling @xmath22 in one dimension for different cluster sizes @xmath50 , with parameters as indicated in the figure . here and in subsequent figures , lines are guides to the eye only , and error bars are smaller than the symbols shown . the data presented in figs . [ fig : sign_1d_n][fig : sign_3d_n ] are for @xmath104 . the inset shows the minimum of @xmath89 as a function of the system size @xmath50 ( see text ) . , scaledwidth=45.0% ] as in fig . [ fig : sign_1d_n ] , but for the case of two - dimensional clusters with ( a ) periodic boundary conditions and ( b ) open boundary conditions in real space . the insets show the minimum of @xmath89 as a function of the linear system size @xmath50.,title="fig:",scaledwidth=45.0% ] + as in fig . [ fig : sign_1d_n ] , but for the case of two - dimensional clusters with ( a ) periodic boundary conditions and ( b ) open boundary conditions in real space . the insets show the minimum of @xmath89 as a function of the linear system size @xmath50.,title="fig:",scaledwidth=45.0% ] dependence of the average sign of @xmath59 on ( a ) phonon frequency @xmath21 and ( b ) inverse temperature @xmath90 on a @xmath105 cluster . the insets show the minimum of @xmath89 as a function of ( a ) phonon frequency @xmath21 , and ( b ) inverse temperature @xmath106 . , title="fig:",scaledwidth=45.0% ] + dependence of the average sign of @xmath59 on ( a ) phonon frequency @xmath21 and ( b ) inverse temperature @xmath90 on a @xmath105 cluster . the insets show the minimum of @xmath89 as a function of ( a ) phonon frequency @xmath21 , and ( b ) inverse temperature @xmath106 . , title="fig:",scaledwidth=45.0% ] dependence of the average sign of @xmath59 on the linear system size @xmath50 in three dimensions . the inset shows the minimum of @xmath89 as a function of the linear system size @xmath50 for two different temperatures.,title="fig:",scaledwidth=45.0% ] + here we are interested in the average sign of the fermionic weight @xmath59 , which is given by @xmath107 where the expectation value @xmath108 has been defined in eq . ( [ eq : ob ] ) . in fig . [ fig : sign_1d_n ] , we show its dependence on electron - phonon coupling and system size . we would like to point out that while the extrapolation to @xmath100 , discussed in sec . [ sec : qmc ] , has been used for the results shown in the following subsection , the calculations for the average sign have been performed for a single , fixed value @xmath104 , for which the trotter error is smaller than statistical errors . some other results for @xmath89 in one dimension have already been reported in i. from the general discussion in sec . [ sec : qmc ] , it is clear that the sign problem encountered in the present approach is of a different nature than in , for example , qmc simulations of the hubbard model . as reported in i , for the holstein polaron problem under consideration , it is most pronounced for _ small _ systems , low temperatures and small phonon frequencies @xmath29 . therefore , the bulk of results presented below will be for such a set of `` worst case '' parameters , including @xmath109 , @xmath110 , and @xmath111 . figure [ fig : sign_1d_n ] shows that , in one dimension , the average sign of @xmath59 in the critical region of intermediate electron - phonon coupling increases quickly as the system size increases from @xmath109 to @xmath112 , which is in strong contrast to eq . ( [ eq : sign_free_energy ] ) . this increase of the minimum as a function of @xmath50 is also shown in the inset of fig . [ fig : sign_1d_n ] . as we have only calculated @xmath89 for a finite number of @xmath22-values , an approximation for the minimum has been determined using a spline interpolation . the minimum of @xmath89 occurs near @xmath113 , where the transition from a large to a small polaron takes place ( see sec . [ sec : holstein ] ) . the increase of statistical errors in this regime as a consequence of the sign problem is similar to the situation encountered with simulations of the untransformed model ( see i ) . however , the use of the transformed model still gives significantly more accurate results for the same number of measurements , in particular for low temperatures and small phonon frequencies . finally , one may be tempted to explain the unusual system - size dependence of the sign problem by ascribing its origin to the periodic boundary conditions in real space . if the latter were indeed the source of the sign problem , the boundary effects would decrease with increasing system size , in accordance with the results of fig . [ fig : sign_1d_n ] . this possibility has been investigated , and we shall see below that the sign problem persists also for the case of open boundary conditions . we now come to the holstein model in two dimensions . in fig . [ fig : sign_2d_n](a ) , we show results for @xmath89 for different lattice sizes , again starting with a very small linear dimension @xmath50 . all other parameters are the same as before , in particular @xmath110 and @xmath111 . obviously , for the smallest cluster size shown , the minimum of the average sign has diminished to a value of approximately 0.1 , so that large numbers of measurements are necessary . however , similar to one dimension , @xmath89 increases with increasing system size , and for the largest system size shown ( @xmath114 ) , we find a rather uncritical minimum value of about 0.5 . the results for open boundary conditions , shown in fig . [ fig : sign_2d_n](b ) , reveal that for small clusters , the average sign increases compared to the case of periodic boundary conditions . however , with increasing system size , @xmath89 quickly converges to the same values , independent of the boundary conditions . this is just what one would expect , since with increasing @xmath50 , the effect of the choice of boundary conditions on the properties of the system diminishes . moreover , we can conclude that the negative weights do not simply result from hopping processes of the electron across the periodic boundaries , since in that case we would expect @xmath115 for open boundary conditions , in contrast to fig . [ fig : sign_2d_n](b ) . similar behavior has been found in one and three dimensions . the influence of the phonon frequency @xmath15 on the average sign is shown in fig . [ fig : sign_2d_par](a ) . clearly , the sign problem is most noticeable for small values of @xmath21 , while it diminishes quickly as we increase the adiabatic ratio ( see inset ) . this is very similar to the small - polaron cross over . as discussed in sec . [ sec : holstein ] , the latter sharpens significantly with decreasing @xmath21 , while an abrupt transition is completely absent in the nonadiabatic regime @xmath35 . moreover , the coupling @xmath38 at which the minimum of @xmath89 occurs increases with @xmath21 , in agreement with the aforementioned small - polaron condition @xmath31 . in fig . [ fig : sign_2d_par](b ) , we report the average sign as a function of @xmath22 , and for different inverse temperatures @xmath90 . again we have taken @xmath109 and @xmath111 , the parameters for which the sign problem is most noticeable . while for @xmath110 , the minimum of @xmath89 lies below 0.1 , the situation is much better already for @xmath116 , as shown in the inset . at even higher temperature @xmath117 , the fermionic weight is always positive so that we have @xmath118 for all @xmath22 . the dependence of the sign problem on temperature is therefore similar to other qmc methods [ see eq . ( [ eq : sign_free_energy ] ) ] , although we do not find a simple exponential relation . finally , we also present results for @xmath89 in three dimensions , for lattices of different linear size @xmath50 , and again for the parameters @xmath111 and @xmath110 as a function of @xmath22 . figure [ fig : sign_3d_n ] reveals that the minimum of @xmath89 in 3d has an even more pronounced form than in two dimensions . the sign problem diminishes slightly as we increase the system size from @xmath109 to @xmath119 . however , accurate simulations in this regime are still quite demanding . as we will see in sec . [ sec : polaron ] , a temperature of @xmath116 is sufficient to study the low - temperature properties of the holstein model . for this higher temperature , and all other parameters unchanged , @xmath89 is close to 1 even for @xmath109 ( see inset in fig . [ fig : sign_3d_n ] ) , so that accurate results can be obtained even for small phonon frequencies in three dimensions ( see sec . [ sec : polaron ] ) . in summary , the investigation of the sign problem has shown that our method works well for a large range of values of phonon frequency , electron - phonon coupling and temperature , as long as the system size is large enough . this contrasts , for example , with the world - line algorithm of kornilovitch,@xcite which is restricted to intermediate and strong coupling , as well as low temperatures and rather large phonon frequencies ( see also sec . [ sec : holstein ] ) , also by a minus - sign problem . although we have investigated the influence of all important parameters on the sign problem , a physical interpretation of its origin has not emerged . nevertheless , it is clear that the negative fermionic weights are a result of the phase factors in the lang - firsov transformed hopping term in eq . ( [ eq : htrans ] ) . this is similar to the sign problem which occurs , e.g. , in simulations of electron - phonon models in an external magnetic field,@xcite but here the phonon fields @xmath120 vary with time ( @xmath121 ) and position ( @xmath6 ) , and are coupled in imaginary time by the bosonic action [ see eqs . ( [ eq : action - w - matrix ] ) and ( [ eq : matrixa ] ) ] . moreover , the dependence of the sign problem on @xmath21 , @xmath22 , @xmath90 and @xmath50 bears striking resemblance to the influence of these parameters on the properties of the holstein polaron . in particular , its reduction with increasing system size may be a consequence of the dilution of the system ( the particle density @xmath122 as @xmath123 , in the one - electron case ) . finally , it is interesting to note that the large statistical fluctuations resulting from the sign problem in the case of the transformed model occur at exactly the same points in parameter space as in the untransformed model . this suggests a strong correlation between the minimum in @xmath89 and the transition to a small polaron , which both occur at or near @xmath38 , similar to simulations of other models , in which the sign problem occurs exactly where the most interesting physics is going on , i.e. , in the vicinity of phase transitions.@xcite normalized kinetic energy @xmath124 [ see eq . ( [ eq : ek ] ) ] as a function of electron - phonon coupling @xmath22 for different values of the phonon frequency @xmath15 on a @xmath125 lattice . the results shown in figs . [ fig : ek_2d_various_w][fig : ek_dimensionality ] have been obtained by extrapolating the qmc data to @xmath100 ( see text).,scaledwidth=45.0% ] the most interesting observable , which is easily accessible with our method ( see discussion in sec . [ sec : algorithm ] ) , and allows us to investigate the small - polaron cross over , is the one - electron kinetic energy @xmath126 , given by the expectation value of the first term in hamiltonian ( [ eq : holstein ] ) . in order to compare results for different dimensions , we define the normalized quantity @xmath127 with @xmath128 for @xmath129 and @xmath130 . due to the large amount of work that has been devoted to the holstein polaron in the past ( see sec . [ sec : holstein ] ) , a lot is known about the transition from a large to a small polaron with increasing electron - phonon coupling . in this work , we therefore concentrate on those aspects which have not been studied in a systematic way so far . to this end , we exploit the advantages of our qmc method which allows us to investigate , in particular , finite - size and finite - temperature effects . the latter have only been touched upon briefly by previous authors employing qmc,@xcite who focused on very large@xcite or infinite systems,@xcite in the ground state@xcite or at two different temperatures.@xcite however , since a large amount of work has been done using ed or other methods based on diagonalization of small clusters , it is essential to study the convergence of the results with increasing system size . additionally , we shall also present a comparison of the kinetic energy in one , two and three dimensions . as pointed out before , the results shown here have been obtained by extrapolating to @xmath100 , thereby eliminating the error due to the trotter discretization . to study the small - polaron cross over , previous authors focused on the effective mass of the electron@xcite and the quasiparticle weight.@xcite unfortunately , as pointed out in sec . [ sec : algorithm ] , these observables can not be calculated directly with our qmc algorithm . nevertheless , we can examine many interesting aspects of the cross over by calculating the kinetic energy of the electron , and comparison will be made to existing work.@xcite we begin with the dependence of the cross over on the phonon frequency . to this end , we present in fig . [ fig : ek_2d_various_w ] results for the kinetic energy calculated for @xmath114 , @xmath110 and different values of @xmath21 . the large range of the adiabatic ratio in fig . [ fig : ek_2d_various_w ] , @xmath131 , shows the ability of our method to give accurate results for almost arbitrary values of the phonon frequency , especially in the adiabatic regime @xmath29 . this is in contrast to the method of kornilovitch,@xcite which is restricted to @xmath132 and @xmath133 by a severe minus - sign problem . in our case , the only limitations regarding the accessible values of @xmath21 are the moderate sign problem , discussed in sec . [ sec : sign - problem ] , which for small systems and low temperatures gives rise to a noticeable increase of statistical errors as @xmath32 , and the increasing trotter error as @xmath134 , which requires the use of more and more time slices ( see discussion in i ) . normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different linear dimensions @xmath50 of the system.,title="fig:",scaledwidth=45.0% ] + normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different linear dimensions @xmath50 of the system.,title="fig:",scaledwidth=45.0% ] normalized kinetic energy @xmath124 as a function of the inverse of the linear size @xmath50 of the system , and for different values of the electron - phonon coupling @xmath22 . as indicated in the legend , some curves have been shifted , in order to allow for a better representation . all curves are monotonic within statistical errors.,title="fig:",scaledwidth=45.0% ] + normalized kinetic energy @xmath124 as a function of the inverse of the linear size @xmath50 of the system , and for different values of the electron - phonon coupling @xmath22 . as indicated in the legend , some curves have been shifted , in order to allow for a better representation . all curves are monotonic within statistical errors.,title="fig:",scaledwidth=45.0% ] normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different inverse temperatures @xmath90.,title="fig:",scaledwidth=45.0% ] + normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different inverse temperatures @xmath90.,title="fig:",scaledwidth=45.0% ] normalized kinetic energy @xmath124 as a function of temperature @xmath135 , and for different values of the electron - phonon coupling @xmath22 . as indicated in the legend , some curves have been shifted , in order to allow for a better representation.,title="fig:",scaledwidth=45.0% ] + normalized kinetic energy @xmath124 as a function of temperature @xmath135 , and for different values of the electron - phonon coupling @xmath22 . as indicated in the legend , some curves have been shifted , in order to allow for a better representation.,title="fig:",scaledwidth=45.0% ] figure [ fig : ek_2d_various_w ] also shows the well known fact that the transition from large to small polaron near @xmath113 sharpens considerably with decreasing phonon frequency . while there is an abrupt decrease in @xmath124 for @xmath111 , the cross over is very smooth for @xmath132 . additionally , we see from fig . [ fig : ek_2d_various_w ] that the cross over position @xmath38 increases with @xmath21 in the nonadiabatic regime . the physics of the transition to a small polaron has been discussed , e.g. , by capone _ et al._@xcite in the adiabatic regime ( @xmath29 ) , the cross over is entirely determined by the balance of kinetic and electron - phonon coupling- or displacement energy , where the latter is given by the polaron binding energy @xmath136 . as soon as the gain in displacement energy outweighs the loss in kinetic energy , the electron localizes in a potential well and forms a polaron . the parameter @xmath22 is defined as the ratio of these two contributions and may be written as @xmath137 ( @xmath138 is the kinetic energy of a free electron at @xmath129 ) . therefore , in the adiabatic regime , the cross over occurs at @xmath139 . with increasing @xmath21 , the lattice energy becomes more and more important , since more energy is required to excite phonons . as a consequence , the distortions of the lattice around the position of the electron giving rise to the large effective mass and low mobility in the small polaron regime are much smaller , and the local oscillators will predominantly be in their ground state . thus , even for @xmath30 , where a trapped state is energetically favored in the adiabatic regime , the electron remains mobile . the decrease of @xmath124 with increasing @xmath22 is a result of the exponentially decreasing overlap of a displaced and an undisplaced harmonic oscillator in its ground state , which reduces the hopping matrix element between neighboring lattice sites . in the nonadiabatic regime , a small polaron is formed if @xmath140 ( see ref . ) , where @xmath41 is the induced lattice distortion at the site of the electron which follows from the lang - firsov transformation.@xcite this is identical to the condition @xmath31 given above . the larger lattice energy also gives rise to the more gradual decrease of @xmath124 for intermediate and large values of @xmath21 . in particular , the kinetic energy is much larger for @xmath30 and @xmath35 than for @xmath34 . finally , like in one dimension,@xcite @xmath124 remains small but finite even at very strong coupling , which is a consequence of the fast and undirected motion of the electron inside the polaron . as pointed out by kornilovitch,@xcite the kinetic energy in the small polaron regime is therefore not related to the effective mass of the electron , the latter exhibiting an exponential decrease as a function of @xmath22 ( refs . , , , and ) . to address the issue of finite - size effects , we have calculated @xmath124 for @xmath110 , @xmath111 and different linear lattice sizes @xmath10912 [ see fig . [ fig : ek_2d_size](a ) ] . the choice of @xmath111 is reasonable since the large polaron , which exists for @xmath141 , is most extended for small phonon frequencies , as discussed in sec . [ sec : holstein ] , so that finite - size effects can be expected to be largest . to illustrate this point , we also present results for a larger phonon frequency @xmath142 [ fig . [ fig : ek_2d_size](b ) ] . in the latter case , the local oscillators can respond very quickly to the motion of the electron , and the extension of the phonon cloud or lattice distortion surrounding the electron is much smaller . since for @xmath35 the transition to a small polaron happens at larger @xmath22 , the polaron will be larger for intermediate values @xmath143 compared to the adiabatic regime , in which @xmath139 . in total , we therefore expect to have smaller finite - size effects for @xmath142 than for @xmath111 as long as we are in the weak - coupling regime @xmath144 , while the opposite should be true for @xmath143 . for @xmath145 , a highly immobile polaron state exists in both cases and results should therefore be virtually independent of @xmath50 . all this is well confirmed by the results shown in fig . [ fig : ek_2d_size ] . figure [ fig : ek_2d_size ] also reveals that the results begin to saturate for @xmath146 , as pointed out previously by kornilovitch.@xcite however , in contrast to the one - dimensional case,@xcite where we performed simulations for @xmath50 as large as 32 , we find a nonnegligible dependence on @xmath50 up to the largest system size ( @xmath114 ) . this is better illustrated in fig . [ fig : ek_finite_size ] , in which we show @xmath124 as a function of @xmath147 , and for several values of @xmath22 . to allow for a better representation , some curves have been shifted , as indicated in the legend . from fig . [ fig : ek_finite_size ] , we see that @xmath124 changes very little for @xmath148 for the case of strong coupling @xmath149 and , in fact , remains constant within the error bars , similar to the 1d results in i. for smaller values of @xmath22 , no such saturation is found on the scale of fig . [ fig : ek_finite_size ] , and the behavior of the kinetic energy for large @xmath50 is almost linear when plotted as a function of @xmath147 . we used a linear fit of the data for @xmath150 , 10 , and 12 to obtain an approximation to the thermodynamic limit . the results are shown in fig . [ fig : ek_2d_size ] . obviously , @xmath124 for @xmath151 has decreased noticeably for small values of @xmath22 ( including @xmath130 ) , while it remains almost unchanged in the small polaron regime . the decrease of @xmath124 for @xmath141 can easily be understood if we consider the fact that our method works at a finite temperature @xmath152 . as a consequence , for very small @xmath50 , the energy gap between the ground state with @xmath153 and the first excited state with @xmath154 is larger than the thermal energy @xmath155 . with increasing system size , thermal population of excited states becomes possible . for @xmath50 large enough ( @xmath156 for @xmath110 and @xmath130 ) , results converge to those for @xmath151 and , in fact , the extrapolated data for @xmath130 , shown in fig . [ fig : ek_finite_size ] , agree well with the results for a free electron on an infinite lattice . we ascribe the smallness of this finite - temperature effect in the strong - coupling regime above @xmath38 to the very small width of the polaron band ( see , e.g. , ref . ) . consequently , the low - energy coherent states with different @xmath157 have very similar energies . figure [ fig : ek_2d_size](a ) also shows that for small phonon frequencies , the influence of the lattice size is largest near @xmath38 . the cross over is sharper for small systems than for the case of large @xmath50 . a similar behavior has been found by marsiglio for the one - dimensional model.@xcite finally , as pointed out by kornilovitch , @xcite despite the good convergence of the results for quantities such as the kinetic energy even for small @xmath158 , other observables , such as the effective mass , rely on the knowledge of the energy of states with infinitely small momenta . therefore , they will show a more pronounced dependence on @xmath50 when calculated on small clusters accessible with , e.g. , ed . we also investigated the effect of temperature on our results , again for @xmath111 and @xmath142 . the results , shown in fig . [ fig : ek_2d_beta ] , indicate that @xmath124 is more affected by the finite temperature of the simulation in the adiabatic case @xmath111 [ fig . [ fig : ek_2d_beta](a ) ] . this is a consequence of the fact that calculations at finite temperatures only give ground - state - like results when @xmath159 . clearly , this condition is much more difficult to meet for @xmath111 , and requires larger values @xmath160 . the changes of @xmath124 with temperature result from an interplay of several effects . for @xmath130 , the kinetic energy approaches its full noninteracting value of @xmath161 ( i.e. , @xmath128 ) as @xmath162 . at finite @xmath163 , however , states with nonzero total quasimomentum @xmath157 will contribute and thereby lead to a decrease of @xmath124 . as discussed above , this effect of temperature on @xmath124 is expected to decrease with increasing @xmath22 , and to be extremely small in the strong - coupling regime . in the adiabatic case [ fig . [ fig : ek_2d_beta](a ) ] , the transition at @xmath139 is smeared out at high temperatures . for both @xmath111 and @xmath142 , a qualitative change in behavior occurs near @xmath38 . @xmath124 decreases with increasing temperature for @xmath141 , whereas the opposite is true for @xmath164 . the behavior above @xmath38 can be understood by considering the electronic hopping amplitude , given by the overlap of the wave functions of a displaced and an undisplaced oscillator at neighboring sites . while the latter is exponentially reduced with increasing electron - phonon coupling at @xmath129 ( see above ) , it increases with temperature since the oscillators can occupy excited states , corresponding to wave functions which are more spread out than the ground state . moreover , the thermal energy allows the electron to overcome the potential barrier more easily . this thermally activated hopping which shows up more clearly for small phonon frequencies since , in this case , phonon excitations require less energy has been studied already a long time ago by holstein.@xcite it is interesting to notice that the dependence of @xmath124 on temperature , as shown in fig . [ fig : ek_2d_beta_scaling ] , is almost linear at low temperatures and for @xmath142 [ see fig . [ fig : ek_2d_beta_scaling](b ) ] , while in the adiabatic case [ fig . [ fig : ek_2d_beta_scaling](a ) ] this is only true for strong electron - phonon coupling . however , as pointed out before , for @xmath111 , an inverse temperature @xmath110 is not sufficient to obtain well - converged results . therefore , a linear dependence , as for @xmath142 , may still be found at lower temperatures . such calculations , with the accuracy of the results presented here , would be more time - consuming . the data for @xmath114 , @xmath165 , @xmath104 , and @xmath166 , i.e. , in the cross over regime where statistical errors are largest ( about 0.5% ) , required about 28 days of cpu time on an intel xeon 2600 mhz computer . however , as has been mentioned previously , the numerical effort may be reduced by a factor of order @xmath167 by linearizing the exponential of the hopping matrix , denoted as @xmath168 ( see sec . [ sec : numerics ] . ) similar to the finite - size scaling performed above , we also linearly extrapolated the data for @xmath110 and 15 to the zero - temperature limit @xmath169 . the results for @xmath142 , for which such a linear scaling is reasonable , are shown in fig . [ fig : ek_2d_beta](b ) . while the general trend agrees well with our expectations based on the finite - temperature data shown , the scaling procedure clearly overestimates the temperature effects , thereby leading to spurious values @xmath170 at @xmath130 . however , this can easily be understood keeping in mind that results saturate at low - enough values of @xmath106 , so that the linear extrapolation used here becomes insufficient . finally , we would like to mention that the results of de raedt and lagendijk@xcite were also given for @xmath116 ( and @xmath171 ) , but the small number of trotter slices ( @xmath172 in their case ) gives rise to relatively large systematic errors.@xcite this is not the case for the results of ref . , in which the same extrapolation to @xmath100 was employed as here . finally , the method of kornilovitch@xcite is free of such errors , but only permits one to calculate ground - state properties for a restricted range of @xmath21 and @xmath22 . normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different linear dimensions @xmath50 of the lattice.,scaledwidth=45.0% ] normalized kinetic energy @xmath124 as a function of electron - phonon coupling @xmath22 for different dimensions d of the lattice.,scaledwidth=45.0% ] in contrast to the two - dimensional case discussed above , less work has been done in three dimensions.@xcite in fact , we are only aware of one calculation of the kinetic energy , which is by de raedt and lagendijk ( drl).@xcite to compare with their work , we chose the same values for the phonon frequency @xmath173 and temperature @xmath116 . as pointed out in sec . [ sec : qmc ] , the numerical effort for calculations with our method , which is proportional to @xmath174 for the algorithm in the form used here but could be reduced to @xmath175 , restricts us to smaller systems than those considered by drl.@xcite for simplicity , we have therefore limited ourselves to a maximum of @xmath119 , for which results can easily be obtained within a reasonable amount of computer time , while the data presented in refs . is for @xmath176 . to be more specific , our calculations for one value of @xmath22 , for @xmath119 and @xmath104 , took about 10 h on an intel xeon 2600 mhz computer . due to the relatively small system size in our work , it is important to study to what extent the results are converged with respect to @xmath50 . to this end , in fig . [ fig : ek_3d_various_n ] , we present @xmath124 as a function of @xmath22 for @xmath177 , and 6 . surprisingly , the results are already satisfactorily converged , keeping in mind the rather small values of @xmath50 . there is a maximal change of less than 20% in the transition region at @xmath113 , while @xmath124 remains almost constant for small and large @xmath22 , as the linear size increases from @xmath109 to @xmath119 . thus , increasing @xmath50 further will not change the results qualitatively , although the finite temperature of our simulations will manifest itself in a way similar to the two - dimensional case . our findings agree well with the results of drl.@xcite the main difference is that for weak coupling , our results are closer to the zero - temperature values ( e.g. , @xmath128 at @xmath130 ) . the reason for this discrepancy despite the fact that we have used the same temperature is the smaller lattice size in our calculations . in contrast to the two - dimensional case considered above , we have not performed a scaling to @xmath151 in 3d , since the clusters under consideration are too small to reveal a systematic power law dependence as a function of @xmath147 . finally , we wish to investigate the effect of dimensionality on the small - polaron cross over . therefore , we compare @xmath124 in one , two and three dimensions using @xmath116 , @xmath119 , and @xmath173 . the dependence on d , which is shown in fig . [ fig : ek_dimensionality ] , is in perfect agreement with previous work . the transition from a mobile large polaron to a small polaron , moving in a very narrow but still coherent band , sharpens considerably with increasing dimension of the system , and while @xmath124 only displays a gradual decrease in 1d without any signs of an abrupt change at @xmath113we find a sharp and well - defined transition in three dimensions . we conclude this section by comparing the accuracy of our results with the qmc methods of drl,@xcite and kornilovitch.@xcite as discussed in sec . [ sec : qmc ] , their main advantage over our method is the fact that they allow one to obtain data which are essentially free of finite - size effects , in any dimension @xmath1783 , and with modest computational effort . however , we have seen above that even in three dimensions , where the limitation of our algorithm is most noticeable , results are reasonably converged . while the approach of refs . and is limited to @xmath129 , drl s method as well as the current approach can , in principle , be used to study any temperature . apart from the sign problem discussed in sec . [ sec : sign ] , the only limitation which occurs is the fact that one has to increase the number of trotter slices as @xmath179 , so as to keep the trotter error smaller than the statistical errors . this situation can be greatly improved by extrapolating results to @xmath100 , as has been done in this work , while the continuous - time algorithm of kornilovitch@xcite is free of any such discretization errors . the accuracy of the results presented here depends on @xmath21 , @xmath106 , @xmath50 and @xmath22 . away from @xmath25 , error bars are use usually smaller than the line width , corresponding to relative errors of less than 0.5% . this is comparable to the accuracy of the results given by kornilovitch@xcite and significantly more accurate than the original results by drl.@xcite extending the work of ref . to the case of the two- and three dimensional holstein model with one electron , we have shown that our qmc approach allows accurate calculations with modest computational effort for a large range of parameters . in particular , the minus - sign problem , due to the lang - firsov transformation , has been found to diminish quickly with increasing system size , and not to have a significant effect on simulations . we have presented a detailed study of the small - polaron cross over and its dependence on the parameters of the system . in particular , we have focused on finite - size and finite - temperature effects , which have not been investigated systematically before . as discussed above , our approach is not as fast as other qmc methods for the holstein polaron,@xcite the main limitation being the restriction to smaller but still reasonably large lattices . this difference in performance is acceptable keeping in mind that it can be extended to the many - electron case ( see below ) , in contrast to the world - line methods of refs . , which work best for the case of a single electron or two electrons of opposite spin , and face a severe sign problem in more than one dimension , similar to other world - line methods.@xcite finally , despite the sign problem , we can perform accurate simulations for virtually all physically interesting values of @xmath21 and @xmath22 , in contrast to the method of refs . and . motivated by the promising results of this work and ref . , our next objective will be a generalization to the many - electron case . to this end , it is important to notice the striking similarity of the qmc method presented here with the determinant method first introduced by blankenbecler _ et al._,@xcite which has been successfully used to study superconductivity and charge - density - wave formation in the holstein model.@xcite in fact , the main difference between our one - electron algorithm and the corresponding grand - canonical method is the calculation of the fermionic weight , which in case of the latter is given by the determinant of @xmath180 . here @xmath181 denotes the unity operator , and @xmath68 is the corresponding generalization of eq . ( [ eq : wf ] ) to many electrons . consequently , the numerical effort is expected to be similar to the simulations with one electron . however , an important open question concerns the dependence of the sign problem on the number of electrons . this work is currently in progress . this work has been supported by the austrian science fund ( fwf ) , project no . one of us ( m.h . ) is grateful to the austrian academy of sciences for financial support . the majority of the results presented here have been obtained using the gigabit - cluster at graz university of technology .

at the nematic quantum critical point that exists in the-wave superconducting dome of cuprates , the massless nodal fermions interact strongly with the quantum critical fluctuation of nematic order . we study this problem by means of renormalization group approach and show that , the fermion damping rate vanishes more rapidly than the energy and the quasiparticle residue in the limit . the nodal fermions thus constitute an unconventional non - fermi liquid that represents an even weaker violation of fermi liquid theory than a marginal fermi liquid . we also investigate the interplay of quantum nematic critical fluctuation and gauge - potential - like disorder , and find that the effective disorder strength flows to the strong coupling regime at low energies . therefore , even an arbitrarily weak disorder can drive the system to become a disorder controlled diffusive state . based on these theoretical results , we are able to understand a number of interesting experimental facts observed in curpate superconductors .

1607.03582

a large amount of experimental and theoretical studies have been devoted to studying the unusual properties of high temperature cuprate superconductors in the past thirty years @xcite . although some consensuses have been reached , many fundamental problems are still in debate , including the microscopic pairing mechanism @xcite , the origin of pseudogap @xcite , and the description of non - fermi liquid behaviors of the normal state @xcite . in the past decade , there have been accumulating experimental evidences for the existence of a strong anisotropy in many of the physical properties of yba@xmath5cu@xmath6o@xmath7 ( ybco ) @xcite and bi@xmath5sr@xmath5cacu@xmath5o@xmath8 ( bscco ) @xcite . such an anisotropy is widely believed to be driven by the formation of a novel electronic nematic order @xcite , which spontaneously breaks the @xmath9 symmetry down to a @xmath10 symmetry . in case the nematic transition line goes across the superconducting transition line and penetrates into the superconducting dome , there exists a zero - temperature nematic quantum critical point ( qcp ) . the nematic quantum phase transition and the associated quantum critical behaviors have been investigated extensively in recent years @xcite . from a theoretical perspective , there are two widely studied scenarios to induce an electronic nematic order . first , the nematic order can be generated by melting a stripe order that spontaneously breaks both translational and rotational symmetry @xcite . the other way is related to pomeranchuk instability which refers to the deformation of the shape of the fermi surface of a metal due to coulomb interaction @xcite . in the simplest case , pomeranchuk instability occurs when the circular fermi surface of a two - dimensional metal becomes ellipse - like via quadrupolar distortion . @xcite . the hubbard model defined on square lattices @xcite provides a pertinent platform to investigate the electronic nematic order . halboth and metzner @xcite studied a two - dimensional hubbard model using functional renormalization group ( rg ) method , and revealed pomeranchuk instability and nematic order . more recently , the square lattice hubbard model is studied by variational cluster approximation @xcite and found to display a local nematic phase under certain circumstances , which might be applied to understand the intra - unit - cell electronic nematicity observed in the scanning tunneling spectroscopy ( sts ) measurements by lawler _ et al _ @xcite . another context of studying nematic order is provided by the superconducting dome of cuprate superconductors . for a pure @xmath0-wave superconductor , the discrete @xmath11 symmetry is preserved . however , when superconductivity coexists with a nematic order , the gap nodes are shifted from their original positions and the @xmath11 symmetry is broken down to @xmath12 @xcite . such an anisotropic superconducting state is physically equivalent to a @xmath13-wave superdconducting state @xcite . at the nematic qcp , the massless fermions excited from the @xmath14-wave gap nodes couple strongly to the quantum critical fluctuation of nematic order parameter , which can be effectively described by a ( 2 + 1)-dimensional field theory @xcite . this model was first analyzed by vojta _ @xcite , who made an @xmath15-expansion and argued that the nematic phase transition is turned to first order . @xcite later tackled the same model by means of @xmath16-expansion , where @xmath17 is a large fermion flavor , and concluded that the transition remains continuous . huh and sachdev @xcite performed a renormalization group ( rg ) analysis , and showed that the fermion velocity ratio @xmath18 in the lowest energy limit , where @xmath19 is the fermi velocity of nodal fermions and @xmath20 the gap velocity @xcite . the unusual velocity renormalization leads to significant changes of a number of quantities , including density of states ( dos ) @xcite , specific heat @xcite , low-@xmath21 thermal conductivity @xcite , superfluid density @xcite , and london penetration depth @xcite . in this article , we revisit the issue of unusual physical properties caused by the strong interaction between massless nodal fermions and critical nematic fluctuation . we will apply the powerful rg approach to calculate the fermion damping rate @xmath1 , where @xmath22 is the imaginary part of retarded fermion self - energy , and the corresponding quasiparticle residue @xmath23 . _ @xcite have previously computed the fermion self - energy and spectral function . interestingly , it will become clear below that the rg analysis give rise to different results once the singular renormalization of fermion velocities is taken into account . in particular , we will show that the fermion damping rate vanishes upon approaching the fermi level more rapidly than the energy @xmath2 , namely @xmath24 . according to the conventional notion of quantum many - particle physics , one would expect the system to behave as a normal fermi liquid . however , by analyzing the rg results , we find that the quasiparticle residue vanishes , i.e. , @xmath25 , in the limit @xmath4 . therefore , the system is actually a non - fermi liquid that represents an even weaker violation of fermi liquid theory comparing to a marginal fermi liquid ( mfl ) . to the best of our knowledge , this type of unconventional non - fermi liquid behavior has not been reported previously . in realistic materials , there are always certain amount of disorders , which may play an important role . as demonstrated earlier by nersesyan _ @xcite , the most important disorder in cuprates behaves like a randomly distributed gauge potential . thus , we will mainly study the influence of random gauge potential on the physical properties of nodal fermions and also the stability of nematic qcp . in the absence of nematic order , the coupling between nodal fermions and random gauge potential has attracted considerable interest @xcite . here , we consider the case in which nodal fermions couple to both the nematic order and random gauge potential , and then study the interplay of these two interactions by means of rg method . we find that the effective strength of gauge potential disorders tends to diverge at the lowest energy . this behavior signals the emergence of a finite zero - energy dos @xmath26 and the happening of quantum phase transition from an unconventional non - fermi liquid to a disorder dominated diffusive state . the nodal fermions acquire a finite scattering rate @xmath27 , which in turn affects the thermodynamic and spectral behaviors of nodal fermions . the rg results for the self - energy of nodal fermions can be used to understand a number of experimental facts observed in cuprate superconductors . we will show that the rg results are qualitatively consistent with some recent measurements of specific heat , fermion damping rate , and temperature dependence of fermion velocities . the rest of the paper will be organized as follows . we first present the effective low - energy field theory for the interaction between nematic order and nodal fermions in section [ sec : modelderivation ] . the random gauge potential is also introduced in this section . in section [ sec : rganalysisclean ] , we make detailed theoretical analysis for the self - consistent rg equations of the fermion velocities in the clean case . based on the rg solutions , we proceed to compute the fermion damping rate , quasiparticle residue , and other physical quantities . we will show that the quantum critical fluctuation of nematic order leads to unconventional non - fermi liquid behaviors of nodal fermions . we consider the influence of random gauge potential in section [ sec : rganalysisdisorder ] and find that the effective disorder strength flows to the strong coupling regime at low energies . therefore , even weak disorders play a significant role and drive the system to enter into a disorder controlled diffusive state . in section [ sec : compaisionexperiments ] , we discuss the possible application of the rg results to some experimental findings of cuprates . in section [ sec : summary ] , we briefly summarize our main results . we start from an effective action @xmath28 . the free action for the nodal fermions is @xcite @xmath29 where @xmath30 denote the pauli matrices . the two component nambu spinors are defined by @xmath31 and @xmath32 where @xmath33 . @xmath34 , @xmath35 , @xmath36 and @xmath37 represent fermions excited from the nodal points @xmath38 , @xmath39 , @xmath40 , and @xmath41 respectively @xcite . the repeated spin index @xmath42 is summed from 1 to @xmath17 , where @xmath17 is the number of fermion spin components with the physical value being @xmath43 . the action @xmath44 describes the nematic order parameter , which is expanded in real space as @xmath45,\end{aligned}\ ] ] where @xmath46 is imaginary time and @xmath47 velocity of field @xmath48 . it is convenient to choose @xmath49 . mass parameter @xmath50 tunes a quantum phase transition from @xmath0-wave superconducting state to a state where the superconducting and nematic orders coexist , with @xmath51 defining the qcp . moreover , @xmath52 is the quartic self - interaction strength . the nematic order couples to fermions via a yukawa - coupling term @xmath53 where @xmath54 is the coupling constant . in addition to the nematic order , the nodal fermions also couple to gauge - potential type disorders , which is described by @xmath55 with @xmath56 and @xmath57 . here , the random gauge potential @xmath58 is assumed to be a gaussian white noise , defined by @xmath59 where @xmath60 is the impurity concentration and @xmath61 measures the strength of a single impurity . we will follow huh and sachdev @xcite and perform rg analysis by employing @xmath16-expansion . the inverse of free propagator of @xmath48 behaves as @xmath62 . after including the polarization , there will be an additional linear - in-@xmath63 term . the @xmath63-term dominates at small @xmath63 over @xmath64-term , so the @xmath64-term can be neglected . near the nematic qcp , we keep only the mass term and rescale @xmath65 and @xmath66 , leading to @xmath67\right\}.\end{aligned}\ ] ] the free propagators for fermions @xmath68 and @xmath69 are written as @xmath70 respectively . at the nematic qcp , @xmath71 , the propagator for the nematic order field @xmath48 is @xmath72 where the polarization @xmath73 , to the leading order of @xmath16-expansion , is given by @xmath74 \nonumber \\ & = & \frac{n}{16v_{f}v_{\delta } } \frac{\omega^{2}+v_{f}^{2}q_{x}^{2}}{\sqrt{\omega^2 + v_{f}^{2}q_{x}^{2}+v_{\delta}^{2}q_{y}^{2 } } } + ( q_{x}\leftrightarrow q_{y}).\end{aligned}\ ] ] after carrying out rg calculations @xcite , we find that all the physical parameters flow with a varying length scale @xmath75 as follows @xmath76 where we have introduced a quantity @xmath77 the expressions of @xmath78 can be found in the [ app : c1c2c3 ] . we emphasize that the impact of disorders is characterized by the quantity @xmath79 , rather than @xmath80 with @xmath81 . it will be shown below that @xmath79 flows to strong coupling at large @xmath75 , leading to remarkable physical properties . it is well established that conventional metals can be described by the standard fermi liquid theory , which states that the fermionic excitations of a normal fermi liquid must have a sufficiently long lifetime and exhibit a sharp quasiparticle peak in their spectral peak despite the existence of coulomb interaction @xcite . the conventional notion is that the fermionic quasiparticles constitute a normal fermi liquid if their zero-@xmath21 damping rate @xmath1 vanishes more rapidly than @xmath2 in the limit @xmath4 . this criterion can be mathematically expressed as @xmath82 another criterion to identify normal fermi liquid is to define an important quantity : the quasiparticle residue , also called renormalization factor , @xmath23 . the residue @xmath23 is usually calculated through the definition @xmath83 where the real part of retarded fermion self - energy @xmath84 is related to @xmath85 via the kramers - kronig ( k .- k . ) relationship . the residue @xmath23 is finite in a normal fermi liquid , but vanishes in a non - fermi liquid . generically , the fermion damping rate in an interacting fermion system can be formally written as @xmath86 where @xmath87 is a constant . with the help of k .- k . relation , the corresponding real part of the fermion self - energy , in the low energy limit , is given by @xmath88 where @xmath89 is a cutoff , and @xmath90 is a function that depends only on @xmath91 . it can be easily checked that @xmath92 for @xmath93 and @xmath94 for @xmath95 . therefore , the above two criteria are actually equivalent because @xmath23 automatically takes a finite value whenever equation ( [ eq : dampingtradi ] ) is fulfilled . , @xmath20 , and @xmath96 flow with a varying length scale @xmath75 respectively in the clean case . the initial values of velocity ratio are @xmath97.[fig : vrgvelocity],title="fig:",width=268 ] , @xmath20 , and @xmath96 flow with a varying length scale @xmath75 respectively in the clean case . the initial values of velocity ratio are @xmath97.[fig : vrgvelocity],title="fig:",width=268 ] , @xmath20 , and @xmath96 flow with a varying length scale @xmath75 respectively in the clean case . the initial values of velocity ratio are @xmath97.[fig : vrgvelocity],title="fig:",width=268 ] however , in the present nodal fermion system , the above two criteria are no longer equivalent . we will show by explicit calculations in this section that the damping rate of nodal fermions vanishes more rapidly than the energy , but the residue @xmath23 vanishes , i.e. , @xmath98 . now we calculate the fermion damping rate and the residue @xmath23 utilizing the solutions of the rg equations ( [ eq : vrgvf])-([eq : vrgvgammagp2 ] ) . the unusual renormalization of fermion velocities need to be taken into account in an appropriate manner . we only present the results obtained in the clean limit @xmath99 in this section , and include the effect of random gauge potential in the next section . after solving rg equations ( [ eq : vrgvf])-([eq : vrgvratio ] ) self - consistently , we show the @xmath75-dependence of fermion velocities @xmath19 , @xmath20 and ratio @xmath96 obtained at different initial values of ratio @xmath100 in figures [ fig : vrgvelocity](a ) , ( b ) and ( c ) respectively . all the quantities @xmath19 , @xmath20 , and @xmath96 decrease with growing @xmath75 and flow eventually to zero as @xmath101 , but apparently @xmath19 decreases much more slowly than @xmath20 . if we use the ratio @xmath96 to characterize the velocity anisotropy , it is clear that the nematic order drives an extreme velocity anisotropy @xmath102 . for later use , it is helpful to extract an approximate analytical expressions for the velocities . considering the leading and sub - leading terms , the solution for @xmath96 is given by @xmath103},\end{aligned}\ ] ] with @xmath104 , which is consistent with the expression in reference @xcite . in the long wavelength limit , @xmath105 , @xmath106 , so the sub - leading term can be ignored . retaining only the leading term , the asymptotic behavior of velocity ratio can be well approximated by @xmath107 substituting the asymptotic form of @xmath96 into equation ( [ eq : vrgvf ] ) , we obtain @xmath108 with @xmath109 for large @xmath75 . solving this equation , we express the renormalized @xmath19 as @xmath110 with @xmath111 . it can also be found that @xmath20 at large @xmath75 behaves approximately as @xmath112^{1 + c_{2}}}.\label{eq : vdeltaasobehavior}\end{aligned}\ ] ] according to equations ( [ eq : vfasobehavior ] ) and ( [ eq : vdeltaasobehavior ] ) , both @xmath19 and @xmath20 vanish as @xmath113 . to examine the impact of singular fermion velocity renormalization and extreme velocity anisotropy on the properties of nodal fermions , we will compute a number of physical quantities . coming first is the quasiparticle residue @xmath23 . apart from the widely used definition given by equation ( [ eq : zdefa ] ) , the residue @xmath23 can also be calculated within the rg framework . the interaction induced renormalization of fermion field @xmath114 is encoded in the residue @xmath23 , which exhibits the following @xmath75-dependence @xmath115 or alternatively @xmath116 -dependence of @xmath23 in the clean case is presented in ( a ) , and of @xmath117 in ( b ) . the initial values of velocity ratio are @xmath97 . [ fig : vrgzf],title="fig:",width=268 ] -dependence of @xmath23 in the clean case is presented in ( a ) , and of @xmath117 in ( b ) . the initial values of velocity ratio are @xmath97 . [ fig : vrgzf],title="fig:",width=268 ] the @xmath75-dependence of @xmath118 can be easily obtained from the above equation , and is presented in figure [ fig : vrgzf](a ) . we find that @xmath118 flows to zero in the limit @xmath113 , which indicates the breakdown of normal fermi liquid and the absence of well - defined landau quasiparticles . however , @xmath118 decreases very slowly with growing @xmath75 , thus the deviation of the system from a normal fermi liquid ought to be quite weak . to see this point , we plot the @xmath75-dependence of @xmath117 in figure [ fig : vrgzf](b ) , and find that @xmath119 therefore , the residue @xmath118 obtained at nematic qcp vanishes with growing @xmath75 more slowly than that of a mfl @xcite , where @xmath120 . according to equation ( [ eq : zdefa ] ) , one can speculate that @xmath121 since a large length scale @xmath75 corresponds to a low energy @xmath2 . this speculation in turn implies that the imaginary part of the retarded fermion self - energy @xmath122 should display the following low - energy behavior , @xmath123 to make the above discussions more quantitative , we are now going to make a detailed analysis of the asymptotic behavior of @xmath118 . at large running scale @xmath75 , the equation of @xmath118 can be approximated by @xmath124 where @xmath125 and @xmath126 . solving this equation leads to the asymptotic behavior @xmath127 in the limit @xmath113 . to proceed , it proves convenient to utilize the relationship between @xmath2 and @xmath75 : @xmath128 where @xmath129 is a cutoff . then the real part of retarded self - energy is approximated by @xmath130\right\}^{c_{5}}\label{eq : realpartselfenergynematicqcp}\end{aligned}\ ] ] as @xmath4 . using the k relation , we obtain the imaginary part of self - energy @xmath131\right\}^{1-c_{5}}}. \label{eq : dampingnematicqcpcleanappro}\end{aligned}\ ] ] from equation ( [ eq : dampingnematicqcpcleanappro ] ) , it is easy to verify that this fermion damping rate is smaller than that in a mfl and manifests the asymptotic behavior @xmath132 , confirming the above analysis based on numerical results . according to the conventional notion of quantum many - body physics , one would expect the system to behave like a normal fermi liquid . however , in the low energy limit @xmath133 , the residue @xmath118 actually vanishes : @xmath134\right\}^{c_{5}}}\rightarrow 0 , \label{eq : znematicqcpcleanappro}\end{aligned}\ ] ] which clearly implies that the system under consideration is actually a non - fermi liquid and the nodal fermions do not have a well - defined quasiparticle peak in the spectral function . we see that the above residue @xmath118 flows to zero at a lower speed than that of a mfl , i.e. , @xmath135 . therefore , the quantum critical fluctuation of nematic order gives rise to an even weaker violation of ordinary fermi liquid theory than a mfl . to the best of our knowledge , this sort of unconventional non - fermi liquid state has not been reported previously . it is now interesting to compare this unconventional non - fermi liquid state with graphene . in graphene , early perturbative calculations revealed that dirac fermions exhibit mfl behavior with damping rate @xmath136 and vanishing @xmath23 @xcite . nevertheless , subsequent careful rg studies @xcite have found that , the fermion damping rate actually depends on energy as @xmath137 at zero temperature due to the long - range coulomb interaction , whereas the corresponding @xmath118 flows to a finite value as @xmath138 . therefore , graphene is a normal fermi liquid . in contrast , the massless nodal fermions constitute a non - fermi liquid at the nematic qcp , because @xmath118 vanishes in the lowest energy limit . the crucial difference between these two cases can be understood as follows . at nematic qcp , the fermion velocities are driven to vanish by the nematic order , so the effective fermion - nematic interaction is significantly pronounced . in graphene , however , the fermion velocity is dramatically enhanced at low energies by coulomb interaction , which then weakens the effective coulomb interaction and guarantees the validity of the fermi liquid description . these two interacting dirac fermion systems provide interesting new insight on the effects of strong electron correlations and also on the criterion of non - fermi liquid states . moreover , an important lesson one can learn from the research experience of graphene is that , rg may lead to qualitatively different spectral properties of fermions from that obtained by ordinary perturbative expansion approach . indeed , this is the main motivation that has promoted us to make an extensive rg analysis of the spectral properties of nodal fermions at the nematic qcp . in the clean limit at different initial values @xmath139 [ fig : rhocl],title="fig:",width=273 ] in the clean limit at different initial values @xmath139 [ fig : rhocl],title="fig:",width=268 ] we have showed in the last subsection that the nodal fermions exhibit unconventional non - fermi liquid behaviors at the nematic qcp . in this subsection , we will compute two important quantities , namely dos and specific heat , on the basis of the rg solutions with the goal to gain a better understanding of the non - fermi liquid state . the fermion dos can calculated from the retarded green functions of nodal fermions via the definition @xcite @xmath140 \nonumber \\ \fl & = & \frac{1}{v_{\delta}v_{f}}\sum_{a=1}^{n } \int\frac{dk_{x}'dk_{y}'}{(2\pi)^{2 } } \frac{1}{\pi}\mathrm{tr}\left[\mathrm{im}g_{1a}^{r}(k_{x}',k_{y}',\omega)+ \mathrm{im}g_{2a}^{r}(k_{x}',k_{y}',\omega ) \right].\end{aligned}\ ] ] in the absence of interactions , the fermion dos is well - known to be linear in @xmath2 , namely @xmath141 . this linear behavior will be changed once the interaction effects are considered . after including the fermion velocity renormalization , employing the method presented in @xcite , the flow equation for @xmath142 is given by @xmath143 in cuprates , the initial value of velocity ratio @xmath100 is known to be much smaller than @xmath144 @xcite . additionally , due to the quantum fluctuation of nematic order , @xmath96 decreases monotonously as the energy scale is lowering . therefore , for a given initial value @xmath145 , we only need to consider the case @xmath146 . in order to show that the conclusion is independent of the condition @xmath145 , we also plot the curves for the case @xmath147 in figures [ fig : rhocl ] and [ fig : cvcl ] . the rg equations of dos and specific heat are also given with a generalized form . in the clean limit , we plot the results for @xmath142 in figure [ fig : rhocl ] . we see from figure [ fig : rhocl](a ) that @xmath142 is apparently not linear in @xmath2 , but displays the asymptotic behavior @xmath148 in the clean limit at different initial values @xmath139 [ fig : cvcl],title="fig:",width=273 ] in the clean limit at different initial values @xmath139 [ fig : cvcl],title="fig:",width=268 ] on the other hand , figure [ fig : rhocl](b ) implies that @xmath149 this dos @xmath142 is qualitatively different from the power law function @xmath150 , where @xmath151 is a small finite value , obtained previously in reference @xcite . in the non - interacting limit , the fermion specific heat depends on @xmath21 as @xmath152 . as shown in [ app : specificheat ] , including the influence of the quantum fluctuation of nematic order and random gauge potential leads us to @xmath153 at low @xmath21 , @xmath154 behaves as @xmath155 which is visualized in figure [ fig : cvcl](a ) . apparently , the original quadratic @xmath21-dependence of @xmath154 obtained in the non - interacting limit is significantly altered . according to figure [ fig : cvcl](b ) , we can express @xmath154 at very small @xmath21 as @xmath156 which is also distinct from the power law @xmath21-dependence @xmath157 , where @xmath158 is a small finite constant , obtained previously @xcite . simple analysis reveal that the three parameters @xmath159 , @xmath12 , and @xmath160 appearing in the rg equations ( [ eq : rhorgeq ] ) and ( [ eq : cvrgeq ] ) all flow to zero in the lowest energy limit @xcite . this asymptotic behavior makes it impossible to express @xmath142 and @xmath161 by power law functions . as given in [ app : approdoscv ] , the low energy behavior of @xmath142 and the low-@xmath21 behavior of @xmath161 can be approximately expressed as @xmath162 , \\ \fl c_{v}(t ) & \sim & \left(\frac{t}{t_{0}}\right)^{2 } \ln\left(\frac{t_{0}}{t}\right ) \left(\ln\ln\left(\frac{t_{0}}{t}\right)\right)^{a_{c } } \exp \left[\frac{1}{2}\left(\ln\ln\ln \left(\frac{t_{0}}{t}\right)\right)^{2}\right],\end{aligned}\ ] ] where @xmath163 and @xmath164 are two negative constants . at @xmath165 , @xmath166 and @xmath167 . we notice that our rg equation for @xmath161 ( [ eq : cvrgeq ] ) is not exactly the same as that obtained by xu _ this difference does not affect our conclusion that @xmath161 is not a power law function at low @xmath21 . indeed , if we start from the rg equation of @xmath161 presented in @xcite , we reach the same conclusion . a more detailed discussion is presented in [ app : specificheat ] and [ app : approdoscv ] . in this section , we investigate the impact of random gauge potential on the quantum critical behaviors near the nematic qcp . we will first show the rg solutions obtained in the presence of random gauge potential , and then re - calculate the fermion dos and specific heat after considering the influence of disorder scattering . to this end , we need to retain a nonzero @xmath168 in the rg equations ( [ eq : vrgvf])-([eq : vrgvgammagp2 ] ) , ( [ eq : zflowdef2 ] ) , ( [ eq : rhorgeq ] ) , and ( [ eq : cvrgeq ] ) , and then solve these rg equations numerically . in order to make a direct comparison to the clean case in which @xmath169 and to explicitly see the influence of a nonzero @xmath168 on the running behaviors of various parameters , we plot the @xmath75-dependence of @xmath19 , @xmath20 , @xmath96 , and @xmath118 in figures [ fig : vrggp](a)-(d ) . moreover , we show the @xmath2-dependence of @xmath142 in figure [ fig : vrggp](e ) and the @xmath21-dependence of @xmath170 in figure [ fig : vrggp](f ) , respectively . we will discuss under what circumstances these rg results are modified in the next subsection . comparing figures [ fig : rhocl](a ) and ( b ) with figures [ fig : vrggp](a ) and ( b ) , we see that the detailed @xmath75-dependence of @xmath19 and @xmath20 are both altered dramatically by the random gauge potential . as shown in figure [ fig : rhocl](c ) and figure [ fig : vrggp](c ) , the velocity ratio @xmath96 exhibits exactly the same @xmath75-dependence in the clean and disordered cases , which originates from the fact that @xmath168 does not enter into the rg equation of velocity ratio @xmath96 . according to figure [ fig : vrgzf](a ) and figure [ fig : vrggp](d ) , the renormalization factor @xmath118 flows to zero in the disordered case much more rapidly than the clean case . as shown in figure [ fig : vrggp](e ) , the dos @xmath142 is divergent in the lowest energy limit due to random gauge potential . this is completely different from the behaviors of clean case presented in figure [ fig : rhocl ] . in figure [ fig : vrggp](f ) , we plot the ratio between @xmath171 , where @xmath172 and @xmath173 are the specific heat obtained in disordered and clean cases , respectively . since the solutions of the rg equations about dos and specific heat are modified substantially in the disordered case , it turns out that random gauge potential is a relevant perturbation in the present system . to verify the relevance of random gauge potential , we should appeal to the rg analysis of the effective disorder strength . in the action term @xmath174 given in equation ( [ eq : actiondisorder ] ) , the parameters that characterize the fermion - disorder coupling seem to be @xmath175 and @xmath176 . it can be seen from the rg equations that @xmath175 flows in precisely the same way as @xmath19 , and that @xmath176 as @xmath20 . thus , both @xmath175 and @xmath176 are strongly renormalized and driven to vanish as @xmath101 . however , this does not mean that the disorders can be simply neglected . indeed , whether disorders are important is determined by the ratio between the interaction energy given by @xmath174 and the fermion energy @xmath177 . we can see that this ratio is defined by @xmath79 , which enters into the rg equations for the parameters @xmath19 , @xmath20 , @xmath175 , and @xmath176 . the effective strength of random gauge potential should be measured by @xmath79 , rather than @xmath175 and @xmath176 . recall that @xmath79 is a function of five parameters , i.e. , @xmath19 , @xmath20 , @xmath175 , @xmath176 , and @xmath60 . among these parameters , @xmath60 is assumed to be a dimensionless constant , but the other four parameters flow strongly with the varying @xmath75 . detailed rg analysis revealed that @xmath168 goes to infinity , namely @xmath178 in the limit @xmath113 as a consequence of the singular renormalization of the fermion velocities ratio @xmath96 . more quantitatively , the large scale behavior of @xmath79 can be described by @xmath179 where @xmath180 . it is necessary to explain here why the effective strength of random gauge potential should be represented by @xmath79 , rather than solely by the coefficients @xmath80 with @xmath181 , and why @xmath182 in the lowest energy limit . in an interacting fermion system , the effective strength of the interaction is characterized by the ratio between the interaction energy scale and the kinetic energy of fermions . this ratio is widely used in condensed matter physics to judge whether an interacting fermion system can be defined as a strongly correlated system or not . for instance , the normal metal with a high density of itinerant electrons is believed to be a weakly interacting system since the energy scale of coulomb potential is much smaller than the fermi energy . in a massless dirac fermion system ( such as graphene ) , the effective strength of long - range coulomb interaction is defined as @xmath183 , where @xmath184 appears in the action of the coulomb interaction as a coupling coefficient and the fermion velocity @xmath19 reflects the order of the kinetic energy @xcite . another example comes from the effective bcs model of dirac fermion systems , where the effective strength of pairing interaction is characterized by @xmath185 with @xmath186 being the coupling coefficient of pairing interaction and @xmath19 fermion velocity @xcite . this criterion also applies to disordered systems . when massless dirac fermions couple to random gauge potential , the effective strength of random gauge potential should be defined by @xmath187 , rather than the coefficient of fermion - disorder coupling @xmath80 @xcite . when @xmath80 flows to zero in the lowest energy limit , @xmath79 does not necessarily vanish since there is a possibility that @xmath19 and @xmath20 may vanish more ( or equally ) rapidly than @xmath80 . when the nodal fermions couple to both the quantum fluctuation of nematic order and random gauge potential , the four parameters @xmath19 , @xmath20 , @xmath175 , and @xmath176 all flow to zero in the lowest energy limit , but the effective strength of random gauge potential becomes very large , namely @xmath188 . this originates from the fact that @xmath189 and @xmath190 are constants , while at the same time the velocity ratio @xmath18 . the behavior @xmath191 at low energies indicates that random gauge potential is relevant . to see this point , we neglect the nematic order and consider only the coupling of nodal fermions to random gauge potential , which results in simplified rg equations : @xmath192 using the two relations @xmath193 and @xmath194 , it is easy to show that @xmath195 in this case , @xmath19 , @xmath20 , @xmath175 , and @xmath176 depend on length scale @xmath75 as @xmath196 which all vanish rapidly in the limit @xmath197 . since @xmath79 is a constant , random gauge potential is marginal . from the above calculations , we can conclude that the behavior @xmath182 obtained in the presence of both nematic fluctuation and random gauge potential is directly related to the extreme velocity anisotropy @xmath18 induced by the quantum fluctuation of nematic order . the flow of @xmath79 towards strong coupling regime is a clear signature that random gauge potential should have substantial physical effects on the low - energy behaviors of nodal fermions , which will be discussed in the next subsection . we notice other interesting correlated electron models in which the coupling coefficient of an interaction vanishes at low energies , but the interaction is not negligible due to the more ( or equally ) rapid decrease of the kinetic energy of electrons . recently , sur and lee studied the influence of quantum fluctuations of an antiferromagnetic ( af ) order at an af quantum critical point in a metal supporting one - dimensional fermi surface @xcite . in particular , they showed that the coupling coefficient of the interaction flows to zero at low energies . however , the fermion velocity also vanishes , thus the interaction can not be simply neglected . actually , sur and lee found that the interaction drives the system to become a so - called quasi - local strange metal that is apparently qualitatively different from the free fermion system . how should we understand the divergence of @xmath79 ? in order to answer this question , we first consider the non - interacting system that contains only nodal fermions and random gauge potential . in this case , the rg equation of fermion dos becomes @xmath198 with @xmath199 if @xmath200 , we have @xmath201 where @xmath151 satisfies @xmath202 . in this case , the rg equation for specific heat is given by @xmath203 the solution of this equation is given by @xmath204 if @xmath205 , it is easy to get @xmath206 in the limit @xmath4 . the divergence @xmath142 indicates the emergence of a disorder dominated diffusive state , in which a finite disorder scattering rate @xmath27 and a finite zero - energy dos @xmath26 are generated @xcite . the value @xmath207 defines a qcp for the quantum phase transition between the ballistic and diffusive states of nodal fermions . therefore , a weak random gauge potential gives rise to power law behavior of @xmath142 , whereas a sufficiently strong disorder can trigger the quantum phase transition to a diffusive state . in the presence of both nematic critical fluctuation and random gauge potential , the fact that @xmath191 in the lowest energy limit signals the development of a disorder dominated diffusive state and the generation of a finite @xmath27 and a finite @xmath26 even in the case of weak random gauge potential . although the perturbative rg method provides a powerful tool to judge whether and when a phase transition takes place , it can not be used to compute the precise value of @xmath27 . to calculate @xmath27 , one usually needs to construct a self - consistent equation for the retarded fermion self - energy by properly considering both the fermion - disorder scattering and the influence of quantum critical fluctuation of nematic order @xcite this is an interesting yet complicated issue , which is beyond the scope of the present work and subjected to future research . here , we use the rg method to make a rough estimation for the energy scale of @xmath27 . as shown in fig . [ fig : vrggp](e ) , for small given values of @xmath208 and @xmath100 , the solution of the rg equation of dos should have the following properties : @xmath142 decreases as the varying energy scale decreases , but tends to increase when the energy scale exceeds a critical value @xmath209 , which is a function of @xmath208 and @xmath100 . the magnitude of @xmath27 should be an increasing function of @xmath210 . in the following calculations of @xmath26 and @xmath161 , we will regard @xmath27 as an undetermined constant . fortunately , the qualitative behaviors of @xmath26 and @xmath154 in the low energy regime is independent of the precise value of @xmath27 . the imaginary part of retarded fermion self - energy can be generically written as @xmath211 where @xmath212 is the contribution induced solely by the nematic order . the disorder - induced scattering rate @xmath27 represents a characteristic energy scale . at energies higher than @xmath27 , namely @xmath213 , @xmath212 dominates over @xmath27 and all the rg results for @xmath19 , @xmath20 , @xmath118 , @xmath142 , and @xmath154 shown in figures [ fig : vrggp](a)-(f ) are still applicable . at @xmath214 , the diffusive motion of nodal fermions and its interplay with critical nematic fluctuation govern the low - energy properties of the system . once a finite @xmath27 is generated , the renormalized velocities @xmath19 and @xmath20 no more vanish at low energies , which is apparently different from the clean case . instead , as the energy scale decreases , both @xmath19 and @xmath20 are saturated to certain finite values , denoted by @xmath215 and @xmath216 , below the energy scale set by @xmath27 . hence , there is no extreme velocity anisotropy in the diffusive state . the fermion dos and specific heat also exhibit different behaviors comparing to the clean case . to demonstrate the difference in dos , we write the retarded propagators of nodal fermions in the forms : @xmath217 calculations find that the fermion dos depends on @xmath2 as @xmath218.\end{aligned}\ ] ] in the case @xmath219 , we have @xmath220 in order to compute the specific heat , it is convenient to invoke the standard matsubara formalism of fermion propagators , i.e. , @xmath221 where @xmath222 with @xmath223 being integer . the fermion free energy is given by @xmath224.\nonumber\end{aligned}\ ] ] summing over @xmath225 leads to @xmath226\right\}.\end{aligned}\ ] ] now the fermion specific heat in the low energy regime can be approximately given by @xmath227 which depends on @xmath21 linearly . we can see that dos and specific heat obtained in the diffusive state exhibit entirely different behaviors from the unconventional non - fermi liquid state below the energy scale @xmath27 . we finally make a brief remark on the behavior of the system staying away from the nematic qcp . suppose the system stays at a distance @xmath50 to the qcp , the rg results are still valid and the fermion velocity ratio is still renormalized at energies higher than the scale corresponding to @xmath50 . however , the renormalization terminates at certain low energy scale . therefore , the effective strength of random gauge potential is moderately enhanced compared its bare value , though @xmath168 does not diverge . in this section , we address the possible connection between the theoretical results obtained in the last sections and the phenomenology of cuprate superconductors . we are particularly interested in three existing experimental findings about some of the unusual properties of the superconducting dome . we first discuss the residual specific heat in cuprates . due to the line nodes of @xmath228-wave gap , the specific heat in the superconducting phase of cuprates is expected to exhibit a @xmath229 behavior , i.e. , @xmath230 , at low @xmath21 . this expectation is generically consistent with experiments @xcite . in the lowest @xmath21 limit , experiments have observed a residual linear @xmath21 term of @xmath161 @xcite , which is usually attributed to the finite fermion dos generated by disorder scattering and is also well consistent with the result given by equation ( [ eq : cvdisorder ] ) . there is , however , an unexpected experimental finding @xcite that the residual specific heat of ybco is obviously larger in magnitude than that of la@xmath231sr@xmath232cuo@xmath233 ( lsco ) , although the former material is known to be cleaner than the latter . apparently , disorder scattering alone is not capable of accounting for this experimental fact . recently , coexistence of @xmath228-wave superconductivity with a loop current order was proposed to give a possible explanation @xcite for the large residual linear-@xmath21 term of specific heat in ybco . in this scenario , when @xmath228-wave superconductivity coexists with a loop current order , two of four nodal points are converted to finite fermi pockets of bougoliubov quasiparticles , which then generates a finite @xmath26 and a residual linear-@xmath21 term of specific heat @xcite . the recent ultrasound measurements showed the possible evidence for the existence of loop current order in ybco@xcite . here we propose an alternative explanation for the above anomalous experimental results of residual specific heat . our rg analysis found that the effective strength of random gauge potential , being the most relevant disorder to cuprates @xcite , is strongly enhanced by the critical nematic fluctuation , which in turn increases the residual value of the specific heat . to understand the role of quantum nematic fluctuation , we firstly consider only the coupling of nodal fermions to random gauge potential . in this special case , @xmath79 does not flow and thus remains a constant , namely @xmath234 . if @xmath208 is very small , which means the system is only slightly disordered , the behavior of @xmath142 and specific heat would be governed by equations ( [ eq : dosonlydisorder ] ) and ( [ eq : specificheatonlydisorder ] ) , respectively . in this case , the system does not develop a finite @xmath26 and there is no residual liner-@xmath21 term of specific heat . if the system is quite disordered such that @xmath235 , it enters a diffusive state with a finite scattering rate @xmath27 , which induces a finite @xmath26 and a residual linear-@xmath21 term of specific heat . apparently , the residual specific heat is larger in more disordered systems , which is not consistent with the aforementioned experiments of residual specific heat . if we consider both the quantum fluctuation of nematic order and random gauge potential , @xmath79 is significantly enhanced and flows to strong coupling at low energies even if @xmath208 takes an arbitrarily small value . this implies that a cleaner compound might acquire a larger amount of @xmath27 and naturally a larger @xmath26 . to make a more careful comparison between theories and experiments , we now briefly discuss the doping dependence of our results . we use @xmath236 to denote the doping concentration and @xmath237 the nematic qcp . at zero temperature , the mass of nematic field @xmath48 is proportional to the difference between @xmath236 and @xmath237 , namely @xmath238 . when the cuprate is at a distance @xmath50 away from the nematic qcp , the quantum fluctuation of nematic order is not critical , but remains important for small @xmath50 . at the energy scales larger than @xmath50 , the fermion velocity anisotropy is still considerably enhanced , which leads @xmath79 to flow to larger values . for a given small value of @xmath208 , @xmath79 can flow to a sufficiently large value to induce a diffusive state and generate a finite scattering rate @xmath27 , provided that @xmath50 is made sufficiently small . in this case , the quantum fluctuation of nematic order can result in a finite @xmath26 and a finite residual linear-@xmath21 term of specific heat . if the bare value @xmath208 is large enough , random gauge potential itself suffices to generate a finite @xmath27 . including the quantum fluctuation of nematic order leads to larger values of both @xmath79 and @xmath27 . in any case , the quantum fluctuation of nematic order tends to amplify @xmath27 , which naturally increase @xmath26 and the residual specific heat . a number of recent experiments provided strong evidence supporting the existence of nematic order in ybco , whereas there is little evidence for nematicity in lsco . although ybco is cleaner than lsco , the quantum fluctuation of nematic order in the former superconductor can drive @xmath79 to flow to much larger values . as a consequence , the residual linear - in-@xmath21 term of specific heat of ybco would be larger than that of lsco . we emphasize that , disorder itself can not explain the anomalous behaviors of the residual specific heat observed in ybco and lsco , and it is necessary to consider the interplay of quantum nematic fluctuation and random gauge potential . for the above elaboration , we know that the roles played by the quantum nematic fluctuation and random gauge potential depends on doping @xmath236 . to simplify discussion , we assume the bare value @xmath208 displays only a weak @xmath236-dependence . since the quantum fluctuation of nematic order is most pronounced at the nematic qcp , the scattering rate @xmath27 and consequently the coefficient of the residual liner-@xmath21 term of specific heat are maximal at the qcp , and decrease as the system moves away from this qcp . this doping dependence is observable , and can be examined by experiments . within the loop current order scenario , both the zero - energy dos @xmath26 and the residual linear-@xmath21 term of specific heat are proportional to the order parameter of the loop current order , which decreases with the growing doping in the underdoped region . early experiments @xcite found that the coefficient of linear-@xmath21 term of specific heat at optimally doped ybco is roughly @xmath43 mj@xmath239mol@xmath240@xmath239k@xmath241 . more recent measurements performed in underdoped ybco @xcite revealed that this coefficient is @xmath242 mj@xmath239mol@xmath240@xmath239k@xmath241 . we feel that the currently available experimental data about the doping dependence of this coefficient are still quite limited . we expect more extensive measurements would be performed in the future to extract a more quantitative doping dependence of the coefficient , which could help to judge whether the scenario proposed in this paper works . we next apply the rg results to understand the observed damping rate of nodal fermions in cuprates . @xcite have performed extensive angle resolved photoemission spectroscopy measurements in optimally doped bi@xmath5sr@xmath5cacu@xmath5o@xmath8 ( bscco ) . their main discovery is that the nodal fermions exhibit a mfl - type damping rate in the normal state above @xmath243 , which is in general consistency with the observed linear resistivity . they further found @xcite that the linear damping rate is not sensitive to the onset of superconductivity and persists well below @xmath243 . this was out of expectation since previous bcs weak coupling analysis @xcite had predicted a quite weak damping rate , i.e. , @xmath244 , in the superconducting phase . several scenarios @xcite were proposed to account for the nearly mfl behavior . in particular , vojta _ et al . _ @xcite and khveshchenko and paaske @xcite have argued that the strong fermion damping may arise from a secondary phase transition from a @xmath0-wave superconducting state to a @xmath245 or @xmath246-wave superconducting state . experimentally , the existence of a nematic phase was observed in bscco by lawler _ more recent experimental studies of fujita _ et al . _ provided further evidence pointing towards the existence of a nematic qcp in the vicinity of optimal doping in bscco @xcite . therefore , it seems natural to account for the experimental finding of valla _ et al . _ by considering the quantum critical fluctuation of nematic order . we have showed through rg analysis that the nodal fermion damping rate caused by critical nematic fluctuation , described by @xmath247 , is slightly weaker than that of a mfl . in realistic experiments , it is difficult to distinguish this non - fermi liquid state from a mfl state . therefore , presence of a nematic qcp provides an alternative scenario for the nearly mfl behavior observed in optimally doped bscco . however , the nearly mfl behavior occurs only at energies higher than the value @xmath248 set by disorder scattering rate @xmath27 . indeed , at @xmath249 , @xmath27 dominates over @xmath122 , thus a finite zero - energy dos @xmath26 is generated . to summarize , our rg results are qualitatively consistent with the quasiparticle self - energy @xmath250 observed in reference @xcite . another experiment is the observation of plumb _ et al . _ @xcite that the fermion velocity @xmath19 along the nodal directions increases as @xmath21 grows in bscco . we know from rg results that the critical nematic fluctuations drive the fermion velocities to vanish in the lowest energy limit . this means the velocities must increase as the energy scale is growing . therefore , the nematic quantum fluctuation will induce increment of fermion velocities when the thermal energy increases with growing @xmath21 , which is qualitatively well consistent with the observation of plumb _ et al_. technically , one can translate the @xmath75-dependence of fermion velocity to a @xmath21-dependence by making the transformation @xmath251 @xcite . with the help of this transformation , it is easy to show that the fermion velocity @xmath252 is an increasing function of @xmath21 . therefore , the singular renormalization of fermion velocities of nodal fermions induced by the quantum fluctuation of namatic order , which is first discovered by huh and sachdev @xcite , agrees with the @xmath21-dependence of @xmath19 observed in reference @xcite . in summary , we have found that the nodal fermions of @xmath228-wave superconductors constitute an unconventional non - fermi liquid , which exhibits a weaker violation of fermi liquid description than a mfl , due to the quantum critical fluctuation of nematic order . this unusual state represents a novel quantum state of matter that can not be well captured by the traditional classification of ( non- ) fermi liquids and thus enriches our knowledge of strong electron correlation effects . we also have calculated the fermion dos and specific heat after incorporating the unusual renormalization of fermion velocities . when a gauge - potential - type disorder is added to the system , we have analyzed its interplay with the quantum fluctuation of nematic order , and found that the effective disorder strength flows to strong coupling , leading to diffusive motion of nodal fermions . therefore , even a weak random gauge potential can drive a quantum phase transition from a unconventional non - fermi liquid state to a disorder dominated diffusive state . however , the unusual fermion damping induced by the nematic order is more important than the disorder scattering at high temperatures , where the nodal fermions still display the unconventional non - fermi liquid behaviors . we finally have discussed the connection between our theoretic results and a number of interesting experiments in the context of some cuprate superconductors . we now would like compare our work to the existing extensive works on the non - fermi liquid behaviors in two - dimensional metals produced by the quantum critical fluctuation of nematic order . at the qcp of pomeranchuk instability in two - dimensional metals , the quantum fluctuation of nematic order can lead to very strong fermion damping @xcite . to the leading order , it is found @xcite the fermion damping rate behaves as @xmath253 and the quasiparticle residue @xmath254 . since @xmath23 vanishes in the limit @xmath4 , this qcp exhibits non - fermi liquid behavior . in this paper , we have considered the interaction between the quantum fluctuation of nematic order and massless nodal fermions in the superconducting dome of cuprate superconductors . it is apparent that the quantum critical fluctuation of nematic order gives rise to a stronger fermion damping effect in metals than in the superconducting dome of cuprates . this difference should be owing to the different forms of the kinetic energies of fermionic excitations . in the context of cuprates , the kinetic energy of the massless dirac fermions excited from the superconducting gap nodes is @xmath255 . in contrast , in a two - dimensional metal with a finite fermi surface , the kinetic energy of fermions can be written as @xmath256 , where @xmath257 is the momentum component perpendicular to the fermi surface and @xmath258 is the momentum component along the tangential direction @xcite . in the low - energy regime , the latter kinetic energy is smaller that the former for the same given values of momenta , which indicates that the interaction plays a more important role in the latter system than the former . to further demonstrate this difference , we consider the different roles of the long - range coulomb interaction in a two - dimensional dirac semimetal and a two - dimensional semi - dirac semimetal . in a dirac semimetal , the kinetic energy of dirac fermions is simply @xmath259 with @xmath260 . rg analysis showed that the residue @xmath118 approaches a finite value at low energies , so the system is a normal fermi liquid@xcite . in a semi - dirac semimetal , the kinetic energy of fermions is written as @xmath261 @xcite . in this case , the long - range coulomb interaction drives @xmath23 to vanish in the lowest energy limit , i.e. , @xmath262 , which apparently implies the breakdown of fermi liquid behavior @xcite . once again , we see that the ratio between the interaction energy scale and the kinetic energy is a crucial quantity to determine the low - energy behaviors of an interacting fermion system . the coupling of nodal fermions to the quantum fluctuation of antiferromagnetic ( af ) order is also an interesting problem @xcite . uemura @xcite considered the coupling of nodal fermions to the @xmath263 af fluctuation , and suggested a possibility that the @xmath263 af fluctuation can connect two nodal charges in different hole pockets and then generate a bound state of two nodal charges . more recently , pelissetto _ et al . _ studied a number of different couplings between nodal fermions and af fluctuations using rg method @xcite . an interesting results is that , though most of these couplings are irrelevant , there emerges a nearly marginal coupling between nodal fermions and an effective , af - order induced nematic fluctuation . this nearly marginal coupling is found @xcite to results in a fermion damping rate that is nearly linear in energy or temperature . the electronic nematic state has been observed not only in some cuprate superconductors , but also in a number of iron - based superconductors @xcite : 122 family , such as hole doped ba@xmath264k@xmath265fe@xmath5as@xmath5 , electron doped ba(fe@xmath264co@xmath5)@xmath5as@xmath5 , and isovalent - doped bafe@xmath5(as@xmath264p@xmath265)@xmath5 ; 111 family , such as nafeas ; 1111 family , such as lafeaso ; 11 family , such as fese @xcite . in most of these compounds , the nematic order emerges in accompany with a spin density wave ( sdw ) order . however , there are also exceptions . for instance , the nematic order is observed in fese without any evidence for sdw order @xcite . whether the nematic order observed in iron - based superconductors is generated by the fluctuation of sdw order or the orbital degrees of freedom is still in fierce debate @xcite . recent experimental studies have unambiguously showed that there is a qcp in the superconducting dome at the optimal doping of bafe@xmath5(as@xmath264p@xmath265)@xmath5 . this qcp may correspond to the critical point for a sdw order or nematic qcp @xcite , and is expected to exhibit rich quantum critical phenomena . moreover , there are clear evidences that the superconducting gap of bafe@xmath5(as@xmath264p@xmath265)@xmath5 has nodal line points @xcite . since the quantum fluctuation of nematic order is peaked at zero momentum @xcite , the nodal fermions excited from the nodal line points might couple strongly to the quantum fluctuation of nematic order at the nematic qcp @xcite . this coupling is physically analogous to the model considered in this work , and it would be interesting to study this coupling by means of rg method . the rg analysis performed in this work could be generalized to study the possible non - fermi liquid behavior and disorder effects in the context of bafe@xmath5(as@xmath264p@xmath265)@xmath5 and other iron based superconductors , where the multi - band effects and different gap symmetry need to be seriously taken into account . the expressions of @xmath159 , @xmath12 , and @xmath160 that appear in equations ( [ eq : vrgvf])-([eq : vrgvgammagp2 ] ) are given by @xmath266 with @xmath267 the free energy density @xmath268 is formally given by @xmath269 where @xmath270 , @xmath271 , and @xmath272 . we can rewrite @xmath273 as @xmath274 which yields the following specific heat : @xmath275 it is then easy to get @xmath276 at a given @xmath21 , the corresponding momentum scale should be determined by the larger component of the fermion velocities @xcite , i.e. , @xmath277 which then gives rise to @xmath278 the scaling equation for @xmath170 is converted to @xmath279 since @xmath280 , we find that @xmath281 substituting equations ( [ eq : vrgvf ] ) and ( [ eq : vrgvdelta ] ) into ( [ eq : cvmiddle ] ) leads us to @xmath282 the quantum fluctuation of nematic order drives the velocity ratio @xmath96 to monotonously decrease as one goes to lower and lower energies . it is known that the bare ratio @xmath283 in cuprate superconductors @xcite , which make it possible to simplify the rg equation to @xmath284 in the clean limit , @xmath99 , so the equation becomes @xmath285 this equation is slightly different from the rg equation for specific heat presented in @xcite , where the equation is @xmath286 the numerical solutions for equation ( [ eq : rgequationcvxu ] ) are shown in figure [ fig : cvclxu ] . obtained from the rg equation of specific heat given in @xcite at different initial values @xmath287 . [ fig : cvclxu],title="fig:",width=270 ] obtained from the rg equation of specific heat given in @xcite at different initial values @xmath287 . [ fig : cvclxu],title="fig:",width=268 ] we notice that the specific heat obtained from equation ( [ eq : rgequationcvxu ] ) also satisfies @xmath288 and @xmath289 , which indicates that the specific heat is enhanced comparing to the case of non - interacting nodal fermion system . however , as will be shown in appendix c , it can not be expressed as a power law function . the reason is that the three parameters @xmath159 , @xmath12 , and @xmath160 appearing in equation ( [ eq : rgequationcvxu ] ) all vanish in the lowest energy limit , rather than approaching certain finite values . in order to show that the dos and specific heat in clean limit do not exhibit power law behaviors , we now derive their approximate analytical expressions in the low - energy regime from the rg results . the rg equation of dos is given by @xmath290 in the absence of disorders . in the lowest energy limit , we know that the velocity ratio @xmath291 . the parameters @xmath159 , @xmath12 , and @xmath160 , being functions of @xmath96 , all vanish @xcite as @xmath291 . thus the rg equation can be approximately written as @xmath292 where we have used three constants @xmath293 , @xmath294 , and @xmath295 @xcite . substituting the approximate low - energy expression of @xmath96 given in equation ( [ eq : vratioapproximate ] ) into ( [ eq : dosapproximatea ] ) , we obtain @xmath296 where @xmath297 . for @xmath165 , @xmath166 . using the relationship @xmath298 , we can solve equation ( [ eq : dosapproximateb ] ) and then obtain the following analytical expression @xmath299,\end{aligned}\ ] ] which is applicable for small @xmath2 . the rg equation for specific heat is shown in equation ( [ eq : rgequationcvclean ] ) . at low energies , it can be approximated as @xmath300 where @xmath301 . at @xmath302 , @xmath303 . using the transformation @xmath304 , we find that the specific heat at low temperature can be well approximated by the expression @xmath305.\end{aligned}\ ] ] by applying the same treatment , the rg equation of specific heat presented in @xcite , shown in equation ( [ eq : rgequationcvxu ] ) , can be approximately expressed as @xmath306 , \label{eq : cvapproximatexu}\end{aligned}\ ] ] where @xmath307 . at @xmath165 , @xmath308 . the above two functions do not exhibit power law dependence on temperature .

a two - dimensional fluid , stirred at high wavenumbers and damped by both viscosity and linear friction , is modeled by a statistical field theory . the fluid s long - distance behavior is studied using renormalization - group ( rg ) methods , as begun by forster , nelson , and stephen [ phys . rev . a * 16 * , 732 ( 1977 ) ] . with friction , which dissipates energy at low wavenumbers , one expects a stationary inverse energy cascade for strong enough stirring . while such developed turbulence is beyond the quantitative reach of perturbation theory , a combination of exact and perturbative results suggests a coherent picture of the inverse cascade . the zero - friction fluctuation - dissipation theorem ( fdt ) is derived from a generalized time - reversal symmetry and implies zero anomalous dimension for the velocity even when friction is present . thus the kolmogorov scaling of the inverse cascade can not be explained by any rg fixed point . the function for the dimensionless coupling is computed through two loops ; the term is positive , as already known , but the term is negative . an ideal cascade requires a linear function for large , consistent with a pad approximant to the borel transform . the conjecture that the kolmogorov spectrum arises from an rg flow through large is compatible with other results , but the accurate scaling is not explained and the kolmogorov constant is not estimated . the lack of scale invariance should produce intermittency in high - order structure functions , as observed in some but not all numerical simulations of the inverse cascade . when analogous rg methods are applied to the one - dimensional burgers equation using an fdt - preserving dimensional continuation , equipartition is obtained instead of a cascade in agreement with simulations .

nlin0410050

the cascade of energy to low wavenumbers in two - dimensional turbulence @xcite , more than other turbulence problems , is suited to the standard methods of statistical field theory and the renormalization group ( rg ) . these methods @xcite , originating in quantum field theory , show that arbitrary short - distance interactions lead to long - distance behavior described by a local , renormalizable action an effective field theory . correlation functions computed from this action contain ultraviolet ( uv ) divergences that can be eliminated by redefining the parameters and fields . the divergences leave their mark , however , in the dependence on the renormalization scale and the resulting anomalous scaling laws . the classic application of statistical field theory is to critical phenomena ( second - order phase transitions ) in condensed matter @xcite . the infrared ( ir ) scale invariance of correlation functions at the transition temperature is explained by a fixed point of the rg flow . the inverse energy cascade of two - dimensional turbulence is likewise believed to be nearly scale invariant , and one might suspect that a similar fixed point is responsible . we will see , however , that no fixed point can reproduce the observed @xmath4 energy spectrum . rather , we will argue that the inverse cascade arises from a nontrivial rg flow and thus is not expected to be completely scale invariant . in the study of turbulence , a deviation from scale invariance ( a dependence of dimensionless physical quantities on scale ) is referred to as intermittency @xcite . while intermittency is recognized as a property of the three - dimensional direct cascade of energy , its existence in the two - dimensional inverse cascade is unsettled . for a nonstationary inverse cascade , in which energy is not dissipated but progresses to ever - lower wavenumbers , intermittency is not observed in numerical simulations @xcite ; a theoretical explanation has been given @xcite . in this paper we deal solely with the stationary regime , where an inverse cascade requires a low - wavenumber energy sink . with few exceptions , simulations of such a cascade confirm the @xmath4 energy spectrum initially predicted @xcite on the basis of scale invariance . but one set of simulations @xcite finds strong intermittency in fourth- and higher - order velocity correlations . other simulations @xcite and experiments @xcite , though , find no significant intermittency . the various studies differ mainly in the precise form of the dissipation terms . the evidence suggests that intermittency in the stationary inverse cascade , permitted by our theory , is at least possibly realized . it is our restricted focus on the inverse cascade that allows us to work with a purely local field theory . the random force that stirs the fluid is correlated over a limited range and is effectively local in a long - distance description . as with the rg treatment of quantum fields and condensed matter , we expect all short - distance details to become irrelevant except as they are manifested in local , renormalizable couplings . the two - dimensional direct cascade of enstrophy to high wavenumbers @xcite thus falls outside our scope . a previous rg analysis of two - dimensional turbulence @xcite is formally similar to ours , but it follows three - dimensional studies by adopting the long - range force correlations necessary for a direct cascade ; even its derivation of the inverse cascade relies on nonlocal forcing . here we apply rg methods in the familiar domain of local field theory , which should allow a physical treatment of the inverse cascade . the explanation of the direct enstrophy cascade may rest on entirely different foundations , such as conformal invariance @xcite . the work most closely aligned with our theoretical approach is due to forster , nelson , and stephen ( fns ) @xcite . at the technical level , our contribution is to add linear friction to fns model a in @xmath5 and compute the rg flow to the next order of perturbation theory , two loops . the inclusion of friction , which dissipates energy at low wavenumbers , makes our theory capable in principle of describing the inverse cascade and its @xmath4 spectrum unlike fns model a , which gives a @xmath6 spectrum corresponding to energy equipartition in @xmath5 . we also note a difference in our viewpoint from that of fns and others @xcite who apply rg methods to turbulence by analogy with critical phenomena . these authors seek a controlled ir - stable fixed point by starting with a logarithmically divergent field theory and then decreasing the dimension of space or the exponent of the stirring - force correlation by @xmath7 . this adds to the @xmath0 function a negative linear term proportional to @xmath7 . with the usual positive one - loop term , there exists an ir - stable fixed point at a coupling that goes to zero with @xmath7 ; the fixed - point theory can then be expanded in @xmath7 instead of the original coupling . like fns , we work in @xmath8 to regulate uv divergences , but we ultimately take @xmath9 , so that the fixed point is trivial . our inverse - cascade model lies not at a fixed point but in the region of large dimensionless coupling . naturally the use of perturbation theory is questionable for strong coupling . our perturbative results will have direct quantitative application only to the extreme ir limit controlled by the trivial fixed point , which is of some interest in itself . nevertheless , we will make reasonable conjectures about the theory s strong - coupling behavior that are consistent with the inverse cascade , bearing in mind the dangers of the nonperturbative regime . besides the concern with the numerical accuracy of extrapolations , there are fundamental difficulties at strong coupling . the anomalous dimensions of operators may be large , and the relevance of terms in the action may differ from the weak - coupling case . furthermore , at strong coupling , there is no simple relation between the couplings in different renormalization schemes , such as the wilsonian cutoff ( useful for physical interpretation ) and minimal subtraction ( convenient for systematic calculation ) . we may hope that these subtleties do not affect the main conclusions even at very strong coupling . at least we know that the theory of critical phenomena in @xmath10 is extrapolated to @xmath11 ( moderate coupling ) with acceptable results . in sec . [ framework ] we describe the basis of our theory and our method of calculation , confirming the one - loop rg flow of fns @xcite . in sec . [ properties ] we present symmetries and other properties of the theory that do not involve a dubious extrapolation to strong coupling . in sec . [ twoloop ] we compute the two - loop term of the @xmath0 function . in sec . [ cascade ] we relate the plausible strong - coupling behavior of the theory to the phenomenology of the inverse cascade . in sec . [ burgers ] , as a test of our methods , we consider a rather different model , the uv - stirred one - dimensional burgers equation . a summary and discussion are presented in sec . [ discussion ] . the navier stokes equation for the velocity field @xmath12 of an incompressible two - dimensional fluid is @xmath13 where @xmath14 is the pressure divided by the density , @xmath15 is the kinematic viscosity , @xmath16 is the friction coefficient , and @xmath17 is the force per unit mass . the incompressibility condition @xmath18 allows the velocity to be expressed as @xmath19 where @xmath20 is a pseudoscalar field called the stream function and @xmath21 is the alternating tensor , which in two dimensions satisfies @xmath22 upon writing eq . ( [ nsv ] ) in terms of @xmath20 and applying the operator @xmath23 , we obtain @xcite @xmath24 here @xmath25 with the notation @xmath26 in the real three - dimensional world , eq . ( [ nsv ] ) is a good approximation for a thin fluid film provided either ( a ) the boundary surface(s ) and the coordinate system are rotating rapidly about a perpendicular axis or ( b ) the fluid is conducting and subject to a strong perpendicular magnetic field @xcite . in either case , boundary - layer effects produce a linear friction parametrized by @xmath16 , which has units of frequency . for convenience , our field - theory calculations will use the method of dimensional regularization @xcite , based on continuation to a noninteger spatial dimension @xmath8 . because in the end we are concerned only with @xmath5 , we adopt a formal continuation of the navier stokes equation that preserves its two - dimensional features . for general @xmath27 , we retain the stream - function representation by choosing a `` physical '' two - dimensional subspace that contains the tensor @xmath21 and the external wavevectors of correlation functions . denoting by @xmath28 the projector onto the physical subspace , we now have @xmath29 we take eq . ( [ nspsi ] ) as the equation of motion for @xmath20 , with @xmath30 interpreted as the @xmath27-dimensional laplacian . in this way we preserve ( for @xmath31 ) the formal conservation of the energy and enstrophy , @xmath32 we assume that the fluid is stirred by a gaussian random force that is uncorrelated in time @xcite , with @xmath33 so that @xmath34 is also gaussian with @xmath35 a classical system with random forcing can be treated in the formalism of quantum field theory @xcite , including the path - integral representation @xcite . upon introduction of a pseudoscalar field @xmath36 conjugate to @xmath34 , correlation functions of @xmath20 are given by the path integral @xmath37\rangle \propto \int \mathcal{d}\psi\ , \mathcal{d}p\ , f[\psi]\ , e^{-s},\ ] ] with the action @xmath38\nonumber\\ = { } & { \int dt}\ , d^d x\ , \bigl[\tfrac{1}{2 } ( -\nabla^2 p)\ , d(-\nabla^2)p + i\nu p\nabla^4 \psi\nonumber\\ \label{sppsi } & - i\alpha p\nabla^2 \psi - ip\nabla^2 \dot\psi - i\epsilon_{ij } \nabla_i \nabla_k p \nabla_j \psi \nabla_k \psi\bigr],\end{aligned}\ ] ] where we have integrated the @xmath39 term by parts . the jacobian determinant from changing variables from @xmath34 to @xmath20 is an unimportant constant by virtue of causality @xcite . in the field theory based on the action ( [ sppsi ] ) , the long - distance behavior is governed by just the renormalizable terms those with coefficients whose scaling dimensions with respect to wavenumber in @xmath5 are zero ( marginal ) or positive ( relevant ) @xcite . we assign scaling dimensions @xmath40 , @xmath41 , @xmath42 to the fields @xmath20 and @xmath36 and to the time @xmath43 by requiring that the highest - derivative quadratic terms in the action have dimensionless coefficients in @xmath5 , since these terms control the asymptotic behavior of propagators and thus the uv convergence or divergence of diagrams . because of the additional derivatives , the viscous term @xmath44 is clearly less relevant than the friction term @xmath45 . in fact , it is commonly said that viscosity is irrelevant to the inverse cascade , but we now show that this can not be taken in the technical sense . if the viscous term is ignored , then the @xmath46 and @xmath47 terms have dimensionless coefficients only if @xmath48 and @xmath49 . for a nontrivial theory , the @xmath39 term must be renormalizable , giving @xmath50 . with @xmath51 , there exists no local renormalizable forcing term quadratic in @xmath36 . let us therefore retain the viscous term and recompute the scaling dimensions . the @xmath52 and @xmath47 terms possess dimensionless coefficients only if @xmath53 and @xmath49 . renormalizability of the @xmath39 term now gives @xmath54 , so that @xmath55 . it remains to specify the forcing term . we assume that the external forcing is confined to a band of high wavenumbers , as in model c of fns @xcite . the effective forcing at low wavenumbers is generated by renormalization ; because the interaction in eq . ( [ sppsi ] ) has two spatial derivatives acting on @xmath36 , at least two derivatives must accompany each factor of @xmath36 in any term so generated . the only renormalizable forcing term is then @xmath56 , whose coefficient is dimensionless for @xmath57 . this effective forcing has @xmath58 , as in model a of fns . all terms in the action are now marginal , except for the friction term , which is relevant ( coefficient of dimension @xmath59 ) . the only other renormalizable terms that could be generated are ones containing only @xmath20 , but these are not generated ( see sec . [ feynman ] ) . next we label the fields and the time in eq . ( [ sppsi ] ) with the subscript `` phys '' and introduce rescaled variables to simplify the action . tentatively seeking to set all coefficients other than forcing and friction equal to @xmath60 , we take @xmath61 the result is @xmath62 where @xmath63 we adopt , however , a different rescaling that will be particularly convenient in sec . [ fdt ] : @xmath64 the final form of the action is then @xmath65 where @xmath66 for general @xmath8 , the scaling dimensions in eq . ( [ sfppsi ] ) are @xmath67 in two dimensions , @xmath68 is a dimensionless coupling and @xmath69 is analogous to a mass parameter in quantum field theory . feynman rules for the action ( [ sfppsi ] ) . ] correlation functions can be calculated for the action ( [ sfppsi ] ) using feynman diagrams whose lines carry both frequencies and wavevectors . we represent the fields @xmath20 and @xmath36 by wiggly and plain lines respectively . the ingredients of the diagrams , shown in fig . [ feynrules ] , are the propagators @xmath70 , @xmath71 , obtained from the quadratic terms in the action , and the vertex factor @xmath72 , obtained from the cubic term . we label these quantities with the subscript @xmath73 because they are the tree - level contributions to the exact two - point correlation functions @xmath74 , @xmath75 and the exact three - point one - particle - irreducible ( 1pi ) function @xmath76 . the remaining contributions arise from diagrams containing loops . for these diagrams , we integrate over each loop frequency and wavevector according to @xmath77 because the integrand is a rational function of the frequencies , the @xmath78 integrations can easily be performed by the contour method before integrating over wavevectors . this contour integration shows that a 1pi diagram ( or subdiagram ) vanishes if all its external lines are @xmath20 @xcite , since there is a closed loop of @xmath70 propagators and the integrand is an analytic function of the loop frequency in the upper half - plane . unlike many field theories in a low number of spatial dimensions , ours does not contain ir divergences even for @xmath79 this is because , after integration over frequencies , internal - line propagators scale as @xmath80 , but at least one further factor of the wavevector arises from each of the two vertices that a line connects . hence the integrand does not diverge as the wavevector of any internal line goes to zero . the frequencies and wavevectors of the external lines act as an ir cutoff . for simplicity , our calculations will adopt another ir cutoff by assuming that @xmath81 ; then 1pi diagrams are analytic at zero external frequencies and wavevectors . one - loop diagrams for the two - point 1pi functions @xmath82 and @xmath83 at zero frequency . ] the coefficients of the quadratic terms in the action ( with frequency @xmath84 and wavevector @xmath85 ) are @xmath86 these are corrected at one loop by the two - point 1pi diagrams in fig . [ 1l1pi ] . because of the external wavevectors in the vertex factors , the diagrams are @xmath87 , and so we can set @xmath88 . we now demonstrate the computation of the @xmath89 diagram , to show the basic methods to be used for two - loop diagrams in sec . [ twoloop ] . the frequency integral of the propagators is @xmath90\nonumber\\ & = \frac{-ig/4k^2}{(g\rho + k^2 - \mathbf{k}\cdot\mathbf{q } + \frac{1}{2}q^2 ) ( g\rho + k^2 - 2\mathbf{k}\cdot\mathbf{q } + q^2)},\end{aligned}\ ] ] obtained conveniently by closing the contour in the upper half - plane and picking up one pole . we next multiply by the vertex factors and expand to @xmath87 : @xmath91 i\nonumber\\ \to { } & \frac{6ig ( \mathbf{k}\cdot\mathbf{q})^2 [ k^2_\parallel q^2 - ( \mathbf{k}\cdot\mathbf{q})^2 ] } { 4(g\rho + k^2)^3}\nonumber\\ \label{q4 } & - \frac{ig [ k^2 q^2 + 4(\mathbf{k}\cdot\mathbf{q})^2 ] [ k^2_\parallel q^2 - ( \mathbf{k}\cdot\mathbf{q})^2]}{4k^2 ( g\rho + k^2)^2},\end{aligned}\ ] ] where @xmath92 is the squared projection of @xmath93 onto the physical subspace and we infer from eq . ( [ epseps ] ) that @xmath94 we omit @xmath95 terms in eq . ( [ q4 ] ) because they will now disappear when we average over directions of @xmath93 . with the @xmath27-dimensional isotropization formulas @xmath96 which imply @xmath97 the value of the diagram to @xmath87 becomes @xmath98 for @xmath8 with @xmath99 , we find @xmath100 where @xmath101 is the euler constant . the result for the other one - loop diagram ( taking into account the symmetry factor ) is very similar : @xmath102 we have carefully obtained the @xmath103 terms , which will be important in sec . [ twoloop ] . the method of minimal subtraction @xcite expresses the bare couplings @xmath68 and @xmath69 , of respective dimensions @xmath104 and @xmath105 , in terms of a renormalization scale @xmath106 ( dimension @xmath60 ) and dimensionless renormalized couplings @xmath107 and @xmath108 as @xmath109,\\ \rho & = \mu^{2-\epsilon/2 } \bar\rho,\end{aligned}\ ] ] where the corrections ( called counterterms ) involve only negative powers of @xmath7 . the coefficient of the @xmath110 counterterm , and the absence of any counterterms for @xmath108 , are obtained by requiring that @xmath111 and @xmath112 to @xmath87 be finite at @xmath9 when expressed in terms of @xmath107 and @xmath108 : @xmath113\span\span\nonumber\\ = { } & q^2 ( \chi + i\mu^{2-\epsilon/2 } \bar\rho)\span\span\nonumber\\ & + { } & & iq^4 \mu^{-\epsilon/2}\nonumber\\ & & & \times \left[\bar g^{-1 } + \bar g \left(\frac{1}{32\pi\epsilon } - \frac{a_1}{\epsilon } + o(\epsilon^0)\right ) + o(\bar g^3)\right ] , \displaybreak[0]\\ \gamma^{pp } = { } & q^4 \mu^{-\epsilon/2}\span\span\nonumber\\ & \times \left[\bar g^{-1 } + \bar g \left(\frac{1}{32\pi\epsilon } - \frac{a_1}{\epsilon } + o(\epsilon^0)\right ) + o(\bar g^3)\right].\span\span\end{aligned}\ ] ] hence @xmath114 and no counterterms of any order are needed for @xmath108 because the @xmath115 term of @xmath111 is not renormalized . as for the three - point 1pi function @xmath76 , we will see in sec . [ galilean ] that it is related by galilean invariance to the @xmath116 term of @xmath111 . because the latter term is not renormalized , we have @xmath117 up to irrelevant terms with more factors of wavevector . we have thus rendered the theory finite at one loop by renormalizing only the coupling @xmath107 , without the need for counterterms to rescale the fields @xmath20 and @xmath36 or the time @xmath43 . crucial for this was the equality of the @xmath118 terms in eqs . ( [ 1lpsip ] ) and ( [ 1lpp ] ) . we conclude that the anomalous dimensions are zero at one loop , and in sec . [ fdt ] we will show that in minimal subtraction they are exactly zero , @xmath119 in a different renormalization scheme , or with a different definition of the fields and the time such as eq . ( [ altpsipt ] ) , the anomalous dimensions would not vanish identically , but at any rg fixed point their values are universal @xcite and so they would be zero there . the @xmath0 functions for the dimensionless couplings are determined by the rg invariance of the bare couplings , @xmath120 we obtain @xmath121 the force correlation adopted by fns @xcite differs from eq . ( [ forcing ] ) by a factor of @xmath59 , and consequently the dimensionless coupling of fns is @xmath122 . thus we have confirmed the fns result @xmath123 in two dimensions , the equation of motion ( [ nspsi ] ) and thus the action ( [ sfppsi ] ) have the important physical property of invariance under a galilean transformation to a reference frame moving with constant velocity @xmath124 @xcite . this property depends on the assumption that the stirring force is uncorrelated in time , since otherwise there would exist a link between a point in space at one time and a `` corresponding '' point in space at a different time . specifically , the action ( [ sfppsi ] ) is invariant under the transformation @xmath125 which induces the familiar transformation of the velocity , @xmath126 the original navier stokes equation ( [ nsv ] ) is not galilean invariant because of the friction term , which introduces a preferred state of rest ; but the differentiation in deriving the stream - function equation ( [ nspsi ] ) eliminates the constant shift in @xmath127 @xcite . we would not expect galilean invariance in @xmath5 to be preserved by our renormalization method unless the theory remains galilean invariant when dimensionally regulated . we now show that our formal stream - function representation in arbitrary @xmath27 is invariant under the transformation ( [ galtrans ] ) , provided that @xmath124 lies in the physical subspace . for convenience we use the infinitesimal form @xmath128 in the corresponding variation of the action ( [ sfppsi ] ) , terms proportional to @xmath43 vanish automatically because they correspond to a simple spatial translation . the interesting terms are those where the @xmath43 is differentiated ( @xmath129 ) and where @xmath20 is varied by @xmath130 . neither of these affects the forcing , viscous , or friction terms , which contain @xmath20 only as @xmath131 and are trivially invariant . we are left with @xmath132 upon integration by parts and use of @xmath133 , the first two terms cancel and the third vanishes . in fourier space , the transformation ( [ infgaltrans ] ) becomes @xmath134 the action is galilean invariant by virtue of the relation @xmath135 both sides equal @xmath136 . the same relation must then hold between the exact 1pi functions : the coefficient of @xmath137 in @xmath111 equals the coefficient of @xmath138 in @xmath139 . since the former is not renormalized , neither is the latter . for zero friction ( @xmath79 ) , the action ( [ sfppsi ] ) is equivalent to model a of fns @xcite , who note that it obeys detailed balance and thus is subject to a classical fluctuation - dissipation theorem ( fdt ) @xcite . a complicated diagrammatic argument demonstrates that the fdt is preserved to all orders of renormalization @xcite . here we reach this conclusion by obtaining the fdt from an exact symmetry of the action in arbitrary @xmath27 . under the formal discrete transformation @xmath140 the action with @xmath79 changes only by reversing the sign of the viscous term . the change in the interaction term vanishes upon integration by parts , just as in deriving conservation of energy . to restore the sign of the viscous term , we further perform a complete time reversal , @xmath141 which naturally reverses the sign of the dissipation . the net effect is the transformation @xmath142 a generalized time reversal that is its own inverse and leaves the action invariant . the fdt is derived from this symmetry by expressing the invariance of @xmath143 : @xmath144 by invariance under time translations and spatial rotations , negating the times in a two - point correlation function is equivalent to interchanging the points . as a first application of the fdt , we make use of the theorem that the equal - time correlation function @xmath145 is exact at tree level @xcite . thus , for @xmath79 , the exact equal - time stream - function correlation is @xmath146 the energy spectrum , in the units implied by eq . ( [ ppsit ] ) , is then @xmath147 this is an equipartition spectrum , in agreement with fns model a @xcite . for @xmath79 , the fdt ( [ fdteq ] ) also implies that , if @xmath74 and thus @xmath148 are made finite by renormalizing @xmath107 , then @xmath75 is likewise finite , without the need for field or time rescalings . as we have seen , the three - point 1pi function is automatically finite . hence the anomalous dimensions vanish exactly for @xmath79 . but in minimal subtraction , the counterterms for rescalings and for dimensionless couplings are independent of the mass parameter @xcite . we conclude that in minimal subtraction , even with friction , @xmath149 and @xmath150 depends only on @xmath107 . indeed , we have seen these statements verified to one loop in sec . [ oneloop ] . we have shown that the anomalous dimensions vanish , and that the renormalized couplings in exactly two dimensions obey @xmath151 thus @xmath108 is very simply related to @xmath106 and can be used to parametrize it . in the rg flow to low wavenumbers , @xmath108 steadily increases as friction becomes more important . meanwhile , @xmath107 flows in its own characteristic way regardless of the value of @xmath108 ; the most we can say is that once @xmath107 becomes small , it decreases further and further , approaching zero in the ir limit . the solution of eq . ( [ rggbar ] ) in this limit is @xmath152 where @xmath153 is the scale at which @xmath107 becomes large . in the ir limit , we expect good accuracy from perturbative results such as the tree - level expression for the energy spectrum , @xmath154 we might suppose that the true asymptotic behavior is given by replacing @xmath68 with the renormalized coupling @xmath155 . this would follow from rg theory if the loop corrections to eq . ( [ ekp ] ) contained @xmath156 . but with @xmath157 providing an ir cutoff , the diagrams are regular as @xmath158 and instead contain @xmath159 . hence , for the purposes of correlation functions , the rg flow effectively halts for wavenumbers below the `` mass '' @xmath160 . if @xmath161 is the suitably renormalized value of this mass , then as @xmath158 we expect that @xmath162 on the other hand , in the bare tree - level result ( [ ekp ] ) , @xmath68 can be interpreted as a coupling renormalized at a very high wavenumber ( the forcing scale or uv cutoff ) . with @xmath163 , we have @xmath164 , and so the rg result ( [ ekp1 ] ) gives a greater @xmath165 at low @xmath166 . whereas the bare tree - level calculation ignores all interactions between scales , the effect of renormalization is to place more energy and dissipation at low @xmath166 ( and therefore less at high @xmath166 ) , consistent with the inverse cascade . from eq . ( [ rgrhobar ] ) , any rg fixed point @xmath167 must have @xmath168 . not only do the anomalous dimensions vanish at any fixed point , but for @xmath169 we have the exact equipartition spectrum @xmath170 , whether @xmath107 is at a fixed point or not . it is clear that , despite the suggestive evidence of scale invariance of the inverse cascade , an rg fixed point in our framework can not be the explanation for the observed @xmath4 spectrum . nevertheless our theory contains all the essential ingredients that have produced the stationary inverse cascade experimentally and numerically . a natural explanation is that the @xmath4 spectrum arises from the nonperturbative behavior of correlation functions at @xmath171 and at large values of @xmath107 that flow rapidly with scale . although it is far from obvious how such an rg flow can produce approximate scale invariance with an effective anomalous dimension , we are motivated to seek hints about the theory s strong - coupling behavior . the first step , which can be useful for more mundane purposes as well , is to extend the renormalization to the next order of perturbation theory . calculating @xmath150 consistently to two loops requires a precise specification of the renormalization scheme . remarkably , though , as long as @xmath150 depends only on @xmath107 , the @xmath0 function to two loops ( but no further ) is independent of the particular scheme chosen @xcite . as in sec . [ oneloop ] , we take @xmath81 , compute two - point 1pi diagrams to @xmath87 expanded about @xmath9 , and renormalize by minimal subtraction . the two - loop expression for the bare coupling in terms of the renormalized coupling is @xmath172.\ ] ] with zero anomalous dimensions , @xmath150 is the only rg function to be determined and we need only compute a single 1pi function to two loops . below we will choose @xmath83 because its two - loop diagrams are technically simpler than those of @xmath82 . we have seen that @xmath157 , which has dimension @xmath59 independent of @xmath7 , acts as the ir cutoff for wavevector integrals . by dimensional analysis , @xmath112 to @xmath87 has the form @xmath173\nonumber\\ & = q^4 & & \biggl[g^{-1 } + g \left(\frac{b_1}{\epsilon } + b_1 ' - \tfrac{1}{2 } b_1 \ln(g\rho ) + o(\epsilon)\right)\nonumber\\ \label{2lpp } & & & + g^3 \left(\frac{b_2}{\epsilon^2 } + \frac{b_2 ' - b_2 \ln(g\rho)}{\epsilon } + o(\epsilon^0)\right ) + o(g^5)\biggr].\end{aligned}\ ] ] to simplify the two - loop calculations we formally set @xmath174 and restore the @xmath159 terms at the end . nonvanishing two - loop diagrams for the two - point 1pi function @xmath83 . along with the overall divergence , each diagram has at most one divergent subdiagram ( bold lines ) . ] multiloop diagrams can be characterized by their overall uv divergence ( as all loop frequencies and wavenumbers go to infinity together ) and their subdivergences ( as a subset of loop frequencies and wavenumbers go to infinity while the rest remain finite ) . overlapping divergences occur when two or more divergent subdiagrams share a propagator ; this is a definite complication in evaluating a diagram , though not insurmountable @xcite . some two - loop diagrams for the 1pi function @xmath82 contain overlapping divergences , so we choose to calculate the diagrams for @xmath83 , which fortunately do not ( fig . [ 2l1pi ] ) . a three - point 1pi subdiagram is divergent only if its external lines are @xmath175 , two wiggly and one plain ; galilean invariance does not eliminate this subdivergence when we treat each two - loop diagram separately , because the finiteness of @xmath176 results from a sum over distinct one - loop diagrams . in fig . [ 2l1pi ] we have omitted one conceivable diagram containing the 1pi subdiagram @xmath177 , which vanishes as noted in sec . [ feynman ] . previous two - loop calculations of similar complexity have been made for different problems using diagrams of the same topology : the burgers kardar - parisi - zhang equation for interface growth @xcite and the navier stokes equation in more than two physical dimensions , where there are fewer divergences @xcite . we have programmed mathematica @xcite to automate the steps in evaluating each two - loop diagram . the loop frequencies are integrated one after the other by adding the residues of all poles in the upper half - plane . the resulting integrand , containing the external wavevector @xmath85 and the loop wavevectors @xmath178 , is expanded to @xmath87 and averaged over directions of @xmath85 in the physical subspace . the numerator of the integrand is now rife with the projector @xmath28 , both from averaging @xmath179 and from applying eq . ( [ epseps ] ) to cross products . to eliminate @xmath28 , we average the integrand over orientations of the physical subspace within @xmath27-dimensional space : we introduce an orthonormal physical basis , write @xmath180 , average over directions of @xmath181 in the @xmath182 dimensions perpendicular to @xmath183 , and then average over directions of @xmath183 in @xmath27-dimensional space . the result is a function of @xmath184 , @xmath185 , and @xmath186 to be integrated over @xmath187 . because we need only the divergent parts of the two - loop diagrams as indicated in eq . ( [ 2lpp ] ) , we eliminate numerator terms that produce neither an overall divergence nor a subdivergence by power counting . we seek to integrate first over the loop wavevector ( say @xmath188 ) associated with the subdivergence ( if any ) . to simplify the denominator , feynman parameters @xcite are introduced , and @xmath188 is translated by a multiple of @xmath189 . after the numerator is isotropized , the @xmath188 integration is done analytically ; there remains an integral over feynman parameters and over the magnitude @xmath190 . from eq . ( [ 2lpp ] ) , the integrand ( excluding @xmath191 ) has dimension @xmath192 . the part that behaves like @xmath193 as @xmath194 is subtracted and separately integrated over @xmath190 , producing another @xmath195 factor ; all the @xmath196 terms arise here and are found analytically , but the @xmath195 terms involve intractable feynman - parameter integrals . meanwhile , the remaining subtracted integral converges at @xmath197 but contains analytically integrable @xmath195 poles from the @xmath188 integration . adding the divergent parts of the eight diagrams gives @xmath198 here @xmath199 is a sum of feynman - parameter integrals of complicated rational functions with integer coefficients ; thus we expect that @xmath199 may be a rational number . though we are unable to calculate @xmath199 analytically , eq . ( [ an2lpp ] ) already displays some important features : if the @xmath196 term were different , the renormalization performed below would become inconsistent @xcite ; and if the coefficients of @xmath101 and @xmath200 were different , these constants would ( contrary to expectation ) appear in the two - loop @xmath0 function . we have evaluated @xmath199 numerically by multidimensional monte carlo integration , using the technique of importance sampling @xcite to select more points in the `` corners '' of feynman - parameter space ( with one parameter near @xmath60 and the others near @xmath73 ) . this technique improves the statistics because the integrands tend to diverge in these corners ( but slowly enough that the integrals converge ) . we treat the integrals that make up @xmath199 separately , since they vary in complexity and in number of feynman parameters . to optimize the precision of the result for @xmath199 in a given computation time @xmath201 , we reason as follows . the contribution to @xmath199 from diagram @xmath202 , evaluated with @xmath203 sample points , is obtained with a precision @xmath204 in a time @xmath205 , where @xmath206 and @xmath207 are characteristics of the integrand and the computer . constrained optimization shows that we achieve the best overall precision @xmath208 in the time @xmath209 by choosing @xmath210 short runs are made to estimate @xmath206 and @xmath207 , and then the high - precision integrations are performed with this plan . our computations for a total of @xmath211 sample points yield @xmath212 strongly suggesting that the exact value is @xmath213 with the one - loop result ( [ 1lpp ] ) and the two - loop result ( [ an2lpp ] ) , we have @xmath214.\end{gathered}\ ] ] by substituting eq . ( [ 2lg ] ) and requiring a finite expression at @xmath9 , we determine @xmath215 the rg invariance of the bare coupling @xmath68 finally gives @xmath216 this is reminiscent of the @xmath0 function in four - dimensional @xmath217 theory , where similarly the one - loop term is positive and the two - loop term is negative @xcite . a positive @xmath0 function that grows too quickly ( faster than linearly ) at large coupling raises the question of whether the coupling becomes infinite at a large but finite renormalization scale . in our theory this would suggest an absolute limit on the extent of the scaling range for any inverse cascade . to avoid such a fate , the expansion of the @xmath0 function must contain many negative terms to slow its initial superlinear growth . it is pleasing to find such a term already at two loops . the initial prediction of the two - dimensional inverse energy cascade @xcite assumed zero friction and small but nonzero viscosity . kinetic energy , continually injected at high wavenumbers , is expected to cascade down through a quasisteady inertial range extending to lower and lower wavenumbers as time passes . within this range , the assumption of scale invariance implies the energy spectrum @xmath218 where @xmath219 is the two - dimensional kolmogorov constant and @xmath220 is the rate of energy injection per unit mass . numerical simulations with zero friction @xcite confirm this spectrum until the cascade approaches the minimum wavenumber associated with a finite system . both nonstationary behavior and finite - size effects , however , lie outside our theoretical framework , which treats a fluid of infinite size that has been stirred for an infinite time . in the absence of friction , such a fluid has an equipartition spectrum as shown in sec . [ fdt ] . when friction is introduced , it is natural to expect a mirror image of the viscous energy dissipation at high wavenumbers in the three - dimensional direct cascade : a stationary inverse cascade should develop , with dissipation by friction at low wavenumbers @xmath221 and with the spectrum ( [ ek ] ) at wavenumbers @xmath222 where dissipation is unimportant . dimensional analysis gives @xmath223 where @xmath16 is the friction coefficient @xcite . several numerical simulations @xcite and laboratory experiments @xcite confirm this picture . other numerical studies obtain similar results but are more difficult to relate to our framework , since they modify the friction term by removing derivatives @xcite or by applying friction only below a cutoff wavenumber @xcite . we are less concerned about the common practice of adding derivatives to the viscous term ( hyperviscosity ) , because such a modified term is irrelevant and the rg flow should introduce a normal viscous term to replace it . but friction modified with inverse derivatives ( hypofriction ) is certainly relevant and is believed to alter the dynamics of the inverse cascade @xcite . in an extreme case , hypofriction with eight inverse laplacians destroys the @xmath4 spectrum @xcite . hypofriction is intended to confine dissipation explicitly to low wavenumbers ; ordinary linear friction accomplishes the same thing more gently , but makes it difficult in practice to achieve an inertial range @xcite . we have used linear friction because it is physically realistic and leads to a local field theory . for convenience in relating the observed inverse cascade to our field theory with the action ( [ sfppsi ] ) , we adopt units of time in which @xmath224 so that the rescalings in eq . ( [ ppsit ] ) are trivial . then the kinematic viscosity is @xmath225 the friction coefficient is @xmath226 and the force correlation is @xmath227 in contrast to the formal methods of dimensional regularization and minimal subtraction , a simple uv cutoff renders our theory finite in a physically meaningful way while still preserving the symmetries noted in sec . [ properties ] . a cutoff would have been inconvenient for the two - loop calculations of sec . [ twoloop ] , but it is appropriate for understanding experimental and numerical results on the inverse cascade . the local , renormalizable action ( [ sfppsi ] ) applies only with a uv cutoff @xmath228 below the wavenumbers of the external forcing @xcite . a corresponding renormalization prescription is obtained by staying in exactly two spatial dimensions and writing the coupling @xmath68 in eq . ( [ sfppsi ] ) as a cutoff - dependent quantity @xmath229 , such that the long - distance behavior is independent of @xmath228 . no cutoff dependence is needed for @xmath69 , because the friction term is not renormalized , as we saw in sec . [ oneloop ] . special properties of minimal subtraction allowed us to conclude ( sec . [ fdt ] ) that in that scheme the anomalous dimensions vanish and @xmath150 depends only on @xmath107 . we may expect that in the cutoff scheme these statements remain approximately true , at least for small values of the dimensionless friction parameter @xmath230 . because the @xmath0 function is scheme independent through two loops @xcite , we have @xmath231 this result is directly useful for weak coupling , but can only hint at the possible strong - coupling behavior . we will now show that a specific strong - coupling form of the @xmath0 function is required for a fully developed inverse cascade , and that it can be naturally interpolated with the two - loop result ( [ lbeta ] ) . the key condition for the ideal inverse cascade is that the dissipation of energy is dominated by friction and is almost totally confined to low wavenumbers . this means that the dissipation rate @xmath232 should be independent of the uv cutoff @xmath228 , because a change in @xmath228 produces neither a renormalization of @xmath69 nor a rescaling of @xmath20 . under stationary conditions , @xmath220 equals the rate of energy injection , given by @xcite @xmath233 in terms of the force correlation ( [ fcg ] ) . for @xmath220 to be independent of @xmath228 , we must have the strong - coupling behavior @xmath234 corresponding to the asymptotically linear @xmath0 function @xmath235 we now attempt to connect this form with the two - loop @xmath0 function ( [ lbeta ] ) . perturbative expansions such as eq . ( [ lbeta ] ) are usually divergent , but a borel transformation is expected to produce a finite radius of convergence about zero coupling @xcite . we see that the true expansion parameter is @xmath236 , and the alternatively normalized action ( [ sippsi ] ) makes it clear that the theory is unstable for @xmath237 . thus we write @xmath238 for the perturbation series @xmath239 the borel transform is @xmath240 in our case , @xmath241 the borel transform often has poles on the negative real axis associated with instantons @xcite , but @xmath242 is well - defined from eq . ( [ bt ] ) as long as @xmath243 is regular on the positive real axis . a simple and suitable rational ( pad ) approximant to @xmath243 is @xmath244 fortunately , the choice @xmath245 agrees with eq . ( [ bser ] ) . we thus take @xmath246 but we do not here attempt to study the possible instanton solutions corresponding to the pole at @xmath247 . ( [ bt ] ) then gives precisely the desired asymptotic behavior @xmath235 we now ask whether the observed @xmath4 energy spectrum ( [ ek ] ) is consistent with our theory , although we are unable to derive it systematically . we conjecture that strong - coupling effects produce the spectrum @xmath248 for @xmath249 from eq . ( [ egl ] ) , the running coupling is @xmath250 for nonperturbative effects to be operative down to the wavenumber @xmath251 , we must have @xmath252 in the borderline case where @xmath253 , we have @xmath254 , and we can match eq . ( [ ek1 ] ) in order of magnitude with the perturbative energy spectrum ( [ ekp ] ) : @xmath255 even when @xmath256 , we may guess from eq . ( [ ekp ] ) that in order of magnitude @xmath257 and this again matches eq . ( [ ek1 ] ) at @xmath251 . in the case @xmath256 , the running coupling eventually becomes @xmath258 at a lower wavenumber @xmath259 below which perturbation theory is applicable and the energy spectrum is given roughly by eq . ( [ ekp1 ] ) . thus we envision a varied but continuous behavior of the energy spectrum in the different regions : @xmath260 as a further check , we note that the contribution to the energy dissipation rate ( [ diss ] ) for each of the three regions is @xmath261 ( up to a logarithmic factor for the middle region ) . this suggests that a fully developed stationary inverse cascade has several distinct dissipation ranges . testing by experiments and simulations is not straightforward , because finite - size effects may become important before the lower - wavenumber ranges are reached . we have described the ideal inverse cascade in terms of external forcing confined to high wavenumbers ( assumed in sec . [ relevance ] ) and energy dissipation practically confined to low wavenumbers ( made plausible in secs.[rg ] and [ strong ] ) . this separation is equivalent to the conventional criterion of an inertial range with a constant energy flux . however , unlike the energy spectrum ( a simple correlation function ) , the energy flux in wavenumber space is not invariant under the rg flow and is not given straightforwardly by our local field theory . renormalization introduces the forcing ( [ fcg ] ) , which is nonzero even for wavenumbers in the inertial range and so makes the flux appear nonconstant . this effective forcing simply represents the energy transfer from wavenumbers @xmath262 that have been integrated out . we can resolve the flux problem by working instead in physical space , where the effective forcing is a differentiated delta function that is zero at finite distances , including the inertial range of lengths . thus we inquire whether the physical - space energy flux @xcite @xmath263 and is independent of @xmath264 by homogeneity . under stationary conditions , the navier stokes equation ( [ nsv ] ) yields @xcite @xmath265 \langle \mathbf{v}(\mathbf{x } ) \cdot \mathbf{v}(\mathbf{x}+\mathbf{r } ) \rangle,\end{aligned}\ ] ] where @xmath266 is the vanishing physical - space force correlation , and we have used eqs . ( [ nug ] ) and ( [ alpharho ] ) . the right - hand side of eq . ( [ er1 ] ) can be interpreted as minus the rate of energy dissipation at length scales larger than @xmath267 . since @xmath268 is an rg - invariant correlation function , only the viscosity @xmath269 introduces cutoff dependence . in accordance with our argument in sec . [ strong ] , as long as the energy dissipation is dominated by friction , the dissipation rate is independent of @xmath228 , giving a well - defined energy flux @xmath270 proportional to the fourier transform of the energy spectrum . and if this spectrum is almost entirely concentrated at low wavenumbers @xmath271 , it follows that in the inertial range ( @xmath272 ) the flux is constant , @xmath273 a striking feature observed in the inverse cascade is the approximate scale invariance of inertial - range velocity correlations ( structure functions ) . the @xmath4 energy spectrum ( [ ek ] ) gives for the second moment @xcite @xmath274 and the constant energy flux ( [ er ] ) gives for the third moment @xcite @xmath275 in our field theory , since the anomalous dimensions vanish , such moments can be written in the form @xmath276 where @xmath277 is the kinematic scaling and @xmath278 is a function of the dimensionless running couplings . ( [ vn ] ) should reduce to the observed forms in the inertial range @xmath279 subject to the condition for a fully developed cascade , @xmath280 conditions ( [ inr ] ) and ( [ gkfr ] ) are equivalent to @xmath281 to achieve complete scale invariance of inertial - range structure functions ( absence of intermittency ) , we must have @xmath282 for the range of arguments ( [ cond ] ) ; then @xmath283 , and @xmath284 eq . ( [ fn ] ) requires a specific nonperturbative behavior of velocity correlations at strong coupling , mimicking the effect of an anomalous dimension . rg theory alone does not constrain the functions @xmath278 . the well - established second moment ( [ vv ] ) and third moment ( [ vvv ] ) strongly suggest that an exact calculation in our theory would yield the required behavior of @xmath285 and @xmath286 . but in the absence of a field - theoretic reason why eq . ( [ fn ] ) should persist for @xmath287 , high - order structure functions may generically be expected to violate scale invariance and produce intermittency . in sum , we would not be at all surprised by observations of intermittency in the inverse cascade , but a seeming total absence of intermittency would raise questions about unknown properties of our theory that enforce effective scale invariance . some numerical studies of higher moments @xcite deal with a nonstationary inverse cascade and find no signs of intermittency . such results are not directly relevant to this paper but have been explained theoretically @xcite based on the growth of the inertial range with time . for the stationary inverse cascade , laboratory experiments @xcite reveal no evidence of intermittency . the results of stationary numerical simulations , however , are mixed : a study using linear friction @xcite confirms the absence of intermittency , but a study using hypofriction with one inverse laplacian @xcite obtains intermittency that is described as strong and as similar to that observed in the three - dimensional direct cascade . surprisingly , results from the latter simulation are also presented in a subsequent paper @xcite where the intermittency is described as insignificant . the evidence on intermittency in the stationary inverse cascade is unclear , and further numerical studies would be useful to resolve the question . results exhibiting strong intermittency are most natural , from the viewpoint of our field theory . the mild hypofriction used in the numerical study obtaining strong intermittency @xcite is unlikely to alter the qualitative structure of our theory . while the inverse laplacian renders the initial navier stokes equation ( [ nsv ] ) nonlocal , the differentiated form ( [ nspsi ] ) and the field - theory action ( [ sfppsi ] ) are still local in terms of the stream function @xmath20 . the hypofriction term certainly violates galilean invariance , but this symmetry holds for zero hypofriction and so its implications for the rg functions persist in minimal subtraction even when hypofriction is added . the same arguments used for linear friction in this paper suggest that intermittency should be expected with hypofriction as well . it would be interesting to perform a direct numerical study of the effect of modified friction on intermittency in the stationary inverse cascade . as an example of an alternative statistical fluid model to which our rg methods can be applied but which exhibits different behavior , let us briefly consider the one - dimensional burgers equation @xcite @xmath288 here @xmath289 is the velocity field of a fluid without pressure , @xmath290 is the force per unit mass , @xmath15 is the kinematic viscosity , and @xmath16 is the friction coefficient ( not normally included in the burgers equation but useful in controlling the long - distance behavior ) . as with the two - dimensional incompressible fluid , we assume that the forcing is confined to a band of high wavenumbers , and we study the response at lower wavenumbers . fns @xcite found that the uv - stirred burgers equation , like the navier stokes equation , has a dimensionless coupling in two spatial dimensions and can be analyzed by an @xmath7-expansion in @xmath291 . but their continuation of the burgers equation to @xmath292 does not preserve important one - dimensional properties , and their @xmath7-expansion is not fully consistent @xcite . with @xmath293 , eq . ( [ b1 ] ) yields conservation of the `` energy '' @xmath294 this @xmath295 is not proportional to the physical energy of a pressure - free fluid , because it does not account for the varying density ; nevertheless , @xmath295 is conserved in @xmath296 , and leads to a fluctuation - dissipation theorem ( fdt ) . it is not easy , however , to generalize eq . ( [ b1 ] ) to @xmath292 so that a similar `` energy '' is conserved , while maintaining other key properties such as galilean invariance . this is the task we now address . if we neglect dissipation and forcing , the conservation of @xmath295 in @xmath296 follows from the relation @xmath297 the analogue in @xmath292 that would yield conservation of @xmath298 is @xmath299\nonumber\\ & \equiv \mathbf{v } \cdot \mathbf{w}.\end{aligned}\ ] ] unfortunately , whatever the choice of the constants @xmath300 and @xmath301 , the quantity @xmath302 is not galilean covariant and is not a suitable generalization of @xmath303 . to remedy this problem , we impose the potential - flow condition @xmath304 as is commonplace when considering a multidimensional burgers equation . then , because @xmath305 a scalar equation of motion can conserve @xmath306 we take @xmath307 if @xmath308 , this expression is covariant under the galilean transformation @xmath309 which induces @xmath126 upon restoring dissipation and forcing , we therefore propose @xmath310 as a fully satisfactory multidimensional burgers equation that reduces to eq . ( [ b1 ] ) in @xmath296 . ( [ bd ] ) is formally very similar to the incompressible stream - function equation ( [ nspsi ] ) , and we will follow our previous analysis closely . as in sec . [ framework ] , we assume gaussian forcing and introduce a path integral with the action @xmath311\nonumber\\ = { } & { \int dt}\ , d^d x\ , \bigl[\tfrac{1}{2 } ( -\nabla^2 p)\ , d(-\nabla^2)p + i\nu p\nabla^4 \psi - i\alpha p\nabla^2 \psi\span\span\nonumber\\ & - ip\nabla^2 \dot\psi - i\nabla_i \nabla_j p\ , \bigl(\tfrac{1}{6 } \delta_{ij } \nabla_k \psi \nabla_k \psi + \tfrac{1}{3 } \nabla_i \psi \nabla_j \psi\bigr)\bigr],\span\span\end{aligned}\ ] ] where we have again integrated the @xmath39 term by parts . analysis of dimensions in @xmath8 proceeds as before , since all terms contain the same numbers of derivatives as for the incompressible fluid . the final burgers action , to be compared with eq . ( [ sfppsi ] ) , is @xmath312.\end{aligned}\ ] ] the scaling dimensions are again given by eq . ( [ deps ] ) . the only change to the feynman rules in fig . [ feynrules ] is the vertex factor , which now becomes @xmath313 thanks to conservation of the `` energy '' @xmath295 , the fdt is obtained just as in sec . [ fdt ] , and together with galilean invariance it implies that the anomalous dimensions vanish . thus , to find the new one - loop @xmath0 function , we need only recompute one of the diagrams in fig . [ 1l1pi ] . we calculate @xmath314 to be compared with eq . ( [ 1lpsip ] ) . curiously , the @xmath315 term is exactly the same and again leads to the one - loop @xmath0 function @xmath316 but the @xmath103 term in eq . ( [ 1lpsipb ] ) , which would enter a calculation of the two - loop @xmath0 function , differs from eq . ( [ 1lpsip ] ) , and we have no reason to expect an identity between the @xmath0 functions beyond one loop . our multidimensional burgers equation ( [ bd ] ) appears not to have been investigated previously , and it is possible that numerical simulations in two or three dimensions could reveal interesting and unexpected behavior . the focus here , however , is on the extrapolation to one dimension ( @xmath317 ) , where we seek to explain the known behavior @xcite of the ordinary burgers equation ( [ b1 ] ) . if @xmath7 is positive and small , then the @xmath0 function ( [ betab ] ) has a nontrivial ir - stable fixed point at @xmath318 by contrast , when a simpler continuation of the burgers equation was used , the fixed point appeared to be ir - unstable near two dimensions @xcite . from our results it is natural to expect an ir - stable strong - coupling fixed point in one dimension ( @xmath11 ) . due to the nontrivial fixed point , the response of the one - dimensional burgers equation to uv forcing is different from that of the two - dimensional navier stokes equation even in the absence of friction . at wavenumbers low enough that the fixed point is reached , burgers correlation functions ( at equal or unequal times ) exhibit purely kinematic scaling , since all anomalous dimensions vanish . for example , from eq . ( [ deps ] ) , the scaling of time and frequency is given by @xmath319 meanwhile , in both models , the fdt implies that the energy spectrum ( an equal - time correlation function ) obeys equipartition . these conclusions agree with previous analytic @xcite and numerical @xcite studies of the frictionless one - dimensional burgers equation . we have argued that the inclusion of friction dramatically alters the behavior of the two - dimensional navier stokes model , from equipartition to a stationary inverse cascade . this is plausible only in the presence of a large , rapidly flowing coupling @xmath1 . we noted in sec . [ cascade ] how this rg flow could give rise to a constant energy flux and a @xmath4 spectrum . no such profound effect is expected for friction in the one - dimensional burgers equation . correlation functions will be modified at very low wavenumbers @xmath320 , but as long as @xmath1 remains close to the fixed point , higher wavenumbers will retain the equipartition spectrum . an open question concerns the strong - coupling behavior of our generalized burgers equation ( [ bd ] ) in two dimensions with friction . if the @xmath0 function happens to grow asymptotically in the same way as the navier stokes one , then the arguments of sec . [ cascade ] can be repeated and it is at least possible that this burgers model could exhibit an inverse energy cascade . if true , this should also be evident if the equation is studied numerically . it is unclear , however , whether this multidimensional galilean - invariant scalar field theory has a direct physical application . we have presented a statistical field theory capable of describing the stationary inverse energy cascade in two - dimensional incompressible turbulence , and computed the rg functions through two loops . by contrast with previous rg studies of turbulence , we have taken advantage of the nature of the inverse cascade , with external forcing confined to high wavenumbers , to work with a conventional local field theory . the consequences of galilean invariance and the fluctuation - dissipation theorem have been systematically derived based on the underlying symmetries of the action . after taking these symmetries into account , we have evaluated the necessary two - loop diagrams in dimensional regularization to obtain the two - loop @xmath0 function @xmath321 which is independent of the renormalization scheme to precisely this order . because the anomalous dimensions vanish identically in minimal subtraction , no rg fixed point can yield the observed @xmath4 energy spectrum of the inverse cascade . instead , we have found that the inverse cascade could plausibly arise from nonperturbative strong - coupling effects in the presence of friction . the apparent anomalous dimension of the velocity must arise from the rapid rg flow of the coupling . cutoff independence of the energy dissipation rate ( or constancy of the energy flux ) requires the strong - coupling behavior @xmath322 and we have also obtained this in a heuristic way from a borel transformation of the two - loop @xmath0 function . the observed energy spectrum in the inertial range has been matched with perturbative results at the wavenumbers where dissipation becomes important . inertial - range intermittency ( violation of scale invariance ) is generically expected because the coupling flows rapidly with scale , but the evidence on intermittency from numerical simulations is mixed . on the other hand , a similar rg analysis of the one - dimensional burgers equation confirms the simpler behavior of that model , controlled by a strong - coupling fixed point . our greatest difficulty is that the inverse cascade appears to be intrinsically a nonperturbative phenomenon . to make quantitative predictions , such as the exponent of the energy spectrum or the value of the kolmogorov constant , it may be useful to combine a high - order perturbative calculation with an appropriate resummation method , as in our analysis of the strong - coupling @xmath0 function . of course evaluating additional diagrams with two or more loops will be very challenging , especially if the finite parts are needed . at the present stage , the most intriguing prediction of our theory is that substantial intermittency should be expected in the stationary inverse cascade . it would be helpful to have more robust and consistent results from experiments and simulations to determine whether this expectation is realized , and if it is not , to identify the field - theoretic explanation . this material is based upon work supported by the national science foundation under grant no . any opinions , findings , and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the national science foundation .

a result of haglund implies that the-bigraded hilbert series of the space of diagonal harmonics is a-ehrhart function of the flow polytope of a complete graph with netflow vector we study the-ehrhart functions of flow polytopes of threshold graphs with arbitrary netflow vectors . our results generalize previously known specializations of the mentioned bigraded hilbert series at , , and . as a corollary to our results , we obtain a proof of a conjecture of armstrong , garsia , haglund , rhoades and sagan about the-ehrhart function of the flow polytope of a complete graph with an arbitrary netflow vector .

1610.08370

the _ space of diagonal harmonics _ @xmath6 \quad \left| \quad\sum_{i=1}^n \frac{\partial^h}{\partial x_i^h } \frac{\partial^k}{\partial y_i^k } f=0 \ ; \text { for all } \ ; h+k>0 \right . \right\}\ ] ] was introduced by garsia and haiman @xcite in their study of macdonald polynomials . haiman @xcite proved using algebro - geometric arguments that it has dimension @xmath7 as a vector space over @xmath8 . the space @xmath9 is naturally bigraded by the degree of the variables @xmath10 and @xmath11 . thus , one can obtain a @xmath12-analogue of @xmath7 by considering the bigraded hilbert series of @xmath9 , which we denote by @xmath13 . this is a symmetric polynomial in @xmath14 and @xmath15 with nonnegative coefficients . the number @xmath7 counts spanning trees of the complete graph on @xmath16 vertices or parking functions of size @xmath17 . a combinatorial model for this bigraded hilbert series in terms of these objects was conjectured by haglund and loehr @xcite in 2002 and settled in 2015 by carlsson and mellit @xcite in their proof of the more general shuffle conjecture @xcite . stated in terms of parking functions , the result is the following ( see ( * ? ? ? * conj . 2 ) ) . [ hlhilb ] @xmath18 where @xmath19 denotes parking functions of size @xmath17 . for more background on @xmath9 and the shuffle conjecture see @xcite . for the definition of @xmath20 and @xmath21 on parking functions see ( * ? ? ? * , 2 ) . special cases of this hilbert series when @xmath22 are combinatorially appealing : @xmath23_q ! , \label{case : q0}\\ q^{\binom{n}{2 } } \operatorname{hilb}_{q , q^{-1}}(dh_n ) & = [ n+1]_q^{n-1 } , \label{case : qqinv } \end{aligned}\ ] ] where @xmath24 is the symmetric group of size @xmath17 , @xmath25 is the number of inversions of the permutation @xmath26 , and @xmath27_q = 1+q+\cdots + q^{k-1}$ ] . the right hand side of evaluated at @xmath28 is trivially the right hand side of . the fact that the right hand side of evaluated at @xmath29 yields the right hand side of follows from ( * ? ? ? * theorem 5.3 ) together with the fact that the major index and number of inversions are equidistributed over @xmath24 @xcite . finally , loehr @xcite showed the case @xmath30 combinatorially . showing directly that the hilbert series has these evaluations is highly nontrivial and is due to haiman @xcite . before the proof of haglund and loehr s hilbert series conjecture in @xcite , haglund @xcite gave an expression for the hilbert series as a weighted sum over certain upper triangular matrices called _ tesler matrices _ ? * http://oeis.org/a008608[a008608 ] ) . in @xcite , mszros , morales , and rhoades noticed that these matrices can be easily reinterpreted as integer flows on the complete graph @xmath31 with netflow vector @xmath32 . with this interpretation , haglund s result states that the hilbert series equals a weighted sum over the lattice points of the polytope of flows on @xmath31 with netflow vector @xmath32 . denoting this sum by @xmath33 ( see section [ sec : preliminaries ] for the precise definition of flows and their weight ) , haglund s result can be restated as follows . [ thm:1 ] @xmath34 combining theorems [ hlhilb ] and [ thm:1 ] , we obtain intriguing combinatorial identities between @xmath35 and @xmath0-analogues ( for @xmath15 , @xmath2 , @xmath36 , and @xmath37 ) of the number of parking functions of size @xmath17 : @xmath38 there are many natural bijections between spanning trees of @xmath31 and parking functions of size @xmath17 . correspondingly , there are various statistics @xmath39 on trees that can be used to rewrite the right hand side of as a sum over spanning trees of @xmath31 : @xmath40 where the sum is over all spanning trees @xmath41 of @xmath31 : see for instance @xcite for one example . equations , , and can be rewritten as @xmath42_q ! , \label{case : q01}\\ q^{\binom{n}{2 } } \operatorname{ehr}_{q , q^{-1}}(\f_{k_{n+1}}(-n,1,\ldots,1))&= [ n+1]_q^{n-1 } , \label{case : qqinv1 } \end{aligned}\ ] ] where on the right hand side of , @xmath41 ranges over all spanning trees of @xmath31 , and @xmath43 is the number of inversions of @xmath41 ( see section [ sec : q1 ] for the definition of @xmath44 statistic and the correspondence to the @xmath20 statistic on parking functions ) . it is then natural to verify these identities directly . doing so in the general case @xmath0 would give an alternative proof of the now settled haglund loehr conjecture . progress in this direction started with levande @xcite who verified the cases @xmath29 using a sign - reversing involution . armstrong et al . @xcite verified the case @xmath28 . we verify directly the @xmath30 case in this paper . more generally , one could extend the identity to flows with other netflow vectors ( in @xcite , these are called _ generalized tesler matrices _ ) or to other graphs . the former was done in @xcite for the @xmath29 case for binary netflows on @xmath31 extending the involution approach of levande . formulas for the @xmath28 case for positive integral netflows were given in @xcite and for integral flows in @xcite . we generalize the known formulas for @xmath33 for @xmath22 ( as in equations , , and ) to a family of graphs called _ threshold graphs _ @xcite with arbitrary positive integral netflows . there are @xmath45 such graphs with @xmath16 vertices including the complete graph . we now summarize our main results . first we state the case @xmath2 for netflow @xmath46 , which implies when @xmath47 is the complete graph . let @xmath47 be a threshold graph . then @xmath48 where @xmath49 is the tutte polynomial of @xmath47 , and @xmath41 ranges over all spanning trees @xmath41 of @xmath47 . this relationship between @xmath50 and the tutte polynomial of @xmath47 extends to general positive flows as follows . for a connected threshold graph @xmath47 and @xmath51 , let @xmath52 be the multigraph obtained from @xmath47 by replacing each edge @xmath53 with @xmath54 parallel edges . then @xmath55 where @xmath56 is the tutte polynomial of @xmath52 . we also state the case @xmath36 , which implies when @xmath47 is the complete graph and @xmath57 . let @xmath47 be a threshold graph with degree sequence @xmath58 and @xmath51 . then @xmath59_q,\ ] ] where @xmath60 is the number of vertices @xmath61 adjacent to @xmath62 . lastly , we state the case @xmath37 , which implies when @xmath47 is the complete graph and @xmath57 . let @xmath47 be a threshold graph with degree sequence @xmath58 and @xmath51 . then @xmath63 where @xmath64 and @xmath65_q&\text{if } d_i > i,\\ [ a_i]_{q^{i+1}}&\text{if } d_i = i,\\ [ a_i]_{q^{d_i+1}}[d_i]_q&\text{if } d_i < i. \end{cases}\ ] ] as a corollary , we prove a conjecture of armstrong et al . @xcite about the @xmath5-ehrhart function of the flow polytope of a complete graph with an arbitrary netflow vector . the case @xmath66 gives . for positive integers @xmath67 we have that @xmath68_q.\ ] ] our proofs are self - contained and inductive on the netflow of the flow polytope without using machinery from symmetric functions . in the case @xmath29 we do not use involutions like levande in @xcite and wilson in @xcite . the outline of this paper is as follows . in section [ sec : preliminaries ] we give the definitions of flow polytopes , @xmath0-ehrhart functions , and threshold graphs . in section [ sec : q1 ] we calculate @xmath69 for flow polytopes of threshold graphs , while in section [ sec : q0 ] we do the same for the evaluation @xmath29 . in section [ sec : qq^-1 ] we calculate the evaluation @xmath30 thereby also proving conjecture 7.1 of armstrong et al . in @xcite . we conclude in section [ sec : conj ] with positivity conjectures regarding the general @xmath0 case of flow polytopes of threshold graphs . in this section , we give some background and preliminary results about flow polytopes and threshold graphs . we first discuss flow polytopes and define the @xmath0-ehrhart functions . let @xmath70 be an acyclic directed graph on @xmath71 , and let @xmath72 . let @xmath73 be the @xmath74 matrix with columns @xmath75 for each directed edge @xmath53 . then the _ flow polytope _ @xmath76 is defined to be @xmath77 in other words , the flow polytope is the set of all nonnegative flows that can be placed on the edges of @xmath47 such that the net flow at each vertex is given by @xmath78 . by convention , we orient the edges from @xmath62 to @xmath79 if @xmath80 . since the sum of the entries of @xmath78 must be 0 for the flow polytope to be nonempty , we will abuse notation and write , for @xmath81 , @xmath82 . we also abbreviate @xmath83 . [ rem : flow2tesler ] a _ tesler matrix _ is an @xmath84 upper triangular matrix @xmath85 with nonnegative integer entries satisfying for @xmath86 , @xmath87 these matrices first appeared in haglund s study of @xmath9 @xcite . by an observation in @xcite , these matrices are in correspondence with integral flows on @xmath31 with netflow @xmath88 . with the conventions on @xmath89 in this paper , the correspondence is as follows : an integral flow @xmath90 in @xmath91 corresponds to the tesler matrix @xmath92 where @xmath93 for example , for @xmath94 the correspondence is the following : this correspondence can be extended to integral flows on subgraphs @xmath47 of @xmath31 by setting the entries corresponding to missing edges of @xmath47 to zero . for any nonnegative integer @xmath95 , define the _ @xmath96 for a lattice point @xmath97 with nonnegative entries , define @xmath98 where @xmath99 denotes the number of nonzero entries of @xmath100 . finally , for an integer polytope @xmath101 , define the _ @xmath0-weighted ehrhart function _ @xmath102 note that if @xmath103 , then any @xmath104 will have at least @xmath17 nonzero entries , so @xmath105 and hence @xmath106 will be _ polynomials _ in @xmath14 and @xmath15 . moreover , this polynomial by construction is _ symmetric _ in @xmath14 and @xmath15 . there is no guarantee , however , that @xmath106 will have nonnegative coefficients , and indeed it will not for general graphs @xmath47 as illustrated in the next example . [ ex : negqtgraph ] if @xmath107 , then there are @xmath108 integer flows on @xmath47 and one can check that @xmath109 we now define threshold graphs , a class of graphs of importance in computer science and optimization . for more information , see @xcite and ( * ? ? ? a _ threshold graph _ @xmath47 is a graph that can be constructed recursively starting from one vertex and no edges by repeatedly carrying out one of the following two steps : * add a dominating vertex : a vertex that is connected to every other existing vertex ; * add an isolated vertex : a vertex that is not connected to any other existing vertex . we say that a threshold graph @xmath47 is labeled by _ reverse degree sequence _ if its vertices are labeled by @xmath110 in such a way that @xmath111 for each pair of vertices @xmath112 , where @xmath113 is the degree of vertex @xmath62 . this family of graphs includes the complete graph and the star graph but excludes paths or cycles of @xmath114 or more vertices . there are @xmath115 threshold graphs with @xmath17 unlabeled vertices . the number @xmath116 of threshold graphs with vertex set @xmath117 $ ] has exponential generating function @xmath118 , and @xmath119 ( e.g. see ( * ? ? ? * http://oeis.org/a005840[a005840 ] ) ) . a threshold graph is uniquely determined up to isomorphism by its degree sequence @xmath120 . by convention , we will assume that all our threshold graphs are labeled by reverse degree sequence and that the edges are directed from @xmath62 to @xmath79 if @xmath80 . alternatively , a graph @xmath47 is a threshold graph if there exist real weights @xmath121 for each vertex @xmath122 and a threshold value @xmath15 such that @xmath62 and @xmath79 are adjacent if and only if @xmath123 . if the vertices are labeled such that @xmath124 , then @xmath47 is labeled by reverse degree sequence . note that if @xmath62 and @xmath79 are adjacent in @xmath47 , then so are @xmath125 and @xmath126 for any @xmath127 and @xmath128 ( provided @xmath129 ) . [ rem : tglabeling ] a threshold graph with @xmath16 vertices can be encoded by a binary sequence @xmath130 where @xmath131 or @xmath3 depending on whether vertex @xmath62 is a dominating or an isolated vertex with respect to vertices @xmath132 . in this labeling , if @xmath62 and @xmath79 are adjacent with @xmath133 , then @xmath134 since all vertices at least @xmath62 are adjacent to @xmath62 and all vertices smaller than @xmath62 are either adjacent to both @xmath79 and @xmath62 or to neither . hence when we relabel the vertices by reverse degree sequence the orientation of the edges is preserved . thus if @xmath135 is a threshold graph with the labeling induced from @xmath136 , and @xmath137 is the graph relabeled by reverse degree sequence , then @xmath138 where @xmath139 is obtained by permuting @xmath140 according to the relabeling of the vertices . in the case when the graph is connected ( @xmath141 ) and @xmath142 , then @xmath143 and the equation above becomes @xmath144 using the correspondence between integral flows on graphs and tesler matrices in remark [ rem : flow2tesler ] , the @xmath84 matrices corresponding to the flows on threshold graph @xmath145 have zero entries above the diagonal in column @xmath146 if @xmath147 . the threshold graph @xmath148 corresponds to the graph @xmath137 with reverse degree sequence @xmath149 . the map between integral flows on @xmath148 and @xmath137 and tesler matrices is the following : in this section , we give a combinatorial formula for the weighted ehrhart function of the flow polytope @xmath150 when @xmath47 is a threshold graph and @xmath2 . we note that one such proof when @xmath47 is the complete graph was given by wilson @xcite . in particular , it will follow that when @xmath151 , the weighted ehrhart function evaluates to the number of spanning trees of @xmath47 , or equivalently , to the number of @xmath47-parking functions . to begin , we will need some background about spanning trees , inversions , and parking functions , particularly in relation to threshold graphs . one important statistic on spanning trees is the number of inversions . the related notion of @xmath152-inversions is due to @xcite . we define both these notions below . let @xmath47 be a graph on @xmath153 , and let @xmath41 be a spanning tree of @xmath47 rooted at @xmath154 . we say that @xmath155 is a _ descendant _ of @xmath156 if @xmath156 lies on the unique path from @xmath154 to @xmath155 in @xmath41 . we say @xmath156 is the _ parent _ of a vertex @xmath155 if @xmath155 is a descendant of @xmath156 in @xmath41 , and @xmath156 and @xmath155 are adjacent in @xmath47 . an _ inversion _ of @xmath47 is a pair of vertices @xmath53 with @xmath157 such that @xmath79 is a descendant of @xmath62 . a _ @xmath152-inversion _ of @xmath47 is an inversion @xmath53 such that @xmath79 is adjacent to the parent of @xmath62 in @xmath47 . we denote the number of inversions of @xmath41 by @xmath43 and the number of @xmath152-inversions by @xmath158 . we will assume our trees are rooted at @xmath159 unless otherwise indicated . we also briefly recall the definition of the tutte polynomial of a graph . let @xmath160 be a multigraph . the _ tutte polynomial _ of @xmath47 is defined by @xmath161 where @xmath162 denotes the number of connected components in the graph @xmath163 . define the _ inversion enumerator _ of @xmath47 to be @xmath164 where @xmath41 ranges over all spanning trees of @xmath47 . gessel shows in @xcite that @xmath165 has the following properties . [ thm : gessel ] @xcite let @xmath47 be a graph on @xmath153 . a. the polynomial @xmath165 does not depend on the labeling of @xmath47 . in fact , @xmath166 . b. for any vertex @xmath167 , let @xmath168 be the number of descendants of @xmath62 in @xmath41 ( including @xmath62 itself ) that are adjacent in @xmath47 to the parent of @xmath62 . then @xmath169_q,\ ] ] where the sum ranges over all spanning trees @xmath41 for which @xmath170 . here we use the standard notation @xmath27_q = 1+q+q^2 + \dots + q^{k-1 } = \frac{q^k-1}{q-1}$ ] . in the case when @xmath47 is a threshold graph , these results specialize as follows . call a spanning tree of @xmath47 _ increasing _ if it has no inversions . [ prop : threshinv ] let @xmath47 be a threshold graph ( labeled by reverse degree sequence ) . then @xmath171_q,\ ] ] where @xmath172 is the number of descendants of @xmath62 in @xmath41 ( including @xmath62 itself ) . for a spanning tree of a threshold graph , any inversion @xmath53 is a @xmath152-inversion : @xmath61 implies that any vertex adjacent to @xmath62 is also adjacent to @xmath79 in @xmath47 , particularly the parent of @xmath62 ( or see ( * ? ? ? * proposition 10 ) ) . for the second equality , if @xmath79 is a descendant of @xmath62 , then the parent of @xmath79 is a descendant of the parent of @xmath62 . thus in an increasing tree , the parent of @xmath62 is at most the parent of @xmath79 , so since @xmath79 is adjacent to the latter , it must also be adjacent to the former in @xmath47 . the result then follows from theorem [ thm : gessel ] . the following notion of a @xmath47-parking function due to postnikov and shapiro @xcite generalizes the usual notion of parking function ( the latter corresponds to the complete graph ) . they are also called _ superstable configurations _ in the context of chip - firing . let @xmath160 be a graph on @xmath71 . @xmath47-parking function _ is a function @xmath173 \to { \mathbf z}_{\geq 0}$ ] such that , for every nonempty set @xmath174 $ ] , there exists @xmath175 such that @xmath176 is less than the number of vertices @xmath177 adjacent to @xmath62 . the _ degree _ of a parking function @xmath178 is defined to be @xmath179 . the _ codegree _ of a parking function @xmath178 is @xmath180 , where @xmath181 . when @xmath47 is the complete graph , @xmath178 is a parking function if and only if , for @xmath182 , there are at least @xmath183 vertices @xmath62 such that @xmath184 . in the context of ordinary parking functions on the complete graph , the @xmath185 statistic is usually referred to as @xmath20 . in general , @xmath47-parking functions are in bijection with the spanning trees of @xmath47 . merino @xcite showed the following relationship ( in the context of chip - firing ) between parking functions and the tutte polynomial of @xmath47 . @xcite let @xmath47 be a graph . then @xmath186 where the sum ranges over all @xmath47-parking functions @xmath178 . in light of gessel s results on the inversion enumerator of @xmath47 , it follows that the @xmath152-inversion statistic on spanning trees of @xmath47 has the same distribution as the codegree statistic on @xmath47-parking functions . ( this was noted in the case of the complete graph by kreweras @xcite . ) the authors of @xcite give an explicit bijection ( called the _ dfs - burning algorithm _ ) between spanning trees @xmath41 and @xmath47-parking functions @xmath178 that sends @xmath158 to @xmath187 . if @xmath47 is a threshold graph , then this bijection sends @xmath43 to @xmath187 . we are now ready to state the main result of this section . [ thm : t=1 ] let @xmath47 be a threshold graph . then @xmath188 where @xmath41 ranges over all spanning trees of @xmath47 , and @xmath178 ranges over all @xmath47-parking functions . note that for any @xmath189 , @xmath190 unless @xmath100 has exactly @xmath17 nonzero entries . hence to compute @xmath191 , we need only sum @xmath192 over such @xmath100 . for any @xmath193 , the set of edges @xmath53 for which @xmath194 forms a connected subgraph of @xmath47 . hence if @xmath100 has exactly @xmath17 nonzero entries , then these edges must form a spanning tree @xmath41 of @xmath47 . we claim that such @xmath100 are in bijection with increasing spanning trees @xmath41 of @xmath47 . indeed , if @xmath41 were not increasing , then there is some vertex @xmath195 that is smaller than its parent but larger than all of its descendants . but then @xmath62 has no outgoing edges in @xmath41 , so there can not be a nonnegative flow supported on @xmath41 with net flow 1 at @xmath62 . given an increasing spanning tree @xmath41 of @xmath47 , there is a unique flow @xmath196 supported on the edges of @xmath41 : we must have that the flow on the edge connecting @xmath62 to its parent is @xmath172 . hence @xmath197_q = i_g(q)\ ] ] by proposition [ prop : threshinv ] . as a corollary , we can specialize to the case when @xmath47 is the complete graph @xmath31 . this gives the @xmath2 case of the haglund - loehr conjecture ( theorem [ hlhilb ] ) via theorem [ thm:1 ] . we have @xmath198 where @xmath41 ranges over all spanning trees of @xmath31 , and @xmath178 ranges over all parking functions of length @xmath17 . we now give a combinatorial formula for @xmath199 for arbitrary @xmath103 as a weighted sum over spanning trees over @xmath47 . this formula is analogous to a result by armstrong et al . * theorem 7.1 ) in the case of the complete graph . note that it is straightforward to give a combinatorial formula for @xmath199 as a weighted sum over increasing spanning trees . for a similar result for the complete graph , see wilson @xcite . [ prop : q1 ] let @xmath47 be a threshold graph . for any vertex @xmath195 , let @xmath200 , where @xmath79 ranges over descendants of @xmath62 ( including @xmath62 itself ) . then @xmath201_q.\ ] ] as in the proof of theorem [ thm : t=1 ] , the only nonzero terms in the sum for @xmath202 come from flows supported on increasing spanning trees of @xmath47 . for any such tree , there is a unique flow in @xmath150 supported on it : the flow on the edge connecting @xmath62 to its parent is @xmath203 . the result follows easily . the following theorem converts this formula from a sum over increasing spanning trees of @xmath47 to a sum over all spanning trees of @xmath47 . for any spanning tree @xmath41 , let @xmath204 denote the edge set of @xmath41 , @xmath205 denote the parent of vertex @xmath62 , and @xmath206 denote the set of inversions of @xmath41 . [ thm : q1a ] let @xmath47 be a threshold graph and @xmath103 . a. let @xmath52 be the multigraph obtained from @xmath47 by replacing each edge @xmath53 with @xmath54 parallel edges . if @xmath47 is connected , then @xmath207 . b. for any spanning tree @xmath41 of @xmath47 , let @xmath208_q \cdot \prod_{(i , j ) \in \operatorname{inv}(t ) } q^{a_{\max\{p_t(i),j\}}}.\ ] ] then @xmath209 , where @xmath41 ranges over all spanning trees of @xmath47 . note that if we set @xmath210 , then @xmath211 , so we recover theorem [ thm : t=1 ] . we will give two proofs of this result . the first is an adaptation of the proof of theorem [ thm : gessel ] above by gessel in @xcite . the second uses known properties of the tutte polynomial . for part ( a ) , let @xmath212 , where @xmath213 ranges over connected sub - multigraphs of @xmath52 . for any such @xmath213 and any fixed vertex @xmath154 , @xmath214 decomposes into connected components , yielding an unordered set partition @xmath215 of @xmath216 . let @xmath217 denote the total number of edges in @xmath52 from @xmath154 to a vertex in @xmath218 . since @xmath213 must have at least one edge from @xmath154 to a vertex in @xmath218 for each @xmath79 , and the induced subgraphs @xmath219 $ ] are all connected , we have @xmath220}(q ) , \tag{$*$}\ ] ] where the sum ranges over set partitions @xmath215 of @xmath216 . to prove that @xmath221 , it suffices to show that , for @xmath159 , the weighted ehrhart sum satisfies the appropriate recursion derived from @xmath222 , namely @xmath223_q \cdot \operatorname{ehr}_{q,1}(\f_{g[v_j]}({\mathbf}a[v_j ] ) ) , \tag{$**$}\ ] ] where if @xmath224 in order , then @xmath225 = ( a_{i_1 } , \dots , a_{i_s})$ ] . ( note that @xmath226 $ ] is still a threshold graph for all @xmath218 ) . if @xmath227 , then @xmath228 , where @xmath41 is any increasing spanning tree of @xmath47 with a subtree supported on @xmath218 . part ( a ) now follows easily from proposition [ prop : q1 ] . in particular , since @xmath222 is satisfied for all @xmath154 , so must @xmath229 also be satisfied for all @xmath154 . we now show part ( b ) by induction on @xmath17in fact , we will show that it holds for any choice of root @xmath154 , not just @xmath159 . ( changing the root will usually change the second factor in @xmath230 . ) to see this , let @xmath41 be a spanning tree of @xmath47 with subtrees @xmath231 on vertex sets @xmath215 . if @xmath232 is a child of the root @xmath154 , then @xmath233_q \prod_{\substack{i \in v_j\\i < v_j } } q^{a_{\max\{r , i\}}}.\ ] ] by induction , for a fixed @xmath234 , @xmath235}({\mathbf}a[v_j]))$ ] , which does not depend on @xmath234 . moreover , as @xmath234 ranges over vertices in @xmath218 adjacent to @xmath154 ( noting that any @xmath236 is also adjacent to @xmath154 since @xmath47 is a threshold graph ) , @xmath237_q \prod_{\substack{i \in v_j\\i < v_j } } q^{a_{\max\{r , i\ } } } = [ \textstyle\sum_{v_j } a_{\max \{r , v_j\}}]_q = [ a(r , v_j)]_q.\ ] ] hence summing over all spanning trees @xmath41 , @xmath238_q \prod_{\substack{i \in v_j\\i < v_j } } q^{a_{\max\{r , i\}}}\\ & = \sum_{v_1 , \dots , v_k } \prod_{j=1}^k [ a(r , v_j)]_q \operatorname{ehr}_{q,1}(\f_{g[v_j]}({\mathbf}a[v_j]))\\ & = \operatorname{ehr}_{q,1}(\f_g({\mathbf}a ) ) . \qedhere \end{aligned}\ ] ] for the second proof , we recall the following properties of the tutte polynomial ( see ( * ? ? ? the first is that the tutte polynomial satisfies the following _ deletion - contraction _ recurrence . let @xmath47 be a multigraph and @xmath239 an edge of @xmath47 . a. if @xmath239 is not a bridge or loop of @xmath47 , then @xmath240 , where @xmath241 and @xmath242 are obtained from @xmath47 by removing edge @xmath239 and contracting edge @xmath239 , respectively . b. if @xmath239 is a bridge of @xmath47 , then @xmath243 . c. if @xmath239 is a loop of @xmath47 , then @xmath244 . the second is that the tutte polynomial can be described in terms of internal and external activity as follows . fix a total order @xmath245 on the edges of @xmath47 . given a spanning tree @xmath41 , we call an edge @xmath246 _ internally active _ if @xmath239 is the smallest edge of @xmath47 joining the two connected components of @xmath247 . we call an edge @xmath248 _ externally active _ if @xmath239 is the smallest edge in the unique cycle of @xmath249 . then the _ internal _ and _ external _ activities @xmath250 and @xmath251 are the total number of internally and externally active edges of @xmath41 , respectively . let @xmath47 be a multigraph . then @xmath252 , where @xmath41 ranges over spanning trees of @xmath47 . ( this does not depend on the choice of total order @xmath245 . ) we are now ready to give a second proof of theorem [ thm : q1a ] . although one can use the method of part ( b ) below to prove part ( a ) as well via proposition [ prop : q1 ] , we present a proof using the deletion - contraction recurrence since a similar recurrence will appear in the proof of theorem [ thm : q-1 ] . for ( a ) , let @xmath253 be the largest neighbor of vertex @xmath17 in @xmath47 . we will use the deletion - contraction recurrence on each of the @xmath254 edges of @xmath52 from @xmath17 to @xmath253 . at most one of these edges can be contracted , and all subsequent edges become loops . let @xmath255 be the graph obtained by contracting any one of these edges and then removing all loops , and let @xmath256 be the graph obtained by deleting all of these edges . then we get @xmath257_q\cdot t_{\widetilde g'}(1,q ) + t_{\widetilde g''}(1,q ) & \text { if $ m>0$,}\\ [ a_n]_q\cdot t_{\widetilde g'}(1,q ) & \text { if $ m=0$. } \end{cases}\ ] ] ( when @xmath258 , the last edge from @xmath17 to @xmath253 is a bridge so it can not be deleted . ) note that for @xmath259 , the number of edges in @xmath255 from @xmath253 to @xmath79 is @xmath260 ; hence @xmath255 comes from the threshold graph @xmath137 ( @xmath47 with vertex @xmath17 removed ) by multiplying edges according to the flow vector @xmath261 if @xmath262 , and @xmath263 if @xmath258 . likewise @xmath256 comes from the threshold graph @xmath264 ( @xmath47 with edge @xmath265 removed ) with flow vector @xmath78 . in fact , @xmath199 satisfies the same recurrence , that is , @xmath266_q \cdot \operatorname{ehr}_{q,1}(\f_{g'}({\mathbf}a ' ) ) + \operatorname{ehr}_{q,1}(\f_{g''}({\mathbf}a ) ) & \text { if $ m>0$,}\\ [ a_n]_q \cdot \operatorname{ehr}_{q,1}(\f_{g'}({\mathbf}a ' ) ) & \text { if $ m=0$. } \end{cases}\ ] ] indeed , as in proposition [ prop : q1 ] , we need only consider flows supported on spanning trees @xmath41 of @xmath47 . if @xmath267 , then it must support a flow of size @xmath254 , which changes the net flow at @xmath253 on the rest of @xmath41 from @xmath268 to @xmath260this gives the first term in the sum . if @xmath262 and @xmath269 , then we get the second term in the sum . it follows that @xmath270 by induction on the number of edges of @xmath47 ( the base case with one edge is trivial ) . for ( b ) , we again prove the claim for any root @xmath154 by induction on @xmath17 . we need to show that @xmath271 where @xmath41 and @xmath272 range over spanning trees of @xmath47 and @xmath52 , respectively . ( recall that the right side does not depend on the choice of total order . ) fix a set partition @xmath215 of @xmath216 and vertices @xmath232 adjacent to @xmath154 . then restrict both sides of @xmath273 to trees @xmath41 and @xmath272 such that the @xmath234 are the children of @xmath154 , and the @xmath218 are the vertex sets supporting the corresponding subtrees @xmath274 and @xmath275 . the left hand side then becomes , by induction , @xmath276_q \prod_{\substack{i \in v_j\\i < v_j}}q^{a_{\max\{r , i\ } } } = \prod_{j=1}^k \sum_{\widetilde t_j}q^{ea(\widetilde t_j ) } \cdot [ a_{\max\{r , v_j\}}]_q \prod_{\substack{i \in v_j\\i < v_j}}q^{a_{\max\{r , i\}}}.\ ] ] we claim this is also what the right hand side of @xmath273 becomes . choose any total order on the edges of @xmath52 that starts with all edges between @xmath154 and @xmath3 , then all edges between @xmath154 and @xmath277 , and so forth . ( the edges not containing @xmath154 can be in any order after that . ) no edges between distinct @xmath278 and @xmath218 are externally active , so the external activity of @xmath272 is the sum of the external activities of its subtrees @xmath275 plus the number of externally active edges containing @xmath154 . of the @xmath279 edges from @xmath154 to @xmath234 , any number from @xmath3 to @xmath280 are externally active depending on which parallel edge lies in @xmath272 . for @xmath281 , all edges from @xmath154 to @xmath62 are externally active if @xmath236 , otherwise none are . the result follows . one special case worth noting is when @xmath151 . in this case , theorem [ thm : q1a ] implies that @xmath282 is the total weight of all spanning trees @xmath41 of @xmath47 , where the weight of any edge @xmath53 is @xmath54 . thus @xmath282 can be expressed as a determinant using the matrix - tree theorem . in fact , one can show that this determinant factors into linear factors . for instance , when @xmath283 , @xmath284 , where @xmath285 multiplying the @xmath62th row by @xmath146 and adding all the lower rows to it for @xmath286 yields a lower triangular matrix , so one can easily recover the result of armstrong et al . @xcite that @xmath287 we will see a generalization of this product formula for general threshold graphs @xmath47 later in section [ sec : qq^-1 ] when we compute @xmath288 . in this section , we give a product formula for the weighted ehrhart function of the flow polytope @xmath150 when @xmath47 is a threshold graph and @xmath36 . in particular , when @xmath289 and @xmath36 , the weighted ehrhart function evaluates to the number of increasing spanning trees of @xmath47 , or equivalently the number of maximal @xmath47-parking functions . for a threshold graph @xmath47 , let @xmath290 be the outdegree of vertex @xmath62 , that is , the number of vertices @xmath79 adjacent to @xmath62 with @xmath61 . it should be noted that these outdegrees are closely related to the number of increasing spanning trees of @xmath47 . let @xmath47 be a threshold graph . the number of increasing spanning trees of @xmath47 is @xmath291 . each vertex @xmath195 has a choice of @xmath292 vertices to be its parent . we now state the main result of this section . observe that when @xmath36 , the weights specialize to @xmath293 [ thm : t=0 ] let @xmath47 be a threshold graph and @xmath103 . then @xmath294_q.\ ] ] note that when @xmath295 , this formula gives a @xmath14-analogue for the number of increasing spanning trees on @xmath47 . we will first need the following lemma . [ lem : t=0 ] for integers @xmath296 and @xmath297 , let @xmath298 . for any @xmath299 , define @xmath300 then @xmath301_q.\ ] ] we induct on @xmath183 . when @xmath302 , @xmath303 has a single point @xmath304 , and both sides equal @xmath305 . we therefore assume @xmath306 . for @xmath307 , write @xmath308 . letting @xmath309 , we have the decomposition @xmath310 ( where @xmath311 is the set containing the single point @xmath312 ) . since @xmath313 we have by the inductive hypothesis that , for fixed @xmath95 , @xmath314_q&\text{if $ b=0$,}\\ q^{(k-1)(c - b-1)}[k-1]_q \cdot q^{(k-1)b}(q^b - q^{b-1})&\text{if $ 0<b < c$,}\\ q^{(k-1)c}\cdot q^{c-1}&\text{if $ b = c$. } \end{cases } \end{aligned}\ ] ] summing over all @xmath95 gives the telescoping sum @xmath315_q + \sum_{b=1}^{c-1 } ( q^{(k-1)(c-1)}[k-1]_q)(q^b - q^{b-1 } ) & + q^{ck-1}\\ & = q^{(k-1)(c-1)}[k-1]_q \cdot q^{c-1 } + q^{ck-1}\\ & = q^{k(c-1)}([k-1]_q+q^{k-1})\\ & = q^{k(c-1)}[k]_q . \qedhere \end{aligned}\ ] ] we now proceed with the proof of the theorem . we may assume that @xmath47 is connected for both sides to be nonzero . we induct on @xmath17 . when @xmath316 , we must have @xmath317 , so both sides equal @xmath318 . now assume @xmath319 . the vertex @xmath17 is adjacent to vertices @xmath320 . for any lattice point @xmath104 , write @xmath321 so that @xmath322 ranges over all lattice points in @xmath323 . for fixed @xmath322 , the remaining flow @xmath324 on the graph @xmath137 obtained from @xmath47 by removing vertex @xmath17 lies in @xmath325 , where @xmath326 note that the outdegree of a vertex @xmath327 in @xmath137 is still @xmath292 . since @xmath328 , we have by induction and lemma [ lem : t=0 ] that @xmath329_q\right)\\ & = \sum_{b \in \delta({\bar d}_n , a_n ) \cap { \mathbf z}^{{\bar d}_n } } q^{b_1 + 2b_2+\cdots + ( { \bar d}_n-1)b_{{\bar d}_n-1}}wt_{q,0}(b)\cdot \prod_{i=1}^{n-1 } q^{{\bar d}_i(a_i-1)}[{\bar d}_i]_q\\ & = q^{{\bar d}_n(a_n-1)}[{\bar d}_n]_q \cdot \prod_{i=1}^{n-1 } q^{{\bar d}_i(a_i-1)}[{\bar d}_i]_q\\ & = \prod_{i=1}^{n } q^{{\bar d}_i(a_i-1)}[{\bar d}_i]_q.\qedhere \end{aligned}\ ] ] specializing to the case when @xmath283 gives the following corollary . . then @xmath330_q!,\ ] ] where @xmath117_q ! = [ n]_q[n-1]_q \cdots [ 1]_q$ ] . in particular , @xmath331_q!$ ] . the case @xmath332_q!$ ] was known by combining theorem [ thm:1 ] with . there is an elegant proof of this result by levande @xcite who defined a function @xmath333 from integer flows on @xmath31 with netflow @xmath46 to permutations in @xmath334 and used a sign - reversing involution to show that @xmath335 . wilson @xcite extended this involution to the case @xmath336 where @xmath337 . in contrast with these proofs , our proof is inductive and does not use involutions . in this section , we give a product formula for the weighted ehrhart function of the flow polytope @xmath150 when @xmath47 is a threshold graph and @xmath37 . when specialized to the case @xmath283 , this proves a conjecture of armstrong et al . @xcite . from theorem [ thm : t=1 ] , we know that @xmath338 should specialize to the number of spanning trees of @xmath47 when @xmath289 and @xmath339 . in fact , for threshold graphs @xmath47 , there is a simple product formula for the number of spanning trees . let @xmath340 . in other words , @xmath341 is the conjugate partition to @xmath342 . [ prop : spt - prod ] let @xmath47 be a threshold graph on @xmath153 . the number of spanning trees of @xmath47 is @xmath343 this is a direct application of the matrix - tree theorem ; see also @xcite for a combinatorial proof . note that when @xmath47 is the complete graph , we recover cayley s formula @xmath7 for the number of spanning trees of @xmath31 . we now state the main result of this section . observe that when @xmath37 , the weights specialize to @xmath344where @xmath345 also recall that @xmath346 is the outdegree of vertex @xmath62 . [ thm : q-1 ] let @xmath47 be a threshold graph and @xmath51 . then @xmath347 where @xmath348 and @xmath349_q&\text{if } d_i > i,\\ [ a_i]_{q^{i+1}}&\text{if } d_i = i,\\ [ a_i]_{q^{d_i+1}}[d_i]_q&\text{if } d_i < i. \end{cases}\ ] ] before we get to the proof , note what happens when we specialize @xmath210 . in this case , @xmath350_q & \text{if $ d_i > i$,}\\1 & \text{if $ d_i = i$,}\\ [ d_i]_{q}&\text{if $ d_i < i$,}\end{cases}\ ] ] so theorem [ thm : q-1 ] gives a @xmath14-analogue of proposition [ prop : spt - prod ] in this case . we may assume @xmath47 is connected and induct on @xmath17 and @xmath351 . when @xmath316 , we have @xmath352 , and @xmath353_{q^2}$ ] , so assume @xmath319 . if @xmath354 , let @xmath137 be the threshold graph obtained by removing vertex @xmath17 . then any flow @xmath104 can be obtained from a flow in @xmath355 by adding the vertex @xmath17 and a single edge with flow @xmath254 from @xmath17 to @xmath3 . hence @xmath356_{q^2 } \prod_{i=1}^{n-1 } b_i'(q),\ ] ] where @xmath357 and @xmath358 are the corresponding values of @xmath359 and @xmath360 for @xmath137 . but @xmath361 for @xmath327 , @xmath362_{q^2}[1]_q = [ a_n]_{q^2}$ ] , and @xmath363 , so the right side is @xmath364 , as desired . now suppose @xmath365 . then vertex @xmath17 is adjacent to @xmath366 . let @xmath137 be the threshold graph obtained by removing vertex @xmath17 , and let @xmath264 be the threshold graph obtained from @xmath47 by removing only the edge from @xmath17 to @xmath253 . choose any @xmath104 , and let @xmath367 . * if @xmath368 , then @xmath369 . * if @xmath370 , then @xmath100 can be obtained from a flow in @xmath371 by adding vertex @xmath17 and flow @xmath254 from @xmath17 to @xmath253 . * if @xmath372 , then @xmath100 can be obtained from a flow in @xmath373 by adding flow @xmath183 from @xmath17 to @xmath253 . it follows that @xmath374 by induction , we may expand each of the terms on the right side . first note that for any of the terms involving @xmath264 , the corresponding value of @xmath360 is , for any @xmath375 , @xmath376 while for the @xmath137 term , the corresponding value of @xmath360 is @xmath377 . next observe that for @xmath378 , the value of @xmath379 is the same in all terms . indeed , this is clear by the definition of @xmath379 if @xmath380 , so assume @xmath381 . then if @xmath382 , vertex @xmath62 is adjacent to @xmath17 , so @xmath383_q$ ] , which is the same in all terms . if instead @xmath384 , then @xmath62 is not adjacent to @xmath17 , so @xmath385_q$ ] does not involve either @xmath254 or @xmath386 , so it is also unchanged . it follows that we need only compare @xmath387 and @xmath388 for each of the terms . therefore to prove the theorem , it suffices to show that , if @xmath389 ( so @xmath390 ) , @xmath391_{q^{d_n+1}}[d_n]_q [ d_na_m + \textstyle\sum_{j = m+1}^n a_j]_q = \\ & q^{a_n } [ a_n]_{q^{d_n}}[d_n-1]_q[d_na_m + \textstyle\sum_{j = m+1}^{n-1 } a_j]_q+q^{a_n-1}wt_{q , q^{-1}}(a_n)[d_n(a_m+a_n ) + \textstyle\sum_{j = m+1}^{n-1 } a_j]_q\\ & { } -q^{a_n}(1-q)(1-q^{-1})\sum_{k=1}^{a_n-1}wt_{q , q^{-1}}(k)[a_n - k]_{q^{d_n}}[d_n-1]_q[d_n(a_m+k)+\textstyle\sum_{j = m+1}^{n-1}a_j]_q , \end{aligned}\ ] ] while if @xmath392 , @xmath393_{q^{n+1}}&[na_{n-1}+ a_n]_q = \\ & q^{a_n } [ a_n]_{q^{n}}[n-1]_q[a_{n-1}]_{q^n}+q^{a_n-1}wt_{q , q^{-1}}(a_n)[a_{n-1}+a_n]_{q^n}\\ & { } -q^{a_n}(1-q)(1-q^{-1})\sum_{k=1}^{a_n-1}wt_{q , q^{-1}}(k)[a_n - k]_{q^{n}}[n-1]_q[a_{n-1}+k]_{q^n}. \end{aligned}\ ] ] both of these follow from lemma [ lemma q ] below : the first follows by letting @xmath394 , @xmath395 , and @xmath396 , while the second follows by letting @xmath394 , @xmath397 , and @xmath398 and dividing both sides by @xmath117_q$ ] ( using the fact that @xmath399_q = [ x]_{q^n}[n]_q$ ] ) . [ lemma q ] let @xmath400 , @xmath401 , and @xmath402 be positive integers . then @xmath403_{q^{d+1}}[d]_q[z+a]_q = { } & q^a[a]_{q^d}[d-1]_q[z]_q + q^{a-1}wt_{q , q^{-1}}(a)[z+da]_q \\ & { } - q^a(1-q)(1-q^{-1})\sum_{k=1}^{a-1}wt_{q , q^{-1}}(k)[a - k]_{q^d}[d-1]_q[z+dk]_{q}. \end{aligned}\ ] ] we compute @xmath404_{q^d}[d-1]_qx^k \\ & = 1+(1-q)(1-q^{d-1})\sum_{k \geq 1}[k]_{q^d}(qx)^k\\ & = 1+\frac{(1-q)(1-q^{d-1})qx}{(1-qx)(1-q^{d+1}x)}\\ & = \frac{(1-q^2x)(1-q^dx)}{(1-qx)(1-q^{d+1}x ) } \end{aligned}\ ] ] and @xmath405_q}{q(1-q)(1-q^{-1 } ) } + \sum_{k \geq 1}q^{k-1}wt_{q , q^{-1}}(k)[z+dk]_qx^k\\ & = \frac{1-q^z}{(1-q)^3 } + \sum_{k \geq 1 } [ k]_{q^2}[z+dk]_qx^k\\ & = \bar g(x ) -q^z \bar g(q^dx ) , \end{aligned}\ ] ] where @xmath406_{q^2}x^k\\ & = \frac{1}{(1-q)^3 } + \frac{x}{(1-q)(1-x)(1-q^2x)}\\ & = \frac{(1-qx)^2}{(1-q)^3(1-x)(1-q^2x)}. \end{aligned}\ ] ] for @xmath407 , the desired right hand side is the coefficient of @xmath408 in @xmath409 . since @xmath410_{q^{d+1}}[d]_q[z+a]_qx^a = h(x ) - q^z h(qx),\ ] ] where @xmath411_{q^{d+1}}[d]_qx^a}{1-q}\\ = \frac{(1-q^d)x}{(1-q)^2(1-x)(1-q^{d+1}x)},\ ] ] it suffices to check that the difference between the two sides , @xmath412 is independent of @xmath413 . indeed , we will show that @xmath414 this is straightforward : @xmath415 and @xmath416 so @xmath417 as well . if we specialize to the case @xmath418 , we arrive at the following corollary , conjectured by armstrong et al . in ( * ? ? ? * conjecture 7.1 ) . [ cor : q-1 ] let @xmath51 . then @xmath419_{q^{n+1}}\prod_{i=1}^{n-1 } [ ( i+1)a_i+\textstyle\sum_{j = i+1}^{n}a_j]_q,\ ] ] where @xmath420 . in this section we look at the weighted ehrhart series of the flow polytope @xmath422 when @xmath47 is a threshold graph with @xmath16 vertices . by haglund s result @xcite , the weighted ehrhart series @xmath35 is the bigraded hilbert series of the space of diagonal harmonics , so it must lie in @xmath423 $ ] . by example [ ex : negqtgraph ] , the polynomial @xmath424 for other graphs @xmath47 sometimes has negative coefficients . however , experimentation suggests some positivity properties of the polynomials @xmath424 for threshold graphs @xmath47 and netflow @xmath46 . in @xcite , it was conjectured that @xmath427 $ ] for integral netflows @xmath140 satisfying @xmath428 . the analogous @xmath12-positivity statement for threshold graphs does not hold even though theorem [ thm : q-1 ] gives product formulas when @xmath37 . along with theorem [ thm : t=1 ] , conjecture [ conj1 ] suggests that there may be some statistic @xmath433 on spanning trees @xmath41 of @xmath47 or on @xmath47-parking functions such that @xmath434 equals @xmath424 . we have so far been unable to find such a statistic ( see section [ subsec : pmaj ] ) . a spanning tree @xmath41 of a connected threshold graph @xmath47 with @xmath16 vertices is also a spanning tree of the complete graph . a stronger positivity result would be that each monomial @xmath435 in @xmath424 appeared also in @xmath436 . calculations up to @xmath426 suggest that this is also the case . the above computations suggest to check positivity of differences of @xmath0-ehrhart functions between a threshold graph and a subgraph that is also a threshold graph . let @xmath19 be the poset of connected threshold graphs with vertices @xmath438 , where @xmath439 if @xmath213 is a subgraph of @xmath47 . this poset is isomorphic to the poset of shifted young diagrams ( or partitions with distinct parts ) contained in @xmath440 , ordered by inclusion . for example the hasse diagrams of the posets @xmath441 and @xmath442 are the following : lastly , one might try to use the poset structure of @xmath19 to refine further each @xmath0-ehrhart series in the following way . for a threshold graph @xmath47 in @xmath19 , let @xmath445 where @xmath446 is the mbius function of @xmath19 . by mbius inversion we then have that @xmath447 one might hope that @xmath448 is @xmath12-positive in general , but this is not the case . one could explore generalizations to other positivity results for tesler matrices . the _ alternant _ @xmath453 is a certain subspace of @xmath9 of dimension @xmath454 @xcite , the @xmath17th catalan number . the bigraded hilbert series of this subspace is the @xmath12-catalan number @xmath455 . gorsky and negut @xcite expressed @xmath455 as a different weighted sum over the integral flows of @xmath456 ( see remark [ rem : flow2tesler ] for translating from integral flows to tesler matrices ) . the gorsky negut weight is not as natural for threshold graphs @xmath47 as it is for the complete graph since @xmath47 might not contain the edge @xmath462 . instead , one could consider a weight @xmath463 where @xmath464 is the largest neighbor of @xmath62 less than @xmath62 . still , summing over this weight does not necessarily yield a polynomial in @xmath465 $ ] . garsia and haglund @xcite gave a weight over certain integral flows in @xmath91 that yields the _ frobenius series _ of the space @xmath9 , a certain symmetric function that in the schur basis has coefficients in @xmath423 $ ] ( for more details see ( * ? ? ? 6 ) ) . using the same weight on flows in @xmath89 for threshold graphs @xmath47 does not give symmetric functions with a schur expansion with coefficients in @xmath423 $ ] . one way to define @xmath471 is as follows . parking functions have a natural partial order : @xmath475 if @xmath476 for all @xmath62 . for the complete graph , the maximal parking functions are those with area 0 , namely the bijections @xmath477 \to \{0 , 1 , \dots , n-1\}$ ] . for a maximal parking function @xmath478 , @xmath479 while for any other parking function @xmath178 , @xmath480 . note that on the maximal parking functions , @xmath481_t!\ ] ] ( and on maximal parking functions , @xmath471 is easily seen to be equidistributed with major index on permutations ) . the @xmath20 and @xmath185 statistics coincide on parking functions , which suggests that , in accordance with conjecture [ conj1 ] , if there exists a statistic @xmath482 on @xmath47-parking functions such that @xmath483 then we should be able to construct @xmath482 to be analogous to @xmath471 . then we might expect to be able to define @xmath482 on @xmath47-parking functions such that @xmath484_t!\ ] ] by theorem [ thm : t=0 ] , while for any other parking function @xmath178 , @xmath485 . this could simplify the task of defining @xmath482 since it would only need to be defined on the maximal @xmath47-parking functions . initial computations suggest that it is possible to find such a statistic for small graphs , though we have not yet found a suitable statistic that works for all threshold graphs .

the existence of an exotic long - lived negatively charged massive particle , i.e. , , during big bang nucleosynthesis can affect primordial light element abundances . especially , the final abundance ofli , mainly originating from the electron capture ofbe , has been suggested to reduce by thebe destruction via the radiative capture ofbe followed by the radiative proton capture of the bound state ofbe and (be ) . we suggest a new route ofbe formation , that is thebe charge exchange at the reaction of ion and . the formation rate depends on the number fraction of ion , the charge exchange cross section of and the probability that produced excited statesbe are converted to the ground state . we estimate respective quantities affecting thebe formation rate , and find that this reaction pathway can be more important than ordinary radiative recombination ofbe and . the effect of the charge exchange reaction is then shown in a latest nuclear reaction network calculation . quantum physical model calculations for related reactions are needed to precisely estimate the efficiency of this pathway in future .

1305.6155

new physics operating during the big bang nucleosynthesis ( bbn ) can be probed by observed light element abundances . possible indications of new physics come from discrepancies between primordial abundances of @xmath4li and @xmath1li predicted in standard bbn ( sbbn ) model and those inferred from observations of metal - poor stars ( mpss ) . these mpss exhibit a plateau - like abundance ratio , @xmath1li / h @xmath5 at low metallicities of [ fe / h]@xmath6 @xcite , and much lower at extremely low metallicities of [ fe / h]@xmath7 @xcite.@xmath8 , where @xmath9 is the number density of @xmath10 and the subscript @xmath11 indicates the solar value , for elements a and b. ] the plateau abundance is a factor of 24 lower than the sbbn prediction for the baryon - to - photon ratio determined from the observation of the cosmic microwave background radiation with _ wilkinson microwave anisotropy probe _ ( wmap ) ( e.g. , @xmath1li / h=@xmath12 @xcite ) . this discrepancy indicates a need of some mechanism to decrease the @xmath1li abundance . although astrophysical processes such as the combination of atomic and turbulent diffusion in stellar atmospheres @xcite may be a cause of the observed abundances , this is not yet established @xcite . in recent spectroscopic observations of mpss , the lithium isotopic ratio of @xmath4li/@xmath1li is also measured . a possible plateau abundance of @xmath4li / h@xmath13 has been suggested @xcite , which is about 1000 times higher than the sbbn prediction . since an effect of convective motions in stellar atmospheres could cause asymmetries in atomic line profiles and consequently leads to an erroneous estimation of @xmath4li abundance @xcite , the effect should be estimated . even including this effect , high @xmath4li abundances have been evidently detected in at most several mpss @xcite . such a high abundance level at the low - metallicity requires processes other than standard galactic cosmic - ray ( cr ) nucleosynthesis models @xcite . to investigate the @xmath14li problems , in this work , we focus on a model to resolve the lithium problems assuming the presence of negatively charged massive particles @xmath0 @xcite with a mass much larger than the nucleon mass , i.e. , @xmath15 gev , during the bbn epoch @xcite . one of candidates for the @xmath0 particle is stau , the supersymmetric partner of tau @xcite . the @xmath0 particles become electromagnetically bound to positively charged nuclides with binding energies of @xmath16 mev . the bound state of a nuclide @xmath17 and an @xmath0 particle is denoted by @xmath18 . because of these low binding energies the bound states can not form until late in the bbn epoch , when nuclear reactions are no longer efficient . hence , the effect of the @xmath0 particles is rather small @xcite . depending upon their abundance and lifetime , however , the @xmath0 particles can affect lithium abundances through the following processes : 1 ) @xmath4li production via the recombination of @xmath19he with @xmath0 followed by the catalyzed @xmath20 transfer reaction @xmath19he@xmath2(@xmath21,@xmath0)@xmath4li @xcite , 2 ) @xmath1be destruction via the recombination of @xmath1be with @xmath0 followed by the radiative proton capture reaction @xmath1be@xmath2(@xmath22,@xmath23)@xmath24b@xmath2 . in the latter process the destruction of @xmath1be occurs through both the first atomic excited state of @xmath24b@xmath2 @xcite with excitation energy of 0.819 mev in the limit of infinite @xmath0 mass @xcite or the atomic ground state ( gs ) of @xmath24b@xmath25(@xmath26,0.770 mev)@xmath2 composed of the @xmath26 nuclear excited state of @xmath24b and an @xmath0 @xcite li is produced mostly as @xmath1be during the bbn epoch , which transforms into @xmath1li by electron capture in late universe . ] . in all previous investigations , all processes are assumed to start from radiative recombination of fully ionized nuclides @xmath17 and @xmath0 as suggested in ref . @xcite . however , we found a possibility that nonradiative charge exchange reactions between hydrogen - like ions and @xmath0 can contribute to change nuclear abundances in the bbn model including an @xmath0 particle . we focus on the @xmath1be nuclide since the effect on its abundance is expected to be important . we introduce a simple model for estimating the significance of the new process in sec . [ sec2 ] , and show the result of @xmath1be@xmath2 formation rate and time evolution of nuclear abundances in example cases in sec . [ sec3 ] . finally we summarize this work in sec . we adopt natural units , @xmath27 , where @xmath28 is the reduced planck constant , @xmath29 is the speed of light , and @xmath30 is the boltzmann constant . we suggest a process of @xmath1be@xmath2 production via the ion formation . this process can be important since @xmath1be@xmath2 is destroyed through the reaction @xmath1be@xmath2(@xmath22 , @xmath23)@xmath24b@xmath2 and the final abundance of @xmath1li changes . the destruction of @xmath1be@xmath2 occurs at the temperature @xmath31 or @xmath32 kev @xcite , so that we focus on this temperature region . in this epoch , there are two nonradiative reactions producing @xmath1be@xmath3 . they are the substitution reaction + @xmath33be@xmath34 , and the three body collisional recombination @xmath1be@xmath35be@xmath3+@xmath36 , where the ground and excited states of fully ionized @xmath1be@xmath37 and @xmath0 are denoted by @xmath1be@xmath2 and @xmath1be@xmath3 as usual , while the partially ionized hydrogen - like ion is denoted by @xmath1be@xmath38 throughout this work . the latter reaction is not considered in this work since the number abundance of @xmath36 is not so large in this late epoch of @xmath36 pair annihilation and consequently the reaction rate is not large in this temperature region . the effective rate for @xmath1be to recombine with @xmath0 particle through the ion formation is given by the product of the following three quantities : 1 ) the number ratio of to @xmath1be@xmath37 ( = @xmath1be ) which is the dominant chemical species containing the @xmath1be nuclide , 2 ) the formation rate of excited states @xmath1be@xmath3 via the reaction @xmath1be@xmath38+@xmath39be@xmath34 , and 3 ) the probability of converting excited states @xmath1be@xmath3 to the gs @xmath1be@xmath2 which is estimated by the ratio of the transition rate producing the gs @xmath1be@xmath2 and the total reaction rate of the excited state @xmath1be@xmath3 . the effective recombination rate is then described as @xmath40 , \label{eq18}\end{aligned}\ ] ] where @xmath9 denotes the number density of species @xmath41be@xmath38 and be@xmath37 , @xmath42 represents the rate for to form excited states @xmath1be@xmath3 . @xmath43 and @xmath44 are rates for the transition to the gs and the destruction of excited states @xmath1be@xmath3 , respectively . the parameters in the square bracket have uncertainties related to the cross sections for formation and destruction of @xmath1be@xmath3 , and the fraction of the produced @xmath1be@xmath3 which decays to the gs @xmath1be@xmath2 through multiple electric dipole transitions . a precise estimation of these uncertainties is , however , beyond the scope of this study . the energy level for the main quantum number @xmath45 is @xmath46 where @xmath47 is the proton number of particle @xmath10 ( @xmath48 ) , @xmath20 is the fine structure constant , and @xmath49(@xmath10 , @xmath50 ) is the reduced mass for the @xmath10+@xmath50 two body system . the charge number of particle @xmath51 or @xmath0 was assumed to be @xmath52 . the binding energy for @xmath45 is related to the energy level as @xmath53 . the expectation value of radius @xmath54 between the @xmath10 and @xmath50 particles composing the hydrogen - like ion is @xmath55 , \label{eq8}\ ] ] where @xmath56 is the azimuthal quantum number . there is a relation between the average radius and the binding energy from eqs . ( [ eq7 ] ) and ( [ eq8 ] ) : @xmath57 the average radius is thus inversely proportional to the binding energy independent of the reduced mass of the system . the binding energy of the gs is given by eq . ( [ eq7 ] ) to be 217.7 ev . in the limit of @xmath58 gev , the reduced mass for the system of @xmath1be and @xmath0 is given by @xmath59 gev . the main quantum number of @xmath1be@xmath3 state whose binding energy is nearly equal to that of the gs is @xmath60113 . we note that the reaction @xmath1be@xmath38+@xmath39be@xmath34 is similar to the reaction h+@xmath61 , where @xmath62 is the protonium consisting of a proton and an antiproton @xcite since the following three characters are shared by the two reactions . firstly , components related to the reaction are two heavy particles and a light electron . the heavy particles are @xmath1be and @xmath0 in the former reaction , and @xmath22 and @xmath63 in the latter . secondly , a charged heavy particle ( @xmath1be and @xmath22 , respectively ) is transfered . thirdly , a bound state of heavy two particles ( @xmath1be@xmath2 and @xmath62 , respectively ) is produced . in the case of the protonium formation , @xmath62 states with the main quantum numbers of @xmath6030 have binding energies nearly equal to that of the hydrogen gs ( 13.6 ev ) . the reaction of the muonic hydrogen formation , h@xmath64 @xcite , is also a similar reaction although the muon mass , 106 mev , is lighter than the nucleon mass by one order of magnitude . the average radius of the gs is given by eq . ( [ eq8 ] ) as @xmath65 the reduced mass @xmath49(@xmath1be , @xmath66 ) is approximately given by the electron mass @xmath67 . the average distance between @xmath36 s at the temperature @xmath68 kev is given by @xmath69 where @xmath70 is the number density of @xmath36 . we assumed that the number density is given by the formula for the non - degenerate non - relativistic fermion @xcite , i.e. , @xmath71 where @xmath72 is the statistical degrees of freedom both for electron and positron . since the average distance is larger than the average radius of the gs , the gs can be regarded as an isolated two - body system . we adopt the rate for the gs formation via the radiative recombination of @xmath1be and @xmath66 from ref . @xcite : @xmath75{\mathstrut \frac{t}{t_0 } } \left(1+\sqrt[]{\mathstrut \frac{t}{t_0}}\right)^{1-b } \left(1+\sqrt[]{\mathstrut \frac{t}{t_1}}\right)^{1+b}\right]^{-1 } , \label{eq4}\ ] ] where parameter values are @xmath76 @xmath77 s@xmath78 , @xmath79 , @xmath80 k , and @xmath81 k. although this is the total rate for transitions to all final states including not only the gs but also excited states , the formation of the gs dominates for the temperature range of @xmath82 k @xcite . the adoption of the rate is thus justified . for the related temperature , the rate is @xmath83 @xmath77 s@xmath78 . the rate for @xmath1be to recombine with @xmath66 is then given by @xmath84 s@xmath78 at @xmath85 . this rate is much larger than the hubble expansion rate , i.e. , @xmath86 s@xmath78 @xmath87 . the recombination and its inverse reaction , i.e. , the ionization , is thus very effective , and the number ratio of to @xmath1be@xmath37 is the equilibrium value . the equilibrium number ratio is described by saha equation : @xmath88 where @xmath89 and @xmath90 are the spin degrees of freedom and the mass , respectively , of species @xmath41be@xmath38 , be@xmath37 , and @xmath66 . the spin factors are @xmath91 $ ] ( for the gs ) , @xmath92 , and @xmath72 , respectively , with @xmath93 the nuclear spin of a given be isotope . using the spin factors , the approximate equality of @xmath94 , and eq . ( [ eq3 ] ) , the number ratio is derived as @xmath95n_e\nonumber\\ & = & 2\exp\left[\frac{i(^7{\rm be}^{3+})-m_e}{t}\right]\sim 2 \mathrm{e}^{-m_e / t } , \label{eq6}\end{aligned}\ ] ] where the ionization potential @xmath96 was defined for species @xmath10 , and @xmath97 ev [ eq . ( [ eq7 ] ) ] . this ratio is @xmath98 for @xmath85 . the equilibrium abundance of @xmath1be@xmath38 ion decays even in this bbn epoch although the fraction of decaying @xmath1be@xmath38 is negligible . the half life for electron capture of @xmath1be@xmath38 is @xmath105 day @xcite = @xmath106 s. it is much longer than the time scale of the universe corresponding to the @xmath1be@xmath2 formation epoch , @xmath107 s @xmath108 with @xmath109 the effective number of relativistic degrees of freedom in terms of energy density @xcite . the thermal reaction rate at a given temperature @xmath110 is given ( eq . ( 2.52 ) of ref . @xcite ) by @xmath111^{1/2 } \frac{1}{t^{3/2 } } \int_0^\infty e \sigma(e ) \exp\left(-\frac{e}{t}\right ) de , \label{eq10}\ ] ] where @xmath112 is the center of mass kinetic energy , and @xmath113 is the reaction cross section as a function of @xmath112 . we assume that the cross section is similar to that of the protonium formation . it is then roughly taken to be @xmath114^{-1/2 } h(i(^7{\rm be}^{3+})-e)$ ] with @xmath115 the cross section at @xmath116 and @xmath117 the step function ( fig . since the ionization , i.e. , @xmath1be@xmath118be@xmath119 , dominates for energies above the ionization threshold , the cross section for @xmath120 is negligibly small . supposing the scaling of the cross section @xmath121 , the cross section for the + @xmath0 reaction is assumed to be @xmath122 b. we note that the reaction of muonic hydrogen formation , @xmath123h(@xmath124 , @xmath66)@xmath125 , is a same kind of reaction as the reaction @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 . the muonic hydrogen formation is the first reaction occurring in experiments of muonic atoms @xcite . observables in the experiments such as energy flux densities of uncaptured muon and x - rays emitted from excited states of muonic atoms , however , do not provide a direct information on the muon capture rate . comparisons of theories and experiments on the cross section is , therefore , difficult for the moment @xcite . then we can only refer to theoretical cross sections . calculations of the cross sections based on several different methods have been performed , and they agrees well at least qualitatively ( fig . 2 of ref . @xcite ) . the cross sections @xmath113 scale as @xmath126 similarly to that for the protonium formation . curve which are caused by a quantum mechanical effect . ] under these assumptions , the reaction rate is given by @xmath127,\nonumber\\ \label{eq11}\end{aligned}\ ] ] where we assumed that @xmath128 since the cross section is nonzero only for @xmath129 . the number density of baryon is given by @xmath130 where @xmath131 and @xmath132 are the baryon density and the present critical density , respectively . @xmath133 is the baryon density parameter , @xmath134 is the redshift of the universe which has a relation with temperature as @xmath135 with @xmath136 k the present radiation temperature of the universe @xcite , and @xmath137/(100 km s@xmath78 mpc@xmath78)@xmath138 is the reduced hubble constant . the cosmological parameters are taken from values determined from the wmap @xcite ( model @xmath139cdm , wmap9 data only ) . the rate for to form excited states of @xmath1be@xmath3 through the charge exchange reaction is then given by @xmath140 , \label{eq13}\end{aligned}\ ] ] where @xmath141 is the number density of the @xmath0 particle , and @xmath142 is the number ratio of @xmath0 to baryon . excited states @xmath1be@xmath3 produced via the @xmath1be charge exchange reaction experience bound - bound transitions including not only the spontaneous emission but also the stimulated emission and the photo - absorption in the early universe filled with the cosmic background radiation ( cbr ) . the spectrum of cbr in bbn epoch is assumed to be the planck function , i.e. , @xmath143 where @xmath144 is the frequency of the photon with @xmath145 the photon energy . dominant process which occurs most frequently is determined as follows . the transition from the upper ( @xmath146 ) to the lower ( @xmath147 ) energy states is considered . energy levels of the upper and lower states are defined by @xmath148 and @xmath149 , respectively . the energy and the frequency of emitted photon is then @xmath150 and @xmath151 , respectively . the spontaneous emission rate for a transition from @xmath146 to @xmath147 is defined with the einstein @xmath17-coefficient , i.e. , @xmath152 . the photo - absorption rate and the stimulated emission rate are described with @xmath153-coefficients as @xmath154 , and @xmath155 , respectively . in addition , the einstein relations are given by @xmath156 the ratio between rates for the stimulated emission and the spontaneous emission then satisfies @xmath157 where eqs . ( [ eq26 ] ) and ( [ eq28 ] ) were used . clearly , transitions of @xmath158 proceeds predominantly through the stimulated emission , while those of @xmath159 proceeds predominantly through the spontaneous emission . the photo - absorption rate is equal to the stimulated emission rate except for the spin factors @xmath160 and @xmath161 [ eq . ( [ eq27 ] ) ] . the einstein @xmath17-coefficient is given ( eq . ( 10.28b ) of ref . @xcite ) by @xmath162 where @xmath163 is the electric dipole matrix element for the initial state @xmath146 and the final state @xmath147 , and the sum is over all magnetic substates of the upper and lower states . the difference in energy between levels with nearest main quantum numbers is @xmath164 $ ] @xmath165 ( for @xmath166 ) . here the typical difference in energy between @xmath146 and @xmath147 is taken as @xmath167 assuming that the main quantum numbers of @xmath146 ( @xmath168 ) , and @xmath147 ( @xmath169 ) are close to each other . here we introduce the absorption oscillator strength @xcite defined as @xmath170 where @xmath171 and @xmath172 are the mass and proton number , respectively , of nuclide @xmath17 bound to @xmath0 ( in the present case , @xmath173be ) . typical amplitude of this strength is known as @xmath174 ( see refs . we note that this definition is different from that for hydrogen - like electronic ions in terms of mass and charge number for the following reason . in the harmonic oscillation model for the latter case , an electron with classical radius @xmath175 is bound to an nucleus . in the case of @xmath1be@xmath3 , on the other hand , an @xmath1be nucleus with classical radius @xmath176 is bound to an @xmath0 . from eqs . ( [ eq14 ] ) and ( [ eq24 ] ) , we find @xmath177 the number of final states is about @xmath178 including @xmath179 levels for azimuthal quantum numbers of final states which are different from that of the initial state by 1 and @xmath52 , respectively . the total spontaneous emission rate of excited state @xmath1be@xmath3 is then roughly estimated as @xmath180\nonumber\\ & & \times \left(\frac{z_a}{4}\right)^2 \left(\frac{m_a}{6.53~{\rm gev}}\right)^{-1 } , \label{eq15}\end{aligned}\ ] ] where the weighted average quantity @xmath181/n_l$ ] was defined . the einstein coefficient @xmath182 is given @xcite by @xmath183 the rate for the deexcitation via the stimulated emission is estimated to be @xmath184 where @xmath185/n_l$ ] was defined . this stimulated emission rate as well as the spontaneous emission rate is larger for states with smaller main quantum numbers . the rate for the excitation via the photo - absorption , @xmath186 is roughly the same as that for the deexcitation . the excited states @xmath1be@xmath3 with @xmath187 produced via the charge exchange reaction have binding energies much less than the temperature at the @xmath1be@xmath2 destruction epoch of @xmath188 . the states , therefore , predominantly transit to other states via cbr - absorptions and the stimulated emission by cbr . their rates are larger for large @xmath189 values as indicated in eq . ( [ eq23 ] ) . the rough assumption of @xmath190 leads to a scaling of @xmath191 . this means that deexcitation rates of low energy states with smaller @xmath45 values are larger than those of high energy states . therefore , initially produced states of @xmath1be@xmath3 with @xmath187 can quickly start transitions to lower states , while transitions between higher states are significantly hindered . the photo - excitation and photo - deexcitation result in an effective population of lower states and an slow population of higher states . when the excited states with @xmath192 corresponding to @xmath193 are formed via the deexcitation , they would transit gradually to the gs through mainly the spontaneous emission ( sec . [ sec261 ] ) . the spontaneous emission rate @xmath194 [ eq . ( [ eq15 ] ) ] of the lower states is much larger than that for states with @xmath187 . time scale for the successive reactions producing the gs is given by @xmath195 since reaction time scales of lower energy states are shorter , and can be neglected . in this work , we do not treat more details on such transitions to the gs , and assume that the transition rate is given by @xmath196 transitions to the @xmath1be@xmath2 gs can realize through multiple stimulated emissions and photo - absorptions induced by cbr and spontaneous emissions . if the destruction rate of an excited state @xmath146 , @xmath197 , is larger than its transition rate @xmath198 , the probability for transitions through multiple ( @xmath199 times ) photo - emissions @xmath200 becomes very small . in this case , therefore , an efficient formation of the @xmath1be@xmath2 gs prefers a direct production of lower energy states at the @xmath1be@xmath3 formation reaction or their production at the first photo - emission after the formation of @xmath1be@xmath3 excited states . the first case is , however , not expected to operate effectively by an analogy of protonium formation dominantly producing states of large main quantum numbers @xmath201 which correspond to small binding energies @xcite . we note that the @xmath1be nucleus has been assumed to be a point charge particle in this work . binding energies of @xmath18 atomic gs can , however , be significantly smaller than those estimated in the assumption of point charge nuclei since bohr radii for the @xmath18 are of nuclear dimensions @xmath202(fm ) @xcite . resultingly , the radiative dipole transition rate of @xmath1be@xmath3 to the gs , which scales with the third power of emitted photon energy , is significantly smaller than that estimated for the point charge nucleus . however , spontaneous emission rates of low energy levels @xmath203 are much larger than those of high energy levels @xmath204 . the total time scale for the transition to the gs is , therefore , contributed mainly from transitions of the highly excited states of @xmath1be@xmath3 . the excited states have bohr radii much larger than the charge radius of @xmath1be , and their binding energies are almost the same as those estimated for a point charge @xmath1be . the effect of a finite size nuclear charge is , therefore , not relevant in this transition rate . the main reaction of @xmath1be@xmath3 destruction is the ionization at collisions with electron and positron , i.e. , @xmath1be@xmath3+@xmath205be+@xmath0+@xmath36 . there is no study about the cross section for this type of reaction . we then assume the cross section by an analogy of cross sections for ionizations of normal hydrogen which has been measured , and of muonic hydrogen - like particle consisting of a muon and a proton @xcite . the energy threshold for the ionization of @xmath1be@xmath3 is determined from the condition that an energy transferred from @xmath36 to @xmath1be at a collision be larger than ionization potential of @xmath1be@xmath3 as @xmath206 kev @xmath207 @xcite . the cross section for energies above the threshold nearly corresponds to the bohr radius squared , i.e. , @xmath208\}^2 $ ] , ( fig . @xcite ; curve 2 ) . we roughly assume that the destruction cross section of @xmath1be@xmath3 is @xmath209\}^2 $ ] independent of energy . using eq . ( [ eq10 ] ) we find @xmath210^{1/2 } \left(1+\frac{e_{\rm th}}{t}\right ) \exp\left(-\frac{e_{\rm th}}{t}\right ) \sigma_{\rm de}\nonumber\\ & \approx&1.66 n^4 \times 10^{-17}~{\rm cm}^3~{\rm s}^{-1 } \left(\frac{t_9}{0.4}\right)^{1/2 } \left(1+\frac{e_{\rm th}}{t}\right ) \nonumber\\ & & \times \exp\left(-\frac{e_{\rm th}}{t}\right ) \left(\frac{\sigma_{\rm de}}{1.35 ~n^4~{\rm mb}}\right ) . \label{eq16}\end{aligned}\ ] ] we note that the equation @xmath211 is satisfied for @xmath212 . the rate of the be@xmath3 destruction through the collisional ionization is then given by @xmath213 where eqs . ( [ eq3 ] ) and ( [ eq16 ] ) were used , and @xmath85 and @xmath214 ( corresponding to the gs ) was assumed for numerical values . the rate depends on the temperature exponentially , and is larger at higher temperature . the rate also significantly depends on the main quantum number through the variable @xmath215 and as the power of @xmath216 . rates for smaller @xmath45 values are smaller because of smaller values of @xmath217 . we can regard excited states of @xmath1be@xmath3 as effective paths to the gs if the transition rates are larger than their destruction rates . this treatment was utilized in an estimation of relic abundances of strongly interacting massive particles ( simps ) after the cosmological color confinement by calculations for the annihilation of massive colored particles ( @xmath218 s ) through a formation of resonant or bound states of @xmath219 @xcite . from a comparison of rates for the @xmath1be@xmath3 destruction [ eq . ( [ eq17 ] ) ] and the transition to the gs [ eqs . ( [ eq29 ] ) and ( [ eq23 ] ) ] , we find that excited states of @xmath1be@xmath3 with @xmath220 produced via the charge exchange reaction of the gs is the effective path in the temperature range of @xmath188 , while those with @xmath221 produced via the reaction of excited state with @xmath222 are not . as will be shown below , the recombination through the charge exchange is insignificant compared with the radiative recombination for @xmath223 . since the gs is the only available path in the epoch of @xmath188 , we take into account only the gs in this work . in this subsection , we show that the @xmath1be charge exchange reaction , @xmath1be@xmath224be@xmath225 , is not an effective destruction process . below , the rate for destruction via the charge exchange @xmath226 is estimated and found to be smaller than the spontaneous emission rate , @xmath227 , or the destruction rate for @xmath36 collisional ionization , @xmath44 . the reaction @xmath1be@xmath228@xmath229 is the inverse of the reaction for @xmath1be@xmath3 production . in thermal environment , the inverse reaction rate @xmath230 is related to the forward reaction rate @xmath231 through the detailed balance @xcite as @xmath232 where @xmath233 is the reaction @xmath233 value , and @xmath56 is the azimuthal quantum number of @xmath1be@xmath3 . since the amplitude of @xmath233 value is of the order of the binding energy of at most , the condition @xmath234 is satisfied . the rate of eq . ( [ eq11 ] ) for the reaction @xmath235be@xmath236 is a total rate as a sum over rates for multiple final states . rates for respective final states are , therefore , smaller than the total rate . as found in studies on the protonium formation @xcite , many final states of @xmath1be@xmath3 with different main and azimuthal quantum numbers are produced . we assume that the effective number @xmath237 of final states are produced with roughly equal partial cross sections . then , the forward rate for one final state @xmath238 is given by eq . ( [ eq11 ] ) divided by @xmath237 . the inverse reaction rate is then estimated as @xmath239 \left(\frac{\mathrm{e}^{-m_e / t}}{2.97\times 10^{-7}}\right ) , \label{eq31}\end{aligned}\ ] ] where eqs . ( [ eq3 ] ) , ( [ eq11 ] ) , and ( [ eq30 ] ) were used , and it was assumed that states of @xmath1be@xmath3 with 30 different main and azimuthal quantum numbers , respectively , are produced in the forward reaction , i.e. , @xmath240 . this charge exchange rate is smaller than the destruction rate via the @xmath36 collisional ionization at @xmath188 , where the destruction significantly affect the effective rate of the @xmath1be@xmath2 formation through ion [ cf . ( [ eq18 ] ) ] . in addition , this rate is smaller than rates of spontaneous and stimulated emissions at @xmath223 so that it is not an effective destruction process . the charge exchange reaction is then safely neglected . in this subsection , we show that the ionization of @xmath1be@xmath3 by cbr is not an effective destruction process . below , the rate for destruction via photoionization @xmath241 is estimated and found to be always smaller than the spontaneous emission rate @xmath227 in the relevant temperature region of @xmath242 . cross sections of atomic photoionization have the following characteristics @xcite . when the kinetic energy in the final scattering state is smaller than the atomic binding energy , i.e. , @xmath243 , the cross section is approximately given by kramer s semi - classical cross section , @xmath244 where @xmath245 is the energy of ionizing photon , @xmath246^{1/2}$ ] @xcite is defined , and the index of the charge number @xmath247 and arguments of the reduced mass @xmath49 are omitted . this cross section is different from that for normal hydrogen - like electronic ion by the factor of @xmath248 . in the current assumption that the mass of @xmath0 is much larger than that of @xmath1be , the effective charge for the electric dipole transition in the center of mass system @xcite is @xmath249 . on the other hand , the effective charge in the electronic ion is necessarily @xmath250 . this difference is reflected in the above equation . the photoionization rate is given by @xmath251 the integral in eq . ( [ eq20 ] ) can be approximately estimated as follows . we divide the integral interval to ( 1 ) @xmath252 and ( 2 ) @xmath253 . as for component ( 1 ) , we assume @xmath254 ^ 2 \sim 1 $ ] , and @xmath255 because of @xmath256 . as for component ( 2 ) , we assume @xmath254 ^ 2 \sim \rho^4 $ ] , and @xmath255 since the region of @xmath256 gives the dominant contribution to the integral . under these approximation , both components of integral are found to be @xmath257 . we then find the photoionization rate from the order of magnitude estimation , @xmath258 this photoionization rate as well as the spontaneous emission rate scales as @xmath194 , and is larger for states with smaller main quantum numbers . the photoionization rate of @xmath1be@xmath3 is always much smaller than the spontaneous emission rate for excited states with @xmath259 at @xmath188 . the photoionization , therefore , never becomes an effective destruction process [ cf . ( [ eq18 ] ) ] , and is neglected . we show an effect of the nonradiative recombination on bbn . figure [ fig1 ] shows effective recombination rates for @xmath1be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 ( thick lines ) and direct radiative capture rate for @xmath1be(@xmath0 , @xmath23)@xmath1be@xmath2 ( eq . ( 2.9 ) of ref . @xcite ; thin solid line ) as a function of temperature @xmath260 . for example we assumed that the effective rate is given by eq . ( [ eq18 ] ) , whose respective terms are described by the rough order of estimates [ eqs . ( [ eq6 ] ) , ( [ eq13 ] ) , ( [ eq23 ] ) , ( [ eq29 ] ) and ( [ eq17 ] ) ] . we fix the physical values except for the @xmath1be@xmath3 formation cross section at @xmath116 , @xmath115 , and its destruction cross section , @xmath261 . in order to conservatively take into account uncertainties in the magnitudes of cross sections , and also to check dependences of the recombination rate on cross sections , five sets of cross section values are assumed : ( @xmath115 , @xmath261)=(17.5 mb , 0.219 mb ) ( standard rate : thick solid line ) , ( @xmath262 mb , 0.219 mb ) and ( 0.175 mb , 0.219 mb ) ( upper and lower dashed lines ) , and ( 17.5 mb , 21.9 mb ) and ( 17.5 mb , @xmath263 mb ) ( lower and upper dotted lines ) . be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 ( thick lines ) and direct radiative capture rate for @xmath1be(@xmath0 , @xmath23)@xmath1be@xmath2 ( thin solid line ) as a function of temperature @xmath260 . the thick solid line shows the standard rate estimated with eqs . ( [ eq18 ] ) , ( [ eq6 ] ) , ( [ eq13 ] ) , ( [ eq23 ] ) , ( [ eq29 ] ) , and ( [ eq17 ] ) . upper and lower dashed lines correspond to rates in which the cross section of charge exchange reaction @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 , @xmath115 , is multiplied by 100 and 0.01 , respectively . lower and upper dotted lines correspond to rates in which the cross section of the @xmath1be@xmath3 destruction reaction , @xmath261 , is multiplied by 100 and 0.01 , respectively . [ fig1],width=302 ] we used a bbn code including many reactions associated with the @xmath0 particle @xcite constructed with the public kawano code @xcite . we then solve the nonequilibrium nuclear and chemical reaction network associated to the @xmath0 particle with improved reaction rates derived from rigorous quantum many - body dynamical calculations @xcite . the neutron lifetime was updated with @xmath264 s @xcite based on improved measurements @xcite . the baryon - to - photon ratio was taken from the value determined by the wmap @xcite for model @xmath139cdm ( wmap9 data only ) : @xmath265 @xcite . the reaction rate for @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 was updated with that in ref . the new reaction suggested in this work , i.e. , @xmath1be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 , was then included . figure [ fig2 ] shows calculated abundances of normal nuclei ( a ) and @xmath266 nuclei ( b ) as a function of @xmath260 . @xmath267 and @xmath268 are mass fractions of @xmath123h and @xmath19he , respectively , while other lines correspond to number abundances with respect to that of @xmath123h . nuclear abundances are shown for all nuclei whose abundances are larger than @xmath269 times the h abundance except for @xmath24li . thick solid lines correspond to the standard rate for @xmath1be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 , while thick dot - dashed and dashed lines correspond to the charge exchange cross section @xmath115 which is higher than that of the standard case by a factor of 10 and 100 , respectively . we have chosen these three cases since the resulting @xmath1be@xmath2 abundance is most sensitive to the cross section especially when the cross section is large . the thin solid line corresponds to the case in which the recombination through @xmath1be@xmath38 is completely neglected . the dotted lines show a result of sbbn calculation . depending on the parameter set for cross sections , time evolutions of @xmath1be and @xmath1be@xmath2 abundances change . therefore , the final abundance of @xmath1be+@xmath1be@xmath2 is affected by the efficiency of the recombination of @xmath1be with @xmath0 through the ion . nuclei ( b ) as a function of @xmath260 . the abundance and the lifetime of @xmath0 particle are taken to be @xmath270 and @xmath271 , respectively . @xmath267 and @xmath268 are mass fractions of @xmath123h and @xmath19he , respectively , while other lines correspond to number abundances with respect to that of @xmath123h . thick lines correspond to the standard reaction rate for @xmath1be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 , while thick dot - dashed and dashed lines correspond to the cross sections of @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 multiplied by a factor of 10 and 100 , respectively . the thin solid lines correspond to the case in which the recombination through @xmath1be@xmath38 is completely neglected . the dotted lines show a result of sbbn calculation . @xmath24be@xmath2 abundances are drawn for both cases with ( thick lines ) and without ( thin lines ) the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 . [ fig2],title="fig:",width=302 ] + nuclei ( b ) as a function of @xmath260 . the abundance and the lifetime of @xmath0 particle are taken to be @xmath270 and @xmath271 , respectively . @xmath267 and @xmath268 are mass fractions of @xmath123h and @xmath19he , respectively , while other lines correspond to number abundances with respect to that of @xmath123h . thick lines correspond to the standard reaction rate for @xmath1be(@xmath66 , @xmath23)@xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath2 , while thick dot - dashed and dashed lines correspond to the cross sections of @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 multiplied by a factor of 10 and 100 , respectively . the thin solid lines correspond to the case in which the recombination through @xmath1be@xmath38 is completely neglected . the dotted lines show a result of sbbn calculation . @xmath24be@xmath2 abundances are drawn for both cases with ( thick lines ) and without ( thin lines ) the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 . [ fig2],title="fig:",width=302 ] we note that @xmath24be@xmath2 is produced in the temperature region of @xmath272@xmath273 through the reaction @xmath1be@xmath2(@xmath21 , @xmath22)@xmath24be@xmath2 @xcite following the production of @xmath1be@xmath2 . there are two operative reactions for @xmath24be@xmath2 production : @xmath1be@xmath2(@xmath21 , @xmath22)@xmath24be@xmath2 and @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 . in fig . [ fig2 ] , @xmath24be@xmath2 abundances are shown for both cases with ( thick lines ) and without ( thin lines ) the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 . when the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 is switched off , @xmath24be@xmath2 is found to be produced via @xmath1be@xmath2(@xmath21 , @xmath22)@xmath24be@xmath2 , and the @xmath1be abundance changes because of the variable recombination rates . however , in this epoch , the reaction @xmath24be@xmath2(@xmath23 , @xmath20)@xmath19he@xmath2 , that is the inverse reaction of the latter reaction above , operates effectively as the destruction reaction of @xmath24be@xmath2 . this destruction is caused by a small reaction @xmath233 value of the forward @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 reaction ( @xmath274 mev in the model of ref . the @xmath24be nuclei is unstable against the @xmath20 emission so that @xmath24be@xmath2 is barely bound with respect to the @xmath20+@xmath19he@xmath2 separation channel although it is stabilized by the coulomb attractive force of @xmath0 . the small @xmath233 value means a small energy threshold of the inverse reaction and resultingly a large inverse reaction rate . at @xmath275 , @xmath24be@xmath2 is produced via the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 because of an increased @xmath19he@xmath2 abundance . since larger exchange rates for @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 lead to earlier enhancement of @xmath1be@xmath2 abundance , @xmath24be@xmath2 production at @xmath272@xmath273 proceeds strongly when the rate is larger although its effect does not seen in curves for the case with the reaction @xmath19he@xmath2(@xmath20 , @xmath23)@xmath24be@xmath2 . we studied effects of a long - lived negatively charged massive particle , i.e. , @xmath0 on bbn . in this bbn model including the @xmath0 , @xmath1be destruction can occur , and the @xmath1be abundance can reduce from the abundance predicted in sbbn model . resultingly the final abundance of @xmath1li can explain the abundances observed in mpss . the @xmath1be destruction proceeds through formation of @xmath1be@xmath2 followed by its radiative proton capture reaction . in this work we suggest a new route of @xmath1be@xmath2 formation , i.e. , the @xmath1be charge exchange between ion and @xmath0 . 1 . the rate for the @xmath1be@xmath2 formation through depends on the number fraction of ion , the charge exchange cross section of and the probability that produced excited states @xmath1be@xmath3 are converted to the gs ( sec . [ sec21 ] ) . 2 . in the @xmath1be exchange reaction , i.e. , @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 , the gs is converted to highly excited states of @xmath1be@xmath3 with main quantum numbers of @xmath276 . the gs and the excited states @xmath1be@xmath3 have almost equal sizes of binding energies and atomic radii ( sec . [ sec22 ] ) . 3 . in the epoch of @xmath1be destruction in the bbn model including the @xmath0 , the ion can be regarded as an isolated system which is separated from particles in thermal bath in the universe ( sec . [ sec23 ] ) . 4 . in the epoch , the rate for the recombination , @xmath1be@xmath277 @xmath74 , is large enough , and the equilibrium abundance ratio of the fully ionized @xmath1be@xmath37 and the partially ionized had been realized ( sec . [ sec24 ] ) . the cross section of the charge exchange reaction @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 is estimated from an analogy of the protonium ( @xmath62 ) formation . the rate for to form @xmath1be@xmath3 through this reaction is then calculated . the rate is proportional to the @xmath0 abundance @xmath278 , and can be much larger than the cosmic expansion rate in the epoch of the @xmath1be destruction ( sec . [ sec25 ] ) . 6 . the bound - bound transition rate of @xmath1be@xmath3 produced via this charge exchange reaction to the gs @xmath1be@xmath2 is estimated by applying the electric dipole transition rate to the exotic atomic system of @xmath1be@xmath3 ( sec . [ sec26 ] ) . 7 . the cross section of the @xmath1be@xmath3 destruction via @xmath36 collisional ionization is estimated from an analogy of the destruction of muonic hydrogen , @xmath279 , at a collision with @xmath66 . the rate for @xmath1be@xmath3 destruction through this reaction is then calculated . the rate is smaller than that for the transition to the gs only for excited states @xmath1be@xmath3 with main quantum number @xmath187 . such excited states are produced via the @xmath1be exchange of the gs . the gs is , therefore , the only effective path of the gs @xmath1be@xmath2 formation ( sec . [ sec27 ] ) . the @xmath1be charge exchange rate of @xmath1be@xmath3 via the @xmath66 collision is estimated with the detailed balance relation . the rate is smaller than the destruction rate via the @xmath36 collisional ionization at @xmath188 , while it is smaller than rates of spontaneous and stimulated emissions at @xmath223 . the charge exchange reaction , therefore , does not work as an effective reaction for @xmath1be@xmath3 destruction ( sec . [ sec29 ] ) . 9 . the photoionization rate of @xmath1be@xmath3 is estimated with a classical cross section . differently from the rate for the destruction in electronic collisional ionization , the photoionization rate is always much smaller than the spontaneous emission rate . the photoionization is , therefore , not an important reaction of @xmath1be@xmath3 destruction ( sec . [ sec28 ] ) . 10 . using physical quantities relevant to the @xmath1be@xmath2 formation through estimated in this work , effective recombination rates are derived as a function of cosmic temperature for several example cases . our primary rate for the @xmath1be@xmath2 formation through is larger than the rate for the direct formation via the radiative recombination of @xmath1be and @xmath0 . uncertainties in cross sections of the charge exchange @xmath1be@xmath38(@xmath0 , @xmath66)@xmath1be@xmath3 and the destruction @xmath1be@xmath3(@xmath36 , @xmath36 @xmath1be)@xmath0 affect the @xmath1be@xmath2 formation rate . the importance of this reaction is shown in bbn calculations with a latest nonequilibrium reaction network code ( sec [ sec3 ] ) . we are grateful to professor masayasu kamimura for discussion on the reaction cross sections . this work was supported by the national research foundation of korea ( grant nos . 2012r1a1a2041974 , 2011 - 0015467 , 2012m7a1a2055605 ) , and in part by grants - in - aid for scientific research of jsps ( 24340060 ) , and scientific research on innovative area of mext ( 20105004 ) . 99 f. spite and m. spite , astron . astrophys . * 115 * , 357 ( 1982 ) . s. g. ryan , t. c. beers , k. a. olive , b. d. fields and j. e. norris , astrophys . j. * 530 * , l57 ( 2000 ) . j. melendez and i. ramirez , astrophys . j. * 615 * , l33 ( 2004 ) . m. asplund , d. l. lambert , p. e. nissen , f. primas and v. v. smith , astrophys . j. * 644 * , 229 ( 2006 ) . p. bonifacio _ et al . _ , astrophys . * 462 * , 851 ( 2007 ) . w. aoki , p. s. barklem , t. c. beers , n. christlieb , s. inoue , a. e. g. perez , j. e. norris and d. carollo , astrophys . j. * 698 * , 1803 ( 2009 ) . j. i. g. hernandez , p. bonifacio , e. caffau , m. steffen , h. -g . ludwig , n. t. behara , l. sbordone and r. cayrel _ et al . _ , astron . astrophys . * 505 * , l13 ( 2009 ) . l. sbordone , p. bonifacio , e. caffau , h. -g . ludwig , n. t. behara , j. i. g. hernandez , m. steffen and r. cayrel _ et al . _ , astron . astrophys . * 522 * , a26 ( 2010 ) . l. monaco , p. bonifacio , l. sbordone , s. villanova and e. pancino , astron . astrophys . * 519 * , l3 ( 2010 ) . l. monaco , s. villanova , p. bonifacio , e. caffau , d. geisler , g. marconi , y. momany and h. -g . ludwig , astron . astrophys . * 539 * , a157 ( 2012 ) . a. mucciarelli , m. salaris and p. bonifacio , mon . not . . soc . * 419 * , 2195 ( 2012 ) . w. aoki , h. ito and a. tajitsu , astrophys . j. * 751 * , l6 ( 2012 ) . w. aoki , memorie della societa astronomica italiana supplementi * 22 * , 35 ( 2012 ) . a. frebel , w. aoki , n. christlieb , h. ando , m. asplund , p. s. barklem , t. c. beers and k. eriksson _ et al . _ , nature * 434 * , 871 ( 2005 ) . w. aoki , a. frebel , n. christlieb , j. e. norris , t. c. beers , t. minezaki , p. s. barklem and s. honda _ et al . _ , astrophys . j. * 639 * , 897 ( 2006 ) . r. h. cyburt , b. d. fields and k. a. olive , jcap * 0811 * , 012 ( 2008 ) . o. richard , g. michaud and j. richer , astrophys . j. * 619 * , 538 ( 2005 ) . a. j. korn _ et al . _ , astrophys . j. * 671 * , 402 ( 2007 ) . k. lind , f. primas , c. charbonnel , f. grundahl and m. asplund , astron . astrophys . * 503 * , 545 ( 2009 ) . r. cayrel _ et al . _ , astron . astrophys . * 473 * , l37 ( 2007 ) . m. asplund and j. melndez , aip conf . proc . * 990 * , 342 ( 2008 ) . m. steffen , r. cayrel , p. bonifacio , h. g. ludwig and e. caffau , in iau symp . 265 , chemical abundances in the universe : connecting first stars to planets , edited by k. cunha , m. spite and b. barbuy ( cambridge univ . press , cambridge , 2010 ) , p. 23 . m. pospelov , phys . lett . * 98 * , 231301 ( 2007 ) . k. kohri and f. takayama , phys . d * 76 * , 063507 ( 2007 ) . r. h. cyburt , j. r. ellis , b. d. fields , k. a. olive and v. c. spanos , jcap * 0611 * , 014 ( 2006 ) . m. kaplinghat and a. rajaraman , phys . d * 74 * , 103004 ( 2006 ) . k. hamaguchi , t. hatsuda , m. kamimura , y. kino and t. t. yanagida , phys . b * 650 * , 268 ( 2007 ) . c. bird , k. koopmans and m. pospelov , phys . d * 78 * , 083010 ( 2008 ) . m. kusakabe , t. kajino , r. n. boyd , t. yoshida and g. j. mathews , phys . d * 76 * , 121302 ( 2007 ) . m. kusakabe , t. kajino , r. n. boyd , t. yoshida and g. j. mathews , astrophys . j. * 680 * , 846 ( 2008 ) . k. jedamzik , phys . d * 77 * , 063524 ( 2008 ) ; k. jedamzik , jcap * 0803 * , 008 ( 2008 ) . m. kamimura , y. kino and e. hiyama , prog . * 121 * , 1059 ( 2009 ) . m. pospelov , arxiv:0712.0647 [ hep - ph ] . m. kawasaki , k. kohri and t. moroi , phys . b * 649 * , 436 ( 2007 ) . t. jittoh , k. kohri , m. koike , j. sato , t. shimomura and m. yamanaka , phys . d * 76 * , 125023 ( 2007 ) ; phys . rev . d * 78 * , 055007 ( 2008 ) ; arxiv:1001.1217 [ hep - ph ] . m. pospelov , j. pradler and f. d. steffen , jcap * 0811 * , 020 ( 2008 ) . m. y. khlopov and c. kouvaris , phys . d * 77 * , 065002 ( 2008 ) . m. kawasaki , k. kohri , t. moroi and a. yotsuyanagi , phys . d * 78 * , 065011 ( 2008 ) . s. bailly , k. jedamzik and g. moultaka , phys . d * 80 * , 063509 ( 2009 ) ; k. jedamzik and m. pospelov , new j. phys . * 11 * , 105028 ( 2009 ) . m. kamimura , y kino and e hiyama , aip conf . proc . * 1238 * , 139 ( 2010 ) . m. kusakabe , t. kajino , t. yoshida and g. j. mathews , phys . rev . d * 81 * , 083521 ( 2010 ) . m. pospelov and j. pradler , ann . nucl . part . sci . * 60 * , 539 ( 2010 ) . k. kohri , s. ohta , j. sato , t. shimomura and m. yamanaka , phys . d * 86 * , 095024 ( 2012 ) . r. h. cyburt , j. ellis , b. d. fields , f. luo , k. a. olive and v. c. spanos , jcap * 1212 * , 037 ( 2012 ) . h. apo , i. boztosun , g. kocak and a. b. balantekin , phys . c * 85 * , 044602 ( 2012 ) . b , v. chist , c. dulieu , e. browne , v. chechev , n. kuzmenmko , r. helmer and a. nichols _ et al . _ , _ table of radionuclides , monographie bipm-5 , vol . 1 _ ( cea / bnm - lnhb , 91191 gif - sur - yvette , france and bipm , pavillon de breteuil , 92312 svres , france , 2004 ) . a. bertin , a. vitale , and a. placci , nuovo cimento rivista serie * 5 * , 423 ( 1975 ) . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 170 * , 377 ( 2007 ) . d. larson _ et al . _ , astrophys . j. suppl . * 192 * , 16 ( 2011 ) . g. hinshaw , d. larson , e. komatsu , d. n. spergel , c. l. bennett , j. dunkley , m. r. nolta and m. halpern _ et al . _ , arxiv:1212.5226 [ astro-ph.co ] . j. a. kunc , phys . rev . a * 49 * , 338 ( 1994 ) . m. s. smith , l. h. kawano and r. a. malaney , astrophys . j. suppl . * 85 * , 219 ( 1993 ) . a. p. serebrov and a. k. fomin , phys . c * 82 * , 035501 ( 2010 ) . g. j. mathews , t. kajino and t. shima , phys . d * 71 * , 021302(r ) ( 2005 ) . a. serebrov , v. varlamov , a. kharitonov , a. fomin , yu . pokotilovski , p. geltenbort , j. butterworth and i. krasnoschekova _ et al . _ , phys . b * 605 * , 72 ( 2005 ) .

building upon earlier observations which demonstrate substantial star - to - star differences in the carbon abundances of m13 subgiants , we present new keck lris spectra reaching more that 1.5 mag below the m13 main - sequence turn - off ( to v 20 ) . our analysis reveals a distribution of c abundances similar to that found among the subgiants , implying little change in the compositions of the m13 stars at least through the main - sequence turn - off . we presume these differences to be the result of some process operating early in the cluster history . additional spectra of previously studied bright m13 giants have been obtained with the hale 5-m . a comparison of c abundances derived using the present methods and those from the literature yield a mean difference of 0.03.14 dex for four stars in common with and 0.14.07 dex for stars also observed by ( if one extreme case is removed ) . we conclude that the lower surface c abundances of these luminous giants as compared to the subgiants and main - sequence stars are likely the result of mixing rather than a difference in our abundance scales . nh band strengths have also been measured for a handful of the most luminous m13 turn - off stars . while molecular band formation in such stars is weak , significant star - to - star nh band strength differences are present . moreover , for the stars with both c and n measurements , differences between stars in these two elements appear to be anticorrelated . finally , the most recent c and n abundances for main - sequence , main - sequence turn - off , and subgiant stars in 47 tuc , m71 , m5 , and the present m13 data are compared .

astro-ph0312315

it has been known since the early 1970 s that the otherwise indistinguishable members of any given galactic globular cluster ( gc ) exhibit significant star - to - star variations in surface abundances of certain light elements ( most notably c and n , as well as o , and often na , al , and mg ) cen and m22 , both of which appear to have experienced some degree of self - enrichment . ] . however , while the abundance patterns commonly observed point to an origin in proton capture nucleosynthesis @xcite , identification of the specific reaction site(s ) and a full theoretical description of the abundance modifying process(es ) remain uncertain . as has been pointed out in numerous reviews ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) two possibilities exist : first , the present day cluster stars may have modified their own surface compositions through some mixing process not included in standard models ( i.e. , an `` in situ '' scenario ) . by far the most promising candidate site in this regard is the region above the h - burning shell after first dredge - up in evolving cluster giants , where conditions for partial cn and possibly on - cycle reactions exist ( see * ? ? ? * for one of the earliest treatments ) . subsequent circulation of this material into the stellar envelope via meridional currents or turbulent diffusion ( for example * ? ? ? * ) will result in decreasing c abundances and increasing n with evolutionary state as has been observed along the red giant branches ( rgb ) of several metal - poor clusters ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for classic examples ) . moreover , the operation of such a mechanism can at least qualitatively explain the o and mg versus na and al anticorrelations found among the most luminous red giants in several clusters ( e.g. * ? ? ? * and references therein ) . common to all models of this process is the prohibition of `` extra mixing '' by the molecular weight gradient left behind by the inward excursion of the convective envelope during first dredge - up . only after the molecular weight discontinuity has been destroyed by the outward moving h - burning shell , an event marked by the rgb luminosity function ( lf ) bump , is mixing expected to take place . this theoretical prediction appears to be borne out by observations of decreased li abundances following the lf bump in ngc 6752 and similar drops in @xmath2c@xmath3c seen by @xcite in ngc 6528 and m4 . however , this can not be the entire picture . as early as 1978 , it was noted by @xcite that the subgiant branch ( sgb ) and likely the main - sequence ( ms ) stars of 47 tuc also possessed star - to - star differences in ch and cn band strengths . this has been most recently followed in 47 tuc to @xmath4 mag below the ms turn - off ( msto ) by @xcite . an analysis of their observed cn and ch band strengths yields factors of 10 variations in n anticorrelated with factors of 3 differences in c @xcite , matching those found among the more evolved members . such cn and ch ( n and c ) variations have also been shown to exist among the ms , msto , or sgb stars of ngc 6752 , m71 , and m5 @xcite moreover , star - to - star variations in na , al , o , and mg , similar to those found among the luminous cluster stars , have been identified among the sgb and msto stars of 47 tuc @xcite , ngc 6752 , m71 @xcite , and m5 @xcite . although the various correlations and anticorrelations among these elements suggest the presence of material exposed to proton - capture reactions , such stars lie well below the lf bump and , particularly in the case of the ms stars , no mechanism is known for circulating significant quantities of cn(o ) nucleosynthesized material to their surfaces . thus the second possible origin of the gc abundance variations - they have been set in place before rgb ascent and are due to the operation of some mechanism early in the cluster history ( sometimes referred to as a `` primordial '' scenario ) . a number of possibilities exist as are discussed extensively by @xcite , including : that the proto - cluster gas was inhomogeneous in these elements ( a true primordial origin ) , that there was an extended period of star formation of sufficient duration to allow some low - mass stars to form with material ejected from more massive already - evolved cluster asymptotic giant branch ( agb ) stars , or that the present day cluster stars have accreted agb ejecta onto their surfaces after their formation . the appeal of agb stars as sites of the proton capture nucelosynthesis lies in their ability to modify the cluster gas in light elements ( including c , n , o , al , na , and mg - see * ? ? ? * ) while not altering the abundances of heavy elements . as the reader has likely noted , observational evidence exists for both mixing and early contamination scenarios , which has led many investigators to conclude that the compositions of the cluster stars we observe today are not the result of one or the other scenario exclusively , but rather both . unfortunately , this leads to difficulties in disentangling the contributions of each process among the more luminous cluster stars - a problem that can only be reconciled by exploring the compositions of a cluster s stars to the msto and below . clearly , abundance trends found among a cluster s ms stars reflect the original makeup of the bright giants , while deviations from this `` baseline '' composition are likely the result of mixing . this was recently demonstrated in the case of m13 by @xcite ( hereafter bcs02 ) - that a large spread in c abundances exists among the sgb stars of m13 , which presumably reflects star - to - star variations in c abundances set early in the cluster history . however , the sgb c abundances also appear larger than those found by other investigators among the more luminous m13 stars , implying the operation of a mixing mechanism on the rgb which has reduced surface c abundances . in the present paper , we return to m13 and extend our sample more than two magnitudes fainter to include ms stars . in addition , we have also obtained spectra of m13 bright giants observed in earlier studies to verify our abundance scale . our results confirm those of bcs02 - that a primordial spread in the distribution of light elements exists in m13 which has further been modified during rgb ascent . measurements of the 3360 nh bands also were obtained for a handful of the more luminous stars in our sample . n abundances calculated from these bands suggest a c versus n anticorrelation at the level of the msto . the initial sample of stars in m13 was aimed to produce subgiants at the base of the rgb . it consisted of those stars from the photometric database ( described by * ? ? ? * ; * ? ? ? * ) located more than 150 arcsec from the center of m13 ( to avoid crowding ) with @xmath5 and with @xmath6 mag . a slitmask with 0.7 arcsec wide slitlets , narrower than normal to enhance the spectral resolution and minimize contributions from adjacent stars in these crowded fields , was designed using jgc s software from this sample and used in may 2001 with lris @xcite at keck . for this slitmask , as for all those used for the m13 stars , the red side of lris was set to include the nad lines and h@xmath7 . we used the highest possible dispersion , 0.64 / pixel ( 29 km s@xmath8/pixel ) or 1.9 / spectral resolution element there , to facilitate radial velocity confirmation of cluster membership . given that the radial velocity of m13 is 246 km s@xmath8 , distinguishing field stars from cluster members is then straightforward . the blue side of lris @xcite was used with the 600 line / mm grism blazed at 5000 . the detector for lris - b at that time was a 2048x2048 ccd not optimized for uv response . the spectra covered the range from @xmath93400 to 5000 , thus including the strongest cn band at 3885 and the g band of ch at 4300 , with a resolution of @xmath94 ( 1.0 / pixel ) . two additional slitmasks were defined from this sample and used in may 2002 during less than ideal weather conditions for 6 exposures of 4800 sec each . the spectra were dithered by moving the stars along the length of the slitlets by 2 arcsec between exposures . these spectra are part of those presented in bcs02 . because of the crowded fields , in addition to the intended stars some slitlets contained additional stars bright enough to provide suitable spectra , and these were utilized as well . as might be expected from the luminosity function , most of the secondary sample consists of stars at or just below the main - sequence turnoff . hence subtraction of sequential exposures was not possible , and they were reduced individually using figaro @xcite scripts , then the 1d spectra for each object were summed . based on the serendipitous main sequence stars found in the 2002 observations ( see the plots in bcs02 ) , we decided to try to reach main sequence stars well below the turnoff in m13 , sufficiently faint to be cool enough to have detectable ch bands . the criteria used to define the sample from the photometric database were 19.3 @xmath10 i @xmath10 19.7 , v - i within 0.06 mag of the main sequence of m13 , taken as 1.26 + 0.28(i-19.4 ) , and located more than 200 arcsec from the center of m13 . a single slitmask with 0.8 arcsec wide slitlets was designed and used at keck with lris june 26 , 2003 . the blue spectra cover the full range from the atmospheric cutoff to 5000 , with 1.0 / pixel and a spectral resolution of @xmath9 4 . four exposures totalling 4200 sec were obtained . the new very high quantum efficiency detector for lris - b consisting of two 2kx4k marconi ccds , with 15 @xmath11 pixels and a readout noise of 4.0 e@xmath12 , was completed and installed into lris in june 2002 , and so was available for these observations . the high uv throughput of lris - b with this new sensitive detector for the first time enabled us to reach the nh bands in the brighter of these stars with some precision . the locations of the faint program stars on the m13 color - magnitude diagram ( cmd ) are shown in figure [ fig_1 ] . there are published surveys @xcite in which ch indices have been used to determine [ c / fe ] values for large samples of the highest luminosity giants in m13 . however , our keck / lris sample of low luminosity stars in m13 has no overlap with these earlier works . to ensure that the merger of our data for faint stars in m13 with these published datasets for ch band strengths in m13 giants is valid , we need to verify the consistency of the different measurements of the ch indices and resulting abundances . to demonstrate this , we obtained new blue spectra of a small sample of bright giants with published ch band strengths from earlier studies , and remeasured their ch indices with the same procedures used for the lower luminosity m13 stars of our main sample ( as described below ) . these spectra were taken in april and may , 2003 at the hale telescope on palomar mountain during observing runs intended primarily for other projects . the blue channel of the double spectrograph @xcite was used with a 1200 line / mm grating and a loral 512x2788 15@xmath11 pixel ccd , giving 0.55 / pixel with a spectral resolution of 1.9 for a 1 arcsec slit . our analysis essentially repeats that of bcs02 and is fully described in @xcite ( hereafter bc01 ) and the reader is referred to these works for details . to summarize : strengths of the 4350 ch ( g ) bands of our program stars were measured via the i(ch ) index of @xcite - an index which compares the flux removed by the g - band to the adjacent continuum on both sides . the resulting indices , calculated using bandpasses corrected for the radial velocity of m13 , are plotted for the sample of faint stars as a function of i magnitude in figure [ fig_2 ] . the decrease in ch band strengths near i @xmath0 18 is due to the higher temperatures of the msto stars ( as pointed out by bc01 ) . however , among the fainter ms stars in the sample ( near i @xmath0 19.5 ) , the surface temperatures have dropped by roughly 300k and again a large and significant scatter in ch band strengths is apparent . the one sigma error bars plotted for the present sample have been determined entirely from poisson statistics in the molecular - band and continuum spectral windows . in a similar manner , the strength of absorption by the 3350 nh band was measured in spectra of the more luminous members of the keck ms / msto sample using the double sided logarithmic @xmath13 index as defined in @xcite . the resulting indices ( and one sigma poisson error bars ) are also plotted in figure [ fig_2 ] . this marks the first time nh bands have been observed among such faint stars in a globular cluster . spectra of two msto stars exhibiting differing nh band strengths , and two ms stars with differing ch band strengths are shown in figure [ fig_3 ] . in order to relate the observed indices to the underlying [ c / fe ] , we employ a series of synthetic spectra based on marcs model atmospheres . our models are those used in bc01 and bcs02 and based on the 16 gyr @xmath14 $ ] = @xmath151.48 o - enhanced isochrone grid of @xcite . the locations of the model points on the m13 i , v@xmath15i cmd are shown in figure [ fig_1 ] assuming @xmath16 = 14.43 and a reddening of e(b@xmath15v ) = 0.02 as in bc01 and bcs02 . from each model and a given set of c / n / o abundances , synthetic spectra were computed using the ssg program ( * ? ? ? * and references therein ) and the line list of @xcite at a step size of 0.02 ( see bc01 for further details ) assuming the average heavy element compositions of @xcite . the result was then convolved with a gaussian kernel to match the resolution of the observed spectra and i(ch ) and @xmath13 indices were measured . the model indices for i(ch ) are illustrated in figure [ fig_2 ] for four c abundances ( as in bcs02 ) : [ c / fe ] = @xmath150.85 and [ c / fe ] = @xmath151.1 , which roughly match the observed compositions of m13 s cn - weak and strong bright giants respectively ( see * ? ? ? * ) , and [ c / fe ] = 0.0 and [ c / fe ] = @xmath150.5 . also plotted in figure [ fig_2 ] are @xmath13 indices for a variety of [ n / fe ] values . note that among these relatively warm ms / msto stars , there is little sensitivity in the ch ( nh ) band strengths to changes in n , o ( c , o ) abundances ( as opposed to the cool giants where molecular equilibrium must be considered , particularly with regard to o ) . as a check of this , table [ tbl-1 ] shows the sensitivity of i(ch ) and @xmath13 to such changes in a cool ms model ( t@xmath17=5601 , log g=4.66 , corresponding to an m13 ms star with @xmath18=19.60 ) . following bcs02 , we have applied the method of @xcite to convert the observed indices to corresponding c and n abundances : the model isoabundance curves were interpolated to the @xmath19 of each program star , and the observed index converted into the corresponding abundance based on the synthetic index at that @xmath19 . resulting c and n abundances are plotted in figures [ fig_4 ] and [ fig_5 ] . note that the large error bars which accompany the stars of figure [ fig_4 ] near i=19 and the stars of low [ c / fe ] ( @xmath20 ) are due to the overall weakness of the ch bands small errors in i(ch ) therefore result in large changes in [ c / fe ] . likewise , a similar situation exists among the msto stars with measured nh band strengths . as with the faint stars , the i(ch ) index was measured from the spectra of the six bright m13 giants . for each star this value was compared to synthetic indices generated from model atmospheres whose stellar parameters were taken from the high resolution analyses of @xcite , and @xcite , including their heavy element and [ o / fe ] abundances . where available , [ n / fe ] values from @xcite were also used . for two stars ( k188 and iii-7 ) , n abundances were not available from the literature , and a value of + 1.0 was assumed . for star iii-7 , an [ o / fe ] of 0.0 was used . the model parameters and the resulting [ c / fe ] abundance which matched the observed i(ch ) indices are listed in table [ tbl-2 ] along with the c abundances from @xcite and @xcite . for the four bright giants in common with @xcite , we find an average offset of 0.03(@xmath1 0.14 ) dex in [ c / fe ] ( present @xmath15 smith ) . we therefore consider our c abundances to be essentially on the same scale , as might be expected considering the similar analysis tools used . the difference between our results and those of @xcite are somewhat larger : 0.25(@xmath10.23 ) . however , almost half of this offset is driven by the result for ii-76 . excluding this star reduces the average difference to 0.14(@xmath10.07 ) . note that ii-76 has both a high [ o / fe ] and a lower [ n / fe ] abundance as might be expected from a star with a lesser amount of cn(o)-cycle material in its atmosphere . the source of this discrepancy is likely the cooler model used for ii-76 by suntzeff ( t@xmath17 = 4220k versus the 4350k used here ) , as well as the lower o abundance ( [ o / fe ] = 0.0 ) and higher microturbulent velocity ( 2.5 km s@xmath8 ) . repeating our analysis with the values used by suntzeff reduces our resulting [ c / fe ] by 0.32 dex to @xmath150.96 . the luminous stars of suntzeff plotted in figure [ fig_4 ] have therefore been shifted by 0.14 in [ c / fe ] to place them on our abundance scale . given the use of the same modeling codes , line lists , and ch indices throughout our analysis , we presume the resulting c abundances from both the faint and bright star samples , as well as those of @xcite and @xcite ( with the appropriate shift ) , to be on the same abundance scale . any systematic differences due to different telescope / spectrograph systems will be minimized by the use of the i(ch ) index which uses continuum bands both blueward and redward of the ch feature to remove slope differences due to variations in instrumental response . there are several points to be made about the present results , which are given in tabular form in tables [ tbl - a1 ] and [ tbl - a2 ] of appendix a. first , as can be seen in figure [ fig_2 ] , significant differences in [ c / fe ] exist among stars at least 1.5 mag fainter than the msto in m13 . this corresponds to a mass of approximately 0.66 @xmath21 using the isochrone of figure [ fig_1 ] . among these old ms stars , cn(o)-cycle reactions are entirely confined to the central core ( see for example figures 4 and 5 of * ? ? ? * ) and as has been pointed out by numerous investigators , ms stars such as these are not thought to possess a mechanism that connects their surface with regions of energy generation ( namely the core ) . indeed , should such mixing take place , the subsequent paths of the stars in the cmd would be radically altered by the infusion of fresh h into the core ( e.g. * ? ? ? one must conclude the source of the observed differences in [ c / fe ] is likely not the stars themselves . moreover , the values of [ c / fe ] among the ms stars are consistent with those found by bcs02 among the m13 sgb stars ( see figure [ fig_4 ] ) and imply little change in composition has occurred from the ms to at least the base of the sgb . figure [ fig_4 ] also includes the [ c / fe ] values of @xcite and @xcite ( shifted upwards by 0.14 dex ) . as discussed in bcs02 , there appears to be a marked decline in [ c / fe ] towards higher luminosities among the m13 giants . clearly an evolutionary change such as this can be best interpreted as the result of a mixing process bringing up c - depleted material from a region in which at least cn - cycle reactions are operating ( see bcs02 for a more detailed discussion ) . also shown in figure [ fig_4 ] is the location of the lf bump in m13 ( from * ? ? ? * ) the event which marks the destruction of the molecular weight gradient thought to inhibit deep mixing . unfortunately , the luminosity at which the onset of c depletion begins is uncertain due to the gap in the available data ( from 15 @xmath10 v @xmath10 17 ) . however , since an extrapolation of the trend in giant - branch [ c / fe ] abundance faintward intersects the magnitude of the lf bump at the average abundance of the fainter stars , it is reasonable to infer that the abundance decline begins near that event ; neither a significant decrease nor a significant increase in carbon abundance with a subsequent recovery to the original value hidden within the gap in our data is reasonably to be expected . the [ n / fe ] values determined from the nh band strengths of the msto stars are plotted in figure [ fig_5 ] . although the error bars are admittedly larger than one would like owing to the weaknesses of the ch and nh bands among the warmer msto stars , a general anticorrelation between [ c / fe ] and [ n / fe ] is suggested . note that these abundances do not suffer from the inherent tendency towards c / n anticorrelations of analyses based on ch and cn band strengths . of course an overall c / n anticorrelation is known to be present among the evolved m13 stars and the values for the bright rgb stars of @xcite are also shown in figure [ fig_5 ] . immediately apparent is the shift of the rgb stars towards lower [ c / fe ] , as is expected from figure [ fig_4 ] . if c - poor / n - rich material is indeed being circulated into the stellar envelopes during rgb ascent , the lack of near solar [ n / fe ] rgb stars is also explained ( although the error bars on the two lower [ n / fe ] msto stars severely limit the weight which can be placed on this statement ) . that higher n abundances do not appear to be found among the rgb stars under these circumstances is perhaps not a surprise if these stars are already leaving the msto with large [ n / fe ] overabundances : an m13 msto star with [ c / fe ] = @xmath150.4 and [ n / fe ] = 1.0 which undergoes a mixing episode reducing [ c / fe ] to @xmath151.2 will experience a rise in [ n / fe ] of only 0.05 dex in essence , the n abundances are already so large , the addition of freshly minted n via the cn - cycle results in only a small fractional change in [ n / fe ] . thus , while the error bars in figure [ fig_5 ] are large , we can at least claim it is not inconsistent with the assertion that we are seeing substantial star - to - star variations in c ( and n ) set early in the cluster history , which are further being modified by mixing during rgb ascent . the possibility of also mixing on - cycle material to the surface is more difficult to assess because of the large n variations among the msto stars . in the example above , an additional reduction in [ o / fe ] from + 0.45 to @xmath22 would increase [ n / fe ] by 0.46 dex and among the bright giants , even larger o depletions ( as much as [ o / fe ] = @xmath23 to @xmath24 ) have been noted . starting with an even larger n overabundance of + 1.4 reduces the change in [ n / fe ] to + 0.25 . however , at least from the small sample of figure [ fig_5 ] , it appears that none of the bright rgb stars possess larger n abundances than their msto counterparts , which in turn suggests the envelopes of at least the initially n - rich stars may not be cycled through a region of on - cycle reactions while on the rgb . clearly knowledge of the o abundances of the m13 msto stars would help settle this question . a similar result was noted in the more metal - poor clusters m92 and m15 by @xcite and @xcite ( respectively ) that substantial n overabundances are present from the sgb to agb that are not necessarily correlated with c abundances . indeed , an analogous situation can been seen in the present results and those of @xcite for m5 ( see figures [ fig_5 ] and [ fig_6 ] ) : the `` higher '' [ c / fe ] msto stars ( at @xmath25 ) span almost a dex in [ n / fe ] . it is clear that if we are to ascribe the same mechanism to the origin of the sgb / msto inhomogeneities in these clusters , it must be operating at the msto or earlier . that significant and correlated star - to - star differences in c and n , as well as o , na , al , and mg have been found among the sgb , msto , and ms stars of several clusters ( see references above ) , implies the operation of some process external to the present stars , presumably having taken place early in the cluster history . the discussion of @xcite includes a comprehensive look at various possibilities . it is worth while however , to revisit a few of the more critical constraints on any theory of the origin of the abundance variations . first , whatever mechanism has altered the light - element compositions of the cluster stars has left the heavy elements essentially untouched , at least to the limits of our ability to determine them the analysis of m5 by @xcite is an excellent example . this alone would seem to exclude the possibility of the light - element variations arising from the merger of two distinct proto - cluster clouds ( as has been pointed out by numerous authors ) . second , these abundance variations appear to be almost ubiquitous among the population of galactic globular clusters . to highlight this , we have plotted in figure [ fig_6 ] the [ c / fe ] and [ n / fe ] values for the present sample of m13 msto stars , the 47 tuc ms stars of @xcite , the m5 sgb stars of @xcite , and the msto stars of m71 from bc01 . note that bc01 did not directly extract c and n abundances from their observed indices we have converted them here following the procedure outlined in @xcite and using the indices and models presented in bc01 ; the values are given in appendix a , table [ tbl - a3 ] . third , the elements which are observed to vary are associated with proton capture nucleosynthesis under conditions of cn and on - cycling . the source / site must process these cno - group elements and return this material to the cluster to be incorporated into the present population of low mass stars either before , during , or after their formation . a popular model which fits these constraints is the incorporation of ejecta from intermediate mass ( 3 - 6@xmath21 ) agb stars undergoing hot bottom burning and third dredge - up ( see * ? ? ? * ) , although difficulties such as the establishment of an o - na anti - correlation remain ( see for example * ? ? ? however , as is discussed in @xcite and bcs02 , the quantities of material required to produce the observed star - to - star differences among the low luminosity stars ( most notably extreme c depletions ) , which are clearly not diluted as the convective envelopes deepen during rgb ascent , rules out any sort of simple accretion model . indeed , for the present m13 stars , roughly 70% of a c - poor ms star s total mass must be captured ejecta if the accreted matter is completely free of c ( see bcs02 ) . it is of course unclear how such an enormous amount of material can be returned to the cluster without appealing to a shallow initial mass function ( see * ? ? ? * ) , nor how the present stars can sweep up the necessary mass of ejecta . we note in figure [ fig_6 ] that the depletions in c do appear smaller in the more metal - rich clusters m71 and 47 tuc in accord with the prediction the of agb ejecta models of @xcite . yet at the same time , if one presumes the highest [ c / fe ] sgb / msto stars in m13 and m5 to represent the original ( accretion free ) c abundance of the cluster stars , they are still some 0.4 dex more c - poor than their 47 tuc / m71 counterparts , implying either truly primordial ( i.e. , pre - accretion ) differences in at least c or that nearly all the present stars in m13 and m5 have undergone at least some accretion of c - poor material . however , the spread in [ n / fe ] is essentially identical among all four clusters . clearly , knowledge of the patterns of [ o / fe ] and [ na / fe ] among the present stars would help constrain the agb ejecta theories . an interesting counterpoint to this model is the scenario suggested by @xcite and @xcite to explain similar results among m92 and m15 sgb stars that the stars of these clusters were inhomogeneously `` polluted '' by an injection of raw c from intermediate mass agb stars which is subsequently converted into n in the present stars before sgb evolution thereby explaining both the c deficiencies and large n enhancements as well as star - to - star differences in ( c+n ) . this has the additional advantage of requiring considerably more modest composition modifications ( a factor of 4 or so in c from star to star ) , which in turn lowers the mass of captured ejecta required . however , to explain the large c depletions already in place by the msto , significant processing of the envelope through a region of cn - cycling must have taken place while the stars occupied the ms . one then returns to the difficulty of mixing in such stars discussed above . another site of the proton - capture reactions has recently been suggested by @xcite , who note that the high mass ( 250 - 300 @xmath21 ) zero metallicity models of @xcite tend to mix he and he - burning products into their h - burning shells during the later stages of he - burning . this fresh c , n , o is partially processed into n while at the same time , the stars expand into red supergiants . if mass loss also occurs at this point , the cluster could be seeded with freshly produced c / o - poor , n - rich material . such a scenario is presented within the context of the cluster formation history of that the gcs formed from primordial material ( zero - metal ) that was subsequently enriched by the supernovae of massive stars before low mass stars could form . however , the problem remains that the production / seeding and mixing of the heavy - elements must be decoupled from that of the light - elements in order to explain the remarkable homogeneity of fe , ti , ca , etc . within the gcs . in the context of gc formation in a well mixed supershell ( e.g. * ? ? ? * ) this is difficult to explain if the cno - modified material is ejected prior to the driving supernovae and subsequent supershell expansion / mixing . the entire keck / lris user community owes a huge debt to jerry nelson , gerry smith , bev oke , and many other people who have worked to make the keck telescope and lris a reality . we are grateful to the w. m. keck foundation , and particularly its late president , howard keck , for the vision to fund the construction of the w. m. keck observatory . we also wish to express our thanks to roger bell whose ssg code was instrumental in this project and the anonymous referee for their suggestions . partial support was provided by the national science foundation under grant ast-0098489 to mmb and grant ast-0205951 to jgc and by the f. john barlow professorship and uw oshkosh faculty development program ( mmb ) . cccccccc 41211_2349 & 17.30 & 17.10 & 0.20 & -0.010 & -0.024 & - & - + 41217_2535 & 17.63 & 16.88 & 0.75 & 0.104 & 0.228 & -0.49 & 1.09 + 41132_2535 & 18.20 & 17.61 & 0.59 & 0.013 & 0.060 & - & 1.37 + 41185_2646 & 18.52 & 17.98 & 0.54 & 0.008 & 0.031 & - & 1.24 + 41165_2813 & 18.64 & 18.10 & 0.54 & 0.034 & -0.026 & -0.25 & -0.14 + 41204_2622 & 18.89 & 18.34 & 0.55 & 0.034 & 0.004 & -0.39 & 0.81 + 41302_2212 & 18.90 & 18.32 & 0.58 & 0.057 & - & 0.30 & - + 41228_2301 & 19.05 & 18.53 & 0.52 & 0.022 & 0.054 & - & 1.20 + 41211_2513 & 19.06 & 18.49 & 0.57 & 0.032 & -0.022 & -0.50 & 0.02 + 41157_2535 & 19.07 & 18.49 & 0.58 & 0.028 & 0.069 & -0.77 & 1.32 + 41170_2333 & 19.07 & 18.50 & 0.57 & 0.024 & 0.094 & - & 1.46 + 41191_2527 & 19.38 & 18.77 & 0.61 & 0.030 & - & -0.91 & - + 41274_2133 & 19.41 & 18.73 & 0.68 & 0.000 & - & - & - + 41265_2254 & 19.55 & 18.94 & 0.61 & 0.061 & - & -0.12 & - + 41197_2648 & 19.58 & 18.97 & 0.61 & 0.038 & 0.175 & -0.66 & 1.47 + 41171_2802 & 19.92 & 19.33 & 0.59 & 0.043 & - & -0.93 & - + 41173_2418 & 19.97 & 19.30 & 0.67 & 0.097 & - & -0.16 & - + 41188_2516 & 19.98 & 19.31 & 0.67 & 0.089 & - & -0.24 & - + 41107_2643 & 20.00 & 19.33 & 0.67 & 0.071 & - & -0.44 & - + 41334_2223 & 20.00 & 19.35 & 0.65 & 0.070 & - & -0.47 & - + 41243_2227 & 20.01 & 19.39 & 0.62 & 0.043 & - & -1.02 & - + 41122_2707 & 20.03 & 19.36 & 0.67 & 0.100 & - & -0.21 & - + 41227_2356 & 20.04 & 19.30 & 0.74 & 0.102 & - & -0.12 & - + 41120_2625 & 20.05 & 19.36 & 0.69 & 0.032 & - & -1.47 & - + 41264_2258 & 20.05 & 19.31 & 0.74 & 0.070 & - & -0.43 & - + 41296_2221 & 20.05 & 19.40 & 0.65 & 0.113 & - & -0.15 & - + 41135_2723 & 20.06 & 19.38 & 0.68 & 0.101 & - & -0.22 & - + 41277_2355 & 20.11 & 19.35 & 0.76 & 0.049 & - & -0.78 & - + 41198_2646 & 20.13 & 19.42 & 0.71 & 0.075 & - & -0.49 & - + 41219_2530 & 20.15 & 19.48 & 0.67 & 0.037 & - & -1.41 & - + 41167_2650 & 20.19 & 19.50 & 0.69 & 0.115 & - & -0.25 & - + 41276_2132 & 20.21 & 19.60 & 0.61 & 0.077 & - & -0.60 & - + 41202_2321 & 20.23 & 19.55 & 0.68 & 0.056 & - & -0.87 & - + 41210_2353 & 20.23 & 19.56 & 0.67 & 0.127 & - & -0.23 & - + 41130_2540 & 20.24 & 19.54 & 0.70 & 0.077 & - & -0.56 & - + 41340_2151 & 20.24 & 19.55 & 0.69 & 0.104 & - & -0.38 & - + 41130_2641 & 20.25 & 19.54 & 0.71 & 0.130 & - & -0.19 & - + 41131_2639 & 20.26 & 19.50 & 0.76 & 0.086 & - & -0.47 & - + 41207_2619 & 20.26 & 19.56 & 0.70 & 0.113 & - & -0.33 & - + 41243_2229 & 20.27 & 19.56 & 0.71 & 0.096 & - & -0.45 & - + 41279_2407 & 20.27 & 19.60 & 0.67 & 0.095 & - & -0.49 & - + 41191_2530 & 20.28 & 19.54 & 0.74 & 0.113 & - & -0.31 & - + 41208_2557 & 20.31 & 19.60 & 0.71 & 0.087 & - & -0.53 & - + 41255_2229 & 20.32 & 19.55 & 0.77 & 0.103 & - & -0.39 & - + 41278_2232 & 20.32 & 19.64 & 0.68 & 0.097 & - & -0.50 & - + 41301_2213 & 20.32 & 19.65 & 0.67 & 0.125 & - & -0.35 & - + 41315_2143 & 20.33 & 19.62 & 0.71 & 0.135 & - & -0.25 & - + 41270_2213 & 20.35 & 19.60 & 0.75 & 0.105 & - & -0.43 & - + 41166_2612 & 20.38 & 19.65 & 0.73 & 0.141 & - & -0.24 & - + 41183_2653 & 20.40 & 19.66 & 0.74 & 0.118 & - & -0.41 & - + 41255_2307 & 20.46 & 19.66 & 0.80 & 0.053 & - & -1.15 & - cccccc 41230_2604 & 16.83 & 16.00 & 0.83 & 0.166 & -0.34 + 41244_2423 & 16.88 & 16.05 & 0.83 & 0.182 & -0.25 + 41224_2734 & 16.92 & 16.11 & 0.81 & 0.154 & -0.40 + 41299_2630 & 16.95 & 16.12 & 0.83 & 0.212 & -0.07 + 41213_2642 & 16.99 & 16.20 & 0.79 & 0.098 & -0.72 + 41259_2821 & 17.03 & 16.23 & 0.80 & 0.182 & -0.22 + 41210_2830 & 17.07 & 16.30 & 0.77 & 0.168 & -0.29 + 41320_2941 & 17.07 & 16.31 & 0.76 & 0.118 & -0.56 + 41207_2719 & 17.09 & 16.31 & 0.78 & 0.136 & -0.47 + 41296_2957 & 17.11 & 16.34 & 0.77 & 0.201 & -0.10 + 41249_2549 & 17.30 & 16.46 & 0.84 & 0.104 & -0.63 + 41210_2834 & 17.05 & 16.48 & 0.57 & 0.128 & -0.48 + 41301_2440 & 17.32 & 16.49 & 0.83 & 0.208 & -0.03 + 41188_2619 & 17.34 & 16.57 & 0.77 & 0.081 & -0.84 + 41260_3026 & 17.35 & 16.60 & 0.75 & 0.112 & -0.55 + 41212_2744 & 17.39 & 16.60 & 0.79 & 0.165 & -0.24 + 41256_2801 & 17.43 & 16.66 & 0.77 & 0.082 & -0.79 + 41260_2850 & 17.43 & 16.69 & 0.74 & 0.127 & -0.43 + 41340_2401 & 17.54 & 16.72 & 0.82 & 0.072 & -0.92 + 41252_2524 & 17.53 & 16.73 & 0.80 & 0.149 & -0.28 + 41282_2908 & 17.49 & 16.76 & 0.73 & 0.094 & -0.62 + 41284_2922 & 17.54 & 16.81 & 0.73 & 0.113 & -0.47 + 41217_2534 & 17.63 & 16.88 & 0.75 & 0.106 & -0.47 + 41260_2459 & 17.69 & 16.88 & 0.81 & 0.116 & -0.40 + 41256_3005 & 17.67 & 16.94 & 0.73 & 0.023 & - + 41249_2548 & 18.09 & 17.03 & 1.06 & -0.032 & - + 42103_2722 & - & 17.03 & - & 0.091 & - + 41253_2521 & 17.79 & 17.07 & 0.72 & 0.068 & - + 42104_2748 & - & 17.15 & - & 0.067 & - + 42108_2808 & - & 17.18 & - & 0.039 & - + 42071_2457 & - & 17.20 & - & 0.051 & - + 42055_2321 & - & 17.23 & - & 0.053 & - + 42097_2610 & - & 17.24 & - & 0.042 & - + 42088_2635 & - & 17.33 & - & 0.046 & - + 42077_2623 & - & 17.40 & - & 0.033 & - + 41112_2621 & 18.05 & 17.44 & 0.61 & 0.012 & - + 41284_2930 & 18.02 & 17.47 & 0.55 & 0.031 & - + 42062_2223 & - & 17.51 & - & 0.030 & - + 42095_2656 & - & 17.52 & - & 0.101 & - + 42104_2854 & - & 17.55 & - & 0.009 & - + 42068_2553 & - & 17.60 & - & 0.017 & - + 42040_2211 & - & 17.63 & - & 0.048 & - + 42073_2606 & - & 17.68 & - & 0.016 & - + 42028_2442 & - & 17.71 & - & 0.006 & - + 42033_2200 & - & 17.72 & - & 0.015 & - + 42072_2737 & 18.32 & 17.74 & 0.58 & 0.035 & - + 42071_2508 & - & 17.74 & - & 0.010 & - + 42068_2541 & - & 17.76 & - & 0.011 & - + 42129_2833 & - & 17.78 & - & 0.039 & - + 42096_2708 & - & 17.79 & - & 0.047 & - + 42071_2515 & - & 17.92 & - & -0.002 & - + 41281_2911 & 18.15 & 17.96 & 0.19 & 0.001 & - + 42051_2422 & - & 17.96 & - & 0.018 & - + 42062_2405 & - & 17.96 & - & 0.044 & - + 42064_2355 & - & 17.96 & - & 0.044 & - + 42078_2253 & - & 17.99 & - & 0.014 & - + 41213_2651 & 18.54 & 18.00 & 0.54 & 0.020 & - + 42060_2650 & - & 18.04 & - & 0.013 & - + 42034_2250 & - & 18.04 & - & 0.043 & - + 42065_2305 & - & 18.05 & - & 0.029 & - + 42029_2445 & - & 18.07 & - & 0.013 & - + 42050_2430 & - & 18.10 & - & 0.028 & - + 42058_2452 & - & 18.17 & - & -0.002 & - + 41099_2615 & 18.75 & 18.19 & 0.56 & -0.006 & - + 42064_2349 & - & 18.21 & - & -0.009 & - + 41341_2405 & 18.83 & 18.22 & 0.61 & 0.004 & - + 42035_2345 & - & 18.24 & - & 0.012 & - + 42099_2820 & - & 18.36 & - & 0.035 & - + 41136_2455 & 19.01 & 18.42 & 0.59 & 0.028 & - + 41081_2630 & 19.05 & 18.46 & 0.59 & 0.039 & - + 41157_2535 & 19.07 & 18.49 & 0.58 & 0.024 & - + 41228_2301 & 19.05 & 18.53 & 0.52 & 0.020 & - + 41121_2548 & 19.12 & 18.54 & 0.58 & 0.009 & - + 41102_2643 & 19.14 & 18.54 & 0.60 & 0.008 & - + 41096_2617 & 19.14 & 18.56 & 0.58 & 0.130 & - + 41173_2525 & 19.15 & 18.56 & 0.59 & 0.040 & - + 41150_2415 & 19.16 & 18.56 & 0.60 & 0.021 & - + 41172_2423 & 19.17 & 18.58 & 0.59 & 0.046 & - + 41216_2412 & 19.18 & 18.61 & 0.57 & 0.034 & - + 41116_2614 & 19.21 & 18.62 & 0.59 & 0.023 & - + 41280_2145 & 19.21 & 18.62 & 0.59 & 0.034 & - + 41139_2533 & 19.21 & 18.62 & 0.59 & 0.041 & - + 41317_2148 & 19.21 & 18.66 & 0.55 & 0.038 & - + 41256_2223 & 19.22 & 18.66 & 0.56 & 0.040 & - + 41204_2446 & 19.27 & 18.66 & 0.61 & 0.054 & - + 41279_2334 & 19.27 & 18.67 & 0.60 & 0.044 & - + 41200_2315 & 19.28 & 18.68 & 0.60 & 0.060 & - + 41263_2211 & 19.28 & 18.69 & 0.59 & 0.021 & - + 41236_2215 & 19.37 & 18.77 & 0.60 & 0.019 & - + 41255_2324 & 19.39 & 18.79 & 0.60 & 0.023 & - + 41174_2358 & 19.39 & 18.79 & 0.60 & 0.030 & - + 41134_2417 & 19.41 & 18.80 & 0.61 & 0.017 & - + 41231_2331 & 19.42 & 18.81 & 0.61 & 0.048 & - + 41094_2540 & 19.43 & 18.82 & 0.61 & 0.050 & - + 41136_2550 & 19.45 & 18.85 & 0.60 & -0.005 & - + 41310_2206 & 19.65 & 18.88 & - & 0.030 & - + 41290_2218 & 19.49 & 18.89 & 0.60 & 0.016 & - + 41219_2401 & 19.50 & 18.96 & 0.54 & 0.010 & - + 41161_2532 & 19.80 & 19.18 & 0.62 & 0.029 & - + 41175_2421 & 19.93 & 19.25 & 0.68 & 0.085 & - + 41286_2327 & 20.13 & 19.40 & 0.73 & 0.090 & - + 41140_2549 & 20.23 & 19.52 & 0.71 & 0.093 & - + 41266_2209 & 20.43 & 19.71 & 0.72 & 0.093 & - ccccccc c51228_3737 & 17.00 & 1.38 & 0.138 & 0.131 & -0.17 & 0.32 + c51265_3739 & 17.01 & 1.31 & 0.122 & 0.141 & -0.25 & 0.49 + c51314_3755 & 17.01 & 1.34 & 0.096 & 0.340 & -0.39 & 1.41 + c51385_4166 & 17.01 & 1.40 & 0.092 & 0.378 & -0.40 & 1.54 + c51312_3634 & 17.03 & 1.35 & 0.156 & 0.388 & 0.04 & 1.19 + c51418_4158 & 17.03 & 1.36 & 0.091 & 0.347 & -0.39 & 1.48 + c51419_3870 & 17.03 & 1.38 & 0.112 & 0.333 & -0.24 & 1.29 + c51291_3655 & 17.05 & 1.33 & 0.125 & 0.296 & -0.13 & 1.11 + c51417_3943 & 17.05 & 1.41 & 0.146 & 0.124 & -0.03 & 0.25 + c51396_4020 & 17.11 & 1.33 & 0.140 & 0.327 & 0.04 & 1.15 + c51285_3749 & 17.12 & 1.29 & 0.120 & 0.142 & -0.10 & 0.55 + c51260_4161 & 17.13 & 1.40 & 0.096 & 0.351 & -0.20 & 1.49 + c51266_3848 & 17.13 & 1.36 & 0.090 & 0.280 & -0.28 & 1.34 + c51254_3957 & 17.14 & 1.42 & 0.143 & 0.143 & 0.06 & 0.43 + c51413_4033 & 17.14 & 1.27 & 0.092 & 0.236 & -0.27 & 1.18 + c51352_4055 & 17.15 & 1.31 & 0.127 & 0.115 & -0.02 & 0.30 + c51424_3823 & 17.15 & 1.31 & 0.086 & 0.279 & -0.29 & 1.38 + c51400_3529 & 17.16 & 1.31 & 0.118 & 0.105 & -0.08 & 0.26 + c51250_3763 & 17.17 & 1.39 & 0.136 & 0.107 & 0.05 & 0.18 + c51373_3631 & 17.17 & 1.31 & 0.128 & 0.102 & -0.01 & 0.17 + c51368_4074 & 17.18 & 1.31 & 0.077 & 0.253 & -0.37 & 1.39 + c51306_3738 & 17.19 & 1.32 & 0.124 & 0.113 & -0.01 & 0.31 + c51277_3950 & 17.20 & 1.34 & 0.094 & 0.255 & -0.20 & 1.25 + c51270_3931 & 17.21 & 1.40 & 0.082 & 0.285 & -0.27 & 1.43 + c51386_3659 & 17.21 & 1.28 & 0.123 & 0.103 & -0.01 & 0.21 + c51416_3834 & 17.22 & 1.31 & 0.118 & 0.124 & -0.03 & 0.45 + c51266_4149 & 17.24 & 1.34 & 0.092 & 0.273 & -0.18 & 1.32 + c51346_4124 & 17.24 & 1.39 & 0.133 & 0.340 & 0.10 & 1.26 + c51267_4025 & 17.25 & 1.35 & 0.127 & 0.121 & 0.04 & 0.38 + c51378_3975 & 17.25 & 1.34 & 0.124 & 0.106 & 0.02 & 0.25 + c51404_3918 & 17.25 & 1.35 & 0.111 & 0.109 & -0.07 & 0.35 + c51308_3765 & 17.26 & 1.27 & 0.099 & 0.283 & -0.12 & 1.31 + c51316_3960 & 17.26 & 1.29 & 0.071 & 0.276 & -0.37 & 1.53 + c51377_3737 & 17.26 & 1.25 & 0.105 & 0.122 & -0.10 & 0.51 + c51430_3648 & 17.29 & 1.34 & 0.083 & 0.234 & -0.25 & 1.27 + c51385_3962 & 17.30 & 1.24 & 0.107 & 0.113 & - & - + c51290_3644 & 17.33 & 1.36 & 0.113 & 0.109 & -0.03 & 0.36 + c51224_4027 & 17.35 & 1.31 & 0.098 & 0.161 & -0.12 & 0.84 + c51279_3957 & 17.35 & 1.36 & 0.120 & 0.139 & 0.02 & 0.57 + c51287_4050 & 17.35 & 1.38 & 0.116 & 0.135 & 0.00 & 0.55 + c51261_4172 & 17.36 & 1.32 & 0.121 & 0.108 & 0.03 & 0.30 + c51281_3638 & 17.37 & 1.28 & 0.111 & 0.113 & -0.04 & 0.41 + c51331_4042 & 17.38 & 1.28 & 0.076 & 0.229 & -0.30 & 1.34 + c51252_4126 & 17.39 & 1.39 & 0.098 & 0.134 & -0.12 & 0.68 + c51279_3842 & 17.39 & 1.28 & 0.124 & 0.082 & 0.05 & -0.14 + c51310_3849 & 17.39 & 1.29 & 0.109 & 0.140 & -0.04 & 0.64 + c51334_3732 & 17.40 & 1.29 & 0.115 & 0.112 & 0.00 & 0.37 + c51262_3874 & 17.42 & 1.33 & 0.114 & 0.103 & -0.01 & 0.29 + c51377_3766 & 17.42 & 1.29 & 0.080 & 0.231 & -0.25 & 1.31 + c51243_4217 & 17.43 & 1.35 & 0.072 & 0.211 & -0.34 & 1.31 + c51412_4143 & 17.43 & 1.27 & 0.089 & 0.170 & - & - + c51287_3658 & 17.44 & 1.30 & 0.114 & 0.099 & -0.01 & 0.23 + c51315_4161 & 17.44 & 1.30 & 0.134 & 0.458 & 0.22 & 1.60 + c51324_3542 & 17.45 & 1.38 & 0.068 & 0.236 & -0.38 & 1.45 + c51405_3749 & 17.45 & 1.24 & 0.072 & 0.234 & -0.31 & 1.39 + c51252_3923 & 17.46 & 1.29 & 0.081 & 0.253 & -0.23 & 1.38 + c51235_3931 & 17.47 & 1.34 & 0.106 & 0.106 & -0.06 & 0.37 + c51244_3757 & 17.47 & 1.35 & 0.116 & 0.126 & 0.01 & 0.50 + c51296_3969 & 17.48 & 1.25 & 0.113 & 0.086 & -0.02 & 0.02 + c51279_4119 & 17.49 & 1.36 & 0.123 & 0.105 & 0.06 & 0.25 + c51289_3768 & 17.49 & 1.23 & 0.139 & 0.261 & 0.18 & 1.01 + c51416_4137 & 17.51 & 1.39 & 0.061 & 0.207 & -0.48 & 1.44 + c51229_3628 & 17.52 & 1.28 & 0.082 & 0.240 & -0.23 & 1.33 + c51391_3723 & 17.52 & 1.26 & 0.132 & 0.163 & 0.12 & 0.65 + c51393_4120 & 17.52 & 1.27 & 0.115 & 0.135 & 0.00 & 0.58 + c51288_4040 & 17.53 & 1.30 & 0.130 & 0.140 & 0.11 & 0.52 + c51292_3964 & 17.54 & 1.42 & 0.107 & 0.117 & -0.05 & 0.48 + c51398_3758 & 17.54 & 1.24 & 0.078 & 0.226 & -0.27 & 1.31 + c51336_3569 & 17.55 & 1.23 & 0.107 & 0.110 & -0.06 & 0.41 + c51338_3624 & 17.55 & 1.28 & 0.107 & 0.137 & -0.05 & 0.64 + c51402_3627 & 17.55 & 1.30 & 0.118 & 0.101 & 0.02 & 0.23 + c51409_4045 & 17.56 & 1.28 & 0.108 & 0.105 & -0.05 & 0.35 + c51275_3873 & 17.57 & 1.32 & 0.073 & 0.227 & -0.33 & 1.37 + c51299_4161 & 17.57 & 1.28 & 0.118 & 0.084 & 0.02 & -0.05 + c51307_3821 & 17.59 & 1.28 & 0.082 & 0.232 & -0.24 & 1.30 + c51419_3843 & 17.59 & 1.29 & 0.074 & 0.239 & -0.32 & 1.41 in @xcite , we studied the ch bands in a large sample of stars in m5 . even taking into account the substantial star - to - star variation seen among the ch band strengths of the stars in our sample , we denoted six of these stars as anomalous . since that time , we have checked the data for these stars yet again . we have found that two of the six stars were misidentified . c18206_0533 with v=18.42 is actually c18188_0733 , with bvi = 17.71 , 17.03 , 16.17 . with this correction , as compared to the bulk of our m5 sample ( see figures 8 and 9 of * ? ? ? * ) the star has normal ch for its @xmath26 , although its uvcn is still anomalously strong , but not as much as previously . also , star c18211_0559 ( v=18.06 ) is actually c18191_0559 , with bvi = 18.27 , 17.57 , 16.74 . its cn is now reasonable for its corrected @xmath27 mag , but its ch index is still unexpectedly strong . in our earlier paper , we presented low accuracy radial velocities from the lris spectra at h@xmath7 which suggested that 4 of the 6 stars classified as anomalous are radial velocity members of m5 . such data was not available for one star , while the radial velocity of c18211_0559 ( now identified as c18191_0559 ) was 25 km / sec higher than that of the cluster . to verify the membership of c18191_0559 in m5 , we obtained low snr spectra with hires @xcite for it and for a second star from the lris sample . a single 1200 sec exposure for each was made on may 1 , 2002 , a night with considerable clouds . the hires slit for one of these two also included a second m5 star . the heliocentric radial velocities for these three stars derived from the nad lines are presented in table [ tbl - b1 ] . the radial velocity for m5 found by @xcite from an extensive high dispersion analysis of stars over a wide range in luminosity is + 55.0 km / sec , so we conclude all three of these stars are members . cccc c18225_0537 & 17.07 & + 58.2 & anomalous star in cohen et al . + c18191_0554 & 17.12 & + 63.1 & in lris sample , but not anomalous + c18191_0558 & 17.57 & + 58.5 & anomalous in cohen et al . as c18211_0559 iv-25/l-954 & 0.166 & 4000 & 0.15 & 2.25 & 12.09 & @xmath150.90 & + 1.22 & @xmath151.31 & @xmath151.36 & - + ii-67/l-70 & 0.165 & 3950 & 0.20 & 2.10 & 12.12 & @xmath150.79 & + 1.33 & @xmath151.32 & @xmath151.34 & - + ii-76/l-96 & 0.200 & 4350 & 1.15 & 1.85 & 12.52 & + 0.46 & + 0.59 & @xmath150.62 & @xmath150.82 & @xmath151.2 + iii-18/l-77 & 0.156 & 4350 & 1.15 & 1.85 & 12.77 & @xmath150.18 & + 1.10 & @xmath151.11 & @xmath150.97 & @xmath151.3 + k188/a1 & 0.238 & 4550 & 1.50 & 1.80 & 13.39 & + 0.45 & + 1.00 & @xmath150.32 & - & @xmath150.5 + iii-7/l-114 & 0.173 & 4600 & 1.65 & 2.00 & 13.45 & 0.00 & + 1.00 & @xmath150.83 & - & -0.9 +

the deluge of networked data motivates the development of algorithms for computation- and communication - efficient information processing . in this context , three data - adaptive censoring strategies are introduced to considerably reduce the computation and communication overhead of decentralized recursive least - squares ( d - rls ) solvers . the first relies on alternating minimization and the stochastic newton iteration to minimize a network - wide cost , which discards observations with small innovations . in the resultant algorithm , each node performs local data - adaptive censoring to reduce computations , while exchanging its local estimate with neighbors so as to consent on a network - wide solution . the communication cost is further reduced by the second strategy , which prevents a node from transmitting its local estimate to neighbors when the innovation it induces to incoming data is minimal . in the third strategy , not only transmitting , but also receiving estimates from neighbors is prohibited when data - adaptive censoring is in effect . for all strategies , a simple criterion is provided for selecting the threshold of innovation to reach a prescribed average data reduction . the novel censoring - based ( c)d - rls algorithms are proved convergent to the optimal argument in the mean - square deviation sense . numerical experiments validate the effectiveness of the proposed algorithms in reducing computation and communication overhead . decentralized estimation , networks , recursive least - squares ( rls ) , data - adaptive censoring

1612.08263

in our big data era , various networks generate massive amounts of streaming data . examples include wireless sensor networks , where a large number of inexpensive sensors cooperate to monitor , e.g. the environment @xcite , or data centers , where a group of servers collaboratively handles dynamic user requests @xcite . since a single node has limited computational resources , decentralized information processing is preferable as the networksize scales up @xcite . in this paper , we focus on a decentralized linear regression setup , and develop computation- and communication - efficient decentralized recursive least - squares ( d - rls ) algorithms . the main tool we adopt to reduce computation and communication costs is data - adaptive censoring , which leverages the redundancy present especially in big data . upon receiving an observation , nodes determine whether it is informative or not . less informative observations are discarded , while messages among neighboring nodes are exchanged only when necessary . we propose three censoring - based ( c)d - rls algorithms that can achieve estimation accuracy comparable to d - rls without censoring , while significantly reducing the computation and communication overhead . the merits of rls algorithms in solving centralized linear regression problems are well recognized @xcite . when streaming observations that depend linearly on a set of unknown parameters become available , rls yileds the least - squares parameter estimates online . rls reduces the computational burden of finding a batch estimate per iteration , and can even allow for tracking time - varying parameters . the computational cost can be further reduced by data - adaptive censoring @xcite , where less informative data are discarded . on the other hand , decentralized versions of rls without censoring have been advocated to solve linear regression tasks over networks @xcite . in d - rls , a node updates its estimate that is common to the entire network by fusing its local observations with the local estimates of its neighbors . as time evolves , all local estimates consent on the centralized rls solution . this paper builds on both @xcite and @xcite by developing censoring - based decentralized rls algorithms , thus catering to efficient online linear regression over large - scale networks . different from our in - network setting where operation is fully decentralized and nodes are only able to communicate with their neighbors , most of the existing distributed censoring algorithms apply to star topology networks that rely on a fusion center @xcite . their basic idea is that each node transmits data to the fusion center for further processing only when its local likelihood ratio exceeds a threshold @xcite ; see also @xcite where communication constraints are also taken into account . information fusion over fading channels is considered in @xcite . practical issues such as joint dependence of sensor decision rules , randomization of decision strategies and partially known distributions are reported in @xcite , while @xcite also explores quantization jointly with censoring . other than the star topology studied in the aforementioned works , @xcite investigates censoring for a tree structure . if a node s local likelihood ratio exceeds a threshold , its local data is sent to its parent node for fusion . a fully decentralized setting is considered in @xcite , where each node determines whether to transmit its local estimate to its neighbors by comparing the local estimate with the weighted average of its neighbors . nevertheless , @xcite aims at mitigating only the communication cost , while the present work also considers reduction of the computational cost across the network . furthermore , the censoring - based decentralized linear regression algorithm in @xcite deals with optimal full - complexity estimation when observations are partially known or corrupted . this is different from our context , where censoring is deliberately introduced to reduce computation and communication costs for decentralized linear regression . the present paper introduces three data - adaptive online censoring strategies for decentralized linear regression . the resultant cd - rls algorithms incur low computation and communication costs , and are thus attractive for large - scale network applications requiring decentralized solvers of linear regressions . unlike most related works that specifically target wireless sensor networks ( wsns ) , the proposed algorithms may be used in a broader context of decentralized linear regression using multiple computing platforms . of particular interest are cases where a regression dataset is not available at a single machine , but it is distributed over a network of computing agents that are interested in accurately estimating the regression coefficients in an efficient manner . in section [ sec : algo ] , we formulate the decentralized online linear regression problem ( section [ sec : algo - problem ] ) , and recast the d - rls in @xcite into a new form ( section [ sec : d - rls ] ) that prompts the development of three censoring strategies ( section [ sec : c - d - rls ] ) . section [ sec : theoretical ] develops the first censoring strategy ( section [ sec : derivation ] ) , analyzes all three censoring strategies ( section [ sec : convergence ] ) , and discusses how to set the censoring thresholds ( section [ sec : threshold ] ) . numerical experiments in section [ sec : numerical ] demonstrate the effectiveness of the novel cd - rls algorithms . _ notation . _ lower ( upper ) case boldface letters denote column vectors ( matrices ) . @xmath0 , @xmath1 , @xmath2 and @xmath3 $ ] stand for transpose , 2-norm , induced matrix 2-norm and expectation , respectively . symbols @xmath4 , @xmath5 and @xmath6 are used for the trace , minimum eigenvalue and maximal eigenvalue of matrix @xmath7 , respectively . kronecker product is denoted by @xmath8 and the uniform distribution over @xmath9 $ ] by @xmath10 , and the gaussian probability distribution function ( pdf ) with mean @xmath11 and variance @xmath12 by @xmath13 . the standardized gaussian pdf is @xmath14 , and its the associated complementary cumulative distribution function is represented by @xmath15 . this section outlines the online linear regression setup over networks , and takes a fresh look at the d - rls algorithm . three strategies are then developed using data - adaptive censoring to reduce the computation and communication costs of d - rls . consider a bidirectionally connected network with @xmath16 nodes , described by a graph @xmath17 , where @xmath18 is the set of nodes with cardinality @xmath19 , and @xmath20 denotes the set of edges . each node @xmath21 only communicates with its one - hop neighbors , collected in the set @xmath22 . the decentralized network is deployed to estimate a real vector @xmath23 . per time slot @xmath24 , node @xmath21 receives a real scalar observation @xmath25 involving the wanted @xmath26 with a regression row @xmath27 , where @xmath0 stands for transposition , so that @xmath28 , with @xmath29 . our goal is to devise efficient decentralized online algorithms to solve the following exponentially - weighted least - squares ( ewls ) problem @xmath30 ^ 2\end{aligned}\ ] ] where @xmath31 is the ewls estimate at slot @xmath32 , and @xmath33 $ ] is a forgetting factor that de - emphasizes the importance of past measurements , and thus enables tracking of a non - stationary process . when @xmath34 , boils down to a standard decentralized online least - squares estimate . the d - rls algorithm of @xcite solves as follows . per time slot @xmath32 , node @xmath21 receives @xmath25 and @xmath27 and uses them to update the per - node inverse @xmath35 covariance matrix as @xmath36 along with the per - node @xmath37 cross - covariance vector as @xmath38 using @xmath39 and @xmath40 , node @xmath21 then updates its local parameter estimate using @xmath41 \label{eq : d - rls - s}\end{aligned}\ ] ] where @xmath42 denotes the lagrange multiplier of node @xmath21 corresponding to its neighbor @xmath43 at slot @xmath44 , that captures the accumulated differences of neighboring estimates , recursively obtained as ( @xmath45 is a step - size ) @xmath46 . \label{eq : d - rls - v}\end{aligned}\ ] ] next , we develop an equivalent novel form of d - rls recursions that is convenient for our incorporation of data - adaptive censoring . detailed derivation of the equivalence can be found in appendix [ sec : eq - drls ] . the inverse covariance matrix is updated as in . however , the update of @xmath47 in is replaced by @xmath48\nonumber\\ & -\rho{\mathbf{\phi}}_j^{-1}(t){\bm{\delta}}_j(t-1)\label{eq : new - d - rls - s}\end{aligned}\ ] ] where @xmath49 stands for a lagrange multiplier conveying network - wide information that is updated as @xmath50 \nonumber \\ & - \lambda \sum_{j'\in{\mathcal{n}}_j}[{{\mathbf{s}}}_j(t-1)-{{\mathbf{s}}}_{j'}(t-1)].\label{eq : new - d - rls - delta}\end{aligned}\ ] ] observe that @xmath49 stores the weighted sum of differences between the local estimate of node @xmath21 , and all estimates of its neighbors . interestingly , if the network is disconnected and the nodes are isolated , then @xmath51 so long as @xmath52 , and the update of @xmath47 in basically boils down to the centralized rls one @xcite . that is , the current estimate is modified from its previous value using the prediction error @xmath53 , which is known as the incoming data _ innovation_. if on the other hand the network is connected , nodes can leverage estimates of their neighbors ( captured by @xmath49 ) , which provide new information from the network other than its own observations @xmath54 . the term @xmath55 can be viewed as a laplacian smoothing regularizer , which encourages all nodes of the graph to reach consensus on their estimates . * remark 1*. in d - rls , incurs computational complexity @xmath56 , since calculating the products @xmath57 and @xmath58 requires @xmath56 multiplications . similarly , incurs _ computational _ complexity @xmath56 , that is dominated by the matrix - vector multiplications @xmath59 and @xmath60 . the cost of carrying out is relatively minor . regarding _ cost per slot @xmath32 , node @xmath21 needs to transmit its local estimate @xmath47 to its neighbors and receive estimates @xmath61 from all neighbors @xmath62 . the computational burden of d - rls recursions is comparable to that of , and , with the cost of being the same as what requires . meanwhile , the original form requires neighboring nodes @xmath21 and @xmath43 to exchange @xmath63 and @xmath64 in addition to @xmath47 and @xmath61 , which doubles the communication cost relative to and . the d - rls algorithm has well documented merits for decentralized online linear regression @xcite . however , its computation and communication costs per iteration are fixed , regardless of whether observations and/or the estimates from neighboring nodes are informative or not . this fact motivates our idea of permeating benefits of data - adaptive censoring to _ decentralized _ rls , through three novel censoring - based ( c)d - rls strategies . they are different from the rls algorithms in @xcite , where the focus is on _ centralized _ online linear regression . our first censoring strategy ( cd - rls-1 ) can be intuitively motivated as follows . if a given datum @xmath65 is not informative enough , we do not have to use it since its contribution to the local estimate of node @xmath21 , as well as to those of all network nodes , is limited . with @xmath66 specifying proper thresholds to be discussed later , this intuition can be realized using a censoring indicator variable @xmath67 if the absolute value of the innovation is less than @xmath68 , then @xmath65 is censored ; otherwise @xmath65 is used . section [ sec : threshold ] will provide rules for selecting the threshold @xmath69 along with the local noise variance @xmath70 , whose computations are lightweight . if data censoring is in effect , we simply throw away the current datum by letting @xmath71 in , to obtain @xmath72 likewise , letting @xmath73 and @xmath74 in , yields @xmath75 cd - rls-1 is summarized in algorithm 1 . if censoring is in effect , computation cost per node and per slot is a fraction @xmath76 of the d - rls in and without censoring . to recognize why , observe that the scalar - matrix multiplication @xmath77 in is not necessary as the update of @xmath39 can be merged to wherever it is needed , e.g. , in and the next slot . in addition , carrying out the @xmath56 multiplications to obtain @xmath59 is no longer necessary , while the @xmath56 multiplications required to obtain @xmath60 remain the same . the first censoring strategy still requires nodes to communicate with neighbors per time slot ; hence , the communication cost remains the same . reducing this communication cost , motivates our second censoring strategy ( cd - rls-2 ) , where each node does not perform extra computations relative to cd - rls1 , but only receives neighboring estimates if its current datum is censored . the intuition behind this strategy is that if a datum is censored , then very likely the current local estimate is sufficiently accurate , and the node does not need to account for estimates from its neighbors . estimates from neighbors , are only stored for future usage . likewise , @xmath78 neighbors do not need node @xmath21 s current estimate either , because they have already received a very similar estimate . cd - rls-2 is summarized in algorithm 2 . the third censoring strategy ( cd - rls-3 ) given by algorithm 3 is more aggressive than the second one . if a node has its datum censored at a certain slot , then it neither transmits to nor receives from its neighbors , and in that sense it remains `` isolated '' from the rest of the network in this slot . apparently , we should not allow any node to be forever isolated . to this end , we can force each node to receive the local estimate from any of its neighbors at least once every @xmath79 slots , which upper bounds the delay of information exchange to @xmath79 . interestingly , the ensuing section will prove convergence of all three strategies to the optimal argument in the mean - square deviation sense under mild conditions . initialize @xmath80 , @xmath81 and @xmath82 all @xmath83 : update @xmath84 using update @xmath47 using update @xmath84 using update @xmath47 using transmit @xmath47 to and receive @xmath61 from all @xmath85 compute @xmath49 using initialize @xmath80 , @xmath81 and @xmath82 all @xmath83 : receives @xmath61 from all @xmath85 set @xmath86 as recently received ones from all @xmath85 update @xmath84 using update @xmath47 using transmit @xmath47 to and receive @xmath61 from all @xmath85 compute @xmath49 using initialize @xmath80 , @xmath81 and @xmath82 all @xmath83 : stay idle set @xmath86 as recently received ones from all @xmath85 update @xmath84 using update @xmath47 using transmit @xmath47 to and receive @xmath61 from all @xmath85 compute @xmath49 using receive @xmath61 this section starts with a criterion - based development of cd - rls-1 . convergence analysis of all three censoring strategies will follow , before developing practical means of setting the censoring threshold @xmath87 . consider the following truncated quadratic cost that is similar to the one used in the censoring - based but centralized rls @xcite @xmath88 ^ 2-\frac{1}{2}\tau^2\sigma_j(t)^2,&|x_j(t)-{\mathbf{h}}_j^t(t){{\mathbf{s}}}| > \tau\sigma_j(t ) . \nonumber \end{cases}\end{aligned}\ ] ] using to replace the quadratic loss @xmath89 ^ 2 $ ] in , our cd - rls-1 criterion is @xmath90 to solve in a decentralized manner , we introduce a local estimate @xmath91 per node @xmath21 , along with auxiliary vectors @xmath92 and @xmath93 per edge @xmath94 . by constraining all local estimates of neighbors to consent , we arrive at the following equivalent separable convex program per slot @xmath32 @xmath95 next , we employ alternating minimization and the stochastic newton iteration to derive our first censoring - based solver of . to this end , consider the lagrangian of that is given by @xmath96\end{aligned}\ ] ] where @xmath97 and @xmath98 are primal variables , while @xmath99 and @xmath100 are dual variables . consider also the augmented lagrangian of , namely @xmath101\end{aligned}\ ] ] where @xmath102 is a positive regularization scale . note that the constraints on @xmath103 are not dualized , but they are collected in the set @xmath104 . to minimize per slot @xmath105 , we rely on alternating minimization @xcite in an online manner , which entails an iterative procedure consisting of three steps . * [ s1 ] local estimate updates : * @xmath106 * [ s2 ] auxiliary variable updates : * @xmath107 * [ s3 ] multiplier updates : * @xmath108\\ { { \mathbf{u}}}_j^{j'}(t ) & = { { \mathbf{u}}}_j^{j'}(t-1)+\rho \big [ { { \mathbf{s}}}_{j'}(t)-{\tilde{\mathbf{z}}}_j^{j'}(t ) \big].\end{aligned}\ ] ] observe that [ s2 ] is a linearly constrained quadratic program , for which if @xmath109 , we always have @xmath110 therefore , the initial values of @xmath111 and @xmath112 in [ s3 ] are selected to satisfy @xmath113 ( the simplest choice is @xmath114 ) . it then holds for @xmath115 that @xmath116 using the latter to eliminate @xmath112 in [ s3 ] , we obtain @xmath117 \nonumber \\ & = { { \mathbf{v}}}_j^{j'}(t-1 ) + \frac{\rho}{2 } \big[{{\mathbf{s}}}_j(t)-{{\mathbf{s}}}_{j'}(t)\big]\end{aligned}\ ] ] where the first equality comes from subtracting the two lines in [ s3 ] , and the second equality is due to @xmath118 . the auxiliary variables @xmath93 and @xmath92 can be also eliminated . when @xmath111 is initialized by @xmath119 , summing up both sides of from @xmath120 to @xmath121 , we arrive , after telescopic cancellation , at @xmath122.\end{aligned}\ ] ] moving on to [ s1 ] , observe that it can be split into @xmath16 per - node subproblems @xmath123^t{{\mathbf{s}}}_j.\end{aligned}\ ] ] before solving with the stochastic newton iteration @xcite , eliminate @xmath111 using to obtain @xmath124^t{{\mathbf{s}}}_j\end{aligned}\ ] ] which after manipulating the double sum yields @xmath125^t{{\mathbf{s}}}_j.\end{aligned}\ ] ] if the update in is initialized with @xmath52 , summing up both sides from @xmath126 to @xmath127 , we find after telescopic cancellation @xmath128.\end{aligned}\ ] ] thus , optimization of @xmath47 reduces to @xmath129 where the instantaneous cost per slot @xmath32 is @xmath130 the stochastic gradient of the latter is given by @xmath131 + \rho { \bm{\delta}}_j(t-1).\end{aligned}\ ] ] in the stochastic newton method , the hessian matrix is given by @xmath132=e[c_j(t){\mathbf{h}}_j(t){\mathbf{h}}_j^t(t)]\ ] ] where the second equality comes from and . a reasonable approximation of the expectation is provided by sample averaging . however , presence of @xmath133 affects attenuation of regressors , which leads to @xmath134 applying the matrix inversion lemma , we obtain @xmath135 \nonumber\end{aligned}\ ] ] and after adopting a diminishing step size @xmath136 , the stochastic newton update becomes @xmath137 for rational convenience , let @xmath138 , and rewrite as ( cf . ) @xmath139 substituting @xmath140 and @xmath39 into the stochastic netwon iteration yields ( cf . ) @xmath141\\ & -\rho{\mathbf{\phi}}_j^{-1}(t){\bm{\delta}}_j(t-1)\end{aligned}\ ] ] which completes the development of cd - rls-1 . here we establish convergence of all three novel strategies for @xmath142 . with @xmath143 , the ewls estimator can even adapt to time - varying parameter vectors , but analyzing its tracking performance goes beyond the scope of this paper . for the time - invariant case ( @xmath34 ) , we will rely on the following assumption . ( * as1 * ) _ observations obey the linear model @xmath144 , where @xmath145 is independent across @xmath21 and @xmath32 . rows @xmath27 are uniformly bounded and independent of @xmath146 . covariance matrices @xmath147 \succ\mathbf{0}_{p\times p}$ ] are constant and positive definite . process @xmath148 is ergodic , while @xmath149 and @xmath150 are independent . _ we are interested in the global mean - square deviation ( gmsd ) metric @xcite , defined as @xmath151.\end{aligned}\ ] ] under ( as1 ) , convergence of cd - rls-1 and cd - rls-2 is asserted as follows ; see appendix [ sec : proof-1 ] for the proof . [ thm : convergence ] for cd - rls-1 and cd - rls-2 algorithms 1 and 2 , set @xmath152 and @xmath153 per node @xmath21 . let @xmath154 , and suppose @xmath155 for cd - rls-1 and correspondingly @xmath156 for cd - rls-2 , while @xmath157 is the network laplacian and the constant @xmath158 depends on @xmath159 , and the upper bound of @xmath160 . under ( as1 ) , there exists @xmath161 for which it holds for @xmath162 that @xmath163 \nonumber \\ \leq & \sum_{j=1}^{j}\frac{\gamma^{-1}||{{\mathbf{s}}}_j(0)-{{\mathbf{s}}}_0||^2 + \gamma t_0\sigma_j^2\mathrm{tr}({\mathbf{r}_{h_j}})}{2q(\tau)\mu t}\nonumber\\ + & \frac{\gamma \sigma_j^2\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})\mathrm{tr}({\mathbf{r}_{h_j}})\ln(t)}{4q^2(\tau)\mu t}. \label{eq : main}\end{aligned}\ ] ] theorem [ thm : convergence ] establishes that the gmsd in converges to zero at a rate @xmath164 . the constant of the convergence rate is related to @xmath165 through @xmath166 , @xmath167 and @xmath11 ; the noise covariance @xmath168 , and the threshold @xmath69 through @xmath169 . theorem [ thm : convergence ] also indicates the impact of the initial states ( determined by @xmath170 and @xmath171 ) , which disappears at a faster rate of @xmath172 . to guarantee convergence , the step size @xmath102 must be small enough . the proof for cd - rls-3 is more challenging . because a node does not receive any information from its neighbors when censoring is in effect , it has to rely on outdated neighboring estimates when the incoming datum is not censored . this delay in percolating information may cause computational instability . for this reason , we will impose an additional constraint to guarantee that all local estimates do not grow unbounded . in practice , this can be realized by truncating local estimates when they exceed a certain threshold . ( * as2 * ) _ local estimates @xmath173 are uniformly bounded @xmath174 . _ convergence of cd - rls-3 is then asserted as follows . similar to cd - rls-1 and cd - rls-2 , the gmsd of cd - rls-3 converges to zero with rate @xmath164 , as stated in the following theorem . the proof is given in appendix [ sec : proof-2 ] . [ thm : convergence3 ] for cd - rls-3 given by algorithms 3 , set @xmath152 and @xmath175 per node @xmath21 . under ( as1 ) and ( as2 ) with @xmath176 as in theorem [ thm : convergence ] , there exists @xmath177 for which it holds @xmath178 , that @xmath163 \leq \frac{a+b\ln(t)}{t } \label{eq : extra}\end{aligned}\ ] ] where @xmath179 and @xmath180 are positive constants that depend on the upper bounds of @xmath160 and @xmath47 , parameters @xmath102 and @xmath69 , the covariance @xmath181 , the laplacian matrix @xmath157 , and @xmath182 . the threshold @xmath69 influences considerably the performance of all cd - rls algorithms . its value trades off estimation accuracy for computation and communication overhead . we provide a simple criterion for setting @xmath69 using the average censoring ratio @xmath183 , which is defined as the number of censored data over the total number of data @xcite . the goal is to choose @xmath69 so that the actual censoring ratio approaches @xmath183 as @xmath32 goes to infinity since we are dealing with streaming big data , such an asymptotic property is certainly relevant . when @xmath32 is large enough , @xmath184 is very close to @xmath26 ; thus , the innovation @xmath185 . as a consequence , @xmath186 , where the last equality holds because @xmath187 . therefore , @xmath188\approx 1 - 2q(\tau)$ ] , which implies that @xmath189 if the variances @xmath190 were known , one could simply choose @xmath152 . however , @xmath191 in practice is often unknown . in this case , we consider the running average @xmath192 ^ 2 = ( t-1)\sigma_j^2(t)/t+[x_j(t+1)-{\mathbf{h}}_j^t(t+1){{\mathbf{s}}}_0]^2/t$ ] , which suggests the recursive variance estimate @xmath193 ^ 2/t\;.\ ] ] this section provides numerical results to validate the effectiveness of our novel censoring strategies . we simulate a network of @xmath194 nodes , which are uniformly randomly deployed over a @xmath195 square . two nodes within communication range @xmath196 are deemed as being neighbors . the resultant network topology is depicted in fig . [ eps : network ] . we compare five algorithms : the centralized adaptive censoring ( ac)-rls that runs in every node independently , d - rls without censoring @xcite , and the three censoring - based d - rls algorithms , namely cd - rls-1 , cd - rls-2 and cd - rls-3 . all algorithms are evaluated on two data sets , one synthetic and one real . the empirical gmsd is used as performance metric . for the synthetic data set , the unknown @xmath26 is @xmath197-dimensional with @xmath198 . the setting is the one in @xcite , where wsn - based decentralized power spectrum estimation is sought for a signal modeled as an autoregressive process . in this context , consider an auxiliary sequence @xmath199 that evolves according to @xmath200 . starting from @xmath199 , the row @xmath27 is formed by taking the next @xmath197 observations , namely @xmath201 $ ] . parameters are selected as @xmath202 , @xmath203 , and also uniformly distributed driving white noise @xmath204 with @xmath205 . observation of node @xmath21 is subject to additive white gaussian noise , with covariance @xmath206 , where @xmath207 . the true signal vector is @xmath208 , for which @xmath34 is set for all algorithms . for d - rls , cd - rls-1 , cd - rls-2 and cd - rls-3 , the step size @xmath209 and @xmath210 where @xmath211 , leading to fastest convergence of d - rls . regarding the four censoring - based algorithms ac - rls , cd - rls-1 , cd - rls-2 and cd - rls-3 , we set the average censoring ratio to @xmath212 , which is approached using @xmath213 . the variances @xmath168 are estimated in an online manner as described in section [ sec : threshold ] . ac - rls uses @xmath210 , where @xmath214 leads to the fastest convergence . for all curves obtained by running the algorithms , the ensemble averages are approximated via sample averaging over 100 monte carlo runs . [ eps : iteration ] depicts the gmsd versus the number of iterations . not surprisingly , since d - rls does not censor data , its convergence rate with respect to the number of iterations is the fastest . among the three proposed cd - rls algorithms , cd - rls-2 and cd - rls-3 are slower than cd - rls-1 , because the former two incur smaller communication cost than the latter . though cd - rls-3 adopts a more aggressive censoring strategy than cd - rls-2 , its convergence does not degrade as confirmed by fig . [ eps : iteration ] . ac - rls is the slowest among all , because it is run at all nodes independently , without sharing information over the network . the merits of censoring are further appreciated when one considers computational costs . recall that the target average censoring ratio is @xmath212 , meaning that @xmath215 of the data are discarded ( actual values are @xmath216 for ac - rls , @xmath217 for cd - rls-1 , @xmath218 for cd - rls-2 , and @xmath219 for cd - rls-3 , averaged over 100 runs ) . as confirmed by fig . [ eps : computation ] , the three cd - rls algorithms consume considerably less computational resources relative to d - rls that does not censor data . indeed , whenever a datum is censored , cd - rls-1 only requires @xmath220 of the computations relative to d - rls , while cd - rls-2 and cd - rls-3 incur minimal computational overhead . although ac - rls is the most computationally efficient algorithm at the beginning , absence of collaboration undermines its performance in steady state . regarding the amount of data exchanged to communicate local estimates in a unicast mode , cd - rls-1 is the worst because nodes need to transmit their local estimate to neighbors , no matter whether local data are censored or not . [ eps : transmission ] corroborates that cd - rls-2 and cd - rls-3 show significant improvement over d - rls , demonstrating their potential for reducing both communication and computation costs in solving decentralized linear regression problems over large - scale networks . next , we vary @xmath221 and evaluate its impact on gmsd , as shown in fig . [ eps : msd - cr ] . when @xmath221 is close to @xmath222 , meaning about @xmath223 of the data is censored , the three proposed cd - rls algorithms are still able to reach gmsd of @xmath224 , which is the limit of d - rls without censoring . among the three algorithms , cd - rls-1 exhibits the best gmsd curve , but its computation and communication costs are the highest . ac - rls does not perform well especially for low censoring ratios due to the lack of network - wide collaboration . cd - rls-2 and cd - rls-3 perform comparably in this experiment . the effectiveness of the novel censoring - based strategies is further assessed on a real data set of protein tertiary structures @xcite . the premise here is that a given dataset is not available at a single location , but it is distributed over a network whose nodes are interested in obtaining accurate regression coefficients while suppressing the communication and computational overhead . again , the graph in fig . [ eps : network ] is used to model the network of regression - performing agents . the number of control variables is @xmath225 . the first @xmath226 ( out of @xmath227 ) observations are normalized and divided evenly into @xmath194 parts , one per node . for cd - rls-1 , cd - rls-2 and cd - rls-3 , we set @xmath228 and @xmath229 , while for ac - rls we choose @xmath230 . the ground truth vector @xmath26 is estimated by solving a batch least - squares problem on the entire data set . similar to what we deduced from fig . [ eps : msd - cr ] in the synthetic data set , the novel cd - rls algorithms outperform ac - rls in terms of gmsd , as one varies the average censoring ratio from @xmath231 to nearly @xmath232 in fig . [ eps : msd - real ] . this paper introduced three data - adaptive censoring strategies that significantly reduce the computation and communication costs of the rls algorithm over large - scale networks . the basic idea behind these strategies is to avoid inefficient computation and communication when the local observations and/or the neighboring messages are not informative . we proved convergence of the resulting algorithms in the mean - square deviation sense . numerical experiments validated the merits of the novel schemes . the notion of identifying and discarding less informative observations can be widely used in various large - scale online machine learning tasks including nonlinear regression , matrix completion , clustering and classification , to name a few . these constitute our future research directions . here we prove that d - rls recursions - are equivalent to , and . it follows from that @xmath233 \nonumber \\ & -\lambda \big[{\bm{\psi}}_j(t-1)-\frac{1}{2}\sum_{j'\in { \mathcal{n}}_j}({{\mathbf{v}}}_j^{j'}(t-2)-{{\mathbf{v}}}_{j'}^{j}(t-2 ) ) \big ] . \nonumber\end{aligned}\ ] ] applying the matrix inversion lemma to yields @xmath234 substituting @xmath235 from and @xmath236 from into , leads to @xmath237 = { \mathbf{h}}_j(t)\big[x_j(t)-{\mathbf{h}}_j^t(t){{\mathbf{s}}}_j(t-1 ) \big ] \nonumber \\ & \hspace{6em } - \frac{1}{2}\sum_{j'\in { \mathcal{n}}_j}({{\mathbf{v}}}_j^{j'}(t-1 ) - \lambda { { \mathbf{v}}}_j^{j'}(t-2 ) ) \nonumber \\ & \hspace{6em } + \frac{1}{2}\sum_{j'\in { \mathcal{n}}_j}({{\mathbf{v}}}_{j'}^{j}(t-1 ) - \lambda{{\mathbf{v}}}_{j'}^{j}(t-2)).\end{aligned}\ ] ] next , we will show that if @xmath238 is defined as @xmath239 then its update is exactly . this can be done by taking the difference between slots @xmath32 and @xmath44 for , and substituting the update of @xmath111 in . due to , it follows that is equivalent to @xmath240 = { \mathbf{h}}_j(t)\big[x_j(t)-{\mathbf{h}}_j^t(t){{\mathbf{s}}}_j(t-1 ) \big ] \nonumber \\ & \hspace{10.3em } - \rho { \bm{\delta}}(t-1).\end{aligned}\ ] ] left multiplying with @xmath39 , yields the update of @xmath91 in , and completes the proof . starting with cd - rls-1 , the proof proceeds in five stages . * stage 1 . * we first investigate the spectral properties of @xmath241 when @xmath32 is sufficiently large . letting @xmath34 and applying the matrix inversion lemma to the censoring form , we have @xmath242 summing up from @xmath120 to @xmath121 and using the telescopic cancellation , yields @xmath243 thanks to the strong law of large numbers , @xmath244 converges to @xmath245 $ ] almost surely as @xmath246 . clearly , @xmath245 \preceq e[{\mathbf{h}}_j(t){\mathbf{h}}_j^t(t ) ] = { \mathbf{r}_{h_j}}$ ] . on the other hand , given @xmath160 , @xmath247 and @xmath248 defined in , the probability of @xmath249 is @xmath250}^{\tau+\sigma_j^{-1}[{\mathbf{h}}_j^t(t)({{\mathbf{s}}}_j(t-1)-{{\mathbf{s}}}_0)]}\phi(x)dx \nonumber \\ \geq & 1-\int_{-\tau}^{\tau}\phi(x)dx = 2q(\tau).\end{aligned}\ ] ] consequently , it holds that @xmath251 \\ = & e\big[{\mathbf{h}}_j(t){\mathbf{h}}_j^t(t)e[c_j(t)|{\mathbf{h}}_j(t),{{\mathbf{s}}}_j(t-1)]\big]\\ = & e[{\mathbf{h}}_j(t){\mathbf{h}}_j^t(t ) \pr(c_j(t)=1|{\mathbf{h}}_j(t),{{\mathbf{s}}}_j(t-1))]\\ \succeq & 2q(\tau ) e[{\mathbf{h}}_j(t){\mathbf{h}}_j^t(t ) ] = 2q(\tau){\mathbf{r}_{h_j}}.\end{aligned}\ ] ] thus , there exists @xmath252,for which it holds @xmath178 that @xmath253 as @xmath32 grows , the eigenvalues of @xmath241 are thus within the interval @xmath254 $ ] . consequently , the eigenvalues of @xmath255 are within the interval @xmath256 $ ] . * stage 2 . * rewrite the update of @xmath91 as @xmath257 \\ & -\rho{\mathbf{\phi}}_j^{-1}(t){\bm{\delta}}_j(t-1).\end{aligned}\ ] ] note also that for @xmath34 , the update of @xmath258 is equivalent to ( cf . ) @xmath259.\end{aligned}\ ] ] letting @xmath260 , the estimation error obeys the recursion @xmath261 \\ & -\rho{\mathbf{\phi}}_j^{-1}(t)\sum_{j'\in{\mathcal{n}}_j } \big [ { \mathbf{e}}_j(t-1)-{\mathbf{e}}_{j'}(t-1 ) \big].\end{aligned}\ ] ] substituting @xmath144 to eliminate @xmath247 , we obtain @xmath262.\end{aligned}\ ] ] left multiplying with @xmath241 yields @xmath263 \nonumber \\ = & { \mathbf{\phi}}_j(t-1){\mathbf{e}}_j(t-1 ) \nonumber\\ + & c_j(t){\mathbf{h}}_j(t)\epsilon_j(t)-\rho\sum_{j'\in{\mathcal{n}}_j } \big [ { \mathbf{e}}_j(t-1)-{\mathbf{e}}_{j'}(t-1 ) \big ] . \nonumber\end{aligned}\ ] ] our convergence analysis result will rely on a matrix form of that accounts for all nodes @xmath21 . define vectors @xmath264^t \in \mathbb{r}^{jp}$ ] , @xmath265^t \in \mathbb{r}^j$ ] , as well as diagonal matrices @xmath266 , @xmath267 , and @xmath268 . then can be written in matrix form as @xmath269{\mathbf{e}}(t-1)+{\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t)\end{aligned}\ ] ] which after left multiplication with @xmath270 yields @xmath271{\mathbf{e}}(t-1)\nonumber \\ + & { \mathbf{\phi}}(t)^{-\frac{1}{2}}{\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t).\end{aligned}\ ] ] from , we have ( @xmath8 denotes kronecker product ) @xmath272 \nonumber\\ = & e[{\mathbf{e}}^t(t-1)({\mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p)^t{\mathbf{\phi}}^{-1}(t)\nonumber\\ & \hspace{1em } \times ( { \mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p){\mathbf{e}}(t-1)]\nonumber\\ + & 2e[{\mathbf{e}}^t(t-1)({\mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p)^t{\mathbf{\phi}}^{-1}(t ) { \mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t)]\nonumber\\ + & e[{\bm{\epsilon}}^t(t){\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t)].\end{aligned}\ ] ] since @xmath273 and @xmath274 are independent under ( as1 ) , the second term on the right hand side is zero ; hence , @xmath275 \nonumber\\ = & e[{\mathbf{e}}^t(t-1)({\mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p)^t{\mathbf{\phi}}^{-1}(t)\nonumber\\ \times & ( { \mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p){\mathbf{e}}(t-1)]\nonumber\\ + & e[{\bm{\epsilon}}^t(t){\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t)].\end{aligned}\ ] ] * stage 3 . * consider the first term on the right hand side of . since @xmath157 is positive semi - definite , we can find a matrix @xmath276 such that @xmath277 . by the matrix inversion lemma , it holds that @xmath278 for @xmath142 , it follows from that @xmath279 . since @xmath280 , it holds that @xmath281 for all @xmath282 , and consequently @xmath283 if @xmath284 , then for all @xmath285 it follows that @xmath286 this implies that the second term of is positive definite . thus , we have @xmath287 and hence , the first term on the right hand side of is bounded by @xmath288\nonumber\\ \leq & e[{\mathbf{e}}^t(t-1)({\mathbf{\phi}}(t-1)-\rho{\mathbf{l}}\otimes\mathbf{i}_p)^t { \mathbf{e}}(t-1)]\nonumber\\ \leq & e[{\mathbf{e}}^t(t-1 ) { \mathbf{\phi}}(t-1)^t { \mathbf{e}}(t-1)].\end{aligned}\ ] ] * stage 4 . * now consider the second term on the right hand side of . manipulating the expectation yields @xmath289\\ = & e[\mathrm{tr}({\bm{\epsilon}}^t(t){\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t))]\\ = & e[\mathrm{tr}({\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t){\bm{\epsilon}}^t(t))]\\ = & e[\mathrm{tr}({\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t)\mathrm{diag}(\{\sigma_j^2\})].\end{aligned}\ ] ] where @xmath290 is a diagonal matrix constructed with @xmath291 on its diagonal . expanding the matrix multiplications and noting that @xmath292 , we obtain @xmath293 \nonumber \\ \leq & \sum_{j=1}^{j}\sigma_j^2 e[{\mathbf{h}}_j^t(t){\mathbf{\phi}}_j^{-1}(t){\mathbf{h}}_j(t)].\end{aligned}\ ] ] since the eigenvalues of @xmath255 are no larger than @xmath294 @xmath162 , it holds that @xmath295 \leq \frac{\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})}{2q(\tau)t } e[{\mathbf{h}}_j^t(t){\mathbf{h}}_j(t)].\end{aligned}\ ] ] which after using that @xmath296 = \mathrm{tr}(e[{\mathbf{h}}_j(t){\mathbf{h}}_j^t(t)])$ ] @xmath297 , allows one to deduce from that @xmath298 \nonumber \\ \leq & \frac{1}{2q(\tau)t}\sum_{j=1}^{j}\sigma_j^2\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})\mathrm{tr}({\mathbf{r}_{h_j}})\end{aligned}\ ] ] holds @xmath178 . for @xmath299 , we have @xmath300 because to , and thus @xmath301 \\ \leq & \sum_{j=1}^{j}\gamma \sigma_j^2 e[{\mathbf{h}}_j^t(t){\mathbf{h}}_j(t)]= \gamma\sum_{j=1}^{j}\sigma_j^2 tr({\mathbf{r}_{h_j}}).\end{aligned}\ ] ] therefore , for @xmath302 yields @xmath303 \nonumber \\ \leq & \gamma\sum_{j=1}^{j}\sigma_j^2 \mathrm{tr}({\mathbf{r}_{h_j}}).\end{aligned}\ ] ] * stage 5 . * substituting , and into implies for @xmath304 that @xmath305\nonumber\\ \leq & e[{\mathbf{e}}^t(t-1){\mathbf{\phi}}(t-1){\mathbf{e}}(t-1)]\nonumber\\ + & \frac{1}{2q(\tau)t}\sum_{j=1}^{j}\sigma_j^2\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})\mathrm{tr}({\mathbf{r}_{h_j}})\end{aligned}\ ] ] while for @xmath302 @xmath306\nonumber\\ \leq & e[{\mathbf{e}}^t(t-1){\mathbf{\phi}}(t-1){\mathbf{e}}(t-1 ) ] + \gamma \sum_{j=1}^{j}\sigma_j^2 \mathrm{tr}({\mathbf{r}_{h_j}}).\end{aligned}\ ] ] summing from @xmath307 to @xmath121 and from @xmath120 to @xmath308 , applying telescopic cancellation , and noticing that @xmath309 , yields for @xmath304 @xmath272 \\ \leq & \gamma^{-1}||{\mathbf{e}}(0)||^2 + ( \gamma t_0 + \sum_{r = t_0}^{t}\frac{\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})}{2q(\tau)t})\sum_{j=1}^{j}\sigma_j^2\mathrm{tr}({\mathbf{r}_{h_j}})\\ \leq & \gamma^{-1}||{\mathbf{e}}(0)||^2+(\gamma t_0+\frac{\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})}{2q(\tau)}\ln(t))\sum_{j=1}^{j}\sigma_j^2\mathrm{tr}({\mathbf{r}_{h_j}}).\end{aligned}\ ] ] using @xmath310 in and the definition @xmath311 , establishes that @xmath312\leq e[{\mathbf{e}}^t(t){\mathbf{\phi}}(t){\mathbf{e}}(t ) ] , \quad t \geq t_0\end{aligned}\ ] ] which in turn implies that @xmath313 \\ \leq & \gamma^{-1}||{\mathbf{e}}(0)||^2+(\gamma t_0 + \frac{\lambda_{\max}({\mathbf{r}_{h_j}}^{-1})}{2q(\tau)}\ln(t))\sum_{j=1}^{j}\sigma_j^2\mathrm{tr}({\mathbf{r}_{h_j } } ) . \nonumber\end{aligned}\ ] ] finally , with @xmath314 this leads to , which completes the proof of cd - rls-1 . consider next cd - rls-2 . stage 1 of the proof remains the same , while for stage 2 , @xmath315 is replaced by @xmath316 $ ] in to arrive at @xmath317 . \nonumber\end{aligned}\ ] ] its matrix form can be expressed as @xmath318 \nonumber\\ = & e[{\mathbf{e}}^t(t-1)({\mathbf{\phi}}(t-1)-\rho({\mathbf{c}}(t){\mathbf{l}})\otimes\mathbf{i}_p)^t{\mathbf{\phi}}^{-1}(t)\nonumber\\ & \hspace{1em } \times ( { \mathbf{\phi}}(t-1)-\rho({\mathbf{c}}(t){\mathbf{l}})\otimes\mathbf{i}_p){\mathbf{e}}(t-1)]\nonumber\\ + & e[{\bm{\epsilon}}^t(t){\mathbf{c}}^t(t){\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t){\mathbf{h}}(t){\mathbf{c}}(t){\bm{\epsilon}}(t)].\end{aligned}\ ] ] observe that the right hand sides of and are only different in their first terms . similar to stage 3 ( cf . ) , we need to show that the first term satisfies @xmath319\nonumber\\ \leq & e[{\mathbf{e}}^t(t-1){\mathbf{\phi}}(t-1){\mathbf{e}}(t-1)].\end{aligned}\ ] ] substituting the update with @xmath142 into , it suffices to prove that @xmath320\nonumber\\ \geq & \rho e[{\mathbf{e}}^t(t-1 ) \mathbf{w } { \mathbf{e}}(t-1 ) ] \nonumber\end{aligned}\ ] ] where @xmath321 for the left hand side of , use the lower bound of the conditional expectation @xmath322 $ ] to eliminate @xmath273 , and arrive at @xmath323 \nonumber \\ \geq & 2q(\tau ) e[{\mathbf{e}}^t(t-1 ) { \mathbf{h}}(t){\mathbf{h}}^t(t ) \nonumber \\ & \hspace{1em } \times ( \mathbf{i}_j+{\mathbf{h}}^t(t){\mathbf{\phi}}^{-1}(t-1){\mathbf{h}}(t))^{-1}\otimes\mathbf{i}_p{\mathbf{e}}(t-1 ) ] . \nonumber\end{aligned}\ ] ] by , it holds that @xmath324 , and thus @xmath325^{-1 } \succeq \left[\mathbf{i}_j+ \gamma { \mathbf{h}}^t(t ) { \mathbf{h}}(t)\right]^{-1}.\end{aligned}\ ] ] by assumption @xmath326 are uniformly bounded . if @xmath327 for all @xmath328 , we find @xmath329^{-1 } \succeq \frac{1}{1+\gamma k^2 } \mathbf{i}_j.\end{aligned}\ ] ] substituting into , we obtain a lower bound for the left hand side of given by @xmath330 \nonumber \\ \geq & \frac{2q(\tau)}{1+\gamma k^2}e[{\mathbf{e}}^t(t-1){\mathbf{h}}(t){\mathbf{h}}^t(t){\mathbf{e}}(t-1 ) ] \nonumber \\ = & \frac{2q(\tau)}{1+\gamma k^2}e[{\mathbf{e}}^t(t-1)\text{diag}\{{\mathbf{r}_{h_j}}\}{\mathbf{e}}(t-1 ) ] \nonumber \\ \geq & \frac{2q(\tau)\mu}{1+\gamma k^2}e[||{\mathbf{e}}(t-1)||^2 ] . \nonumber\end{aligned}\ ] ] as for the right hand side of , it is upper bounded by @xmath331 \\ \leq & \rho e[(2||\mathbf{w}_1||_2+||\mathbf{w}_2||_2 + 2||{\mathbf{l}}||_2 \nonumber \\ & \hspace{1em } + \rho||{\mathbf{l}}||_2 ^ 2||{\mathbf{\phi}}^{-1}(t-1)||_2)||{\mathbf{e}}(t-1)||^2 ] \nonumber\end{aligned}\ ] ] where we used that all the diagonal elements @xmath248 of @xmath273 are within the range @xmath332 $ ] while @xmath333 is upper bounded by @xmath334 , @xmath335 by assumption , @xmath336 , @xmath337 and @xmath338 , we find that @xmath339 similarly , @xmath340 is upper bounded by @xmath341 therefore , reduces to @xmath342 \\ \leq & \rho(2\gamma\lambda_{\max}({\mathbf{l}})k^2+\gamma^2\lambda_{\max}({\mathbf{l}})^2k^2 + 2\lambda_{\max}({\mathbf{l } } ) \nonumber \\ + & \rho \gamma \lambda_{\max}({\mathbf{l}})^2)e[||{\mathbf{e}}(t-1)||^2 ] . \nonumber\end{aligned}\ ] ] considering a positive constant @xmath343 and combining with , we see that if @xmath102 is chosen within @xmath344 $ ] , then holds for all @xmath282 ; and so does . the update of @xmath347 for cd - rls-3 is ( cf . for cd - rls-1 ) @xmath348 per time @xmath32 , @xmath349 is the latest time slot when node @xmath21 received information from its neighbor @xmath43 . therefore , @xmath350 can be viewed as network delay caused by the censoring strategy . then we have @xmath351||\\ \leq & ||{\mathbf{\phi}}_j^{-1}(t)||_2 \big[||{\mathbf{h}}_j(t)||^2||{\mathbf{e}}_j(t-1)||+||{\mathbf{h}}_j(t)|||\epsilon_j(t)|\\ + & \rho\sum_{j'\in{\mathcal{n}}_j}(||{\mathbf{e}}_j(t-1)||+||{\mathbf{e}}_{j'}(t - d_j^{j'}(t))||)\big].\end{aligned}\ ] ] in deriving the inequality we use the fact that @xmath352 . according to in the proof of theorem [ thm : convergence ] , which also holds true for cd - rls-3 , there exists @xmath252 , such that @xmath310 for all @xmath162 and @xmath353 . thus , @xmath354 is upper bounded by @xmath355 where @xmath356 is a positive constant determined by @xmath169 and the smallest eigenvalue of @xmath241 . by ( as1 ) and ( as2 ) , @xmath357 , @xmath358 and @xmath359 are also upper bounded . therefore , there exist constants @xmath360 , such that @xmath361.\end{aligned}\ ] ] taking expectations on both sides yields . rewrite the update of @xmath347 for cd - rls-3 in to @xmath362 multiplying @xmath241 on both sides , we have @xmath363 using the same notations as in the proof of theorem [ thm : convergence ] , we obtain an matrix form @xmath364 where @xmath365 and its @xmath21th block is @xmath366 . observe that @xmath367 contains the differences between the local estimates and their delayed values , and hence plays a critical role in the convergence proof . below we look for an upper bound for @xmath368 $ ] . by the cauchy - schwarz inequality , we have latexmath:[\ ] ] observing that @xmath398 $ ] is bounded because @xmath399 is bounded by ( as2 ) , the right hand side of is in the order of @xmath400 . following the argument in step 5 of the proof of theorem [ thm : convergence ] , @xmath401 is in the order of @xmath389 when @xmath162 . therefore , @xmath402 $ ] is in the order of @xmath403 , which completes the proof of theorem [ thm : convergence3 ] . r. arroyo - valles , s. maleki , and g. leus , `` a censoring strategy for decentralized estimation in energy - constrained adaptive diffusion networks , '' _ proc . of intl . work . on signal processing advances in wireless communications _ , germany , june 2013 . d. berberidis , v. kekatos , and g. b. giannakis , `` online censoring for large - scale regressions with application to streaming big data , '' _ ieee transactions on signal processing _ , vol . 64 , pp . 38543867 , aug . v. cevher , s. becker , and m. schmidt , `` convex optimization for big data : scalable , randomized , and parallel algorithms for big data analytics , '' _ ieee signal processing magazine _ , vol . 31 , pp . 3243 , sept . g. b. giannakis , q. ling , g. mateos , i. d. schizas , and h. zhu , `` decentralized learning for wireless communications and networking , '' _ splitting methods in communication and imaging , science and engineering _ , r. glowinski , s. osher , and w. yin ( eds . ) , springer , 2016 . r. jiang , y. lin , b. chen , and b. suter , `` distributed sensor censoring for detection in sensor networks under communication constraints , '' _ proc . of asilomar conf . on signals , systems and computers _ , pacific grove , ca , nov . 2005 . g. mateos , i. d. schizas and g. b. giannakis , `` distributed recursive least - squares for consensus - based in - network adaptive estimation , '' _ ieee transactions on signal processing _ , vol . 57 , pp . 45834588 , nov . 2009 . c. rago , p. willett , and y. bar - shalom , `` censoring sensors : a low - communication - rate scheme for distributed detection , '' _ ieee transactions on aerospace and electronic systems _ , vol . 554568 , apr . m. a. sharkh , m. jammal , a. shami , and a. ouda , `` resource allocation in a network - based cloud computing environment : design challenges , '' _ ieee communications magazine _ , vol . 51 , pp . 4652 , nov . k. slavakis , s. j. kim , g. mateos , and g. b. giannakis , `` stochastic approximation vis - a - vis online learning for big data analytics , '' _ ieee signal processing magazine _ , vol . 124129 , nov . 2014 . z. wang , z. yu , q. ling , d. berberidis , and g. b. giannakis , `` distributed recursive least - squares with data - adaptive censoring , '' _ proc . conf . on acoustics , speech , and signal processing _ , new orleans , march 2017 .

we consider the question when the so called _ spectral condition _ for hermite subdivision schemes extends to spaces generated by polynomials and exponential functions . the main tool are convolution operators that annihilate the space in question which apparently is a general concept in the study of various types of subdivision operators . based on these annihilators , we characterize the spectral condition in terms of factorization of the subdivision operator . * keywords * subdivision schemes ; hermite schemes ; factorization ; annihilators * msc * 65d15 ; 41a05 ; 42c15

1312.1776

subdivision schemes are efficient iterative procedures based on the repeated application of _ subdivision operators _ which might differ at different levels of iteration . whenever convergent , they generate functions that hopefully resemble the data used to start the iterative procedure . subdivision operators act on bi - infinite sequences @xmath0 by means of a _ finitely _ supported _ mask _ @xmath1 in the convolution like form @xmath2 this type of operators has been generalized in various ways , considering multivariate operators , operators with dilation factors other than @xmath3 or subdivision operators acting on vector or matrix data by means of matrix valued masks . there is such a vast amount of literature meanwhile that we do not even attempt to give references here . it has been observed from very early on that preservation of _ polynomial data _ is an important property of subdivision operators . for example , the preservation of constants , @xmath4 , is necessary for the convergence of the subdivision schemes which iterate the same operator @xmath5 . more generally , the preservation of polynomial spaces , @xmath6 , plays an important role in the investigation of the differentiability of the limit function of subdivision schemes . in addition , there has been interest in also preserving functions other than polynomials , see for example @xcite , and it is natural that such functions must be exponential , i.e. , of the form @xmath7 , cf . @xcite . in this paper we will consider preservation of such exponentials by _ hermite subdivision operators _ which act on vector data but with the particular understanding that these vectors represent function values and consecutive derivatives up to a certain order . we will study the preservation capability of such operators by means of a _ cancellation operator _ , a concept that applies to subdivision schemes in quite some generality . this is why , before we get to the main technical content of the paper , we want to illustrate the idea and the concept through a few examples . the simplest example deals with the preservation of constants , @xmath8 . note that constant sequences are exactly the kernel of the difference operator @xmath9 , defined as @xmath10 ; in other words : the difference operator is the simplest _ cancellation operator _ or _ annihilator _ of the constant functions . now , whenever @xmath5 _ preserves _ constants , then @xmath11 is a subdivision operator that _ annihilates _ the constants . as it can easily be shown , any such operator can be written as @xmath12 for some other finitely supported mask @xmath13 , hence we get the factorization @xmath14 . switching to the calculus of _ symbols _ which associates to a finitely supported sequence @xmath15 the _ laurent polynomial _ @xmath16 the factorization is equivalent to @xmath17 or , equivalently , to the famous `` zero at @xmath18 '' condition @xmath19 . for a slightly more sophisticated example , suppose that now the subdivision operator provides preservation of the subspace @xmath20 in the sense that @xmath21 where @xmath22 , see , for example , @xcite . again we approach this problem in terms of cancellation , therefore determining an operator @xmath23 such that @xmath24 . assuming that @xmath23 is a _ convolution operator _ ( or _ lti filter _ in the language of signal processing , cf . @xcite ) with _ impulse response _ @xmath25 , it is easily seen and well known that cancellation of the polynomials of degree at most @xmath26 implies that @xmath27 , @xmath28 , hence cancellation of the polynomial part of @xmath29 implies that @xmath30 . cancellation of an exponential sequence @xmath7 , on the other hand , leads to @xmath31 hence , the annihilation of the space implies that @xmath32 summarizing , the simplest cancellation operator for @xmath29 takes the form @xmath33 and the associated factorization by means of cancellation operators @xmath34 is easily verified to be equivalent to the symbol factorization @xmath35 given in @xcite . note that in ( [ eq : hfactscalar ] ) one really has to consider different spaces , hence a preservation property of the form @xmath21 because the result of the subdivision operator corresponds to a sequence on the grid @xmath36 . the last example considers _ hermite subdivision schemes _ which we will investigate in more detail in the rest of this paper . in hermite subdivision , the data are vector valued sequences @xmath37 with the intuition that the @xmath38th component of such a sequence represents a @xmath38th derivative . then , as considered for example in @xcite , one defines , for @xmath39 , a sequence @xmath40 , \qquad \alpha \in { \mathbb{z}},\ ] ] and asks when a subdivision operator @xmath41 with _ matrix valued _ masks @xmath42 annihilates all @xmath43 for @xmath44 which , by the aforementioned machinery , can again be used to describe the _ spectral condition _ , a `` polynomial preservation '' rule introduced by dubuc and merrien in @xcite . note that it is no mistake or accident that the letter @xmath45 appears for the maximal order of derivatives and the maximal degree of polynomial cancellation the space dimension and the order of derivatives are closely tied . it was then shown in @xcite that whenever @xmath46 for all @xmath47 , then there exist a finitely supported @xmath48 such that @xmath49.\ ] ] since the operator @xmath50 acts for @xmath51 , @xmath52 , as @xmath53 hence measures the difference between a function and its taylor polynomial approximation at the neighboring point , it is called the _ ( complete ) taylor operator _ of order @xmath45 . that @xmath50 annihilates all @xmath43 , @xmath54 , is immediate from ( [ eq : taylorformula ] ) . it should have become clear by now that there is an obvious common structure behind all these examples . _ preservation _ of a subspace that can be written as the kernel of a convolution operator is related to a commuting property provided that the convolution operator factorize - s or `` divides '' _ any _ annihilator of the subspace . this can be seen as a minimality property with respect to the partial ordering given by divisibility and justifies the following terminology where we identify any function @xmath55 with the sequence @xmath56 . a linear operator @xmath57 is called a _ convolution operator _ for a space @xmath58 if there exists a matrix sequence @xmath59 , called the _ impulse response _ of @xmath60 , such that @xmath61 a convolution operator @xmath57 is called a _ minimal annihilator _ for a space @xmath58 with respect to 1 . _ subdivision _ , if for any @xmath62 such that @xmath63 there exists @xmath64 with @xmath65 . _ convolution _ , if for any @xmath62 such that @xmath66 there exists @xmath64 with @xmath67 , respectively . if @xmath60 satisfies both properties it is simply called a _ minimal annihilator_. the goal of this paper is to use this general concept to understand preservation of exponentials and polynomials by hermite subdivision schemes where the subdivision operators will have to vary with the iteration level ; some call this _ nonstationary _ , some _ nonuniform _ operators , but the problem is too interesting to dwell on such niceties here and therefore we omit it as the name of a property that is not fulfilled anyway is simply irrelevant . in more technical terms , we will derive the analogy of the taylor operator for the case of preservation of exponentials and prove in theorem [ thm : convfactorization ] that it is again a minimal annihilator . we will see that even the cancellation operator depends only on the space @xmath29 and on the level . we will also see that the existence of the annihilator operator is strongly connected with the factorization of the subdivision operator satisfying specific preservation properties . the organization of the paper is as follows . after providing the necessary notation and terminology , the main results on hermite subdivision schemes and their reproduction capabilities will be derived in section [ sec : hermite ] . to better explain the underlying ideas , we will first consider the case of adding a single frequency to the polynomial space and then extend the results and methods to an arbitrary number of frequencies . these descriptions will be in terms of appropriate cancellation operators . thereafter , in section [ sec : factorization ] we will use these cancellation operators to derive factorization properties which will also verify that the cancellation operators are minimal . finally , we will illustrate our results with specific examples . we begin by fixing the notation and recalling some known facts about subdivision schemes . we denote by @xmath68 and @xmath69 the linear spaces of all sequences of @xmath70vectors and @xmath71 matrices , respectively . operators acting on that spaces are denoted by capital calligrafic letter . sequences in @xmath68 and @xmath72 will be denoted by boldface lower case and upper case letters , respectively . in particular , @xmath73 is @xmath74 , while @xmath75 stands for @xmath76 , indexing @xmath77 as @xmath78 $ ] . as usual , @xmath79 and @xmath80 will denote the subspaces of finitely supported sequences , and @xmath81 denotes the set @xmath82 . for @xmath83 and @xmath84 we define the associated _ symbols _ as the _ laurent polynomials _ @xmath85 for @xmath86 and @xmath87 the _ convolution _ @xmath88 in @xmath89 is defined as usually as @xmath90 the _ subdivision operator _ @xmath91 based on the matrix sequence or _ mask _ @xmath92 is defined as @xmath93 alternatively , using symbol calculus notation , we can also decribe the action of the subdivision operator in the form @xmath94 though , in strict formalism , ( [ lab1 ] ) is only valid for @xmath95 . a _ subdivision scheme _ consists of the successive application of potentially different subdivision operators @xmath96}}$ ] , constructed from a sequence of masks @xmath97 } \;:\ ; n \in { \mathbb{n}}_0 \right)$ ] where @xmath98}=\left ( a^{[n ] } ( \alpha ) \;:\ ; \alpha \in { \mathbb{z}}\right ) \in \ell_{00}^{m \times m}({\mathbb{z}})$ ] is called the _ level @xmath99 subdivision mask _ and is assumed to be of finite support . accordingly , a sequence of matrix _ symbols _ @xmath100})^*(z ) \;:\ ; n \in { \mathbb{n}}_0 \right)$ ] characterizes such schemes . for some initial sequence @xmath101 } \in \ell^m ( { \mathbb{z}})$ ] the subdivision scheme iteratively produces sequences @xmath102 } : = { { \cal s}}_{{{\boldsymbol{a}}}^{[n]}}{{\boldsymbol{c}}}^{[n ] } , \qquad n \in { \mathbb{n}}_0,\ ] ] whose elements can be interpreted as function values at @xmath103 , from which one can define convergence in the usual way . as already mentioned , hermite subdivision schemes act on vector valued data @xmath104 , whose @xmath38-th component corresponds to a @xmath38th derivative . we are interested in studying the exponential and polynomial preservation capabilities of such kind of schemes . a preliminary simple observation is that for @xmath105 and for @xmath106 we clearly have @xmath107 , @xmath108 . hence @xmath109= d^n \left [ f^{(j ) } ( 2^{-n}\cdots):\ , j=0,\dots , d\right ] , \ ] ] where @xmath110.\ ] ] since the sequence @xmath111}$ ] is related to evaluations on the grid @xmath112 , we consider hermite subdivision schemes with the @xmath99-th iteration of the following type : @xmath113}=\sum_{\beta\in { \mathbb{z } } } a^{[n]}(\cdot -2\beta ) d^n c^{[n]}(\beta),\ ] ] where in `` usual '' hermite subdivision the mask is the same over all levels , i.e. , @xmath98 } = { { \boldsymbol{a}}}$ ] , @xmath114 . setting @xmath115 } : = d^{-n-1 } { { \boldsymbol{a}}}^{[n ] } d^n$ ] , ( [ eq : hermsubd0 ] ) fits into the framework of section 2 with the @xmath99-th subdivision operator of the form @xmath116 } = s_{\widetilde a^{[n ] } } { { \boldsymbol{c}}}^{[n ] } = \sum_{\beta\in { \mathbb{z } } } \widetilde a^{[n]}(\cdot -2\beta ) \ , c^{[n]}(\beta).\ ] ] in the first step of our analysis of the stepwise reproduction capability of a hermite subdivision scheme of type ( [ eq : hermsubd0 ] ) , we add only _ a single _ pair of exponentials @xmath117 and consider the space @xmath118 to keep notation simple and to better explain the underlying ideas , we will first carefully investigate this situation and then extend it in a quite straightforward fashion to the general case . as can be seen in ( [ eq : vlambdadef ] ) , the addition of an exponential frequency @xmath119 always means the addition of the pair @xmath120 of functions . on the one hand , this is motivated by the fact that choosing @xmath121 equals reproduction of the trigonometric functions @xmath122 and @xmath123 . moreover , our approach to construct the annihilator and the factorization actually strongly depends on the presence of this pair of frequencies . whether or not similar results will be available for the case where only @xmath7 but not @xmath124 , we do not know at present . for any function @xmath105 and any integer @xmath114 we consider the two vector sequences @xmath125 , @xmath126 , defined , for @xmath127 , as @xmath128 , \qquad v_{f,{n } } ( \alpha ) : = d^n \widetilde v_{f,{n } } ( \alpha ) = \left [ \begin{array}{c } f ( 2^{-n } \alpha ) \\ 2^{-n } f ' ( 2^{-n } \alpha ) \\ \vdots \\ 2^{-nd } f^{(d)}(2^{-n } \alpha ) \end{array } \right].\ ] ] we simply write @xmath129 when @xmath130 . [ def : spectrvd ] a mask @xmath98}\in \ell_{00}^{(d+1)\times(d+1)}({\mathbb{z}})$ ] or its associated subdivision operator @xmath131}}$ ] satisfies the @xmath132-spectral condition if @xmath133 } } { { \boldsymbol{v}}}_{f , n } = { { \boldsymbol{v}}}_{f,{n+1}},\qquad f \in v_{d,\lambda}.\ ] ] equivalently , the mask @xmath115 } = d^{-(n+1 ) } { { \boldsymbol{a}}}^{[n ] } d^n$ ] satisfies the @xmath132-spectral condition if @xmath134 } } { \widetilde { { \boldsymbol{v}}}}_{f , n } = { \widetilde { { \boldsymbol{v}}}}_{f,{n+1 } } , \qquad f\in v_{d,\lambda}.\ ] ] it is important to observe that definition [ def : spectrvd ] is fully consistent with ( * ? ? ? * definition 1 ) though formulated in a slightly different way taking into account the stronger form of level dependency needed for the reproduction of exponentials . since we plan to extend difference operators and taylor operators , we next recall their formal definitions . [ def : taylorop ] the taylor operator @xmath135 of order @xmath45 , acting on @xmath136 is defined as @xmath137,\ ] ] where @xmath9 is the forward difference operator . the symbol of the taylor operator then takes the form @xmath138.\ ] ] [ def : annop ] a _ level-@xmath99 cancellation operator _ @xmath139 } : \ell^{d+1}({\mathbb{z } } ) \to \ell^{d+1}({\mathbb{z}})$ ] for a linear function space @xmath140 is a convolution operator such that @xmath141 } { { \boldsymbol{v}}}_{f , n } = \sum_{\alpha\in { \mathbb{z } } } h^{[n ] } ( \cdot-\alpha ) v_{f , n}(\alpha)=0 , \qquad f\in v.\ ] ] by @xmath139}_{d,\lambda}$ ] we denote a level-@xmath99 cancellation operator for the function space spanned by @xmath132 . [ lem : annihilop ] an operator @xmath139}_{d,\lambda}$ ] is a level-@xmath99 cancellation operator for the space @xmath132 if it satisfies @xmath142}_{d,\lambda } \right)^*(z ) = \left [ \begin{array}{cc } t^*_{d-2}(z ) & * \\ 0 & * \end{array } \right]\ ] ] and @xmath143}_{d,\lambda } \right)^ * \left ( e^{\mp 2^{-n } \lambda } \right ) d^n \left [ \begin{array}{c } 1 \\ \pm \lambda \\ \vdots \\ ( \pm \lambda)^d \end{array } \right ] = 0.\ ] ] to annihilate polynomials of degree @xmath144 , condition ( [ eq : annop ] ) has to be satisfied for the vector sequences @xmath145^t,\qquad j=0,\ldots , d-2,\ ] ] and since these sequences are exactly annihilated by the complete taylor operator as shown in @xcite , any decomposition of the form ( [ eq : annihilop1 ] ) annihilates polynomials of degree at most @xmath144 . to describe cancellation of exponentials , we first observe that @xmath146 , \ ] ] so that the condition becomes @xmath147 } ( \cdot - \alpha ) \ , e^{\pm 2^{-n } \lambda \alpha } d^n \left [ \begin{array}{c } 1 \\ \pm \lambda \\ \vdots \\ ( \pm \lambda)^d \end{array } \right ] \\ & = & \sum_{\alpha \in { \mathbb{z } } } h_{d,\lambda}^{[n ] } ( \alpha ) \ , e^{\pm 2^{-n } \lambda ( \cdot - \alpha ) } d^n\left [ \begin{array}{c } 1 \\ \pm \lambda \\ \vdots \\ ( \pm \lambda)^d \end{array } \right ] \\ & = & e^{\pm 2^{-n } \lambda \cdot } \left ( h_{d,\lambda}^{[n ] } \right)^ * ( e^{\mp 2^{-n } \lambda } ) \ , d^n \left [ \begin{array}{c } 1 \\ \pm \lambda \\ \vdots \\ ( \pm \lambda)^d \end{array } \right ] \end{aligned}\ ] ] which yields ( [ eq : annihilop2 ] ) . [ rem_marianto ] if we are able to find an operator @xmath148 that satisfies ( [ eq : annihilop1 ] ) and @xmath149 = 0,\ ] ] then we automatically obtain level-@xmath99 cancellation operators @xmath139}_{d,\lambda}$ ] for any @xmath114 by setting @xmath150}_{d,\lambda}={{\cal h}}_{d,2^{-n}\lambda}.\ ] ] in fact , this follows from the simple observation that the identity @xmath151 = h_{d,2^{-n}\lambda}^*\left ( e^{\mp 2^{-n } \lambda } \right ) \left [ \begin{array}{c } 1 \\ \pm \frac{\lambda}{2^n } \\ \vdots \\ ( \pm \frac{\lambda}{2^n})^d \end{array } \right ] = 0,\ ] ] is equivalent to ( [ eq : annihilop2n0 ] ) , as can be verified by just replacing @xmath119 with @xmath152 . in view of remark [ rem_marianto ] we see that to generate a level-@xmath99 cancellation operator we just need to construct a level-@xmath153 cancellation operator . therefore we continue with the analysis of @xmath154}_{d,\lambda}$ ] which will be simply denoted by @xmath148 . the next step is now to construct a cancellation operator which will eventually even turn out to be a minimal one . based on lemma [ lem : annihilop ] , the structure of the cancellation operator @xmath148 for the space @xmath132 can now be derived . indeed , we write its symbol in the general form @xmath155,\quad q^*(z)\in { \mathbb{r}}^{(d-2)\times 2},\ , r^*(z)\in { \mathbb{r}}^{2\times 2}\ ] ] and determine the remaining part of @xmath156 , namely the laurent polynomial matrices @xmath157 and @xmath158 . to this aim , we begin to explicitly compute the first line @xmath159 , where the `` @xmath160 '' is to be understood in the sense of matlab notation . [ lem : line1compl ] the condition @xmath161 = 0\,,\ ] ] can be fulfilled by setting @xmath162 and @xmath163 due to ( [ eq : symbtaylorop ] ) , the identity ( [ eq : line1complcond ] ) can be written as @xmath164 ( 1 ) + ( \pm \lambda)^{d-1 } h_{0,d-1 } + ( \pm \lambda)^d h_{0,d } , \end{aligned}\ ] ] where @xmath165 = \sum_{j=0}^k \frac{f^{(j)}(0)}{j ! } ( \cdot)^j\,,\ ] ] denotes the taylor polynomial of @xmath166 of order @xmath38 expanded at @xmath153 . adding and subtracting the above conditions we get @xmath167 ( 1 ) \\ & & \quad + \left ( \lambda^{d-1 } \pm ( -\lambda)^{d-1 } \right ) h_{0,d-1 } + \left ( \lambda^{d } \pm ( -\lambda)^{d } \right ) h_{0,d}. \end{aligned}\ ] ] if @xmath45 is even , this implies that @xmath168 ( 1)}{2 \lambda^{d-1 } } , \\ h_{0,d } & = & -\frac{e^{-\lambda } + e^\lambda - t_{d-2 } \left [ e^{-\lambda \cdot } + e^{\lambda \cdot } \right ] ( 1)}{2 \lambda^d } , \end{aligned}\ ] ] while for odd @xmath45 we get @xmath169 ( 1)}{2 \lambda^{d-1 } } , \\ h_{0,d } & = & \frac{e^{-\lambda } - e^\lambda - t_{d-2 } \left [ e^{-\lambda \cdot } - e^{\lambda \cdot } \right ] ( 1)}{2 \lambda^d}. \end{aligned}\ ] ] since @xmath170 we have that @xmath171 ( 1 ) = 2\lambda + \frac23 \lambda^3 + \cdots = 2 \sum_{2j+1 \le d-2 } \frac{\lambda^{2j+1}}{(2j+1)!}\,,\ ] ] and @xmath172 ( 1 ) = 2 + \lambda^2 + \cdots = 2 \sum_{2j \le d-2 } \frac{\lambda^{2j}}{(2j)!}.\ ] ] substituting these identities readily gives ( [ eq : line1complcond0d-1 ] ) and ( [ eq : line1complcond0d ] ) . taking into account the structure of @xmath156 , we can now easily give also the entries of the other lines . + for @xmath173 , we have that @xmath174 in particular , @xmath175 . to complete the construction of @xmath156 , we have to define the lower right block @xmath158 as @xmath176 \lambda \displaystyle\frac{e^{-\lambda } - e^\lambda}{2 } & z^{-1 } - \displaystyle\frac{e^\lambda + e^{-\lambda}}{2 } \end{array } \right]\\ & = & l_{d,\lambda}\,\left[\begin{array}{cc } z^{-1}-e^{\lambda } & 0\\ 0 & z^{-1}-e^{-\lambda } \end{array}\right ] \ , l_{d,\lambda}^{-1},\end{aligned}\ ] ] where @xmath177,\ ] ] for which the validity of ( [ eq : annihilop2 ] ) is easily verified by direct computations . as an example , we provide the explicit structures of @xmath178 , @xmath179 for the spaces @xmath180 and @xmath181 : @xmath182 0&\displaystyle{z^{-1}-\frac { e^{-\lambda}+e^{\lambda}}2 } & \displaystyle{\frac { e^{-\lambda}-e^{\lambda}}{2\lambda } } \\[5 mm ] 0&\displaystyle{\lambda\frac { e^{-\lambda}-e^{\lambda}}{2 } } & \displaystyle{z^{-1}-\frac { e^{-\lambda}+e^{\lambda}}2 } \end{array } \right]\,,\ ] ] and @xmath183 0 & z^{-1}-1 & \displaystyle{\frac { e^{-\lambda}-e^\lambda}{2\lambda } } & -\displaystyle{\frac { e^{-\lambda}+e^\lambda-2}{2\lambda^2 } } \\[5 mm ] 0&0&\displaystyle{z^{-1}-\frac { e^{-\lambda}+e^{\lambda}}2 } & \displaystyle{\frac { e^{-\lambda}-e^{\lambda}}{2\lambda } } \\[5 mm ] 0&0&\displaystyle{\lambda\frac { e^{-\lambda}-e^{\lambda}}{2 } } & \displaystyle{z^{-1}-\frac { e^{-\lambda}+e^{\lambda}}2 } \end{array } \right]$$ $ $ = \left[\begin{array}{cccc } z^{-1}-1 & -1 & \displaystyle{\frac{2-e^{-\lambda}-e^{\lambda}}{2\lambda^2 } } & \displaystyle{\frac{2\lambda+e^{-\lambda}-e^{\lambda}}{2\lambda^3}}\\[5 mm ] 0 & & & \\[5 mm ] 0&&h^*_{2,\lambda}(z ) & \\[5 mm ] 0 & & & \end{array } \right]\,.\ ] ] of course , the above construction of @xmath148 is only one of many possibilities to construct a cancellation operator for @xmath132 . however , our construction is well chosen in the sense that it includes the taylor operator as action on the polynomials and that it in fact naturally extends the taylor operator . [ thm : hdlconvtd ] @xmath184 it is obvious from ( [ eq : rzdefinition ] ) that @xmath185\,,\ ] ] as @xmath186 , hence it suffices to show that @xmath187 and @xmath188 as @xmath186 . suppose that @xmath189 is even in which case we get @xmath190(1)}{2 \lambda^{d-1-k } } = \frac{1}{2 \lambda^{d-1-k } } \sum_{j = d-1-k}^\infty \left ( ( -1)^j - 1 \right ) \frac{\lambda^j}{j ! } \\ & = & -\frac{1}{(d-1-k ) ! } + \lambda^2 \sum_{j = d+1-k } \frac{(-1)^j + 1}{2 j ! } \lambda^{j-(d+1-k ) } , \end{aligned}\ ] ] which converges as desired when @xmath186 . the arguments for @xmath191 and the case of odd @xmath189 are identical . having understood the case of a single frequency @xmath119 , it is not hard any more to extend the construction to arbitrary sets of frequencies . to that end , let @xmath192 consist of @xmath193 different frequencies , all either real or purely imaginary , and let us construct a cancellation operator @xmath23 for the space @xmath194 the conditions for cancellation extend in a straightforward way . [ lem : hlambdacancel ] the operator @xmath23 with symbol @xmath195,\quad q^*(z)\in { \mathbb{r}}^{p\times 2r},\ , r^*(z)\in { \mathbb{r}}^{2r\times 2r},\ ] ] annihilates @xmath29 if and only if @xmath196 = 0 , \qquad j=1,\dots , r.\ ] ] since the taylor part of @xmath23 annihilates the polynomials , we only need to perform the computations used to derive ( [ eq : annihilop2 ] ) for any @xmath197 to show that cancellation of the exponential polynomials is equivalent to ( [ eq : hlambdacancel ] ) . the construction of @xmath23 now follows the same lines as before , namely by determining the matrix symbol @xmath157 . for the first row we now get , for @xmath198 , the conditions @xmath199 ( 1 ) + \sum_{\ell=1}^r ( \pm \lambda_j)^{p+2\ell-1 } h_{0,p+2\ell-1 } + ( \pm \lambda_j)^{p+2\ell } h_{0,p+2\ell}.\end{aligned}\ ] ] again , we add and subtract to obtain @xmath200 ( 1 ) \\ & & + \sum_{\ell=1}^r \left ( \lambda_j^{p+2\ell-1 } \pm ( - \lambda_j)^{p+2\ell-1 } \right ) h_{0,p+2\ell-1 } + \left ( \lambda_j^{p+2\ell } \pm ( - \lambda_j)^{p+2\ell } \right ) h_{0,p+2\ell}.\end{aligned}\ ] ] this again decomposes depending on the parity of @xmath26 . supposing that @xmath26 is even , we get for @xmath198 @xmath201 ( 1)}{2 \lambda_j^{p+1 } } , \\ \sum_{\ell=0}^{r-1 } \lambda_j^{2\ell } \ , h_{0,p+2\ell+2 } & = & - \frac{\left ( e^{\lambda_j } + e^{-\lambda_j } \right ) - t_p \left [ e^{\lambda_j \cdot } + e^{-\lambda_j \cdot } \right ] ( 1)}{2 \lambda_j^{p+2}},\end{aligned}\ ] ] and since the polynomials @xmath202 form a chebychev system on @xmath203 , this system of equations has a unique solution . defining the vectors @xmath204 ( 1)}{2 \lambda_j^{p+2 } } \;:\ ; j=1,\dots , r \right ] , \\ w_- & : = & \left [ -\frac{e^{\lambda_j } - e^{-\lambda_j } - t_p \left [ e^{\lambda_j \cdot } - e^{-\lambda_j \cdot } \right ] ( 1)}{2 \lambda_j^{p+1 } } \;:\ ; j=1,\dots , r \right],\end{aligned}\ ] ] and the _ vandermonde matrices \in { \mathbb{r}}^{r\times r},\ ] ] we can therefore write down the construction of the cancellation operator explicitly . the condition ( [ eq : hlambdacancel ] ) can be satisfied by setting @xmath206 = \left\ { \begin{array}{ccl } l_\lambda^{-1 } w_- , & \quad & p \in 2 { \mathbb{n } } , \\ l_\lambda^{-1 } w_+ , & \quad & p \in 2 { \mathbb{n}}+1 , \end{array } \right . \\ \left [ h_{0,p+2\ell+2 } \;:\ ; \ell = 0,\dots , r-1 \right ] = \left\ { \begin{array}{ccl } l_\lambda^{-1 } w_+ , & \quad & p \in 2 { \mathbb{n } } , \\ l_\lambda^{-1 } w_- , & \quad & p \in 2 { \mathbb{n}}+1 . \end{array } \right . \end{aligned}\ ] ] the completion of @xmath23 by means of @xmath207 is now an obvious extension of ( [ eq : rzdefinition ] ) , namely @xmath208 where @xmath209 \in { \mathbb{r}}^{2r \times 2r}\,,\ ] ] and @xmath210 \\ \label{eq : rzgeneraldef2 } & = & \left [ \begin{array}{ccccc } z^{-1 } - e^{\lambda_1 } \\ & z^{-1 } - e^{-\lambda_1 } \\ & & \ddots \\ & & & z^{-1 } - e^{\lambda_r } \\ & & & & z^{-1 } - e^{-\lambda_r } \end{array } \right].\end{aligned}\ ] ] since @xmath211 is the transpose of a vandermonde matrix , it is nonsingular . the main result for the use of cancellation operators is related to the factorization of any subdivision operator that satisfies the spectral condition . [ thm : factorization ] if the subdivision operator @xmath96}}$ ] satisfies the @xmath132-spectral condition , then there exists a mask @xmath212}\in \ell_{00}^{(d+1)\times ( d+1)}({\mathbb{z}})$ ] such that @xmath213 } } = { { \cal s}}_{{{\boldsymbol{b}}}^{[n ] } } { { \cal h}}_{d,2^{-n } \lambda},\ ] ] or , in terms of symbols , @xmath214 } \right)^*(z ) = \left ( b^{[n ] } \right)^*(z ) \ , h_{d,2^{-n } \lambda}^*(z^2).\ ] ] in order to prove this theorem , we first give some results about the factorization of ( subdivision and convolution ) operators which annihilate the space @xmath29 . [ thm : cfactorization ] if @xmath215 is a finitely supported mask such that @xmath216 , then there exists a finitely supported mask @xmath217 such that @xmath218 . we first recall from @xcite that whenever @xmath219 , then there exists @xmath220 such that @xmath221,\ ] ] and @xmath222 has a symbol with structure @xmath223 : = \left [ b_0^ * ( z),\dots , b_p^ * ( z ) , c_{p+1}^ * ( z ) , \dots , c_d^ * ( z ) \right],\ ] ] where @xmath224 are columns of the original @xmath225 . we define the matrix sequence @xmath226 \in { \mathbb{r}}^{(d+1 ) \times 2r}.\ ] ] by assumption , @xmath227 and @xmath228 { { \boldsymbol{w}}}= 0 $ ] , and we thus get @xmath229 { { \boldsymbol{w}}}= { { \cal s}}_{{\boldsymbol{b}}}\left ( \left [ \begin{array}{cc } { { \cal t}}_{p } & 0 \\ 0 & i \end{array } \right ] - { { \cal h}}_{d,\lambda } \right ) { { \boldsymbol{w}}}\\ & = & { { \cal s}}_{{\boldsymbol{b}}}\left [ \begin{array}{cc } 0 & -{{\cal q}}\\ 0 & i - { { \cal r}}\end{array } \right ] { { \boldsymbol{w}}}= { { \cal s}}_{{\boldsymbol{b}}}\left [ \begin{array}{cc } 0 & -{{\cal q}}\\ 0 & i \end{array } \right ] { { \boldsymbol{w}}}= { { \cal s}}_{{\boldsymbol{b}}}\left [ \begin{array}{c } -{{\cal q}}l_{d,\lambda } \\ l_{d,\lambda } \end{array } \right ] \mbox{diag}\ , \left ( e^{\pm \lambda\cdot } \right ) \\ & = & \sum_{\alpha \in { \mathbb{z } } } \left ( - b_p ( \cdot - 2\alpha ) q ( \cdot - 2\alpha ) + c_{2r } ( \cdot - 2\alpha ) \right ) l_{d,\lambda } \ , \mbox{diag } \left ( e^{\pm \lambda\cdot } \right ) , \end{aligned}\ ] ] where @xmath230.\ ] ] this implies that for @xmath231 and @xmath198 we must have @xmath232 with @xmath233 the standard @xmath234-th unit vector in @xmath235 , from which it follows that @xmath236 hence , @xmath237 and @xmath238 setting @xmath239 $ ] , ( [ eq : thmfactorizationpf1 ] ) and ( [ eq : thmfactorizationpf2 ] ) can be conveniently combined into @xmath240 which leads to @xmath241 and consequently @xmath242 = \left [ b_p^ * ( z ) , b_{2r}^ * ( z ) l_{d,\lambda}^{-1 } \ , l_{d,\lambda } \delta_{\lambda}^ * ( z^2 ) l_{d,\lambda}^{-1 } + b_p^ * ( z ) q^*(z^2 ) \right ] \\ & = & \left [ b_p^ * ( z ) , b_{2r}^ * ( z ) l_{d,\lambda}^{-1 } \right ] \left [ \begin{array}{cc } i & q^*(z^2 ) \\ 0 & r^ * ( z^2 ) \end{array } \right ] . \end{aligned}\ ] ] this eventually gives @xmath243 \\ & = & \left [ b_p^ * ( z ) , b_{2r}^ * ( z ) l_{d,\lambda}^{-1 } \right ] \left [ \begin{array}{cc } i & q^*(z^2 ) \\ 0 & r^ * ( z^2 ) \end{array } \right ] \left [ \begin{array}{cc } t_{p}^ * ( z^2 ) & 0 \\ 0 & i \end{array } \right ] \\ & = & \left [ b_p^ * ( z ) , b_{2r}^ * ( z ) l_{d,\lambda}^{-1 } \right ] \left [ \begin{array}{cc } t_{d-2}^ * ( z^2 ) & q^*(z^2)\\ 0 & r^ * ( z^2 ) \end{array } \right ] \\ & = & \left [ b_p^ * ( z ) , b_{2r}^ * ( z ) l_{d,\lambda}^{-1 } \right ] h_{d,\lambda}^ * ( z^2)\ , , \end{aligned}\ ] ] and completes the proof . as a consequence of theorem [ thm : cfactorization ] and remark [ rem_marianto ] we get the desired result that extends the observations made in the introduction . [ cor : cfactorization ] if @xmath244 } \in \ell^{(d+1 ) \times ( d+1)}_{00 } ( { \mathbb{z}})$ ] is such that @xmath245 } } { { \boldsymbol{v}}}_{f , n}= 0 $ ] , @xmath246 , then there exists a finitely supported mask @xmath212}\in \ell^{(d+1 ) \times ( d+1)}_{00 } ( { \mathbb{z}})$ ] such that @xmath245 } } = { { \cal s}}_{{{\boldsymbol{b}}}^{[n ] } } { { \cal h}}_{d,2^{-n}\lambda}$ ] . using this result , theorem [ thm : factorization ] is now easy to prove theorem [ thm : factorization ] set @xmath245 } } : = { { \cal h}}_{d,2^{-n-1}\lambda } \ , { { \cal s}}_{{{\boldsymbol{a}}}^{[n]}}$ ] . since for @xmath246 we have @xmath247 } } { { \boldsymbol{v}}}_{f , n } = { { \cal h}}_{d,2^{-n-1}\lambda } \ , { { \cal s}}_{{{\boldsymbol{a}}}^{[n]}}{{\boldsymbol{v}}}_{f , n } = { { \cal h}}_{d,2^{-n-1}\lambda } \ , { { \boldsymbol{v}}}_{f,{n+1 } } = 0,\ ] ] it follows from corollary [ cor : cfactorization ] that there exists @xmath212}$ ] such that @xmath248 } } = { { \cal s}}_{{{\boldsymbol{b}}}^{[n ] } } \ , { { \cal h}}_{d,2^{-n}\lambda},\ ] ] as claimed . note that the factorization ( [ eq : factsa ] ) of @xmath96}}$ ] is equivalent to the following factorization of @xmath249}}$ ] : @xmath250 } { { \cal s}}_{\widetilde { { \boldsymbol{a}}}^{[n ] } } = { { \cal s}}_{\widetilde { { \boldsymbol{b}}}^{[n ] } } \widetilde { { \cal h}}_{d,\lambda}^{[n]},\ ] ] where @xmath251 } : = d^n \widetilde { { \boldsymbol{h}}}_{d,2^{-n}\lambda } d^{-n},\qquad \widetilde { { \boldsymbol{b}}}^{[n ] } : = d^{-n-1 } \widetilde { { \boldsymbol{b}}}^{[n ] } d^{n}.\ ] ] a careful inspection of the proof of theorem [ thm : cfactorization ] shows that the factorization can also be extended to convolution operators . [ thm : convfactorization ] if @xmath252 is such that @xmath253 , then there exists a finitely supported mask @xmath254 such that @xmath255 . the proof follows exactly the lines of the one of theorem [ thm : factorization ] except that ( [ eq : thmfactorizationpf2a ] ) and ( [ eq : thmfactorizationpf2b ] ) become @xmath256 that is , @xmath257 from there on the arguments can be repeated literally to yield that @xmath258 finally , observe that in the same way the argument from @xcite can be modified to give the initial factorization by means of the taylor operator . since @xmath23 is a convolution operator itself and since ( [ eq : convopfactor ] ) can be reformulated as the fact that for _ any _ @xmath259 that annihilates @xmath29 , the laurent polynomial @xmath260 must be divisible by @xmath261 , this operator is a particular annihilator of @xmath29 . [ cor : specialanni ] the operator @xmath23 is a minimal annihilator for @xmath29 . the taylor operator @xmath135 is a minimal annihilator for @xmath262 . to illustrate the results of the preceding sections , we construct two matrix subdivision schemes which reproduce , by construction , polynomials and exponential from the spaces @xmath263 and explicitly verify for these cases the factorization property via the annihilators in ( [ esempio : h_2 ] ) and ( [ esempio : h_3 ] ) . to construct the first vector hermite subdivision scheme , we start with a sufficiently smooth real valued function @xmath166 and define the initial sequence of vector data @xmath264^t \;:\ ; \alpha \in { \mathbb{z}}\right)$ ] from which we construct in each interval the functions @xmath265 such that they solve hermite interpolation problems at @xmath266 and @xmath267 based on the data @xmath268 and @xmath269 , respectively . it is easy to verify that these interpolation problems admit a unique solution in @xmath270 . this leads to the general interpolatory subdivision rules @xmath271 it turns out that matrix masks of the interpolatory hermite subdivision scheme defined as in ( [ eq : hermsubd0 ] ) consist of three nonzero @xmath272 matrices . the symbol of the scheme at the @xmath99-th iteration is @xmath273})^*(z)= \frac 1{16z}\left [ \begin { array}{ccc } 8\ , \left ( z+1 \right ) ^{2 } & \displaystyle{\frac { 4}{\lambda_n } \left ( z^2 - 1\right ) \sinh \frac{\lambda_n}{2 } } & \displaystyle{\frac { 4}{\lambda_n^{2 } } \left ( 1+{z}^{2 } \right ) \left ( \cosh \frac{\lambda_n}{2}-2 \right ) } \\\noalign{\medskip } 0&2(1+z^2)\cosh \frac{\lambda_n}{2}+8\,z & \displaystyle{\frac { 2}{\lambda_n } \left ( z^2 - 1 \right ) \ \sinh \frac{\lambda_n}{2 } } \\\noalign{\medskip}0&\lambda_n\ , \left ( z^2 -1\right ) \sinh \frac{\lambda_n}{2 } & ( 1+z^2)\ \cosh \frac{\lambda_n}{2}+4\,z \end { array } \right],\ ] ] with the abbreviation @xmath274 . observe that the determinant of @xmath275})^*(z)$ ] , @xmath114 , factorizes into @xmath276})^*(z ) = \frac{(z+1)^2 e^{-\lambda_n } \left(e^{\frac{\lambda_n}{2}}+z\right)^2 \left(z e^{\frac{\lambda_n}{2}}+1\right)^2}{64 z^3}.\ ] ] the resulting subdivision scheme appears to be convergent since , when starting the subdivision iterations by applying column - wise the subdivision rules to the delta matrix sequence the result after @xmath277 iterations stabilizes on the matrix function shown in fig . [ fig : a3x3 ] , but a specific convergence analysis is not in the scope of this paper . by construction this scheme satisfies the @xmath270-spectral condition and according to theorem [ thm : factorization ] it is possible to find a subdivision operator @xmath278}}$ ] such that the factorization ( [ eq : factsa ] ) holds true . at the @xmath99-th iteration , its symbol is given by : @xmath279 } \right)^*(z)=\frac{1}{16 } \left [ \begin { array}{ccc } 8 + 8\,z&-4\,\displaystyle{\frac { \sinh \frac{\lambda_n}{2}}{\lambda_n}}&4\,\displaystyle{\frac { \cosh \frac{\lambda_n}{2}-2}{{\lambda_n}^{2 } } } \\\noalign{\medskip}0&2\,\displaystyle{\cosh \frac{\lambda_n}{2}+4\,z}&-2\,\displaystyle{\frac { \sinh \frac{\lambda_n}{2}}{\lambda_n}}\\\noalign{\medskip}0&-\lambda_n\ , \sinh \frac{\lambda_n}{2 } & \cosh \frac{\lambda_n}{2}+2\,z\end { array } \right].\ ] ] the corresponding subdivision scheme seems to be zero - convergent , see figure [ fig : b3x3 ] , and hence contractive as one should expect . the symbol at level @xmath99 can be computed explicitly as @xmath282 } \right)^*(z)=\frac 1{32 z } \left [ \begin { array}{cccc } 16\ , \left ( 1+z \right ) ^{2 } & 8\ , \left(z^2 - 1 \right ) & \displaystyle{\frac { 8}{\lambda_n^2}\ , \left ( 1+{z}^{2 } \right ) \left(\cosh \frac{\lambda_n}{2}-2 \right ) } & \displaystyle{\frac { 8}{\lambda^3}\ , { \left ( z^2 - 1\right ) \left ( \lambda+\sinh \frac{\lambda_n}{2 } \right ) } } \\\noalign{\medskip } 0&8\ , \left ( z+1 \right ) ^{2 } & \displaystyle{\frac { 4}{\lambda_n } \left ( z^2 - 1\right ) \sinh \frac{\lambda_n}{2 } } & \displaystyle{\frac { 4}{\lambda_n^{2 } } \left ( 1+{z}^{2 } \right ) \left(\cosh \frac{\lambda_n}{2}-2 \right ) } \\\noalign{\medskip } 0 & 0&2(1+z^2)\ \cosh \frac{\lambda_n}{2}+8\,z & \displaystyle{\frac { 2}{\lambda_n } \left ( z^2 - 1 \right ) \ \sinh \frac{\lambda_n}{2 } } \\\noalign{\medskip } 0 & 0&\lambda_n\ , \left ( z^2 - 1 \right ) \sinh \frac{\lambda_n}{2 } & ( 1+z^2)\cosh \frac{\lambda_n}{2}+4\,z \end { array } \right],}\ ] ] the determinant of @xmath283 } \right)^*(z)$ ] factorizes into @xmath284 } \right)^*(z ) = \frac{(z+1)^4 e^{-\lambda_n } \left(e^{\frac{\lambda_n}{2}}+z\right)^2 \left(z e^{\frac{\lambda_n}{2}}+1\right)^2}{1024 \ , z^4}.\ ] ] evidence for the convergence of this scheme is given in fig . [ fig : a4x4 ] , where we show the plot of @xmath277 iterations of the scheme applied to the delta matrix sequence . this scheme satisfies the @xmath286-spectral condition and therefore admits the factorization ( [ eq : factsa ] ) with @xmath287 } \right)^*(z ) = \frac 1{32}\left [ \begin { array}{cccc } 16 + 16\,z&-8&8\,\displaystyle{\frac { \cosh \frac{\lambda_n}{2}-2}{{\lambda_n}^{2}}}&8\,\displaystyle{\frac { \lambda-\sinh \frac{\lambda_n}{2}}{{\lambda_n}^ { 3}}}\\\noalign{\medskip}0&8 + 8\,z&-4\,\displaystyle{\frac { \sinh \frac{\lambda_n}{2}}{\lambda_n}}&4\,\displaystyle{\frac { \cosh \frac{\lambda_n}{2}-2}{{\lambda_n}^{2 } } } \\\noalign{\medskip}0&0&2\,\cosh \frac{\lambda_n}{2}+4\,z&-2\,\displaystyle{\frac { \sinh \frac{\lambda_n}{2}}{\lambda_n}}\\ \noalign{\medskip } 0 & 0 & \displaystyle{\lambda_n\ , \sinh \frac{\lambda_n}{2}}&\displaystyle{\cosh \frac{\lambda_n}{2}+2\,z}\end { array } \right],\ ] ] which again seems to be a contraction , see figure [ fig : b4x4 ] . we conclude this section by observing that , as @xmath99 goes to infinity , the symbols @xmath288 } \right)^*(z)$ ] in ( [ def : aes1 ] ) and ( [ def : aes2 ] ) tend to @xmath289 } \right)^*(z ) = \frac 1{16z}\left [ \begin { array}{ccc } 8\ , \left ( z+1 \right ) ^{2 } & 4\ , \left ( z^2 - 1 \right ) & ( { z}^{2}+1 ) \\\noalign{\medskip } 0&4\ , \left ( z+1 \right ) ^{2}&2\ , \left ( z^2 - 1 \right ) \\\noalign{\medskip}0&0&2\ , \left ( z+1 \right ) ^{2}\end { array } \right],\ ] ] and @xmath289 } \right)^*(z ) = \frac 1{96z}\left [ \begin { array}{cccc } 48\ , \left ( z+1 \right ) ^{2}&24\ , \left ( z^2 - 1 \right ) & 6\,({z}^{2}+1 ) & \left ( z^2 - 1 \right ) \\\noalign{\medskip}0&24\ , \left ( z+1 \right ) ^{2 } & 12\ , \left ( z^2 - 1 \right ) & 3\,({z}^{2}+1 ) \\\noalign{\medskip}0&0&12\ , \left ( z+1 \right ) ^{2}&6\ , \left ( z^2 - 1 \right ) \\\noalign{\medskip}0&0&0&6\ , \left ( z+1 \right ) ^{2}\end { array } \right],\ ] ] respectively . these are the symbols of hermite schemes satisfying a polynomial space spectral condition . in particular , they reproduce polynomials up to the degree 2 and 3 , respectively . merrien and t. sauer , _ from hermite to stationary subdivision schemes in one and several variables _ , advances comput . math . * 36 * ( 2012 ) , 547579 . c. a. micchelli , _ interpolatory subdivision schemes and wavelets _ , j. approx . theory * 86 * ( 1996 ) , 4171 . m. unser and th . blu , _ cardinal exponential splines : part i theory and filtering algorithms _ , ieee trans . * 53 * ( 2005 ) , 14251438 .

we report on the methodology and first results from the deep lens survey ( dls ) transient search . we utilize image subtraction on survey data to yield all sources of optical variability down to 24 magnitude . images are analyzed immediately after acquisition , at the telescope and in near real time , to allow for followup in the case of time critical events . all classes of transients are posted to the web upon detection . our observing strategy allows sensitivity to variability over several decades in timescale . the dls is the first survey to classify and report all types of photometric and astrometric variability detected , including solar system objects , variable stars , supernovae , and short timescale phenomena . three unusual optical transient events were detected , flaring on thousand second timescales . all three events were seen in the passband , suggesting blue color indices for the phenomena . one event ( ot 20020115 ) is determined to be from a flaring galactic dwarf star of spectral type dm4 . from the remaining two events , we find an overall rate of on thousand second timescales , with a confidence limit of . one of these events ( ot 20010326 ) originated from a compact precursor in the field of galaxy cluster abell 1836 , and its nature is uncertain . for the second ( ot 20030305 ) we find strong evidence for an extended extragalactic host . a dearth of such events in the passband yields an upper confidence limit on short timescale astronomical variability between of . we report also on our ensemble of astrometrically variable objects , as well as an example of photometric variability with an undetected precursor .

astro-ph0404416

characterization of the variable optical sky is one of the new observational frontiers in astrophysics , with vast regions of parameter space remaining unexplored . at the faint flux levels reached by this optical transient search , previous surveys were only able to probe down to timescales of hours . an increase in observational sensitivity at short timescale and low peak flux holds the promise of detection and characterization of rare , violent events , as well as new astrophysics . the detection of transient optical emission provides a window into a range of known astrophysical events , from stellar variability and explosions to the mergers of compact stellar remnants . known types of catastrophic stellar explosions , such as supernovae and gamma ray bursts ( grbs ) , produce prompt optical transients decaying with timescales of hours to months . some classes of grbs result from the explosion of massive stars these hypernovae are known to produce bright optical flashes decaying with hour long timescales due to emission from a reverse shock plowing into ejecta from the explosion . in addition , grbs produce optical afterglows , decaying on day to week timescales , resulting from jet like relativistic shocks expanding into a circumstellar medium . even more interesting are explosive events yet to be discovered , such as mergers among neutron stars and black holes . these may have little or no high - energy emission , and hence may be discoverable only at longer wavelengths @xcite . finally , there is the opportunity to find rare examples of variability , as well as the potential for discovering new , unanticipated phenomena . one of the primary science goals of this transient search is the rate and distribution of short timescale astronomical variability . however , we catalogue and report all classes of photometric or astrometric variability with equal consideration . a weakness in many surveys with targeted science is that serendipitous information is discarded as background a counterexample is the wealth of stellar variability information being gleaned from microlensing survey data . the dls survey geometry and observation cadence are not optimal for maximizing the overall number of detected transients ( e.g. ) , being driven instead by weak lensing science requirements . however , given the current lack of constraints at short timescales and to significant depth , the potential remains for the discovery of new types of astronomical variability . the search for variability without population bias is one of the primary goals of the large - aperture synoptic survey telescope ( lsst , ) . efforts such as the dls transient search are a useful and necessary testing ground for the software design and implementation needed to reduce , in real time , lsst s expected data flow of 20 tb nightly . the field of optical variability surveys has been revolutionized by the adaption of ccd devices to astronomy , and by moore s progression of computing power . the former has yielded wide field imaging systems on large aperture telescopes , and the latter ensures their data can be reduced in quick order , if not real time . such advances have lead to modern variability surveys characterized by successively more extreme combinations of depth , cadence , and sky coverage . the advent of microlensing surveys advanced by orders of magnitude the known number of variable objects , driving the forefront of wide field ( 0.11 deg@xmath8 ) imaging on meter class telescopes ( e.g. eros : ; macho : ; mega : ; moa : ; ogle i , ii , iii : ; point - agape : ; supermacho : ) . the cadences are typically of order 1 image per day per field , although various strategies have been employed for sensitivity to short events or fine structure in longer events . such surveys are traditionally directed at crowded systems of stars serving to backlight a foreground microlensing population . this stellar background , however , also serves as a foreground that obfuscates sources at cosmological distances . surveys for supernovae at cosmological distances avoid such crowded foregrounds , employing similar resources and attaining similar sky coverage , but reaching deeper than the microlensing searches ( e.g. high - z : ; scp : ) . the original surveys cadence of field reacquisition was typically of order a month , short enough to avoid confusion by active galactic nuclei but long enough to ensure a handful of new supernova per observing run . these surveys were designed only for event discovery , with subsequent lightcurve coverage provided by narrower field followup resources . given the successes of the original supernovae surveys , subsequent efforts have been allocated resources for an advanced observing cadence allowing both detection and self followup , such as the ifa deep @xcite and essence @xcite supernova surveys . the nearby supernova factory is expecting @xmath9 type ia supernova per year from their moderate aperture ( 1.2meter ) telescope and large ( 9 deg@xmath8 ) field of view @xcite . however , dedicated systems with a small field of view ( 0.01 deg@xmath8 ) , similar depth , and a cadence of several days can lead to a wealth of supernova detections ( e.g. loss : ) . asteroid and near earth object ( neo ) searches cover of order thousands of degrees per night , primarily in the ecliptic , and near opposition ( e.g. loneos : ; linear : ; neat : ) . their photometric depths are shallower than supernova surveys and the cadence more rapid , given the requirement of recovering objects in motion . the images are typically of poorer seeing and image sampling , given less stringent photometric requirements . truly wide field systems ( 10 to 100 deg@xmath8 ) incorporating small apertures and limiting survey depths of 1217@xmath0 magnitudes are able to cover the entirety of the sky in a given night , or a given large region of sky multiple instances per night ( e.g. grosce : ; rotse - iii : ; raptor : ; asas-3 : ) . such systems are optimized for the detection and followup of fast optical transients , such as optical counterparts to grbs . these systems compromise nyquist sampling of the point spread function ( psf ) and photometric depth for breadth and real time agility . raptor s advanced implementation includes a narrow field fovea to optimally study localized transients . the northern sky variability survey ( nsvs : ) makes use of rotse - i survey data , which sampled the entire local sky up to twice nightly and provides variability information on @xmath10 million objects between 815.5@xmath0 magnitudes . additional variability surveys include the quest rr lyrae survey @xcite , which is to cover 700 deg@xmath8 to @xmath11 magnitude in the galactic equator , with sensitivity to several timescales of variability . the sloan digital sky survey has a sequence of multi color temporal information for over 700 deg@xmath8 at multiple timescales @xcite . the palomar - quest synoptic sky survey drift scans 500 deg@xmath8 per night , with scans separated by time baselines of days to months @xcite . the deep ecliptic survey monitors 13 deg@xmath8 per night near the ecliptic in a search for kuiper belt objects ( kbos ) , using 4meter telescopes and 2 exposures separated by 23 hours and a third within a day @xcite . finally , the faint sky variability survey ( fsvs : ) is a study of overall optical and astrometric variability at faint magnitudes on a 2.5meter telescope . point source limiting magnitudes reach to 25 over a field of @xmath12 deg@xmath8 , and the cadence of field reacquisitions is staggered for sensitivity at several timescales , and thus to multiple phenomena . of all on going variability surveys , the fsvs is most similar to the dls transient search . the deep lens survey ( dls ) is a 5year noao survey operating on the 4meter blanco and mayall telescopes + mosaic imagers at the cerro tololo and kitt peak observatories , respectively @xcite . the survey is undertaking very deep multicolor imaging of five @xmath13 fields chosen at high galactic latitude , and composed of 9 subfields apiece . exposure times and filters are chosen to reach limiting magnitudes of @xmath14 to @xmath15 mag per square arcsecond surface brightness , with images typically acquired near new moon . at these limiting magnitudes , we expect to measure accurately the shapes and color redshifts of @xmath16 galaxies per square degree for weak lensing science . an overview of the survey and public data releases are available at http://dls.bell - labs.com/. the survey is expected to conclude observations in march , 2005 . full exposure depth requires 20 exposures per passband in each of our 45 subfields . the images are obtained , with some exceptions , in sets of 5pointing dither sequences , with offsets of @xmath17 and @xmath18 around the initial central pointing to fill gaps between mosaic chips and enable super sky flat construction . typical exposure times are 600 seconds in @xmath1 , @xmath19 , and @xmath20 , and 900 seconds in @xmath5 , yielding limiting magnitudes per exposure for point sources of approximately @xmath21 . the weak lensing science goals of the survey require co adding these images . in the co addition process , however , variable sources are clipped out , or averaged away . here we pursue an orthogonal approach , which instead analyzes _ differences _ in these data , at the pixel level , to extract this variability . this paper summarizes our transient search to date . starting in december 1999 , we began analysis of our acquired images in near real time , which we define as within the duration of a dither sequence of observations . true real time reductions ideally occur within the timescale it takes to acquire the subsequent observation , or , given a target population of events , shorter than the timescale of the variability being searched for . recognizing variability in near real time with current ccd mosaics requires data analysis of several gb nightly of images on site . to this end , we maintain computers in the control rooms of both the kitt peak and cerro tololo 4meter telescopes . at kitt peak , we have a quad processor 700 mhz machine , and at cerro tololo a quad processor 550 mhz machine . both systems run under the linux operating environment , and execute the custom data reduction pipeline described in section [ sec - pipeline ] . the ensemble of deep lens survey images considered for this publication was taken from 14 observing runs between 2002 november and 2003 april where the transient pipeline was operating efficiently . figure [ fig - samp ] displays the distribution ( _ dotted _ line ) of temporal intervals between subfield reacquisitions , integrated over all subfields . the _ solid _ line shows this same quantity , but only for reacquisitions in the same filter as the previous observation . since variability is detected through comparison of observations taken in similar filters , the integral under this _ solid _ subfield filter histogram up to timescale @xmath22 defines our overall sensitivity to transients at timescale @xmath22 . the integral under the _ dotted _ histogram represents our followup capability at a given @xmath22 . figure [ fig - samp ] indicates our primary sensitivity is to variability on thousand second timescales . in this subset we have a total integrated exposure time of 10.2 days distributed over 425 @xmath1band , 464 @xmath19band , and 393 @xmath5band images . at @xmath23 per image , this yields a total exposure @xmath24 of @xmath25 deg@xmath8days for sensitivity to @xmath26 second events down to 24@xmath0 magnitude . we have a factor of 17 smaller @xmath24 for inter night observation intervals at @xmath27 seconds , and also at inter month observation intervals at @xmath28 seconds . a planned re analysis of all survey data for transients will allow for a precise determination of @xmath24 at all available timescales . our typical subfield observation comes as a 5image dither pattern controlled using the iraf script mosdither . standardized information is written in each of the images fits headers , and is used in the pipeline to automatically associate each quintuple of images . basic image calibration is done before each image is sent to the transient pipeline . the iraf task mscred.ccdproc is used to correct for cross talk , apply overscan correction and trim the image , and bias and flat field correct the images ( we generally take calibration sequences in the afternoon to allow for real time reductions ) . in the case of @xmath20band data , we need to apply a fringing correction , requiring an accumulation of images that generally precludes real time reductions . finally , the images are registered to the world coordinate system ( wcs ) on the sky using mscred.msccmatch with the noao : usno - a2 @xcite catalog . the images are then copied to a directory that is continuously searched by the transient pipeline image reaper . our transient pipeline uses the opus environment @xcite as a backbone sequencing a series of python and perl language scripts , each defining a `` stage '' in our pipeline . these stages are defined as follows : * in : copy an image via the scp protocol from the iraf reduction machine to the transient pipeline machine . * ri : register and pixel wise resample an incoming image to a fiducial template image of the subfield . wcs information is used as a starting point for determining the registration coefficients . the astrometric template is the same for all passbands , and is constructed from the very first observation of a subfield . * ci : detect and catalog objects in the registered image using sextractor @xcite . for our detection thresholds , we use sextractor parameters detect_thresh = analysis_thresh = 2.5 . * hi : difference each dither against the first pointing , using a modified version of the @xcite algorithm . modifications include internal methods to register and remap pixels based upon wcs information , comprehensive discrimination of appropriate regions to determine the image convolution kernel , figures of merit to determine the direction of convolution , robust methods to ensure complete spatial constraint on the variation of the kernel , and the propagation of noise through the convolution process * fi : catalog the difference image and filter object detections , including the calculation of adaptive second moments @xcite . these moments , originally designed for accurate shape measurements in our weak lensing pipeline , are used here to verify the integrity of each candidate . cuts on the results of this analysis ( e.g size criteria for cosmic ray rejection ) help to reject a significant fraction of the noise and false positives , as well as a fraction of the actual transients present . this stage yields the short timescale transients in each dither sequence , primarily solar system objects , but also including stellar and unknown sources of variability . + to algorithmically cull false detections from our object lists , we first construct exclusion zones around bright stars , using the noao : usno - a2 @xcite catalog , and along bad ccd columns , using bad pixel masks supplied with the mscred package . given a list of candidates sextracted from the difference image , we further analyze each object in both the original input image and the difference image . in the input image , we require the object s semi minor axis , as computed from adaptive second moments , to be greater than 1 pixel ( @xmath29 ) as a cut against cosmic rays . in the difference image , we further require second moments greater than 0.9 pixels , uncertainties in the orthogonal ellipticity components less than 0.5 pixels , and integrated flux measurements greater than 0 counts . we also examine all pixels within 20 pixels of a candidate , and require @xmath30 of all pixels to be neither negative nor masked , and @xmath31 of non masked pixels to be positive valued . these additional cuts safeguard against bad registrations or subtractions , which tend to leave a number of contiguous pixels above the detection threshold , but also regions of systematic dipole residuals . this does impact our efficiencies for slowly moving objects such as kbos . by correctly propagating noise arrays through the remapping and difference imaging process , it is possible to compare projected noise properties of the image to empirical measurements , and automatically identify and reject such regions . at this point , the individual images are co added to make a deeper representation of the subfield and searched for longer timescale transients : * si : co add the dither sequence of images into a single deep image of the subfield , providing complete spatial coverage by filling in mosaic chip gaps . * it : identify prior templates of the same subfield , in the same filter , to use as templates for detecting variability on longer timescales . * hs : difference the deep image stack constructed in stage si against the prior templates . * fs : catalog and filter the deep difference images . longer timescale transients are revealed at this stage , primarily supernova and active galactic nuclei ( agn ) , and stellar variability . we employ similar object cuts as in stage fi . * cl : clean up and archive data files , and inform the observers that the sequence of observations is ready to be reviewed . our automated cuts typically reduce the number of candidates per image from several hundreds to several tens . after the last stage of the pipeline , the observer must manually categorize the remaining candidates . a graphical user interface displays the sequence of difference images . each is presented as a triplet of postage stamps centered on the candidate and showing the prior template , current image , and difference between them . the nature of the transient is determined using a variety of factors , which can be divided into temporal and contextual information . temporal discriminants include astrometric motion of the transient , indicative of solar system origin , as well as the timescale of photometric variability for stationary objects , which can help discriminate between populations of variability with known timescales , or to single out an object varying with unexpected rapidity . contextual discriminants include the existence of a precursor or host object , and the proximity and relative location of the transient event . during visual classification of a transient , a set of data files are automatically generated locally on disk , after cross checking for spatial coincidence with previous transients , or proximity to approximate ephemerides for moving objects . an ephemeris is generated for moving objects , and for stationary objects , a finding chart . these files are copied to our remote web server where a periodic daemon executes a command to regenerate our web page based upon the data files present . thus , every 15 minutes our web page is brought into sync with the transients detected at the telescope . at the conclusion of each run , all moving objects are reported to the minor planet center and the most convincing supernova candidates to the central bureau for astronomical telegrams . the magnitudes resulting from photometry on difference images are denoted by the symbol @xmath32 . this represents the magnitude of a static object whose brightness is equal to the measured change in flux @xmath33 , @xmath34 , where @xmath35 is the photometric zero point of the template image . this quantity can be related to the change in magnitude @xmath36 of an object with quiescent magnitude @xmath37 through the relation @xmath38 in the limit where the variability comes from a precursor fainter than the detection limit , such as for supernovae , then @xmath32 represents the magnitude of the transient itself . in the following , we report all magnitudes derived from difference imaging in units of @xmath32 . we are further pursuing algorithms that might automatically and optimally discriminate between transient classes , which will be a necessity for larger scale transient searches one such example is the genetic imagery exploration software package ( genie : ) . genie uses genetic algorithms to assemble a sequence of image operators that optimally extract features from multi spectral data , and might be naturally be applied instead to our multi temporal sequences . all archived transients are made available on the internet at http://dls.bell-labs.com/transients.html , generally within a few hours of observations . transients are broadly categorized as `` moving '' ( slow or fast objects are highlighted ) , `` variable star '' , `` supernova '' , `` agn or supernova '' ( if this can not be distinguished based on the host morphology and transient location ) , or `` unknown '' , a broad category indicating objects whose astrophysical nature is uncertain . this includes very rapidly varying objects , or transients with no obvious precursor or host . roughly 75% of the transients detected in the dls are astrometrically variable solar system objects . four of the five dls fields are within 15@xmath39 of the ecliptic , and in these fields we generally find of order ten moving objects per subfield visit . the total number of moving objects detected through spring 2003 is 3651 , with 900 detected in @xmath1 , 1508 detected in @xmath19 , and 1487 detected in @xmath5 ( some were detected in multiple filters ) . our detection efficiency for moving objects has not been modeled . because of our observing pattern , most of the moving objects are observed on a single night and never recovered . a main belt asteroid moving at @xmath4030@xmath41 hr@xmath42 near opposition leaves a subfield within a few days . as a result , only about one - third of our moving objects have been observed on multiple nights , and 333 of these have received minor planet center provisional designations , which are listed on our website . similarly , we are able to measure colors for only a fairly small fraction of these objects . therefore , we do not attempt a thorough analysis of orbital parameters , albedos , and sizes . rather , we give an overview of the dataset and then detail a few especially interesting objects . figure [ fig - astmag ] shows the magnitude distribution of all moving objects discovered , separately for @xmath1 , @xmath19 , and @xmath5 . note that the magnitude range is displaced from the magnitude range in which our efficiency for stationary point sources is nonzero ( figure [ fig - eff ] ) . this is because the vast majority of moving objects are significantly trailed during our 600 or 900 second exposures . an asteroid at the bright end can therefore be much brighter than a point source before saturation sets in , and at the faint end must also be brighter to rise above the detection threshold . figure [ fig - veldist1 ] shows the moving objects velocity distribution in ecliptic coordinates , highlighting a pair of kuiper belt objects ( kbos ) detailed below . figure [ fig - veldist2 ] provides tentative classifications into asteroid families ( main belt , trojans , centaurs , neos , etc . ) based upon ecliptic velocities , adopting the boundaries of @xcite . the 2003 april ctio run yielded two kuiper belt objects ( kbos ) . our pipeline does not include an orbit calculator , so kbos are not automatically flagged . these two examples were selected manually on the basis of their small ( @xmath43@xmath41 hr@xmath42 ) angular motions near opposition . transient 207 from the 2003 april ctio run was discovered on 2003 april 1 ( ut ) , moving at 2.9@xmath41 hr@xmath42 near opposition , with a median @xmath5 magnitude of 21.5 . using the method of @xcite , we find a semimajor axis , @xmath44 , of 41.0 au , eccentricity @xmath45 of 0.022 , and inclination @xmath46 of 20.1@xmath39 , making this a classical kbo . assuming an albedo of 0.1 , the diameter of this object is @xmath40475 km , making it one of the larger kbos known . transient 587 from the same run was discovered on 2003 march 31 ( ut ) at a @xmath1 magnitude of 23.3 , moving at 2.6@xmath41 hr@xmath42 , also near opposition . our orbit calculation yields @xmath47 , @xmath48 , and @xmath49 , making it a likely scattered kbo . for an albedo of 0.1 , this object is @xmath40410 km in diameter . our observations provide orbital arcs of 5 and 4 days for these kbos , respectively . the detection efficiency for fast movers declines with velocity , as their reflected light is trailed over larger area . the dls , with its long exposures , is therefore far from optimal for detecting near - earth objects ( neos ) . nevertheless , we have detected a sizable sample , as shown in figure [ fig - veldist2 ] . here we highlight one that was moving with an unusual velocity , transient 1363 from the 2003 march kpno run . this object is the fastest solar system object seen during the transient search , moving at 80@xmath41 hr@xmath42 . the box near the top of figure [ fig - veldist2 ] indicates this object s remote position in the ecliptic velocity diagram . we focus here on those transients that have been noted to flare on thousand second timescales and are detected in at least two images , as well as on longer term variability with no detectable originating host or precursor . however , our transient catalog also includes a selection of several hundred variable stars , as well as over 100 supernova candidates , 18 of which pass the requirements for recognition as defined by the central bureau for astronomical telegrams . these include : 2000bj ( ) , 2000fq ( ) , 2002aj , 2002ak , 2002al , 2002am ( ) , 2002ax , 2002ay , 2002az , 2002ba , 2002bb , 2002bc , 2002bd , 2002be ( ) , 2003bx , 2003by , 2003bz , 2003ca ( ) . we have re run the difference imaging algorithm for this paper , using 2k x 2k ( @xmath50 x @xmath50 ) arrays centered on the selected optical transients ( ots ) . we have generally been able to construct deeper , higher signal to noise ( s / n ) template images than those available to us at the time of detection . for photometric calibration , we need only calibrate the zero points of these template images , as the difference imaging convolution normalizes each input image to the template s photometric scale . the @xmath1 , @xmath19 , and @xmath5 templates are normalized to the @xcite system , with additional errors ( typically @xmath51 ) added in quadrature to account for instrumental differences between the kpno , ctio , and @xcite systems . all @xmath20 templates are initially calibrated to the @xcite @xmath52 system , and presented here in the vega system by adopting an offset of @xmath53 magnitudes ( @xmath54 ) . the transformation between photometric systems depends on the instrumental filter response and on vega s spectral energy distribution , adding an additional @xmath55 magnitude uncertainty to our @xmath56 measurements . the template brightnesses of the precursor , or host , objects were determined using mag_best from sextractor @xcite , except in the case of ot 20010326 ( section [ sec - trans52 ] below ) . for this precursor , which appears unresolved , we first used the iraf noao.digiphot.daophot.psf package to determine the local psf from nearby isolated stars , and subtracted these stars and the transient precursor from the image using noao.digiphot.daophot.allstar . for those template images where we are not able to detect the ot precursor / host , we define the point source detection limit in accordance with the noao archive definition of photometric depth ( t. lauer , private communication ) , as @xmath57 where @xmath35 is the magnitude of one adu , @xmath58 is the seeing in pixels , @xmath59 is the dispersion of the image around its modal sky value , and we choose @xmath60 @xmath61 as our detection limit . we determine detection limits in the difference images by measuring directly our ability to recover input stellar psfs . for the difference imaging , we choose the best fit convolution kernel that varies spatially to order one across our 2k x 2k subimage , and a sky background with a similar degree of variation . we fit these spatial variations over a 20 x 20 grid , using a convolution kernel with a half width of 10 pixels . the function of the convolution kernel is to degrade the higher quality image to match the psf of the lesser quality image . objects are detected in each grid element to constrain locally the convolution kernel through a pixel by pixel comparison between images . a global fit is next performed , using each element as a constraint on spatial variation of the kernel . several sigma clipping iterations ensure that sources of variability , such as variable stars or asteroids , are not used as constraints for this kernel . if such an element is rejected from the global fit , a replacement object within this element is chosen to ensure maximal spatial constraint . application of the final convolution kernel to the entire image and pixel by pixel subtraction yields the difference between the two images , with a resulting point spread function of the lesser quality image , and photometric normalization to the template image . the transient fluxes were determined using the iraf noao.digiphot.apphot.phot package . the relative transient magnitudes are typically determined to better than @xmath62 , with the remaining uncertainty due to absolute calibration of the template zero point . we have detected several optical transients with no obvious hosts , and which are seen to vary over the timescale of months ( e.g. ) , similar to supernovae and grb afterglows . one of the more luminous examples of this class of objects is ot 20020112 @xcite , originally reported as transient 139 from our 2002 january run . information on this event is summarized in table [ tab - short ] , including the dls subfield in which it was detected ( f4p13 ) , and position on the sky in j2000 coordinates . we include limits on a precursor or host for this event , which is not detected to deeper than @xmath63 magnitudes in the @xmath1 , @xmath19 , and @xmath5 passbands , and @xmath64 magnitude in @xmath20 . table [ tab - short ] also lists limits on the precursor / host brightness after taking into account galactic reddening , assuming a @xmath65 extinction curve @xcite . ot 20020112 was detected in observations taken 2002 january 12 . previous subfield observations were taken 2001 march 27 , so the age of the transient at detection is poorly constrained . information on this transient was quickly posted to the variable star network ( vsnet ) , prompting a sequence of followup observations from the 1.3 m mcgraw hill telescope ( mdm ) . the lightcurve for this transient is shown in figure [ fig - ot20020112 ] , and represents a daily averaged lightcurve of all observations from the dls and mdm . the event was re imaged 35 days after detection as part of dls survey observations , and had faded and reddened significantly . detailed photometry for this event is listed in table [ tab - phot ] . the magnitudes listed are in differential flux magnitudes @xmath32 however , given the non detection of a precursor or host , this also represents the traditional magnitude @xmath37 of this transient . ot 20020112 was also observed spectroscopically with the las campanas observatory 6.5-m baade telescope ( + dual imager / spectrograph ldss2 ) on 2002 january 18 . the spectrum in figure [ fig - spec20020112 ] is characterized by a blue continuum with no obvious broad features , and marginal evidence for emission lines h@xmath66 at @xmath67 and [ nii ] 6583 at @xmath68 . the continuum features and lack of short timescale variability are consistent with a type ii sn caught close to maximum brightness . however at @xmath67 , the observed brightness ( @xmath69 ) is @xmath70 magnitudes dimmer than a typical type ii sn near maximum . the blue continuum makes a reddened sn explanation unlikely . examination of the first 1.4 ghz radio catalog @xcite yields a limit of 0.94 mjy / beam at this position , with no object closer than @xmath71 . a search of the rosat source catalog @xcite , and a more comprehensive search of archival x ray and gamma ray data through heasarc yields no archival sources within @xmath72 . thus the nature of this object remains unknown , a situation compounded by the lack of an obvious host galaxy to @xmath73 . we note the observed @xmath74 color change during the event is inconsistent with a spectral energy distribution of the form @xmath75 , such as that expected for grb afterglows ( e.g. , ) . however , evolution of the index @xmath76 could lead to the observed behavior . future ot searches will require facilities dedicated to followup of such events , to associate or distinguish them from the expected supernova and grb afterglow populations . over the course of the survey we have also detected three optical transients on thousand second timescales that were present in multiple images , and whose precursors were not immediately recognized as stellar in all available passbands . in all cases , we were able to identify a precursor or host object after the fact . these are designated ot 20010326 , ot 20020115 , and ot 20030305 . table [ tab - short ] lists the observed characteristics of the precursors or hosts for these transients in their quiescent state . table [ tab - short ] also includes magnitudes corrected for galactic reddening for ot 20010326 and ot 20030305 , in the case that they lie outside the galactic dust layer . when putting our subsequent results in a cosmological context , we assume a wmap cosmology of @xmath77 @xcite . ot 20010326 was detected 2001 march 26.2 as the @xmath78 transient during that particular run . photometry for this event lightcurve , including the constraining observations immediately preceding and following the event , is listed in table [ tab - phot ] . after verification , an alert was dispersed via e mail on mar 26.4 requesting immediate followup observations . the event lightcurve is shown in figure [ fig - trans52 ] , and is characterized by a null detection in @xmath1 , followed by a @xmath1band peak and immediate fall off . given the range of allowed turn on times , from @xmath79 before the start of the detection observation ( to account for read out time between images ) to @xmath80 before the shutter closes in the detection image , we can limit a power law index for the flux decay @xmath81 of @xmath82 . radio observations of this ot were undertaken at the vla on 2001 march 30 . limits on radio flux at 8.5 ghz are @xmath83 mjy . subsequent observations of the event revealed a host galaxy or precursor object visible in the @xmath19 , @xmath5 , and @xmath20 passbands , and a limit of @xmath84 . this event occurred in the field of galaxy cluster abell 1836 , at a redshift of @xmath85 . figure [ fig - trans52 ] also shows an @xmath50 x @xmath50 @xmath5band template image centered on the transient event . the precursor / host is displayed at higher resolution in the upper right corner as the dim object @xmath86 north west of the brighter star . a single archival hst image of cluster galaxy pks 1358 - 11 also covers the location of the observed transient . this @xmath87 integration in the f606w filter was obtained jan 25 , 1995 as part of a nearby agn survey by @xcite . the precursor appears unresolved in this image . we compare this precursor position with the location of two background galaxies , and over a baseline of 7.0 years find proper motion limits of @xmath88 . it is particularly difficult to photometer and calibrate this ot precursor / host , as it occurs @xmath89 from a bright star that is saturated , or nearly so , in all of our images . there is also a very strong background gradient from the nearby elliptical galaxy pks 1358 - 11 and spiral galaxy lcrs b135905.8 - 112006 . if the ot host is a member of abell 1836 , it lies only 49 kpc in projection from the core of pks 1358 - 11 , and 53 kpc in projection from the center of lcrs b135905.8 - 112006 . the hst resolution limit at the distance of abell 1836 is around 70 parsecs , which does not immediately preclude a globular cluster host . however , at a distance modulus of 36.0 , the host absolute magnitude of @xmath90 would be considerably brighter than the globular cluster systems in our own galaxy @xcite . in addition , the observed @xmath91 color of the object makes it unlikely to be an unresolved globular cluster system ( k.a.g . olsen , private communication ) . our second fast optical transient , ot 20020115 , was initially reported as transient 337 from the run ongoing 2002 january 15.3 . the lightcurve , whose data are presented in table [ tab - phot ] and displayed in figure [ fig - trans337 ] , is very similar to that of ot 20010326 , including @xmath1band detection , but it reaches a brighter @xmath92 , and decays much more rapidly . the flux decay power law index @xmath66 is constrained to @xmath93 . this event was quickly reported as a potential grb optical counterpart to the grb circular network ( gcn , ) . we obtained spectra of the source soon thereafter figure [ fig - trans337spec ] shows 3 x 10 minutes of exposure on the source with the magellan 1 telescope + ldss2 dual image / grism spectrograph , taken 2002 january 18.3 . the appearance of zero - redshift emission and absorption features in these spectra are evidence for galactic origin , and the gcn alert was amended ( ) . subsequent observations in @xmath5 and @xmath20 revealed a very bright , red precursor object , consistently saturated in @xmath5band images . we also recover the precursor object in the @xmath1 and @xmath19 bands . with a quiescent magnitude of @xmath94 and measured peak @xmath95 , equation [ eqn - dm ] indicates the precursor varied by more than 1 magnitude , averaged over the duration of our @xmath96 exposure . the analysis of the photometric observations and object spectrum ( figure [ fig - trans337spec ] ) shows ot 20020115 has the colors and spectrum of a dm4 post uv ceti type flare star . the spectrum shows the clear presence of late type m - dwarf spectral indicators : molecular bands of tio , caoh ( 5530 - 60 and centered at 6250 ) , plus cah at 6385 , in addition to weak h@xmath66 and h@xmath76 in emission . with a @xmath19 magnitude of @xmath97 , and colors @xmath98 and @xmath99 , these data are consistent with an m dwarf of spectral type dm4 , i.e. a late type flare star in the quiescent state . assuming a typical @xmath100 , ot 20020115 would lie at a distance of approximately 350 pc , and 250 pc above the galactic plane , well within the scale height for disk population m dwarfs . the most dynamic transient , ot 20030305 , was detected as transient 153 from our 2003 march run . this is the only event detected in the @xmath19band , and we immediately acquired a sequence of @xmath1band observations after the @xmath19band series . the event lightcurve is shown in figure [ fig - trans153 ] . also displayed are the series of images of this event . the top row represents the precursor / host of the event in its quiescent state , in the @xmath5 , @xmath101 , and @xmath20bands , from left to right . included in each panel is the magnitude of this precursor / host in its quiescent state , except for the @xmath101 band where we have no calibration data . this precursor / host is not visible in @xmath1 or @xmath19 , and appears unresolved in the @xmath20 images but measurably elliptical in both @xmath101 and @xmath5 . the subsequent panels , reading from left to right and top to bottom , show the emergence of the ot in our time series of images , as well as ot magnitudes and time elapsed from the detection observation . all images are separated by approximately @xmath102 except for the final image , which was obtained more than 2 days after the event . the event was also announced via the iau circulars ( ) . the ot lightcurve shows more complex behavior than the previous events , and can not be described as simple power law flux decay . the second temporal peak indicates either a second round of brightening or spectral evolution of the flare . as noted above , the precursor / host object appears resolved and elliptical in the @xmath5 and @xmath101 passbands . the object s position angles in @xmath5 and @xmath101 , derived with the adaptive second moment method of @xcite , agree to within @xmath103 , indicating an extended object . we have characterized the expected psf at the location of the host object by first identifying candidate stars , based on their position in the size magnitude plane , in the @xmath5band template image . we then fit for the spatial variation of the moments of these candidates , clipping at 3@xmath61 to reject interloping small galaxies . a @xmath104 test comparing the shape of the host with the predicted shape of the psf at its position yields @xmath105 for 3 degrees of freedom , for a confidence level of @xmath106 that a point source would not have the measured shape of the precursor object . we consider the above as strong evidence for a resolved , extragalactic host for ot 20030305 . a primary discriminant between galactic stellar events , such as ot 20020115 , and extragalactic events is the measured shape of the precursor or host object . ot 20010326 is shown to be unresolved in hst imaging , implying galactic origin , although potentially originating from a compact extragalactic host . ot 20030305 has an apparent resolved host that is inconsistent with the @xmath5band psf interpolated at its position . this argues strongly for an extragalactic origin . it is suggestive that all of our ots are detected in the @xmath1 passband . we consider various stellar and extragalactic possibilities for the origin of these events in turn . since we are certain ot 20020115 is from a flaring dwarf star , we investigate the possibility that one , or even both , of the other ots are similar in nature . we examine the peak of the disk dwarf luminosity function , which is occupied by dm4 dm6 ( @xmath107 ) stars @xcite . this region is occupied by the most active uv ceti type flare stars , with event timescales of order minutes and flare @xmath74 colors of approximately 0.0 to 0.3 @xcite . @xcite further characterizes the `` type i '' outburst lightcurves expected of uv ceti type stars , with rise times of seconds to a few minutes , and decay times of minutes to about one hour , qualitatively similar to our observed events . from figure 2 of @xcite , and figure [ fig - cmds ] here , the precursor @xmath108 color of ot 20010326 implies a spectral type of dm4 were it a flare star . with @xmath109 , this object lies @xmath110 from the color defined locus of disk main sequence stars . however , using a @xmath19band dm4 distance modulus of between 11.5 and 12.5 magnitudes , this object would be approximately 2.03.2 kpc distant , and 1.52.4 kpc above the galactic plane . this instead suggests association with the sdm halo subdwarf population , consistent with our proper motion limits . however , these stars are not known to flare . the situation is similar for ot 20030305 , which is far too faint at @xmath111 mag to belong to the disk population . assuming an @xmath108 derived spectral type of approximately dm6 , its distance modulus yields @xmath112 kpc . for a typical dls field at b(ii ) = @xmath113 , we expect to find approximately 100 dm6 disk stars per subfield out to their galactic scale height of 350 pcs , which is verified through direct examination of several subfields . since flare events do not follow poisson statistics , and show large scatter in total energy , peak light and duration , it is difficult to calculate the exact number expected in each sequence of b exposures . in addition , selection effects in most previous studies have biased discovery towards stars of greatest flare visibility . until recently , one of the only ways to estimate the expected rate of flare star activity was to extrapolate equation 5 from @xcite . this relation was derived using data for a selected few of the most active flare stars in the immediate solar neighborhood . scaling kunkel s relation to our @xmath1 passband , and assuming that @xmath51 of all dm46 stars exhibit strong flaring activity ( i.e. @xmath114 of 1 mag or greater ) , we may place an upper limit of approximately 0.08 flare events expected per subfield in each 600 sec exposure . given the extreme selection bias in kunkel s sample , it is not unexpected to discover that this estimate ( @xmath115 ) is much different than what we find ( @xmath116 ) . we also examine the stellar flare survey data of @xcite . utilizing astrographic plates covering @xmath117 deg@xmath8 at low galactic latitude to apparent @xmath1 magnitude of 1014 , they find @xmath118 of their stars show flare events of @xmath119 mag over 2030 minutes . if half of these stars are m dwarfs , of which approximately @xmath120 are types dm46 based on the volume of the sample limited by the luminosity of those spectral types , we expect an upper limit of 0.01 flare of @xmath119 mag per 100 stars in each 600 sec @xmath1band exposure . having analyzed 425 accumulated @xmath1band images for this publication , we expect an upper limit of 6.4 flaring events @xmath119 magnitudes . the distribution function of the amplitudes of flaring events in these stars is not well known , but in general the brighter flares make up a smaller fraction of the total number of events @xcite . thus the discovery of only one certain flare event of @xmath121 mag is consistent with this upper limit . given the lack of unbiased flare star samples down to our limiting magnitudes , a direct comparison with other surveys requires the analysis of more modern variability datasets ( e.g fsvs : ) . @xcite have reported a highly luminous ot event from the sloan digital sky survey , with an underlying host galaxy at @xmath122 . @xcite have determined , using in part archival optical and radio data , that the event is likely due to a radio loud agn . heeding the suggestions of @xcite , we search archival data sources for observations of the fields of ot 20010326 and ot 20030305 . we include in these searches the first 1.4 ghz radio catalog @xcite , 1.4 ghz nvss radio catalog @xcite , and 4.8 ghz pmn catalogs @xcite . the only survey that covers the field of ot 20010326 is the pmn tropical survey , which detects no sources closer than @xmath123 . in combination with our vla detection limits 4 days after the event , this argues against a radio variable source . null detections are also reported for ot 20030305 in the first , nvss , and pmn equatorial catalogs , with an explicit limit from first of 0.94 mjy / beam . overall , these null detections seem to reject radio loud agn as sources for this variability . we have also searched the archival x ray and gamma ray catalogs through heasarc , and while ot 20010326 returns several matches corresponding to components of abell 1836 , no matches are found within @xmath124 . a search around ot 20030305 finds no matches within @xmath125 . this weighs against a variety of accretion scenarios , including agn and other galactic x ray sources . there is an expected population of optical transients associated with the observed rate of gamma ray bursts . however , the connection between the two event rates is very uncertain , given the variety of means to draw optical emission from hydrodynamic evolution of a grb event . these means include `` orphan '' grbs , where a highly collimated grb points at large enough angle away from the observer that gamma ray emission is avoided but optical afterglow radiation is not @xcite ; and `` dirty '' or `` failed '' fireballs whose ejecta comprise a significant amount of baryons and/or have a small lorentz factor ( @xmath126 100 , e.g. , ) . distinguishing between these possibilities requires fine constraints on the time decay slope @xmath66 , radio observations , and/or multiwavelength observations at the time of the so called grb `` jet break '' @xcite . given the our poor temporal resolution , and lack of simultaneous multi wavelength coverage , we are not uniquely sensitive to the micro physics expected to drive the early evolution of grb lightcurves , and whose resolution might distinguish between the above possibilities . within the resolution and timescale of our observations , all grb variants effectively yield the same characteristic isolated fading object . the thousand second timescale of the observed events is of similar duration to the early lightcurve break observed in grb 021211 @xcite . after this break , the power law flux decay index @xmath66 was seen to decrease from 1.8 to 0.8 before this time , the transient dimmed by more than 2.5 magnitudes . if our already faint , rapidly declining optical transients represent a similar early phase of gamma ray dark grb lightcurves , their subsequent evolution would be below our detection limit . we also note that the spectral energy distribution ( sed ) of the grb afterglow component is modeled as a synchrotron emission spectrum ( e.g. , ) . the expected spectral shape scales like @xmath127 . thus it is surprising that our distribution of @xmath1:@xmath19:@xmath5detected transients is 2:1:0 ( where ot 20030305 contributes to both the @xmath1 and @xmath19 event rate ) if these events are drawn from a population of events characterized by a synchrotron sed . recent submillimeter and radio observations of localized grb host galaxies show they are bluer than selected galaxies at similar redshifts , indicative of star formation or relatively low dust content ( e.g. , ) . host galaxies for localized x ray flashes ( xrfs ) also appear to exhibit the same characteristics @xcite . however , our putative hosts are considerably redder than these samples of grb or xrf selected host galaxies , and are unlikely to be drawn from the same population . the @xmath108 colors of our ot hosts are , however , loosely consistent with the observed population of extremely red objects ( eros ) , which comprise old or dusty star forming galaxies . if eros represent highly obscured starburst activity at moderate redshift ( @xmath128 ) , they contribute a significant fraction of the overall star formation at that epoch @xcite . assuming our population of ots traces star formation , we expect some association with eros . the absence of extremely red grb host galaxies suggests detection of grbs is biased against dusty galaxies . if our ots arise from dust enshrouded galaxies , this bias would seem lesser for optically detected events . we map our detection of these objects into a quantitative statement describing the rate of short timescale astrophysical variability . we derive general constraints without bias towards a particular ot model . to account for inefficiencies in our pipeline , we model our efficiency @xmath129 at recovering transients throughout a range in brightness . we calibrate the efficiency via monte carlo runs in which we add point source transients . we define an efficiency run as the generation of 32 random amplitude point sources , with the integrated flux chosen as a uniform deviate in the exponent of @xmath130 , which are placed within a dither sequence of survey data . positions are randomly selected within the limits of the primary exposure , and are placed at the same astronomical position within each of the subsequent 4 dithers . we do not bias our results by assuming a temporal shape for the variability . instead , we generate an overall efficiency for recovering a given amount of input flux , averaged over the spatial coverage of our dither sequence . we include in our analysis @xmath1 , @xmath19 , and @xmath5band survey data ( @xmath20band data are infrequently reduced in real time due to fringing complications ) . we account for variations in observing conditions by choosing 19 night subfield combinations of survey data to submit for efficiency analysis . in total , we initiated @xmath131 efficiency runs yielding @xmath132 total efficiency points , approximately 85000 per passband . the magnitudes of these transients are calibrated to the @xcite system . image zero point offsets were calculated from directly calibrated images of each subfield . one subfield was calibrated using publicly available noao deep wide field survey data @xcite . after placing additional flux in each image , we pass them through our difference imaging pipeline . the number of input objects recovered is tallied , searching both the unfiltered and filtered object catalogs generated at the fi stage in our pipeline . this yields an unfiltered detection efficiency ( _ dotted _ histogram in figure [ fig - eff ] ) as well as the efficiency of passing our cuts ( _ solid _ histogram in figure [ fig - eff ] ) . the difference between the two histograms demonstrates the rejection of true positives in our filtering process , yielding an overall decrease in efficiency . we use the _ solid _ histogram in the subsequent event rate analysis . we note that approximately @xmath133 of our sky coverage lands in gaps between mosaic chips , and the dithering procedure shifts @xmath134 and @xmath135 of the template images common sky area off of the subsequent dithers . thus our maximum possible efficiency is @xmath136 . our overall ot detection efficiency of @xmath137 indicates our filtering efficiency is @xmath138 . we expect substantial improvements in the context of a future automated classification environment . figure [ fig - eff ] shows our efficiency is fairly constant between the saturation limit on the bright end , and our detection limits on the faint end . we thus expect little bias against faint transient detection in this given range . on the bright end , there appears to be non zero @xmath129 for objects brighter than saturation . these are systematic artifacts that result from the detection of unsaturated and unmasked wings of saturated efficiency objects , in close enough proximity to the input object itself to warrant a positional match . in the following analysis , we set @xmath139 brighter than the average saturation limits of @xmath140 and @xmath141 . to average over unknown inefficiencies in the human element of our transient pipeline , we require that a transient be confirmed in a subsequent image , which we consider a human observer @xmath142 efficient at classifying as real . thus our limits on short timescale variability correspond to twice our typical exposure time , plus @xmath79 of readout between images . our efficiencies in figure [ fig - eff ] are relatively constant between the bright and faint limits , which we can approximate by a constant efficiency @xmath143 without loss of specificity . however , due to the differing @xmath144 per passband , differences in typical exposure times , and in number of events detected , we also quote event rates for each passband individually . the observed rate for each passband , averaged over all combinations of detectable variability amplitudes , is @xmath145 where @xmath146 is the number of observed events , @xmath24 the appropriate exposure , and @xmath147 is the averaged efficiency , squared due to the requirement of two detections . we note that ot 20030305 contributes to both the @xmath19 and @xmath1 event rate , as it was discovered in both sets of images . for the @xmath1 passband with 3 detected transients , our overall rate for variability on @xmath148 timescales , between @xmath149 , is @xmath150 events deg@xmath151 . we recognize at least one of these events as galactic in origin . having detected no more than two cosmological events , poisson statistics exclude at the @xmath3 level any ot model that predicts a mean number of detectable cosmological events @xmath152 , implying @xmath153 events deg@xmath151 . table [ tab - rate ] lists , for a given number of considered events , experimental rates and limits on @xmath154 . we have detected zero short timescale events in the @xmath5 passband , constraining overall @xmath155 astronomical variability between @xmath156 to @xmath157 events deg@xmath151 . in our total summed exposure of @xmath25 deg@xmath8days from all passbands , assuming @xmath158 and 4 events ( 3 unique events ) , we find an overall rate of short timescale astronomical variability of @xmath159 , with @xmath160 . these are the first general constraints on short timescale variability at such depths . we have reported on the structure of and first results from our wide field image subtraction pipeline . an important characterization of our transient survey is the exposure @xmath24 at a given timescale and to a given depth . the dls transient search is primarily sensitive to @xmath161 variability from @xmath162 to deeper than @xmath163 magnitudes in @xmath1 , @xmath19 , and @xmath5 . within this envelope of sensitivity , we have detected three short timescale optical transient events . ot 20010326 occurred in the field of galaxy cluster abell 1836 . one archival hst image of the cluster includes this region , and the transient precursor is present and unresolved . this indicates a compact precursor , stellar in nature if it resides in our galaxy . the colors of this object are consistent at the @xmath110 level with those expected of galactic dwarf stars ( figure [ fig - cmds ] ) , whose flaring activity presents a known background . however , the object would be too far out of the galactic plane to belong to the disk population of dwarf stars , and halo subdwarfs are not known to flare . it is also possible the precursor resides in , or behind , the abell cluster , but overall its nature remains uncertain . ot 20020115 is identified as a galactic m dwarf of spectral type dm4 , exhibiting classical flare star activity . finally , the host for ot 20030305 appears consistently elliptical in the @xmath5 and @xmath101 passbands , and inconsistent with the @xmath5band stellar psf , which we consider a strong argument for a resolved host , extragalactic in nature . we find ot 20030305 the strongest candidate so far for optically detected , short timescale cosmological variability . the precursor or host objects for ots 20010326 and 20030305 are definitively redder than galaxies that are known to host grbs and xrfs . if our events are cosmological in nature , this suggests that optical and high energy events arise from different mechanisms , or , given the dearth of grbs from dusty star forming galaxies , a stronger bias against grbs and xrfs from a dusty environment . if our lightcurves evolve analogous to the prompt stage of grbs emission , the rapid decline coupled with their intrinsic faintness would make them difficult to monitor beyond several hours finally , we emphasize the diversity of variable objects in the stellar menagerie , and thus it is not straightforward to rule out galactic stars as the sources for our ots . spectroscopic information will ultimately help to clarify their nature , and such followup observations are planned . our search has also yielded sn like events that appear to have no host galaxy to significant ( @xmath164 ) limiting magnitude ot 20020112 is one such example . in addition , our catalog of sn candidates , classified as supernovae primarily due to their proximity to a host galaxy , might also contain sources with unusual temporal evolution . overall , the dls transient search is well suited to explore the parameter space of ots with small energy budgets , a population that could plausibly have escaped detection by gamma ray and x ray satellite missions . this raises the possibility that the phenomena detected represent a new class of astronomical variability . coordinated photometric followup of future optical transients is absolutely necessary to reveal fully the diversity of variability at faint optical magnitudes . a primary goal of future variability surveys must be to enable the photometric and spectroscopic followup of detected events through timely release of information and ease of access to available data . the wealth of information that can be gleaned from real time synoptic transient science , only a subset of which is covered in this paper , strengthens the science case for the expansion of deep , wide astronomical surveys into the short timescale regime . expected future surveys like the lsst and pan starrs @xcite will survey thousands of square degrees per night , and will benefit greatly from modern development in this field . we thank the many observers who have assisted in reviewing transients during the course of our survey , including m. lopez - caniego alcarria , p. baca , h. khiabanian , j. kubo , d. loomba , c. navarro , and r. wilcox . we thank g. bernstein for orbit calculations , r. sari and s. hawley for useful discussions , and g. bravo , m. hamui , d. kirkman , c. smith , and c. stubbs for assistance . we are very grateful for the skilled support of the staff at ctio and kpno . the deep lens survey is supported by nsf grants ast-0307714 and ast-0134753 . ac acknowledges the support of conicyt , through grant fondecyt 1000524 . this work made use of images and/or data products provided by the noao deep wide - field survey , which is supported by the national optical astronomy observatory ( noao ) . noao is operated by aura , inc . , under a cooperative agreement with the national science foundation . this research has made use of nasa s astrophysics data system bibliographic services . this work made use of images obtained with the magellan i ( baade ) telescope , operated by the observatories of the carnegie institution of washington . includes observations made with the nasa / esa hubble space telescope , obtained from the data archive at the space telescope science institute . stsci is operated by the association of universities for research in astronomy , inc . under nasa contract nas 5 - 26555 . , g. , adam , g. , antilogus , p. , astier , p. , bacon , r. , bongard , s. , bonnaud , c. , copin , y. , hardin , d. , henault , f. , howell , d. a. , lemonnier , j. , levy , j. , loken , s. c. , nugent , p. e. , pain , r. , pecontal , a. , pecontal , e. , perlmutter , s. , quimby , r. m. , schahmaneche , k. , smadja , g. , & wood - vasey , w. m. 2002 , in survey and other telescope technologies and discoveries . edited by tyson , j. anthony ; wolff , sidney . proceedings of the spie , volume 4836 , pp . 61 - 72 ( 2002 ) . , 6172 llll|llll ot 20020112 & f4p13 & 10:49:32.8 & -04:31:45 & @xmath165 & @xmath166 & @xmath167 & @xmath168 + & & & & @xmath166 & @xmath169 & @xmath170 & @xmath171 + & & & & & & & + ot 20010326 & f5p31 & 14:01:42.2 & -11:35:15 & @xmath172 & @xmath173 & @xmath174 & @xmath175 + & & & & @xmath176 & @xmath177 & @xmath178 & @xmath179 + & & & & & & & + ot 20020115 & f4p23 & 10:48:56.1 & -05:00:41 & @xmath180 & @xmath97 & @xmath181 & @xmath182 + & & & & & & & + ot 20030305 & f4p31 & 10:53:45.8 & -05:37:44 & @xmath166 & @xmath183 & @xmath184 & @xmath185 + & & & & @xmath169 & @xmath169 & @xmath186 & @xmath187 + lllll ot 20020112 & 52285.3 & v & @xmath188 & @xmath189 + & 52286.2 & b & @xmath190 & @xmath191 + & 52287.2 & r & @xmath192 & @xmath193 + & 52287.4 & v & @xmath194 & @xmath195 + & 52288.2 & r & @xmath196 & @xmath197 + & 52288.2 & b & @xmath198 & @xmath199 + & 52288.3 & v & @xmath200 & @xmath201 + & 52290.2 & b & @xmath202 & @xmath203 + & 52290.3 & v & @xmath204 & @xmath200 + & 52319.2 & v & @xmath205 & @xmath206 + & 52321.3 & b & @xmath207 & @xmath208 + ot 20010326 & 51994.2315949 & b & @xmath171 & @xmath209 + & 51994.2397245 & b & @xmath210 & @xmath211 + & 51994.2478576 & b & @xmath212 & @xmath213 + & 51994.2646898 & r & @xmath214 & @xmath215 + ot 20020115 & 52289.2963074 & b & @xmath216 & + & 52289.3044363 & b & @xmath217 & + & 52289.3125651 & b & @xmath218 & + & 52290.2594999 & b & @xmath171 & + ot 20030305 & 52703.1732501 & v & @xmath219 & @xmath220 + & 52703.1813791 & v & @xmath221 & @xmath222 + & 52703.1895094 & v & @xmath223 & @xmath224 + & 52703.1995017 & b & @xmath225 & @xmath226 + & 52703.2076305 & b & @xmath227 & @xmath228 + & 52703.2157589 & b & @xmath229 & @xmath230 + & 52703.2249089 & b & @xmath231 & @xmath227 + & 52703.2330383 & b & @xmath232 & @xmath233 + & 52705.3304513 & v & @xmath234 & @xmath219 llclllcccc b & 1300s & 1.1 & 18.6 & 23.8 & 0.65 & 3 & 7.8 & 6.5 & 17 + & & & & & & 2 & 6.3 & 4.3 & 14 + & & & & & & 1 & 4.7 & 2.2 & 10 + & & & & & & 0 & 3.0 & 0.0 & 6.5 + v & 1300s & 1.2 & 18.8 & 23.3 & 0.63 & 1 & 4.7 & 2.1 & 9.9 + & & & & & & 0 & 3.0 & 0.0 & 6.3 + r & 1900s & 1.5 & 19.5 & 23.4 & 0.62 & 0 & 3.0 & 0.0 & 5.2 + b+v+r & & 3.7 & & & 0.63 & 4 & 9.2 & 2.7 & 6.3 + & & & & & & 3 & 7.8 & 2.0 & 5.3 + & & & & & & 2 & 6.3 & 1.4 & 4.3 + & & & & & & 1 & 4.7 & 0.7 & 3.2 + & & & & & & 0 & 3.0 & 0.0 & 2.0

we critically compare the practicality and accuracy of numerical approximations of phase field models and sharp interface models of solidification . particular emphasis is put on stefan problems , and their quasi - static variants , with applications to crystal growth . new approaches with a high mesh quality for the parametric approximations of the resulting free boundary problems and new stable discretizations of the anisotropic phase field system are taken into account in a comparison involving benchmark problems based on exact solutions of the free boundary problem . = 1 phase field models , parametric sharp interface methods , stefan problem , anisotropy , solidification , crystal growth , numerical simulations , benchmark problems 35k55 , 35r35 , 35r37 , 53c44 , 65m12 , 65m50 , 65m60 , 74e10 , 74e15 , 74n20 , 80a22 , 82c26

1212.2077

the solidification of a liquid or the melting of a solid lead to complex free boundary problems involving many different physical effects . for example , latent heat is set free at the interface and a competition between surface energy and diffusion leads to instabilities like the mullins sekerka instability . the resulting model is a stefan problem with boundary conditions taking surface energy effects and kinetic effects at the interface into account , see e.g. @xcite . crystals forming in an undercooled melt lead to very complex patterns and , in particular , dendritic growth can be observed since the growth is typically diffusion limited , see @xcite . the numerical simulation of time - dependent stefan problems , or variants of it , is a formidable task since the evolving free boundary has to be computed . hence , direct front tracking type numerical methods need to adequately capture the geometry of the interface and to evolve the interface approximation , often with a coupling to other physical fields . this coupling , in particular , represents a significant initial hurdle towards obtaining practical implementations , and thus numerical simulations for the problem at hand . examples of the implementation of such direct methods can be found in e.g. @xcite . a further drawback of direct front tracking methods has been the inability of most direct methods to deal with so - called mesh effects , or to prevent them altogether . when a discrete approximation of an interface , for example a polygonal curve in the plane , evolves in time , then in general it is possible for the approximation to deteriorate or to break down . examples of such pathologies are self - crossings and vertex coalescence . while for simple isotropic problems in the plane these issues can be dealt with , for example by frequent remeshings or by using clever formulations that only allow equidistributed approximations , see e.g. @xcite , until very recently there has been no remedy for fully anisotropic problems in two and three space dimensions . however , building on their work for isotropic problems in @xcite , the present authors recently introduced stable parametric finite element schemes for the direct approximation of anisotropic geometric evolution equations in @xcite , for which good mesh properties can be guaranteed . in particular , even for the simulation of interface evolutions in the presence of strong anisotropies , no remeshing or redistribution of vertices is needed in practice . these schemes , in which only the interface without a coupling to bulk quantities is modelled , have been extended to approximations of the stefan problem with fully anisotropic gibbs thomson law and kinetic undercooling in @xcite . the novel method from @xcite can be shown to be stable and to have good mesh properties . we remark that these approaches extend earlier ideas from @xcite . here we recall the pioneering work of schmidt @xcite , where the full stefan problem in three dimensions was solved within a sharp interface framework for the first time . phase field methods are an alternative approach to solve solidification phenomena in the framework of continuum modelling . in phase field approaches a new non - conserved order parameter @xmath0 is introduced , which in the two phases is close to two different prescribed values and which smoothly changes its value across a small diffuse interfacial region . a parabolic partial differential equation for @xmath0 is then coupled to an energy balance , which results in a diffusion equation for the temperature taking latent heat effects into account . we refer to @xcite and to the five review articles @xcite for further details . in particular , it can be shown that solutions to the phase field equations converge to classical sharp interface problems , see e.g. @xcite and the references therein . the popularity of phase field methods , often also called diffuse interface methods , can be explained by two characteristic features that they share with the level set method , which is another sharp interface computational tool . firstly , phase field methods naturally allow for changes of topology . and secondly , computing simulations for phase field models only requires the solution of partial differential equations on standard cartesian domains . the fact that these can usually be implemented and solved in a relatively straightforward way makes the phase field method particularly appealing . it is the aim of this article to investigate the practical merits of phase field models compared to the recently introduced sharp interface algorithm for the approximation of stefan problems from @xcite ; see also @xcite . of particular interest will be the relative accuracy of the two methods , in situations where a true solution to the sharp interface problem is known . in a phase field simulation the observed error is made up of contributions from * the asymptotic error , * the spatial and temporal discretization errors , * rounding errors and solver tolerances . here the asymptotic error is induced by the choice of interfacial parameter @xmath1 . in general one can formally show that the asymptotic error converges to zero as @xmath2 , see e.g. @xcite . for certain phase field models and under certain conditions it can be rigorously shown that the asymptotic error vanishes as @xmath3 , see e.g. @xcite . in computations for sharp interface approximations , on the other hand , the observed error is made up of the latter two contributions only , i.e. of discretization and rounding errors . a disadvantage of phase field models is that the resulting pdes become stiff for decreasing @xmath4 , leading to a requirement for very fine spatial and temporal discretizations . hence it becomes computationally challenging to produce very accurate phase field simulations . in any case , the available computational resources will often set a limit on the smallest interfacial parameter @xmath4 that one can compute for . hence another aspect that needs to be taken into account in an objective comparison between phase field simulations and sharp interface approximations is the overall cpu time that is needed to obtain the results . while it can often be formally shown that phase field computations can attain an arbitrarily high accuracy , the existing limitations on computer hardware often mean that in practice very fine computations can not be performed . in addition , as discussed in @xcite , the early computational approaches were limited as they could only be used in the presence of interfacial kinetics in the gibbs thomson law . historically these limitations of the phase field model have long been known , and as a result a different underlying interpretation of the model , the so - called `` thin interface limit '' , has been introduced and analyzed by karma and rappel @xcite . their approach made it possible to do efficient computations with a smaller capillary length to interface thickness ratio , and to study the physically relevant case of small or zero kinetic coefficients . later the findings of karma and rappel were reinterpreted as second order convergence with respect to the interfacial thickness , see @xcite . the first successful phase field computations of dendritic growth were performed by kobayashi @xcite , and his computations demonstrated the importance of anisotropy for dendritic growth . since then many successful improvements with respect to numerical simulations have appeared in the literature . we refer to @xcite and the references therein . finally , we would like to mention work on the numerical analysis of phase field and sharp interface approaches . numerical analysis of discretizations of phase field models can be found in e.g.@xcite . numerical analysis of discretizations of sharp interface models can be found in @xcite . we also remark that level set methods are another possible way to solve stefan problems and related free boundary problems . we refer to @xcite for more details on how the level set method can be used to solve free boundary problems . the remainder of the paper is organized as follows . in section [ sec:2 ] we state the sharp interface formulation of the two phase stefan problem with kinetic undercooling and an anisotropic gibbs thomson law . in section [ sec:3 ] we state the corresponding phase field model and recall the finite element algorithms from @xcite . in section [ sec:4 ] we numerically compare the sharp interface method from @xcite with the phase field algorithms from section [ sec:3 ] for some isotropic benchmark problems with known true solutions . computations for a phase field model with a correction term , for which a second order convergence property can formally be shown , are presented in section [ sec:43 ] . finally , we compare the sharp interface and phase field methods for a variety of anisotropic model problems in section [ sec:5 ] . in this paper we concentrate on interfacial problems in materials science in which one driving force is due to capillary effects . in applications the interface often separates a solid and a liquid phase , say , or a solid phase and a gas phase . let @xmath5 , @xmath6 , denote this sharp interface . then the surface energy of @xmath7 is defined as @xmath8 where @xmath9 denotes the unit normal of @xmath7 , and where the anisotropic density function @xmath10 with @xmath11 is assumed to be absolutely homogeneous of degree one , i.e.@xmath12 with @xmath13 denoting the gradient of @xmath14 . for all the considerations in this paper we assume that @xmath14 is of the class of anisotropies first introduced by the authors in @xcite ; see also @xcite . relevant for our considerations is the first variation , @xmath15 , of ( [ eq : egamma ] ) , which can be computed as @xmath16 where @xmath17 is the tangential divergence of @xmath18 , see e.g.@xcite . note that @xmath19 reduces to @xmath20 , the sum of the principal curvatures of @xmath18 , in the isotropic case , i.e. when @xmath14 satisfies @xmath21 then the full stefan problem we want to consider in this paper is given as follows , where @xmath22 is a given fixed domain with boundary @xmath23 and outer normal @xmath24 . find @xmath25 \to { { \mathbb r}}$ ] and the interface @xmath26}$ ] such that for all @xmath27 $ ] the following conditions hold : @xmath28_{\gamma(t ) } & = -\lambda\,{\cal v } \qquad & & \mbox{on } \gamma(t)\ , , \label{eq:1b } \\ \frac{\rho\,{\cal v}}{\beta({{}{\rm n } } ) } & = \alpha\,\kappa_\gamma - a\,u \qquad & & \mbox{on } \gamma(t)\ , , \label{eq:1c } \\ \frac{\partial u}{\partial { } \nu } & = 0 \qquad \mbox{on } \partial_n\omega , \qquad u = { u_d}\qquad & & \mbox{on } \partial_d \omega \ , , \label{eq:1d } \\ \gamma(0 ) & = \gamma_0 \ , , \qquad\qquad \vartheta\,u(\cdot,0 ) = \vartheta\,u_0 \quad & & \mbox{in } \omega \ , . \label{eq:1e } \end{aligned}\ ] ] in the above @xmath29 denotes the deviation from the melting temperature @xmath30 , i.e. @xmath30 is the melting temperature for a planar interface . in addition , @xmath31 is the solid region , with boundary @xmath32 , so that the liquid region is given by @xmath33 . moreover , here and throughout this paper , for a quantity @xmath34 defined on @xmath35 , we use the shorthand notations @xmath36 and @xmath37 . the parameters @xmath38 , @xmath39 , @xmath40 , @xmath41 , @xmath42 are assumed to be constant , while @xmath43 is assumed to be constant in each phase . the mobility coefficient @xmath44 is assumed to satisfy @xmath45 for all @xmath46 and to be positively homogeneous of degree one . we note that in the isotropic case ( [ eq : iso ] ) it is often also assumed that @xmath47 in addition @xmath48_{\gamma(t)}({}{z } ) : = ( \conduct_+\,\frac{\partial u_+}{\partial { { } { \rm n } } } - \conduct_-\,\frac{\partial u_-}{\partial { { } { \rm n } } } ) ( { } { z})$ ] for all @xmath49 , and @xmath50 is the velocity of @xmath7 in the direction of its normal @xmath9 , which from now on we assume is pointing into @xmath51 . finally , @xmath52 with @xmath53 , @xmath54 is the applied supercooling at the boundary , and @xmath55 and @xmath56 are given initial data . here we use the convention that @xmath57 if @xmath58 . the model ( [ eq:1a]e ) can be derived for example within the theory of rational thermodynamics and we refer to @xcite for details . we remark that a derivation from thermodynamics would lead to the identity @xmath59 . we note that ( [ eq:1b ] ) is the well - known stefan condition , while ( [ eq:1c ] ) is the gibbs thomson condition , with kinetic undercooling if @xmath60 . the case @xmath61 , @xmath60 , @xmath62 leads to the stefan problem with the gibbs thomson law and kinetic undercooling . in some models in the literature , see e.g. @xcite , the kinetic undercooling is set to zero , i.e. @xmath63 . setting @xmath64 but keeping @xmath41 leads to the mullins sekerka problem with the gibbs thomson law , see @xcite . we recall from @xcite that for a solution @xmath29 and @xmath18 to ( [ eq:1a]e ) it can be shown that the following equality holds @xmath65 where @xmath66 and where @xmath67 denotes the @xmath68inner product over @xmath35 , with the corresponding norm given by @xmath69 , and where @xmath70 . traditional front tracking methods for sharp interface problems had a major drawback , in that the meshes used to describe the interface seriously deteriorated during the evolution . in addition , introducing mesh smoothing during the evolution is difficult , see e.g. the discussion in @xcite . for interfaces in the plane it is possible to formulate a non - trivial method such that mesh points are nearly equally distributed during the evolution , see @xcite . the present authors introduced a novel parametric finite element method for problems involving curves and surfaces evolving in time , which has a simple variational structure and which leads to good mesh properties , see @xcite . in fact , for curves a semi - discrete variant leads to equally distributed mesh points in the isotropic case , while in the general anisotropic setting equidistribution with respect to some anisotropic weight function is obtained , see @xcite . for surfaces the resulting meshes have also good properties , which in the isotropic case can be explained by using ideas from conformal geometry . in particular , no remeshing is needed during the evolution , even in the general anisotropic situation . an example triangulation obtained during the simulation of dendritic growth in three space dimensions can be seen in figure [ fig : mesh ] , below . in addition , as the mesh for the parameterization of the interface is decoupled from the bulk mesh , no deformation of the bulk mesh is required in order to contain the interface at predefined locations on it . the novel and stable parametric finite element approximation of ( [ eq:1a]e ) in the case @xmath71 has been introduced by the present authors in @xcite , and this scheme has been extended to the more general case @xmath72 in @xcite . throughout this paper we will refer to these variants as pfem . the algorithm pfem features the discretization parameters @xmath73 , @xmath74 , @xmath75 and @xmath76 . here @xmath73 refers to the fineness of the triangulated approximation of @xmath7 , for which isoparametric piecewise linear finite elements are employed . in particular , a simple mesh refinement strategy allows for the natural growth of the interface , i.e. elements of the triangulated approximation of @xmath7 are refined when they become too large . moreover , the temperature in the bulk is approximated with standard continuous piecewise linear finite elements , and @xmath74 and @xmath75 refer to bulk mesh parameters for fine regions close to the interface and coarser regions far away from it . for all the computations presented in this paper we fix @xmath77 and , unless stated otherwise , we let @xmath78 . finally , @xmath76 denotes a uniform time step size . the linear discrete systems of equations are solved with preconditioned conjugate gradient solvers of suitable schur complement formulations . we refer to @xcite for more details . as indicated earlier , no remeshing of the discrete interface is necessary for the scheme pfem , and all the numerical results presented in this paper for this scheme are performed without any redistribution of mesh points . we now state the phase field model that we are going to consider in this paper . to this end , for @xmath79 , let @xmath80 and define @xmath81 where @xmath82 is a constant satisfying @xmath83 . moreover , let @xmath84 be the phase field variable , so that the sets @xmath85 are approximations to @xmath86 , with the zero level set of @xmath87 approximating the interface @xmath7 . on introducing the small interfacial parameter @xmath1 , it can be shown that @xmath88 for @xmath4 sufficiently small , where @xmath89 here @xmath90 $ ] is a double well potential , which we assume to be symmetric and to have its global minima at @xmath91 . the canonical example is @xmath92 another possibility is to choose @xmath93 see e.g. @xcite . clearly the obstacle potential ( [ eq : obstacle ] ) , which in contrast to the smooth quartic potential ( [ eq : quartic ] ) forces @xmath0 to stay within the interval @xmath94 $ ] , is not differentiable at @xmath91 . hence , whenever we write @xmath95 in the case ( [ eq : obstacle ] ) in this paper , we mean that the expression holds only for @xmath96 , and that in general a variational inequality needs to be employed . our phase field model for ( [ eq:1a]e ) is then given by the coupled system @xmath97 with @xmath98 where @xmath99 and where the function @xmath100 is such that @xmath101\,,\quad \int^1_{-1 } \varrho(y ) \ ; { \rm d}y = 1 \quad \text{and } \quad { { \rm p}}(s ) : = \int^s_{-1 } \varrho(y ) \ ; { \rm d}y\,.\ ] ] we note that @xmath102 , which is a monotonically increasing function over the interval @xmath94 $ ] with @xmath103 and @xmath104 , is often called the interpolation function . in this paper , we follow the convention from @xcite , where @xmath105 is called the shape function . possible choices of @xmath106 that will be considered in this paper are @xmath107 more details on interpolation functions @xmath102 , respectively shape functions @xmath106 , can be found in e.g.@xcite . in particular , if one also assumes that @xmath106 is symmetric , i.e. @xmath108 and that @xmath109 then a faster rate of convergence of the phase field model to the sharp interface limit , as @xmath3 , can be shown in the isotropic case ( [ eq : iso ] ) , ( [ eq : betaiso ] ) on replacing @xmath110 in ( [ eq : aca ] ) with the first order correction @xmath111 where @xmath112 is defined in ( [ eq : rho1 ] ) in section [ sec:43 ] , below . the condition ( [ eq : varrhoroot ] ) is one motivation for the latter two choices in ( [ eq : varrho ] ) , with the choice ( [ eq : varrho])(iv ) also satisfying the stronger condition ( [ eq : varrhopsi ] ) , below , for the quartic potential ( [ eq : quartic ] ) . an error analysis for a fully discrete approximation of the phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) with the quartic potential ( [ eq : quartic ] ) and the shape function ( [ eq : varrho])(i ) in the isotropic case ( [ eq : iso ] ) , ( [ eq : betaiso ] ) with @xmath113 and @xmath114 has been performed in @xcite . these authors also show convergence of the phase field discretizations to the underlying sharp interface problem as @xmath115 , where @xmath116 and @xmath76 denote the discretization parameters in space and time , respectively . however , to our knowledge , no convergence rates are known for the convergence of discretizations of the phase field model to the sharp interface problem ( [ eq:1a]e ) . here we recall that for the much simpler situation of planar curvature flow , as the sharp interface limit of the isotropic allen cahn equation , such convergence rates have been obtained in @xcite . in particular , it can be shown that the zero level sets of discretizations of @xmath117 for the obstacle potential ( [ eq : obstacle ] ) converge with @xmath118 to the sharp interface limit moving by @xmath119 if @xmath120 while no such result is known for the full phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) even in the isotropic setting , it is natural to expect constraints of the form ( [ eq : nv97a ] ) in order to observe @xmath118 in practice . we remark that the phase field analogue of the sharp interface energy identity ( [ eq : testd ] ) is given by the formal energy bound @xmath121 for the phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) , where @xmath122 phase field models that satisfy such an inequality , in analogy to the sharp interface energy identity ( [ eq : testd ] ) , are often called thermodynamically consistent , see @xcite . unconditionally stable , fully practical finite element approximations of ( [ eq : heata]d ) , with either ( [ eq : quartic ] ) or ( [ eq : obstacle ] ) have been introduced by the authors in @xcite . here stable means that they satisfy a discrete analogue of the formal energy bound ( [ eq : pf2lyap ] ) . throughout this paper we will refer to the approximations from @xcite for ( [ eq : quartic ] ) and ( [ eq : obstacle ] ) as pf@xmath124-fem and pf@xmath123-fem , respectively , where the inclusion of a subscript refers to the choice of shape function in ( [ eq : varrho ] ) , e.g. pf@xmath125-fem . we recall from @xcite that a side effect of the interpolation function @xmath102 in ( [ eq : fg ] ) is that the function @xmath126 need no longer have local minima at @xmath127 if @xmath128 . this can result , for example , in undesired , artificial boundary layers for strong supercoolings , i.e. when @xmath129 is large . for the smooth potential @xmath130 from ( [ eq : quartic ] ) , sufficient conditions for @xmath131 to be local minimum points of @xmath132 are @xmath133 which is evidently satisfied by ( [ eq : varrho])(iii ) . in fact , in applications phase field models for solidification almost exclusively use this shape function ; see e.g. @xcite . for the obstacle potential ( [ eq : obstacle ] ) the situation is similar , although there is more flexibility in the possible choices of @xmath106 . in particular , here a sufficient condition for @xmath132 to have local minima at @xmath131 is given by @xmath134 on recalling that @xmath135 we see that for ( [ eq : obstcond ] ) to hold it is sufficient to require that @xmath136 , which is evidently satisfied by ( [ eq : varrho])(ii ) , ( [ eq : varrho])(iii ) and ( [ eq : varrho])(iv ) . a major advantage of ( [ eq : varrho])(ii ) over ( [ eq : varrho])(iii ) and ( [ eq : varrho])(iv ) is that for ( [ eq : varrho])(ii ) it is possible to derive almost linear finite element approximations that are unconditionally stable . the corresponding unconditionally stable schemes for the nonlinear shape functions ( [ eq : varrho])(iii ) and ( [ eq : varrho])(iv ) , on the other hand , turn out to be highly nonlinear . see @xcite for more details . we remark that even when ( [ eq : quarticcond ] ) and ( [ eq : obstcond ] ) are satisfied for the potentials ( [ eq : quartic ] ) and ( [ eq : obstacle ] ) , respectively , it is possible that mushy interfacial regions are observed in practice for approximations of the phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) ; see e.g. figure [ fig : stefanii_16pi ] , below . that is particularly the case in situations where the instability of the moving free boundary is strong , i.e. when @xmath137 is large , recall ( [ eq : f ] ) and see e.g. @xcite . then @xmath4 needs to be chosen small , recall ( [ eq : fg ] ) , so that the phase field variable @xmath0 admits well - defined interfacial regions that approximate the sharp interface @xmath7 . this gives rise to a formal constraint of the form @xmath138 for the choice of the interfacial parameter @xmath4 in terms of the physical parameters for the sharp interface problem ( [ eq:1a]e ) , irrespective of the choice of @xmath106 . the reason for this is that in the estimate ( [ eq : pf2lyap ] ) the double well term @xmath139 in @xmath140 is for large @xmath4 not strong enough to bound the unstable term involving @xmath141 , which encourages the growth of the diffuse interface . the two algorithms pf@xmath123-fem and pf@xmath124-fem , which use continuous piecewise linear finite elements in space , feature the discretization parameters @xmath74 , @xmath75 and @xmath76 . here @xmath74 and @xmath75 are mesh parameters for fine triangulations inside the diffuse interfacial region and coarser triangulations far away from it . meaningful phase field simulations need to resolve the interfacial regions , whose width is of the order @xmath4 , and so a constraint of the form @xmath142 needs to be enforced . for ( [ eq : obstacle ] ) the asymptotic interface width is @xmath143 in the isotropic case ( [ eq : iso ] ) , and in this paper we always choose @xmath144 with @xmath145 . unless otherwise stated we let @xmath146 . finally , @xmath76 denotes a uniform time step size . here we recall that the schemes pf@xmath123-fem and pf@xmath124-fem employ a semi - implicit discretization in time , which utilizes convex / concave splittings of the nonlinearity arising from the potential @xmath130 and from the interpolation function @xmath102 . such a splitting for @xmath130 was first proposed in @xcite , see also @xcite , and the idea generalizes naturally to @xmath102 ; see section [ sec:32 ] , below , for details . this means that for the shape function choices ( [ eq : varrho])(i ) and ( [ eq : varrho])(ii ) almost linear schemes are obtained , while the choices ( [ eq : varrho])(iii ) and ( [ eq : varrho])(iv ) give rise to more nonlinear finite element approximations ; see @xcite for details . the discrete systems of linear equations and variational inequalities arising from the schemes pf@xmath123-fem are solved with the uzawa - multigrid solver from @xcite , while the systems of nonlinear equations arising from pf@xmath124-fem are solved with a newton method . we refer to @xcite for more details . for completeness we briefly describe the choice of the initial profile @xmath147 in ( [ eq : acc ] ) in our numerical computations . given the initial interface @xmath148 from ( [ eq:1e ] ) , we let @xmath149 denote the signed distance function of @xmath148 . then , on recalling the asymptotic phase field profiles from e.g. @xcite , we define @xmath150 for the obstacle potential ( [ eq : obstacle ] ) , while for the smooth quartic potential ( [ eq : quartic ] ) we use @xmath151 for simplicity we use the profiles ( [ eq : phi0],b ) also in the anisotropic setting , where it would be more appropriate to replace @xmath152 with a suitably defined anisotropic distance function @xmath153 , see @xcite for details . finally , if @xmath154 , we fix @xmath155 . in section [ sec:4 ] we will investigate the accuracy and the efficiency of several discretizations of the phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) in the isotropic case ( [ eq : iso ] ) , ( [ eq : betaiso ] ) . in addition to the schemes pf@xmath123-fem and pf@xmath124-fem from @xcite , which use a semi - implicit discretization in time , we will also look at a more implicit discretization and at a fully explicit discretization . for later reference , we now state the three different time discretizations , and for simplicity we do so on ignoring spatial discretization . a strong formulation of the time discretization from @xcite is given as follows . let @xmath156 , with @xmath157 being convex on @xmath158 and @xmath159 being concave , and let @xmath160 be a similar splitting that is convex / concave on a suitable superset of @xmath94 $ ] , where we recall that @xmath0 need not remain in @xmath94 $ ] for the quartic potential ( [ eq : quartic ] ) . we also define @xmath161 . for the schemes pf@xmath123-fem and pf@xmath124-fem we set @xmath162 and @xmath163 for the choices of @xmath106 in ( [ eq : varrho ] ) . the semi - implicit time discretization employed by the schemes pf@xmath123-fem and pf@xmath124-fem can then be formulated as : @xmath164 with this time discretization existence of a unique solution @xmath165 to the fully discrete scheme can be shown for arbitrary time step sizes @xmath76 if @xmath166 , where @xmath167 may not be unique if @xmath168 and @xmath169 in very rare circumstances . moreover , any solution to the semi - implicit schemes pf@xmath123-fem and pf@xmath124-fem is stable ; see @xcite for details . the semi - implicit time discretization in ( [ eq : semi1],b ) can be modified to an implicit time discretization by replacing ( [ eq : semi2 ] ) with @xmath170 which then gives rise to a time step size constraint of the form @xmath171 in order to ensure the existence of a unique , stable solution in the case @xmath166 and @xmath172 . in the situation @xmath173 and @xmath174 , with @xmath169 , which has been treated by the present authors in @xcite , a stronger time step constraint of the form @xmath175 arises for the implicit discretization ( [ eq : impl ] ) ; see also @xcite . here we recall that it is often observed that implicit time discretizations of allen cahn and cahn hilliard type equations yield a better accuracy in time compared to semi - implicit time discretizations as in ( [ eq : semi1],b ) ; see e.g. @xcite . we will present several computations for a variant of pf@xmath124-fem with the implicit time discretization ( [ eq : semi1 ] ) , ( [ eq : impl ] ) in sections [ sec:4 ] and [ sec:5 ] . finally , fully explicit approximations , as advanced in e.g. @xcite , can also be considered . here we replace ( [ eq : semi1],b ) by @xmath176 if @xmath154 and @xmath172 , then the above fully explicit time discretization is well - defined , and in this case stability of the fully discrete scheme can be shown if @xmath177 in particular , in the case of the obstacle potential ( [ eq : obstacle ] ) , if @xmath178 , then the solutions to the fully discrete variant of ( [ eq : expl1],b ) are stable if @xmath179 where @xmath180 } |\varrho(s)|$ ] , @xmath181 and where @xmath182 is a constant only depending on the spatial mesh . the advantage of ( [ eq : expl1],b ) over ( [ eq : semi1],b ) is that the discretized systems of equations now decouple in space , which leads to huge efficiency gains when the computations are performed in parallel on a large cluster . however , in practice this advantage is often negated because ( [ eq : tauh ] ) , together with e.g. ( [ eq : hf ] ) , enforces that very small time steps need to be taken . in section [ sec:42 ] and in section [ sec:5 ] we will present computations for a variant of pf@xmath123-fem with an explicit time discretization as in ( [ eq : expl1],b ) . a standard validation used for phase field models in the literature is the comparison of tip velocities between the computed phase field discretizations and real world measurements from the laboratory , see e.g. @xcite . here the physical parameters in the phase field model have to be chosen appropriately , so that they correspond to the physical properties of the material in question . however , often the exact values of these parameters are unknown or they are themselves based on measurements . here we propose a much simpler quantitative validation , which makes use of known radially symmetric solutions to the underlying sharp interface problem in the isotropic case . we would argue that such a simple comparison should be part of the validation of every phase field method to be proposed in the literature . it gives an indication of the accuracy of the overall method and it helps to fine - tune the discretization parameters that should be used for the anisotropic physical applications . in particular , in this section we consider the following isotropic variant of ( [ eq:1a]e ) . find @xmath25 \to { { \mathbb r}}$ ] and the interface @xmath26}$ ] such that for all @xmath27 $ ] it holds that : @xmath183_{\gamma(t ) } & = -\mathcal{v } \qquad & & \mbox{on } \gamma(t ) , \label{eq : i1b } \\ \rho\,\mathcal{v } & = \alpha\,\kappa - u \qquad & & \mbox{on } \gamma(t ) , \label{eq : i1c } \\ \frac{\partial u}{\partial { } \nu } & = 0 \qquad \mbox{on } \partial_n\omega , \qquad u = { u_d}\qquad & & \mbox{on } \partial_d \omega \ , , \label{eq : i1d } \\ \gamma(0 ) & = \gamma_0 \ , , \qquad\qquad \vartheta\,u(\cdot,0 ) = \vartheta\,u_0 \quad & & \mbox{in } \omega \ , . \label{eq : i1e } \end{aligned}\ ] ] here @xmath184 \to { { \mathbb r}}$ ] in ( [ eq : i1a ] ) is a given spatially homogeneous forcing term . in the phase field approximation ( [ eq : heata]d ) , ( [ eq : aca]c ) this forcing appears analogously as a right hand side term @xmath185 in ( [ eq : heata ] ) . note that for @xmath186 the above system ( [ eq : i1a]e ) corresponds to ( [ eq:1a]e ) with ( [ eq : iso ] ) , ( [ eq : betaiso ] ) and @xmath187 . we first consider the quasi - static case @xmath64 . to this end we take the known solution of an annular region @xmath31 , for which the inner boundary shrinks to a point so that @xmath31 becomes a disk for sufficiently large @xmath188 . here we take @xmath189 and let @xmath169 . in addition , @xmath190 consists of two concentric circles / spheres . it is then not difficult to show that the two radii @xmath191 satisfy the following system of nonlinear odes : in the case @xmath192 we have @xmath193_t = - \frac1{r_1}\frac{\frac1{r_1 } + \frac1{r_2}}{\ln\frac{r_2}{r_1 } } \quad\mbox { and } \quad [ r_2]_t = - \frac1{r_2}\frac{\frac1{r_1 } + \frac1{r_2}}{\ln\frac{r_2}{r_1 } } = \frac{r_1}{r_2}\,[r_1]_t \qquad \forall\ t \in[0,t_0)\ , , \label{eq : odea}\ ] ] while for @xmath194 it holds that @xmath193_t = - \frac2{r_1 ^ 2}\frac{r_1 + r_2}{r_2-r_1 } \quad\mbox { and } \quad [ r_2]_t = - \frac2{r_2 ^ 2}\frac{r_1 + r_2}{r_2-r_1 } = \frac{r_1 ^ 2}{r_2 ^ 2}\ , [ r_1]_t \qquad \forall\ t \in[0,t_0)\ , , \label{eq : odeb}\ ] ] where @xmath195 is the extinction time of the smaller sphere , i.e. @xmath196 , see e.g. @xcite . note that the corresponding solution @xmath29 satisfying ( [ eq:1a]e ) is given by the radially symmetric function @xmath197 as ( [ eq : odea],b ) does not appear to be analytically solvable , it needs to be integrated numerically to compute the solution @xmath198 , for @xmath199 $ ] , where @xmath200 . possible strategies to integrate ( [ eq : odea],b ) to a high accuracy are described in @xcite . we visualize the evolution of @xmath7 over the time interval @xmath201 $ ] in figure [ fig : mstrue ] . the above true solution forms the basis of our first benchmark problem : true solution ( [ eq : odea ] ) , ( [ eq : ode_u ] ) to ( [ eq : i1a]e ) with @xmath202 and @xmath203 . + initial data @xmath204 . + domain @xmath205 with @xmath169 . + time interval @xmath206 $ ] with @xmath207 so that @xmath208 . in figure [ fig : tuneqi ] we compare the energy @xmath209 of the true sharp interface solution , recall ( [ eq : f ] ) , to the corresponding energies @xmath210 , recall ( [ eq : fg ] ) , of the finite element approximations from the algorithm pf@xmath211-fem on the time interval @xmath212 $ ] . here we recall that for the given data and the given evolution it holds that latexmath:[\[\mathcal{f}(\gamma , u ) = very similar energy plots can be obtained for the other variants of pf@xmath124-fem and those of pf@xmath123-fem . we note that for decreasing @xmath4 , the time step size @xmath76 needs to be chosen smaller and smaller in order to capture the correct time scaling of the evolution . we compare this computation for pf@xmath211-fem , which uses a semi - implicit discretization in time , now with one computation for the implicit time discretization as in ( [ eq : impl ] ) . here we recall that better accuracy for such discretizations has been reported in @xcite for the isotropic cahn hilliard equation , i.e. ( [ eq : heata ] ) , ( [ eq : aca ] ) for ( [ eq : iso ] ) with ( [ eq : obstacle ] ) , ( [ eq : varrho])(i ) and with @xmath202 . in fact , here we observe a similar behaviour . see figure [ fig : tuneqi_impl ] , where even for very large choices of @xmath76 the time evolution of the phase field model seems to be captured accurately . finally , we show some discrete energies @xmath214 from the parametric scheme pfemin figure [ fig : tunep ] . here we choose rather crude discretization parameters , since otherwise the discrete energies @xmath215 would lie virtually on top of the true energy @xmath209 . in table [ tab : tuneil2 ] we present the errors in @xmath216 and in @xmath29 for the scheme pf@xmath125-fem for a selection of interface parameters @xmath4 and for a range of discretization parameters @xmath74 and @xmath76 for the benchmark problem [ bm : ms ] . the displayed error quantities are defined as @xmath217 and @xmath218 , with @xmath219 being the first unit vector in @xmath220 , denotes the phase field approximation of the inner radius . we also show the number of degrees of freedom ( dofs ) for the calculation of the discrete solution for the final time step at time @xmath221 . the presented overall cpu times are for a single - thread computation on an intel i7 - 860 ( 2.8 ghz ) processor with 8 gb of main memory . .benchmark problem [ bm : ms ] for pf@xmath125-fem . [ cols="^,^,^,^,^,^,>",options="header " , ] what can be clearly seen from table [ tab : tscqiv ] is that once again we have convergence of order at least @xmath222 in @xmath223 . in order to visualize the performances of all the considered methods for the benchmark problem [ bm : stefan1 ] with @xmath224 , i.e. including the computations in tables [ tab : tsciii1 ] and [ tab : tscqiv ] for the second order accurate phase field model ( [ eq : egsa],b ) , we present a log - log plot of the errors in the radius @xmath225 against the necessary cpu time for all the entries in the appropriate tables in figure [ fig : scatter8all ] . the plot appears to confirm that computations for the phase field model ( [ eq : egsa],b ) are on average more efficient than computations for the standard phase field model , i.e. ( [ eq : egsa],b ) with @xmath226 . however , the finer meshes needed for computations for ( [ eq : egsa],b ) , recall table [ tab : tsciii1 ] , mean that due to cpu time constraints we can not choose @xmath4 as small as in the standard phase field computations . finally , the plot in figure [ fig : scatter8all ] once again underlines the superiority of the sharp interface algorithm pfem over all the phase field methods . in this section we present numerical simulations for the anisotropic stefan problem . here we always let @xmath227 and @xmath228 , where we recall ( [ eq : mu ] ) . moreover , we always choose @xmath229 and , unless otherwise stated , we let @xmath230 . in order to appreciate the computational challenges involved with the different experiments , we recall from ( [ eq : epscond ] ) , ( [ eq : hf ] ) and ( [ eq : nv97a ] ) that for accurate phase field calculations the following implications arise : @xmath231 moreover , we note that for the fully anisotropic situation a formally second order accurate phase field model similarly to section [ sec:43 ] involves a parameter @xmath232 in place of ( [ eq : rho1 ] ) that depends on @xmath233 , @xmath14 and on @xmath234 , see @xcite . we stress that these approaches are not well analyzed so far , e.g. in the spirit of @xcite . in particular , to our knowledge there are no formal or rigorous results in the literature on the second order convergence of phase field models for the fully anisotropic gibbs thomson law . moreover , our numerical results in the isotropic case showed that the second order models do not give a large gain in computational efficiency . that is why all of our phase field computations in this section are for the standard phase field model ( [ eq : heata]d ) , ( [ eq : aca]c ) . for the first simulations that we present we choose as anisotropy the regularized @xmath235-norm @xmath236 the radius of the initially circular seed @xmath148 is chosen as @xmath237 , while we set @xmath168 , @xmath238 and @xmath239 . the supercooling at the boundary @xmath227 of @xmath240 is set to @xmath241 . three numerical simulations for the scheme pf@xmath242-fem with the interfacial parameter @xmath243 can be seen in figure [ fig : bsrhoii_4pi ] . it can be seen that varying the time discretization parameter @xmath76 from @xmath244 to @xmath245 has a significant impact on the observed numerical results . however , the observed changes for the smallest value of @xmath76 are small , which indicates that the simulation appears to be converging . very similar results can be obtained for pf@xmath246-fem with the implicit time discretization from ( [ eq : impl ] ) , and so we omit them here . we compare the above numerical experiments for the phase field method with three simulations for the sharp interface approximation pfem in figure [ fig : bsrho ] , where we fix the spatial discretization parameters as @xmath247 . here we observe that even for a very crude time discretization , the evolution is captured remarkably well , and there is very little variation in the numerical results from pfem when @xmath76 is decreased . also note that it takes ( less than ) a second of cpu time with pfemin order to get a good idea about the evolution of the growing crystal , while the phase field methods pf@xmath123-fem and pf@xmath124-fem take at least @xmath248 times as long . for the next set of numerical experiments we use the hexagonal anisotropy @xmath249^{\frac12}\,,\ ] ] and where @xmath250 denotes a clockwise rotation through the angle @xmath251 . moreover , we use the parameters @xmath252 , @xmath253 , @xmath239 and @xmath254 on the boundary @xmath227 of @xmath255 . the radius of the initially circular seed is again chosen as @xmath256 , and we set @xmath257 with @xmath258 . three numerical simulations for the scheme pf@xmath242-fem with the interfacial parameter @xmath259 can be seen in figure [ fig : stefanii_16pi ] . observe that here we use a much smaller value of @xmath4 than previously , because for larger values of @xmath4 large mushy interfacial regions develop , which means that the phase field simulations hold no value for the investigation of the underlying sharp interface problem . the creation of small localized mushy interfacial regions can be observed in figure [ fig : stefanii_16pi ] for the run with @xmath260 , while the run with @xmath261 shows larger such regions . in addition , in the latter run the phase field approximation of the growing crystal s surface reaches the external boundary @xmath23 , which results in the creation of artificial , nonphysical interfaces . repeating these simulations for a smaller interfacial parameter @xmath4 yields the results shown in figure [ fig : stefanii_32pi ] . now for sufficiently small values of the time step size @xmath76 , the numerical results appear to be converging . we also repeat the last computation for the scheme pf@xmath242-fem with the explicit time discretization from ( [ eq : expl1],b ) . here any computation with a time step size @xmath262 was unstable , and so in figure [ fig : stefanii_32piex ] we only show a run for @xmath263 . we recall from section [ sec:42 ] , see tables [ tab : tsii1_expl ] and [ tab : tsii1_expl_unif ] , that in the interest of accuracy uniform meshes should be employed for an explicit method . however , the large cpu times associated with a uniform grid mean that we are unable to complete the evolution within a reasonable amount of time . hence in figure [ fig : stefanii_32piex ] we use the same adaptive mesh strategy as in figure [ fig : stefanii_32pi ] for the semi - implicit scheme pf@xmath242-fem . note that while the finest run in figure [ fig : stefanii_32pi ] agrees well with the results shown in figure [ fig : stefanii_32piex ] , the very small time step size used for the latter means that the explicit scheme takes about @xmath264 times as long as the semi - implicit scheme to compute the evolution . hence , without further code optimizations , the explicit scheme would need to be run in parallel on a cluster with at least @xmath264 nodes to become competitive . when we repeat the simulations shown in figures [ fig : stefanii_16pi ] and [ fig : stefanii_32pi ] for the scheme pf@xmath246-fem , then in the run for @xmath259 large mushy interfacial regions appear , which quickly reach the boundary @xmath23 . hence we only present the simulations for @xmath265 , see figure [ fig : qvstefan_32pi ] , where we observe similar , but qualitatively quite different , results to the ones shown in figure [ fig : stefanii_32pi ] for the scheme pf@xmath242-fem . in particular , the side arms in figure [ fig : qvstefan_32pi ] appear to be thinner than in figure [ fig : stefanii_32pi ] , and the convergence as @xmath76 gets smaller appears to be slower . we also repeat the computation from figure [ fig : qvstefan_32pi ] for the implicit time discretization from ( [ eq : impl ] ) , see figure [ fig : qvstefan_32pii ] . we observe that , in contrast to the conclusions that could be drawn from the isotropic experiments in section [ sec:4 ] , for the strongly anisotropic situation treated here there does not seem to be an advantage in using the implicit time discretization from ( [ eq : impl ] ) over the standard semi - implicit discretization ( [ eq : semi1],b ) from @xcite . in addition , we present three simulations for the same physical problem for the sharp interface approximation pfem in figure [ fig : stefan ] , where we fix the spatial discretization parameters as @xmath266 . here we observe once again that even for a very crude time discretization , the evolution is captured remarkably well , and there is very little variation in the numerical results from pfem when @xmath76 is decreased . we also draw particular attention to the differing cpu times between the sharp interface calculations in figure [ fig : stefan ] and the phase field simulations in figures [ fig : stefanii_32pi][fig : qvstefan_32pii ] . in the remainder of this section we consider two simulations for the scheme pfem , which on present computer hardware are virtually impossible to repeat to a desirable accuracy with the phase field method . the first experiment is with the physical parameters from ( * ? ? ? 7 ) , and so is for the one - sided quasi - stationary problem ( [ eq:1a]e ) with @xmath267 and with @xmath14 as in ( [ eq : hexgamma2d ] ) . the remaining parameters are chosen as @xmath268 , @xmath269 and @xmath270 on the boundary @xmath227 of @xmath271 . the radius of the initially circular seed is chosen as @xmath272 . see figure [ fig : jcgtau ] for the results for different choices of the time step sizes @xmath76 , and with @xmath273 fixed . we see that , as before , there is hardly any variation in the numerical results obtained from the three simulations with different values of @xmath76 for pfem . moreover , we note that due the choice of the physical parameters much finer side branches appear in figure [ fig : jcgtau ] compared to the simulations in figure [ fig : stefan ] . to precisely capture these small structures within a phase field computation would require very small values for the interfacial parameter @xmath4 , as well as correspondingly small discretization parameters @xmath74 and @xmath76 ; recall ( [ eq : hf ] ) , and ( [ eq : nv97a ] ) . taken together this means that we are currently unable to present phase field computations for an evolution as shown in figure [ fig : jcgtau ] . the next computation is similar to the simulation shown in ( * ? ? ? 14 ) , where here we take as anisotropy @xmath274 ^ 9 + [ g(r_1\,{}{p})]^9 + [ g(r_2\,{}{p})]^9 \right)^\frac19\ , , \quad\text{where}~~ g({}{p } ) : = \left [ p_1 ^ 2 + \tfrac14\,(p_2 ^ 2 + p_3 ^ 2 ) \right]^{\frac12}\,,\ ] ] with the two rotation matrices defined as @xmath275 and @xmath276 . the wulff shape of the anisotropy ( [ eq : gocta ] ) can be seen on the right of figure 3 in @xcite . moreover , we use the parameters @xmath252 , @xmath277 , @xmath239 and @xmath254 on the boundary @xmath227 of @xmath278 . the radius of the initially spherical seed is chosen as @xmath256 , and we let @xmath279 be defined by ( [ eq : u0exp ] ) with @xmath280 . three simulations for these parameters , with the spatial discretization parameters fixed as @xmath281 , are presented in figure [ fig : dendritic ] . we observe that the three simulations all show the same general shape of the growing six - armed crystal , and for the smallest value of @xmath76 the results appear to have converged . we also note that the small oscillations in the solution for the simulation with the largest time step size disappear as @xmath76 is decreased . in order to demonstrate the good mesh properties of the parametric method pfem , we show in figure [ fig : mesh ] two details of the triangulated approximation of @xmath7 at times @xmath282 and @xmath283 for the finest time discretization in figure [ fig : dendritic ] . we recall that the algorithm pfemdoes not employ any mesh - redistribution or mesh - smoothing methods . rather it relies solely on local mesh refinements , where individual elements of the triangulation become too large , see @xcite for more details . the quality of the meshes shown in figure [ fig : mesh ] is excellent . + we recall that a simulation for the phase field algorithm pf@xmath242-fem for the same physical parameters has recently been performed in ( * ? ? ? 23 ) . as a comparison to the sharp interface calculations from figure [ fig : dendritic ] , we present the results for the scheme pf@xmath242-fem in figure [ fig : vch ] . we note that the evolution shown for the phase field approximation in figure [ fig : vch ] is qualitatively very different from the sharp interface simulations in figure [ fig : dendritic ] . in all likelihood the physically challenging parameters for the computation in figure [ fig : vch ] mean that , both in terms of the discretization parameters for the given @xmath259 , e.g. the time step size @xmath76 , as well as in terms of the interfacial parameter @xmath4 itself , the shown numerical results are still far from the true underlying solution to the sharp interface problem ( [ eq:1a]e ) . of course , a detailed numerical study into this question is not yet possible due to the long time that such computations would take . while numerical simulations for phase field models in general show qualitatively correct behaviour , often such numerical results are far away from the true sharp interface evolution . in order to obtain accurate simulations , the interface width @xmath4 , as well as the spatial and temporal discretization parameters need to be chosen sufficiently small . however , reducing these parameters to reach an acceptable accuracy often requires very large computing times on even the most advanced of today s desktop computers . direct sharp interface approximations , on the other hand , can provide a computationally cheap method to compute interface evolutions in materials science accurately . an example of such an algorithm is pfem from @xcite . in the computations presented in this paper we have seen that even for very crude discretization parameters , the algorithm pfemprovides surprisingly accurate approximations . the computational time needed to compute these sharp interface approximations is often negligible compared to the cpu times necessary for a corresponding phase field simulation . the main problem of phase field methods is that the asymptotic error in @xmath4 , which in general is not known , plays a significant role in determining the accuracy of phase field simulations . small values of @xmath4 , in turn , require very small discretization parameters . similarly , in second order convergent isotropic phase field models with a correction term , where the asymptotic error in @xmath4 may be assumed to be relatively smaller than in classical phase field models , very small discretization parameters need to be employed in order to benefit from the smaller asymptotic error in practice . all of these issues do not arise in sharp interface approximations . the main advantage of phase field methods over direct front tracking methods is that they intrinsically allow for topological changes . however , for the problem of solidification and dendritic growth as considered in this paper , topological changes in general do not occur during the simulation of physically relevant evolutions for single crystals . in the past , researchers and scientists may have been discouraged from applying front tracking methods because of the difficulties in implementing such methods and because of the deterioration of the mesh quality as the approximated sharp interface evolves in time . however , assembling the system matrices in parametric finite element methods for evolving manifolds is not much different from the assembly in standard cartesian problems , see e.g. @xcite . of course , the coupling between a lower dimensional parametric mesh and a bulk mesh is nontrivial , but successful implementations have been used in e.g.@xcite . moreover , the good mesh properties of the scheme pfem from @xcite mean that a good mesh quality is maintained throughout the numerical simulations , and no remeshing is required in practice . in fact , all the simulations presented in this paper were performed without any remeshing , see @xcite for more details . * the parametric front tracking method pfem is more accurate and computationally more efficient than phase field methods . * for isotropic problems , implicit time discretizations for phase field models are often more accurate than semi - implicit time discretizations . * explicit time discretizations for phase field models need very small time steps in practice , and hence computations with explicit schemes are only competitive if run in parallel on large clusters . * second order accurate phase field models need finer discretization parameters than classical phase field models in order to demonstrate their superior approximation properties in practice . finally we note that while the focus of this paper has been the problem of dendritic solidification , it is to be expected that similar conclusions can be drawn when considering the respective merits of phase field models and sharp interface methods for other free boundary problems in materials science , physics and biology . as possible examples we refer to epitaxial growth , surface diffusion , thermal grooving , sintering , vesicle dynamics and two phase flow . it is our hope that the comparisons presented in this paper encourage a discussion about the merits of phase field methods in general , and of the possible advantages of using sharp interface approximations instead . height 2pt depth -1.6pt width 23pt , _ on stable parametric finite element methods for the stefan problem and the mullins sekerka problem with applications to dendritic growth _ , j. comput . phys . , 229 ( 2010 ) , pp . 62706299 . height 2pt depth -1.6pt width 23pt , _ finite element approximation of one - sided stefan problems with anisotropic , approximately crystalline , gibbs thomson law _ , adv . differential equations , ( 2012 ) . ( to appear ) , see also http://arxiv.org/abs/1201.1802 . height 2pt depth -1.6pt width 23pt , _ a phase - field model with a double obstacle potential _ , in motion by mean curvature and related topics ( trento , 1992 ) , g. buttazzo and a. visintin , eds . , de gruyter , berlin , 1994 , pp . 122 . , _ approximation of curvature dependent interface motion _ , in the state of the art in numerical analysis ( york , 1996 ) , i. s. duff and g. a. watson , eds . 63 of inst . , oxford univ . press , new york , 1997 , pp . 407440 . , _ double obstacle phase field computations of dendritic growth _ , 1996 . university of sussex cmaia research report 96 - 19 , http://homepages.warwick.ac.uk/staff/c.m.elliott/papers/doubleobstaclephasefield/ellgar96.pdf . , _ models of pattern formation in first - order phase transitions _ , in directions in condensed matter physics , vol . 1 of world sci . condensed matter phys . , world sci . publishing , singapore , 1986 , pp . 165186 . , _ phase - field models of solidification _ , in recent advances in numerical methods for partial differential equations and applications ( knoxville , tn , 2001 ) , vol . 306 of contemp . soc . , providence , ri , 2002 , pp . 107145 . height 2pt depth -1.6pt width 23pt , _ approximation of crystalline dendrite growth in two space dimensions _ , in proceedings of the algoritmy97 conference on scientific computing ( zuberec ) , j. kaur and k. mikula , eds . 67 , bratislava , 1998 , slovak university of technology , pp .

by analyzing the gas temperature maps created from the archive data , we reveal the prevailing existence of temperature substructures on kpc scales in the central regions of nine intermediate - redshift ( ) galaxy clusters , which resemble those found in the virgo and coma clusters . each substructure contains a clump of hot plasma whose temperature is about kev higher than the environment , corresponding to an excess thermal energy of erg per clump . since if there were no significant non - gravitational heating sources , these substructures would have perished in yrs due to thermal conduction and turbulent flows , whose velocity is found to range from about 200 to 400 km , we conclude that the substructures can not be created and sustained by inhomogeneous radiative cooling . we also eliminate the possibilities that the temperature substructures are caused by supernova explosions , or by the non - thermal x - ray emission due to the inverse - comptonization of the cmb photons . by calculating the rising time of agn - induced buoyant bubbles , we speculate that the intermittent agn outbursts ( erg per burst ) may have played a crucial role in the forming of the high temperature substructures . our results are supported by recent study of mcnamara & nulsen ( 2007 ) , posing a tight observational constraint on future theoretical and numerical studies .

0905.3933

in the frame of hierarchical formation theories , it is believed that the mass clustering develops due to the gravitational instability caused by the intrinsic , tiny gaussian density perturbations that existed in the early universe ( see , e.g. , frenk et al . 1996 for a review ) . since the initial conditions and input physics are assumed to be scale - free , dark halos are expected to have evolved self - similarly in time , which can be described by the scaling laws over many fundamental parameters ( navarro et al . 1995 and references therein ) , as have been advocated by the numerical simulations in which spherical collapse and adiabatic gas dynamics are assumed ( e.g. , navarro et al . 1996 ; eke et al . 1998 ) . on the observational aspect , tight correlations between x - ray luminosity , gas temperature , gas entropy , x - ray isophotal radius , and virial mass of galaxy clusters have been discovered successively with _ einstein _ , _ asca _ , and _ rosat _ ( e.g. , mushotzky 1984 ; mohr & evrard 1997 ; horner et al . these correlations , however , often show notable deviations from theoretical predictions . for example , in early 1990s , a significant scatter on the measured luminosity - temperature curves was reported by , e.g. , edge & stewart ( 1991 ) and fabian et al . ( 1994b ) , who showed that the cool - core galaxy clusters are more luminous at a given temperature . the observed mass - temperature relation , on the other hand , shows a lower ( @xmath8 ) normalization and a steeper slope with respect to the prediction of adiabatic models ( e.g. , evrard et al . 1996 ; xu et al . 2001 ) . moreover , an excess in the gas entropy was found in the central regions ( @xmath9 ) of many low temperature systems ( ponman et al . 1999 , 2003 ) , which forms a platform of extra energy ( @xmath10 kev per particle ; tozzi & norman 2001 ) on the entropy - temperature curve . these deviations and mismatches , which have been confirmed with and observations recently ( e.g. , arnaud et al . 2005 ; kotov & vikhlinin 2006 ) , clearly indicate that our understanding of the thermal history of galaxy clusters is far from complete , and the roles played by non - gravitational heating processes need to be evaluated properly ( e.g. , kaiser 1991 ; evrard & henry 1991 ; kauffmann et al . 1993 ) , which will also help solve the cooling flow problem ( fabian 1994a ; makishima et al . 2001 ; peterson et al . 2003 ) . as of today , a number of heating sources have been proposed , which include merger shocks , cosmic rays , supernovae , thermal conduction , turbulent dissipation and diffusion , and agn feedbacks ( e.g. , mcnamara & nulsen 2007 ; markevitch & vikhlinin 2007 ) . however , none of the proposed candidates is all - purpose for halting the gas cooling and breaking the self - similarities on both galactic and cluster scales . for example , recent observations ( e.g. , croston et al . 2005 ; jetha et al . 2007 ) showed that , agn activity , the most representative heating source of today , tends to deposit most of its energy into the vast intergalactic space , rather than within the host galaxy ( @xmath11 ) . clearly , detailed mappings of gas temperature gradients in the icm is necessary as a crucial observational constraint in order to correctly investigate the gas heating history . two - dimensional gas temperature variations have been observed in many galaxy clusters with and _ xmm - newton_. by analyzing the temperature map generated with the high - quality data , fabian et al . ( 2000 ) identified a pair of cool gas shells that coincide with two radio bubbles , whose radii are @xmath12 and 8 @xmath1 kpc , respectively , in the innermost region of the perseus cluster . the authors suggested that the cool gas shells were formed as the buoyant gas detached from the terminals of agn jets and inflated in pressure equilibrium with the surrounding gas . similar phenomenon was also observed in , e.g. , centaurus cluster ( fabian et al . 2005 ) , abell 2052 ( blanton et al . 2003 ) , and hydra a ( nulsen et al . 2002 ) . also , in clusters that show clear merger signatures , such as 1e 0657 - 56 ( govoni et al . 2004 ) , abell 3667 ( vikhlinin et al . 2001 ) , abell 665 , and abell 2163 ( markevitch & vikhlinin 2001 ) , conspicuous temperature variations were copiously found , which are presumably caused by the rapid adiabatic expansion of the shock - heated gas . so far , however , in most of the existing works , little has been done to measure the morphologies of the temperature substructures and to investigate their origins in a quantitative way . one of few exceptional works was presented by shibata et al . ( 2001 ) , who calculated the two point correlation function of the hardness ratio map of the virgo cluster , and found that there is a characteristic scale of @xmath13 @xmath1 kpc for the two - dimensional temperature variations . the authors associated this characteristic scale with the size of the gas blobs arising during the infalling of sub - groups . by analyzing the two - dimensional temperature distribution with the data , schuecker et al . ( 2004 ) reported the presence of a turnover at @xmath0 @xmath1 kpc on the pressure fluctuation spectrum of the coma cluster , and attributed it to the existence of kolmogorov / oboukhov - type turbulent flows . as of today , due to the lack of more such quantitative studies , it is not clear if there does exist a prevalence for a characteristic scale @xmath0 @xmath1 kpc in galaxy clusters and groups , and how far it can be used to constrain the heating models in the icm . in this work we analyze the _ chandra _ archive data to search for possible characteristic scales of temperature variations in a sample of nine intermediate - redshift clusters . in 2 & 3 , we describe the sample selection criteria and data analysis , respectively . in 4 , we discuss the physical implications of the results . in 5 , we summarize our work . we assume @xmath14 km s@xmath15 mpc@xmath15 , a flat universe for which @xmath16 and @xmath17 , and adopt the solar abundance standards of grevesse & sauval ( 1998 ) . in order to balance between the angular resolution and detector coverage of the targets , we construct our sample with nine intermediate - redshift ( @xmath18 ) galaxy clusters , so that the central 400 @xmath1 kpc regions of the clusters can be fully , or nearly fully covered by the s3 ccd of the advanced ccd imaging spectrometer ( acis ) instrument . all the selected clusters have a @xmath19 kev flux greater than 5.0 @xmath20 erg @xmath6 @xmath21 . we list some basic properties of these clusters in table 1 , which are arranged in the orders of source names ( col . [ 1 ] ) , names of the cd galaxies ( col . [ 2 ] ) , redshifts ( col . [ 3 ] ) , richness classes ( col . [ 4 ] ) , right ascension and declination coordinates ( j2000 ) of the cluster optical centroids ( col . [ 5 ] & [ 6 ] ) , and notes ( col . [ 7 ] ) . on the optical and infrared images extracted from the dss ( fig . 1 ) and 2mass archives , we do not find any strong evidence for pronounced signatures caused by major mergers on cluster scales , such as off - center optical and/or infrared substructures . a478 , a1068 , a1650 , a2244 , and a3112 possess a central galaxy that has been classified as a cd , whose size is @xmath22 times that of any other member galaxy ( rood & sastry 1971 ; takizawa et al . the remaining four clusters are dominated by a pair or a clump of brightest member galaxies ( struble & rood 1987 ) . in addition , schombert et al . ( 1989 ) reported the existence of a low - velocity ( the radial velocity difference is @xmath23 50 km @xmath6 ) companion galaxy dwelling deep inside the envelope of the cd galaxy of a2244 . strong radio emission associated with agn activity has been detected in four clusters . in a478 , two symmetric radio lobes are identified at 1.4 ghz , which extend about @xmath24 towards northeast and southwest , respectively ( sun et al . 2003 ) . in a1068 , two compact radio sources are resolved , which are consistent with the locations of the cd galaxy and sdss j104043.45 + 395705.3 ( a member galaxy located at about @xmath25 southwest of the cluster center ) , respectively ( mcnamara et al . a2204 and a3112 are reported to possess an extended radio halo that roughly traces the spatial distribution of the x - ray gas , respectively ( sanders et al . 2005 ; takizawa et al . in the remaining five clusters , neither radio lobes nor luminous radio halos has been found in the central region in previous works , suggesting that they have not experienced strong agn outbursts during past few tens of myr , if the life time of such radio sources can be estimated by @xmath26 and @xmath27 @xmath28 g ( takahashi & yamashita 2003 ; donahue et al . 2005 ) . as a unique case in our sample , a2556 harbors an off - center extended radio source at about @xmath29 northeast of the center , which is often speculated to be the relic of a merger - driven shock ( e.g. , govoni et al . 2004 ) . all the x - ray data analyzed in this work were acquired with the acis s3 ccd ( table 2 ) , with the focal plane temperature set to be 120 @xmath30 . using the ciao v4.0 software and caldb v3.5.0 , we removed the bad pixels and columns , as well as events with grades 1 , 5 , and 7 , and then executed the gain , cti , and astrometry corrections . in order to identify possible strong background flares , lightcurves were extracted from regions sufficiently far away , i.e. , @xmath31 from the x - ray peaks . time intervals during which the count rate exceeds the average quiescent value by 20 percent were excluded . the s1 chip data were used to crosscheck the results when it was available . figure 1 shows the intensity contours of the x - ray emission in @xmath32 kev of our sample clusters , which overlay the corresponding optical dss images . in all cases , the x - ray morphology appears to be smooth and symmetric on @xmath33 @xmath1 kpc scales , showing no remarkable twists , fronts , or edges that indicate recent major mergers . on smaller scales , however , there exist substructures , such as x - ray cavities coinciding with the agn lobes ( sun et al . 2003 ) , stripes , filaments , and arc - like sharp fronts ( wise et al . 2004 ; sanders et al . 2005 ) . for each cluster , we examine the vignetting - corrected surface brightness profile ( sbp ) extracted in @xmath34 kev , after filtering all the point sources that are detected above the 3@xmath35 level of the local background by using both the celldetect and wavdetect tools with their default settings ( fig . 2 ) . by applying the @xmath36-test , we find that , except for the sbp of a2204 , which can be well described with a single-@xmath37 model @xmath38 , all the obtained sbps are consistent with the empirical two-@xmath37 model @xmath39 ( e.g. , ikebe et al . 1996 ) , because in the central @xmath40 @xmath1 kpc ( a1650 , a2556 , and a3112 ) or @xmath41 @xmath1 kpc ( a478 , a1068 , a1201 , a2104 , and a2244 ) there exists an emission excess beyond the @xmath37 model that can best describe the sbps of outer regions . such an excess is often seen in clusters that host a central dominating galaxy , which is usually classified as a cd ( makishima et al . 2001 ) . in the spectral analysis that follows , we utilized the blank - sky templates for the s3 ccd as the background . the templates were tailored to match the actual pointings , and the background spectra were extracted and processed identically to the source spectra ( 3.2.2 ) . after rescaling each background spectrum by normalizing its high energy end to the corresponding observed spectrum , we further enhance the accuracy of the background by adding an unabsorbed apec model with fixed temperature ( 0.2 kev ) and metal abundance ( 1 @xmath42 ; snowden et al . 1998 ) to account for variations in the galactic x - ray background ( gxb ; vikhlinin et al . 2005 ) . we note that the gxb component is significant only in a2204 , where it has a flux of 2.0 @xmath43 erg @xmath6 @xmath21 @xmath44 in @xmath45 kev and is nearly uniform over the s3 chip . this was also reported by sanders et al . ( 2009 ) , who indicated that it is probably due to the excess galactic foreground . here we calculate the azimuthally averaged temperature profiles , which will be used to create the reference temperature maps applied in wavelet detection ( 3.3.2 ) and calculation of excess energies in the temperature substructures ( 4 ) . we divided the inner 400 @xmath1 kpc of each cluster into concentric annuli , masked all the detected point sources ( @xmath46 point sources per cluster ) , and extracted the spectrum from each annulus . the corresponding spectral redistribution matrix files ( rmfs ) and auxiliary response files ( arfs ) were generated using the ciao tool mkwarf and mkacisrmf . the spectra were grouped with @xmath47 counts per bin to allow the use of @xmath36 statistics . we used xspec 12.4.0 package to fit the spectra , by modeling the gas emission with an absorbed apec component and employing the projct model to correct the projection effect . the absorption column density @xmath48 was fixed to the galactic value ( dickey & lockman 1990 ) in all cases , except for a478 , for which @xmath48 was set to be about twice the galactic value ( sanderson et al . 2005 ; table 3 ) ; since the absorption excess in a478 exhibits a spatial extension beyond the cluster s core region , pointecouteau et al . ( 2004 ) argued that it is likely to be associated with a local absorption substructure in our galaxy . in order to calculate the error ranges of the model parameters , we executed the xspec command steppar and carried out multiple iterations at every step , to ensure that the actual minimum @xmath36 is found . the best - fit gas temperature and abundance profiles are illustrated in figure 3 , in which multiple sets of annuli are used for each cluster to crosscheck the results . we note that the inclusion of an additional gas component could only slightly improve the fits for the central regions of a478 , a1650 , a2204 , a2244 , and a3112 , which is insignificant as indicated by the @xmath49-test . a significant temperature decline by a factor of @xmath50 is observed in the central regions of a478 , a1068 , a2204 , a2556 , and a3112 , four of which ( except a2556 ) possess radio halos and/or lobes in the center ( 2 ) . in the remaining four clusters without apparent radio emission , the inward temperature drop is somewhat ambiguous , as the difference between the core temperature and mean cluster temperature , which refer to the temperatures measured in the central 100 @xmath1 kpc ( @xmath51 ) and @xmath52 @xmath1 kpc regions ( fig . 3 ) , respectively , is found within 3@xmath35 error ranges . if we apply the classification criteria of sanderson et al . ( 2006 ) , the latter four clusters are non - cooling - core clusters , whose central regions have not yet fully condensed . these results reflect the trend for the cooling core clusters to host central radio sources ( burns 1990 ) . as shown in figure 3 , we also adopt eqs.@xmath53 of vikhlinin et al . ( 2006 ) to describe the obtained radial temperature profiles in a smooth form , which will be used in 3.3.2 to calculate the reference temperature maps in order to examine the significances of detected temperature substructures , and in 4 to calculate the excess energies in these substructures . in a478 , a1068 , a1650 , a2204 , a2556 , and a3112 , the abundance increases inwards significantly ( 68% confidence level ) from @xmath54 @xmath42 in the outermost annuli to @xmath55 @xmath42 in the innermost regions ( @xmath56 @xmath1 kpc ) . emission measure - weighted gas temperature and abundance of each cluster are also calculated ( table 3 ) , which are in good agreement with previous results obtained with and ( sun et al . 2003 ; takahashi & yamashita 2003 ; takizawa et al . 2003 ; pointecouteau et al . 2004 ; wise et al . 2004 ; donahue et al . 2005 ; sanders et al . 2005 ; sanderson et al . 2005 ) . following the methods of , e.g. , osullivan et al . ( 2005 ) , maughan et al . ( 2006 ) , and kanov et al . ( 2006 ) , we define a set of knots @xmath57 ( @xmath58 knots per cluster ) in the central 300 @xmath1 kpc region , which are randomly distributed with a separation of @xmath59 @xmath60 between any two adjacent knots @xmath61 and @xmath62 . to each knot we assign an adaptively sized circular cell , which is centered on the knot and contains more than 800 counts in @xmath63 kev after the background is subtracted . the typical cell radius ranges from @xmath64at the center to about 10 at r @xmath65 @xmath1 kpc , which is always larger than the associated @xmath59 . we calculate the projected temperature of each cell @xmath66 , which is assigned to the corresponding knot , by fitting the spectrum extracted from the cell with an absorbed apec model , using the absorption column given in table 3 ; adding another thermal component can neither improve the fit significantly , nor make the parameters constrained more tightly . in the fittings , the source spectrum is grouped to have @xmath47 photons in each energy bin . in addition to the standard @xmath36 statistics , we also have applied the maximum likelihood statistics , and find that the best - fit parameters obtained with the two statistics agree nicely with each other within the @xmath67 error ranges . the calculated @xmath67 errors of gas temperature are typically @xmath68% in the central 100 @xmath1 kpc , which increase progressively to @xmath69 15% ( a478 , a1650 , a2244 , and a3112 ) or @xmath69 20% ( the remaining five clusters ) in @xmath70 @xmath1 kpc . for any given position @xmath71 , we define a scale @xmath72 , so that there are at least 800 counts contained in a circular region centered at @xmath71 , whose radius is @xmath72 . we calculate the projected temperature at @xmath71 by integrating over all the knots @xmath73 located in the circular region , @xmath74 where @xmath75 is the distance from @xmath71 to @xmath73 , and @xmath76 is the gaussian kernel whose scale parameter @xmath35 is fixed at @xmath77 . since @xmath72 is essentially proportional to the square root of the local counts , the obtained temperature map @xmath78 ( fig . 4 ) is less affected by the statistical uncertainties caused by surface brightness fluctuations . in addition , angular resolutions of @xmath79 are guaranteed by the use of compact gaussian kernel . maps for the upper and lower limits of @xmath78 are calculated in a similar way . * wavelet transform * a visual inspection of the obtained temperature maps ( fig . 4 ) indicates that there exist significant substructures , which typically have linear sizes of @xmath0 @xmath1 kpc , in all of the sample clusters . in order to describe these substructures in a quantitative way , we analyze the statistical properties of the temperature maps by applying the wavelet transform , a powerful tool in imaging process that identifies and disentangles small embedded features as a function of scales ( e.g. , vikhlinin et al . 1997 ; starck & pierre 1998 ) . in an analogue study , bourdin & mazzotta ( 2008 ) conducted the wavelet transform to achieve an optimized spatial binning of temperature maps . we employ a non - orthogonal version of discrete wavelet transform , the trous wavelet ( shensa 1992 ; starck et al . 1995 ; vikhlinin et al . 1998 ) , which defines a straightforward decomposition rule . to be specific , at position @xmath80 and scale level @xmath61 ( @xmath81 ) , the trous wavelet @xmath82 equals the signal difference between two smoothed maps @xmath83 , where @xmath84 is the obtained temperature map @xmath85 , and @xmath86 connects with @xmath87 via @xmath88 , in which @xmath89 is the convolution kernel , and @xmath90 is its grid . here @xmath89 is chosen to be a compact gaussian @xmath91 . each temperature map is thus decomposed into a set of subimages @xmath82 on nine scales ( @xmath92 pixels , @xmath93 ) , respectively . based on the obtained @xmath82 maps , we calculate the significant wavelet coefficient @xmath94 to detect the temperature substructures on different scales . for each cluster , we create a set of random knots @xmath73 as used in 3.3.1 , and assign a randomized temperature @xmath95 to each knot , which is constrained by the measured temperature profile @xmath96 and its error ranges ( fig . 3 ) . by applying eq.(1 ) to the @xmath97 map , we obtain a reference temperature map @xmath98 , in which the effects of the radial temperature gradients and knot tessellation are both involved . a series of such reference temperature maps are created by repeating this random process , and each reference map is decomposed into subimages @xmath99 on nine scales by performing the same trous wavelet transform as described above . on each scale @xmath61 , we determine the significant coefficient as @xmath100 when @xmath101 , and @xmath102 when @xmath103 ( e.g. , slezak et al . 1994 ; starck et al . 2003 ) , where @xmath104 is the standard deviation of the simulated @xmath99 maps so that 3@xmath104 is equivalent to a confidence interval of @xmath105 for gaussian noise . a similar method was used to determine the actual significances of diffuse sources on images ( e.g. , rosati et al . 1995 ; grebenev et al . 1995 ; damiani et al . 1997 ) . since the trous transform is essentially a non - orthogonal algorithm , the obtained coefficient @xmath82 , and thus @xmath106 , may be systematically biased from the exact solution ( starck & pierre 1998 ) . this bias can be corrected by using an iterative method ( van cittert 1931 ; biviano et al . 1996 ; starck et al . 2003 ; bourdin et al . 2004 ) , the iteration routine of which is defined as @xmath107 where @xmath108 is the corrected detection on scale @xmath61 after @xmath109 iteration(s ) ( @xmath110 is the initial run ) , and @xmath111 is a non - linear synthesized operator of inverse wavelet transform , wavelet transform , and filtering operation using @xmath112 as the threshold . typically the calculation converges within five iterations , which yields a corrected @xmath106 map as the unbiased reconstruction of the temperature substructures on scale @xmath61 ( fig . we define the wavelet spectrum as @xmath113 ( torrence & compo 1998 ) , and plot it in figure 5 as a function of scale for each cluster . as a crosscheck , we also employ the two - dimensional b - spline function ( e.g. , bourdin & mazzotta 2008 ) as the scaling function @xmath114 to perform a compact trous transform in the fourier domain @xmath115 , so that the fourier transform of the wavelet coefficient @xmath82 is directly calculated as @xmath116 , where the filter @xmath117 when @xmath118 ( in this paper the fourier transform of any function @xmath119 is written as @xmath120 ) . by carrying out the inverse fourier transform to @xmath121 , @xmath122 maps are obtained , and then used to calculate the corrected @xmath106 maps following the procedure described above . the corresponding wavelet spectra are illustrated in figure 5 for comparison , which agree nicely with those obtained with the gaussian wavelet . we find that the most popular scales of the projected temperature substructures are in the range of about @xmath123 @xmath1 kpc , and about 70% of these substructures are located at 100 200 @xmath1 kpc from the cluster center , exhibiting irregular shapes and spatial distributions ( fig . * power spectrum * as a straightforward approach , we have also studied the power spectra of the two - dimensional gas temperature distributions . adopting the method of schuecker et al . ( 2004 ) , we determine the flat fields @xmath124 by applying a low - pass wavelet filter with a scale of @xmath125 pixels on the original temperature maps , and calculate the normalized temperature fluctuation distributions as @xmath126 . by carrying out the fourier transform of @xmath127 @xmath128 where @xmath129 is the projected area , we obtain the power spectrum @xmath130 , @xmath131 following the method of schuecker et al . ( 2004 ) , we carry out monte - carlo simulations for each cluster to estimate the temperature fluctuation at any given knot @xmath73 based on the temperature error range measured at the same knot ( 3.3.1 ) . after a randomized value @xmath132 is assigned to each knot , a random temperature error map @xmath133 is created by applying eq.(1 ) to the @xmath132 map , in which the uncertainties introduced by data statistics , spectral model fitting , and cell tessellation are all involved . then , by combining @xmath78 and @xmath133 maps we create a random temperature map @xmath134 . the @xmath67 error bars shown in figure 6 are calculated from the variances of power spectra obtained from ten such @xmath135 maps for each cluster . in general , the power spectra @xmath130 show a significant peak in @xmath123 @xmath1 kpc , and drop quickly beyond @xmath136 @xmath1 kpc . this strongly supports our conclusion derived with the wavelet analysis ( fig . we present robust evidence for the prevailing existence of hot gas clumps , whose characteristic scales are @xmath0 @xmath1 kpc , in the central regions of a sample of nine intermediate - redshift galaxy clusters . these substructures are not caused by uncertainties in background subtraction , response calibration , and spectral model fittings , nor are they related to background flares , the effects of which have been largely corrected and/or fixed . the possibility of the temperature substructures arising from the non - thermal emission of low - mass x - ray binaries ( lmxbs ) can also be eliminated , since the icm is only sparsely filled with galaxies . our results confirm the previous findings in the coma cluster ( schuecker et al . 2004 ) and the virgo cluster ( shibata et al . 2001 ) . we estimate the excess thermal energy @xmath137 in each high temperature substructure ( table 4 ) , where the gas temperature is typically higher than environment by @xmath138 kev ( fig . 4 ) , by defining the thermal energy as the product of gas pressure @xmath111 and volume @xmath139 , and assuming a spherical or elliptical geometry for the volume of the hot gas clump . we obtain @xmath140 @xmath141 erg for each clump , where @xmath142 is the electron density determined in the deprojected spectral analysis ( 3.2.2 ) . to account for azimuthal variations in the electron density , we include a 10% systematic error ( sanders et al . 2004 ) in the calculations . as the projection effect is not corrected for @xmath143 , the obtained excess energy may have been underestimated by a factor of about two . the origin of this energy excess is discussed as follows . the detected temperature substructures may have been formed due to the spatially inhomogeneous cooling in the icm . since for a unit volume , the radiative cooling rate is @xmath144 , where @xmath145 is the cooling function calculated by including both continuum and line emissions , the radiative cooling time of the substructures can be estimated as @xmath146 typically , the calculated cooling time is @xmath147 yrs , and is thus nearly an order of magnitude longer than the characteristic time for thermal conduction alone to smear out the temperature substructures ( table 4 ) , which is given by @xmath148 where @xmath149 is the enthalpy excess per unit volume in the hot substructure , @xmath150 is the conductive flux , @xmath151 is the characteristic scale length of the substructure , @xmath119 = 0.2 is the factor to assess the suppression caused by magnetic fields ( e.g. , narayan & medvedev 2001 ) , and @xmath152 is the thermal conductivity of hydrogen plasma , @xmath153\ ] ] for coulomb gas ( spitzer 1962 ) . this indicates that , unless the conduction process is extremely hindered ( i.e. , when @xmath154 ; maron et al . 2004 ; markevitch et al . 2003 ) , the effect of the radiative cooling is less important in the formation and evolution of temperature substructures . hence the possibility of the substructures being formed by inhomogeneous cooling can be excluded . the destruction of a temperature substructure via conduction can be accelerated by turbulence ( dennis & chandran 2005 ) , whose effect can be evaluated by introducing a modified conductivity ( cho et al . 2003 ; voigt & fabian 2004 ) @xmath155 where @xmath156 is the ratio of the turbulence velocity @xmath157 to the local adiabatic sound speed @xmath158 ( @xmath159 ) , and @xmath160 is the turbulence length scale . to calculate @xmath161 , the turbulence velocity @xmath157 can be estimated as follows . as there is evidence that heating continuously compensates a large fraction of radiative cooling since z@xmath162 ( bauer et al . 2005 ; mcnamara & nulsen 2007 ) , we may assume an approximate energy balance between heating and cooling for the cluster ( e.g. , kim & narayan 2003 ; zakamska & narayan 2003 ; dennis & chandran 2005 ) as @xmath163 where @xmath164 is the radiative loss , and @xmath165 , @xmath166 and @xmath167 are the heating rates of thermal conduction , viscous dissipation and turbulent diffusion , respectively . since @xmath168 @xmath169 and @xmath170 where @xmath171 is the gas mass density , @xmath172 is the specific entropy , @xmath173 is a dimensionless constant ( dennis & chandran 2005 and references therein ) , and @xmath174 @xmath175 @xmath6 is the diffusion coefficient ( rebusco et al . 2006 ) . by subtracting eqs.@xmath176 into eq.(9 ) , we have @xmath177\frac{l}{c_{\rm diss } \rho}\}^{\frac{1}{3}},\ ] ] if we define @xmath178 and @xmath179 ( dennis & chandran 2005 ) . using the observed deprojected gas density and temperature profiles ( fig . 3 ) , we calculate @xmath157 and show the results as a function of turbulence scale @xmath160 in figure 7 . if we follow schuecker et al . ( 2004 ) to adopt the lengths indicated by the turnover points on the wavelet spectra ( fig . 5 ) as the turbulence scale , we find that @xmath157 commonly ranges from 200 to 400 @xmath180 @xmath181 , or @xmath182 , for our sample . by using eqs.(6 ) and ( 8) , we find that the inclusion of turbulence reduces the characteristic time @xmath183 to the modified characteristic time @xmath184 by a factor of about 3 ( table 4 ) , which sets an upper limit ( @xmath185 yrs ) on the duty cycles of the dominating heating sources . the calculated @xmath137 ( table 4 ) suggests that the hot clumps may be dying or remnants of buoyant bubbles , into which the central agn has injected energy via shocks and turbulences ( e.g. , mcnamara & nulsen 2007 ) . such agn - induced hot gas clumps are predicted in numerical simulations ( e.g. , quilis et al . 2001 ; dalla vecchia et al . 2004 ; vernaleo & reynolds 2006 ) and theoretical calculations ( e.g. , pizzolato & soker 2006 ) , whose scales are expected to be about 100 kpc by the end of the buoyancy process . by examining the temperature maps ( fig . 4 ) , we note that some hot clumps appear in pairs on both sides of the central galaxies ( @xmath1861 and @xmath1862 in a1201 ; @xmath1861 and @xmath1862 in a2204 ; @xmath1861 and @xmath1863 in a3112 ) , which suggests possible agn activity taking place at the cluster cores . however , can an agn - heated temperature substructure survive conduction and turbulence destructions ? to answer this question , we estimate the characteristic rising time of a bubble as @xmath187 , where @xmath188 is the travel distance measured from the cluster center to the location of a hot clump , and @xmath189 is the travel velocity . in terms of the keplerian velocity @xmath190 , a large bubble is expected to travel through the ambient gas at @xmath191 where @xmath192 is the drag coefficient and @xmath193 is the bubble radius in kpc ( churazov et al . this yields @xmath194 smaller or very close to @xmath184 for @xmath195% of the substructures ( table 4 ) , indicating that such substructures can survive in most cases . even for the clumps with @xmath196 , the agn heating scenario may still stand , if the distance traveled by the bubble is shorter ( e.g. , gardini 2007 ) . it should be noted that no hot gas clump is found associated with any known radio source in all sample clusters . this , however , does not contradict with the agn heating scenario , because the lack of radio sources can be explained by the fact that the relativistic electrons may have lost most of their energy when the bubbles arose , as inferred by @xmath197 ( @xmath198 yrs ; 2 ) . numerical and theoretical works show that shocks caused by supersonic galaxy motion in the icm can generate observable temperature substructures via gas compression and friction ( e.g. , el - zant et al . 2004 ) , and even the low - speed galaxy infalling may also be a heating source candidate due to the efficient energy conversion via the magnetohydrodynamic turbulence and magnetic reconnection ( makishima et al . 2001 ; reviewed by markevitch & vikhlinin 2007 ) . the energy dissipation rate in the central region of a cluster is as large as @xmath199 erg @xmath6 ( fujita et al . 2004 ) , which assembles @xmath200 erg during the merger process and is thus sufficient to account for the excess energy in the observed temperature substructures . nevertheless , no convincing evidence is given in numerical studies that the non - filamentary morphologies of the observed temperature substructures can appear commonly in merger process ( e.g. , poole et al . 2006 ) . we also have estimated the contribution of supernova feedback to the icm heating over the substructure lifetime ( @xmath184 ) using the observed rate of type ia supernovae ( sne ia ; sharon et al . 2007 ) , which implies that about @xmath201 sn ia explosions have occurred in the central 400 @xmath1 kpc region of each sample cluster . since the averaged dynamic energy injection into the environment is @xmath202 erg per sn ia event ( spitzer 1978 ) , and @xmath79 % of this energy is assumed to have been used to heat the gas ( e.g. , thornton et al . 1998 ) , the total sn ia heating is found to be @xmath203 erg per cluster per @xmath204 yrs . the contribution of type ii supernovae ( sne ii ) will not significantly increase the total supernova heating , since even in the ir - luminous cluster a1068 , where the ratio of sn ii rate to sn ia rate is @xmath205 ( wise et al . 2004 ) , the sn ii heating is still lower than @xmath206 erg . therefore , we conclude that the supernova heating is insufficient to create the observed high temperature substructures . high temperature substructures may also occur as a result of the inverse - comptonization ( ic ) of the cosmic microwave background ( cmb ) photons that are scattered by the relativistic electrons in the icm , since in a acis spectrum it is difficult to distinguish between such an ic component and a high temperature thermal component ( petrosian et al . if in the icm @xmath207% of the electrons are relativistic ( @xmath208 ; eilek 2003 ) , we will have an ic energy conversion rate of @xmath209 erg @xmath6 , where @xmath210 is the thomson cross section and @xmath211 is the cmb energy density at the location of the cluster ( sarazin 1999 ) , which implies that the flux of the ic component is sufficiently high to bias the temperature measurements by about @xmath212 kev . however , a tight correlation between the high temperature substructures and radio synchrotron sources is required in this scenario , which is actually not observed , possibly due to the lack of relativistic electrons . therefore , the ic mechanism for the formation of the temperature substructures can be eliminated . we reveal the prevailing existence of temperature substructures on @xmath0 @xmath1 kpc scales in the central regions of nine intermediate - redshift galaxy clusters . by comparing the characteristic destruction times of the temperature substructures with the gas cooling times and the rising times of agn - induced bubbles , we conclude that the agn outbursts may have played a crucial role in the forming of the observed temperature substructures . our results agree with those found earlier in the virgo and coma clusters ( shibata et al . 2001 ; schuecker et al . 2004 ) . we thank kazuo makishima , kazuhiro nakazawa , peter schuecker , and herv@xmath213 bourdin for their helpful suggestions and comments . this work was supported by the national science foundation of china ( grant no . 10673008 and 10878001 ) , the ministry of science and technology of china ( grant no . 2009cb824900/2009cb24904 ) , and the ministry of education of china ( the ncet program ) . abell , g. o. , corwin , h. g. , & olowin , r. p. 1989 , apjs , 70 , 1 arnaud , m. , pointecouteau , e. , & pratt , g. w. 2005 , a&a , 441 , 893 bauer , f. e. , fabian , a. c. , sanders , j. s. , allen , s. w. , & johnstone , r. m. 2005 , mnras , 359 , 1481 biviano , a. , durret , f. , gerbal , d. , le fevre , o. , lobo , c. , mazure , a. , & slezak , e. 1996 , a&a , 311 , 95 blanton , e. l. , sarazin , c. l. , & mcnamara , b. r. 2003 , apj , 585 , 227 bourdin , h. , sauvageot , j .- l . , slezak , e. , bijaoui , a. , & teyssier , r. 2004 , a&a , 414 , 429 bourdin , h. , & mazzotta , p. 2008 , a&a , 479 , 307 bregman , j. n. , & david , l. p. 1989 , apj , 341 , 49 burns , j. o. 1990 , aj , 99 , 14 cho , j. , lazarian , a. , honein , a. , knaepen , b. , kassinos , s. , & moin , p. 2003 , apj , 589 , l77 churazov , e. , br@xmath214ggen , m. , kaiser , c. r. , b@xmath215hringer , h. , & forman , w. 2001 , apj , 554 , 261 croston , j. h. , hardcastle , m. j. , & birkinshaw , m. 2005 , mnras , 357 , 279 dalla vecchia , c. , bower , r. g. , theuns , t. , balogh , m. l. , mazzotta , p. , & frenk , c. s. 2004 , mnras , 355 , 995 damiani , f. , maggio , a. , micela , g. , & sciortino , s. 1997 , apj , 483 , 350 dennis , t. j. , & chandran , b. d. g. 2005 , apj , 622 , 205 dickey , j. m. , & lockman , f. j. 1990 , ara@xmath216a , 28 , 215 donahue , m. , voit , g. m. , odea , c. p. , baum , s. a. , & sparks , w. b. 2005 , apj , 630 , l13 edge , a. c. , & stewart , g. c. 1991 , mnras , 252 , 414 eilek , j. a. 2003 , phys . plasmas , 10 , 1539 eke , v. r. , cole , s. , frenk , c. s. , & patrick h. , j. 1998 , mnras , 298 , 1145 el - zant , a. a. , kim , w .- t . , & kamionkowski , m. 2004 , mnras , 354 , 169 evrard , a. e. , & henry , j. p. 1991 , apj , 383 , 95 evrard , a. e. , metzler , c. a. , & navarro , j. f. 1996 , apj , 469 , 494 fabian , a. c. 1994a , ara&a , 32 , 277 fabian , a. c. , crawford , c. s. , edge , a. c. , & mushotzky , r. f. 1994b , mnras , 267 , 779 fabian , a. c. , sanders , j. s. , ettori , s. , taylor , g. b. , allen , s. w. , crawford , c. s. , iwasawa , k. , johnstone , r. m. , & ogle , p. m. 2000 , mnras , 318 , l65 fabian , a. c. , sanders , j. s. , taylor , g. b. , & allen , s. w. 2005 , mnras , 360 , l20 frenk , c. s. , baugh , c. m. , & cole , s. m. 1996 , iau symposia 171 , 247 fujita , y. , matsumoto , t. , & wada , k. 2004 , apj , 612 , l9 gardini , a. 2007 , a&a , 464 , 143 govoni , f. , markevitch , m. , vikhlinin , a. , vanspeybroeck , l. , feretti , l. , & giovannini , g. 2004 , apj , 605 , 695 grebenev , s. a. , forman , w. , jones , c. , & murray , s. 1995 , apj , 445 , 607 grevesse , n. , & sauval , a. j. 1998 , space sci . rev . , 85 , 161 horner , d. j. , mushotzky , r. f. , & scharf , c. a. 1999 , apj , 520 , 78 ikebe , y. , ezawa , h. , fukazawa , y. , hirayama , m. , ishisaki , y. , kikuchi , k. , kubo , h. , makishima , k. , matsushita , k. , ohashi , t. , takahashi , t. , & tamura , t. 1996 , nature , 379 , 427 jetha , n. n. , ponman , t. j. , hardcastle , m. j. , & croston , j. h. 2007 , mnras , 376 , 193 kaiser , n. 1991 , apj , 383 , 104 kanov , k. n. , sarazin , c. l. , & hicks , a. k. 2006 , apj , 653 , 184 kauffmann , g. , white , s. d. m. , & guiderdoni , b. 1993 , mnras , 264 , 201 kim , w .- t . , & narayan , r. 2003 , apj , 596 , 889 kotov , o. , & vikhlinin , a. 2006 , apj , 641 , 752 makishima , k. , ezawa , h. , fukuzawa , y. , honda , h. , ikebe , y. , kamae , t. , kikuchi , k. , matsushita , k. , nakazawa , k. , ohashi , t. , takahashi , t. , tamura , t. , & xu , h. 2001 , pasj , 53 , 401 markevitch , m. , & vikhlinin , a. 2001 , apj , 563 , 95 markevitch , m. , mazzotta , p. , vikhlinin , a. , burke , d. , butt , y. , david , l. , donnelly , h. , forman , w. r. , harris , d. , kim , d .- w . , virani , s. , & vrtilek , j. 2003 , apj , 586 , l19 markevitch , m. , & vikhlinin , a. 2007 , phys . , 443 , 1 maron , j. , chandran , b. d. , & blackman , e. 2004 , phys . lett . , 92 , 045001 maughan , b. j. , ellis , s. c. , jones , l. r. , mason , k. o. , c@xmath217rdova , f. a. , & priedhorsky , w. 2006 , apj , 640 , 219 mcnamara , b. r. , wise , m. w. , & murray , s. s. 2004 , apj , 601 , 173 mcnamara , b. r. , & nulsen , p. e. j. 2007 , ara&a , 45 , 117 mohr , j. j. , & evrard , a. e. 1997 , apj , 491 , 38 mushotzky , r. f. 1984 , phys . , 7 , 157 narayan , r. , & medvedev , m. v. 2001 , apj , 562 , l129 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1995 , mnras , 275 , 720 nulsen , p. e. j. , david , l. p. , mcnamara , b. r. , jones , c. , forman , w. r. , & wise , m. 2002 , apj , 568 , 163 osullivan , e. , vrtilek , j. m. , kempner , j. c. , david , l. p. , & houck , j. c. 2005 , mnras , 357 , 1134 peterson , j. r. , kahn , s. m. , paerels , f. b. s. , kaastra , j. s. , tamura , t. , bleeker , j. a. m. , ferrigno , c. , & jernigan , j. g. 2003 , apj , 590 , 207 petrosian , v. , bykov , a. , & rephaeli , y. 2008 , space sci . , 134 , 191 pizzolato , f. , & soker , n. 2006 , mnras , 371 , 1835 pointecouteau , e. , arnaud , m. , kaastra , j. , & de plaa , j. 2004 , a@xmath216a , 423 , 33 ponman , t. j. , cannon , d. b. , & navarro , j. f. 1999 , nature , 397 , 135 ponman , t. j. , sanderson , a. j. r. , & finoguenov , a. 2003 , mnras , 343 , 331 poole , g. b. , fardal , m. a. , babul , a. , mccarthy , i. g. , quinn , t. , & wadsley , j. 2006 , mnras , 373 , 881 quilis , v. , bower , r. g. , & balogh , m. l. 2001 , mnras , 328 , 1091 rebusco , p. , churazov , e. , b@xmath215hringer , h. , & forman , w. 2006 , mnras , 372 , 1840 rood , h. j. , & sastry , g. n. 1971 , pasp , 83 , 313 rosati , p. , della ceca , r. , burg , r. , norman , c. , & giacconi , r. 1995 , apj , 445 , l11 sanders , j. s. , fabian , a. c. , allen , s. w. , & schmidt , r. w. 2004 , mnras , 349 , 952 sanders , j. s. , fabian , a. c. , & taylor , g. b. 2005 , mnras , 356 , 1022 sanders , j. s. , fabian , a. c. , & taylor , g. b. 2009 , mnras , 393 , 71 sanderson , a. j. r. , finoguenov , a. , & mohr , j. j. 2005 , apj , 630 , 191 sanderson , a. j. r. , ponman , t. j. , & osullivan , e. 2006 , mnras , 372 , 1496 sarazin , c. l. 1999 , apj , 520 , 529 schombert , j. m. , west , m. j. , zucker , j. r. , & struble , m. f. 1989 , aj , 98 , 1999 schuecker , p. , finoguenov , a. , miniati , f. , b@xmath215hringer , h. , & briel , u. g. 2004 , a&a , 426 , 387 sharon , k. , gal - yam , a. , maoz , d. , filippenko , a. v. , & guhathakurta , p. 2007 , apj , 60 , 1165 shensa , m. j. 1992 , ieee trans . sig . proc . , 40 , 2464 shibata , r. , matsushita , k. , yamasaki , n. y. , ohashi , t. , ishida , m. , kikuchi , k. , b@xmath215hringer , h. , & matsumoto , h. 2001 , apj , 549 , 228 slezak , e. , durret , f. , & gerbal , d. 1994 , aj , 108 , 1996 snowden , s. l. , egger , r. , finkbeiner , d. , freyberg , m. j. , & plucinsky , p. p. 1998 , apj , 493 , 715 spitzer , l. 1962 , physics of fully ionized gases ( new york : interscience ) spitzer , l. 1978 , physical processes in the interstellar medium ( new york : john wiley & sons ) starck , j .- murtagh , f. , & bijaoui , a. 1995 , adass conf . series , 77 , 279 starck , j .- l . , & pierre , m. 1998 , a@xmath216as , 128 , 397 starck , j. l. , donoho , d. l. , & cand@xmath218s , e. j. 2003 , a&a , 398 , 785 struble , m. f. , & rood , h. j. 1987 , apjs , 63 , 555 sun , m. , jones , c. , murray , s. s. , allen , s. w. , fabian , a. c. , & edge , a. c. 2003 , apj , 587 , 619 takahashi , s. , & yamashita , k. 2003 , pasj , 55 , 1105 takizawa , m. , sarazin , c. l. , blanton , e. l. , & taylor , g. b. 2003 , apj , 595 , 142 thornton , k. , gaudlitz , m. , janka , h .- th . , & steinmetz , m. 1998 , apj , 500 , 95 torrence , c. , & compo , g. p. 1998 amer . met . soc . , 79 , 61 tozzi , p. , & norman , c. 2001 , apj , 546 , 63 van cittert , p. h. 1931 , z. phys . , 69 , 298 vernaleo , j. c. , & reynolds , c. s. , 2006 , apj , 645 , 83 vikhlinin , a. , forman , w. , & jones , c. 1997 , apj , 474 , l7 vikhlinin , a. , mcnamara , b. r. , forman , w. , jones , c. , quintana , h. , & hornstrup , a. 1998 , apj , 502 , 558 vikhlinin , a. , markevitch , m. , & murray , s. s. 2001 , apj , 551 , 160 vikhlinin , a. , markevitch , m. , murray , s. s. , jones , c. , forman , w. , & van speybroeck , l. 2005 , apj , 628 , 665 vikhlinin , a. , kravtsov , a. , forman , w. , jones , c. , markevitch , m. , murray , s. s. , & van speybroeck , l. 2006 , apj , 640 , 691 voigt , l. m .. & fabian , a. c. 2004 , mnras , 347 , 1130 wise , m. w. , mcnamara , b. r. , & murray , s. s. 2004 , apj , 601 , 184 xu , h. , jin , g. , & wu , x .- p . 2001 , apj , 553 , 78 zakamska , n. l. , & narayan , r. 2003 , apj , 582 , 162 a0478 & 2masx j04132526 + 1027551 & 0.0881 & 2 & 04 13 20.7 & + 10 28 35 & 1,3 + a1068 & 2masx j10404446 + 3957117 & 0.1375 & 1 & 10 40 47.1 & + 39 57 19 & 2,3 + a1201 & @xmath219 & 0.1688 & 2 & 11 13 01.1 & + 13 25 40 & + a1650 & 2masx j125841490145410 & 0.0845 & 2 & 12 58 46.2 & @xmath21901 45 11 & + a2104 & @xmath219 & 0.1554 & 2 & 15 40 06.8 & @xmath21903 17 39 & + a2204 & @xmath219 & 0.1523 & 3 & 16 32 45.7 & + 05 34 43 & 2,3 + a2244 & first j170242.5 + 340337 & 0.0968 & 2 & 17 02 44.0 & + 34 02 48 & + a2556 & @xmath219 & 0.0871 & 1 & 23 13 01.6 & @xmath21921 37 59 & 4 + a3112 & eso 248- g 006 & 0.0750 & 2 & 03 17 52.4 & @xmath21944 14 35 & 3 + llllcc a0478 & 1669 & 235678 & faint & 27/01/2001 & 42.9/41.9 + a1068 & 1652 & 236789 & faint & 04/02/2001 & 27.2/26.0 + a1201 & 4216 & 35678 & vfaint & 01/11/2003 & 40.2/25.4 + a1650 & 4178 & 35678 & vfaint & 03/08/2003 & 27.6/26.0 + a2104 & 895 & 235678 & faint & 25/05/2000 & 49.8/48.1 + a2204 & 499 & 235678 & faint & 29/07/2000 & 10.2/9.3 + a2244 & 4179 & 35678 & vfaint & 10/10/2003 & 57.7/56.0 + a2556 & 2226 & 35678 & vfaint & 05/10/2001 & 20.2/19.8 + a3112 & 2516 & 35678 & vfaint & 15/09/2001 & 17.2/15.1 + a0478 & 405.4 & @xmath220 & @xmath221 & @xmath222 + a1068 & 418.7 & @xmath223 & @xmath224 & 1.40 + a1201 & 425.5 & @xmath225 & @xmath226 & 1.61 + a1650 & 390.5 & @xmath227 & @xmath228 & 1.56 + a2104 & 397.6 & @xmath229 & @xmath230 & 8.69 + a2204 & 391.0 & @xmath231 & @xmath232 & 5.67 + a2244 & 396.9 & @xmath233 & @xmath234 & 2.14 + a2556 & 401.3 & @xmath235 & @xmath236 & 2.05 + a3112 & 385.5 & @xmath237 & @xmath238 & 2.61 + cccccccccc a0478[1 ] & 95 & 0.016 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & 1.6 + [ 2 ] & 141 & 0.010 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & 2.0 + [ 3 ] & 155 & 0.009 & @xmath247 & @xmath248 & @xmath249 & @xmath246 & 2.2 + [ 4 ] & 157 & 0.009 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & 2.3 + a1068[1 ] & 184 & 0.004 & @xmath254 & @xmath255 & @xmath256 & @xmath257 & 2.7 + [ 2 ] & 237 & 0.002 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & 3.2 + [ 3 ] & 244 & 0.002 & @xmath262 & @xmath263 & @xmath264 & @xmath265 & 3.3 + a1201[1 ] & 123 & 0.005 & @xmath266 & @xmath267 & @xmath268 & @xmath269 & 2.0 + [ 2 ] & 171 & 0.003 & @xmath270 & @xmath271 & @xmath272 & @xmath273 & 2.7 + [ 3 ] & 185 & 0.003 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & 2.9 + a1650[1 ] & 72 & 0.013 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & 2.4 + [ 2 ] & 108 & 0.007 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & 2.9 + [ 3 ] & 113 & 0.007 & @xmath286 & @xmath287 & @xmath288 & @xmath253 & 3.0 + [ 4 ] & 171 & 0.005 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & 3.6 + a2104[1 ] & 132 & 0.007 & @xmath293 & @xmath294 & @xmath295 & @xmath296 & 2.8 + [ 2 ] & 253 & 0.003 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & 4.0 + [ 3 ] & 261 & 0.003 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & 4.1 + a2204[1 ] & 94 & 0.017 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & 1.4 + [ 2 ] & 117 & 0.013 & @xmath309 & @xmath310 & @xmath311 & @xmath312 & 1.6 + [ 3 ] & 172 & 0.009 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & 2.0 + [ 4 ] & 220 & 0.006 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & 2.4 + a2244[1 ] & 74 & 0.014 & @xmath321 & @xmath322 & @xmath323 & @xmath324 & 1.3 + [ 2 ] & 100 & 0.010 & @xmath325 & @xmath326 & @xmath327 & @xmath324 & 1.7 + [ 3 ] & 155 & 0.006 & @xmath328 & @xmath329 & @xmath330 & @xmath331 & 2.3 + a2556[1 ] & 108 & 0.005 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & 1.8 + [ 2 ] & 141 & 0.004 & @xmath336 & @xmath337 & @xmath338 & @xmath339 & 2.3 + [ 3 ] & 144 & 0.004 & @xmath340 & @xmath341 & @xmath342 & @xmath343 & 2.3 + a3112[1 ] & 73 & 0.014 & @xmath344 & @xmath345 & @xmath346 & @xmath347 & 1.2 + [ 2 ] & 121 & 0.006 & @xmath348 & @xmath349 & @xmath350 & @xmath351 & 1.8 + [ 3 ] & 125 & 0.006 & @xmath352 & @xmath353 & @xmath354 & @xmath355 & 1.9 + [ 4 ] & 141 & 0.006 & @xmath356 & @xmath357 & @xmath358 & @xmath359 & 2.0 + kev x - ray intensity contours are plotted . to calculate the x - ray contours we subtract the background , correct the exposure , and smooth the image with a minimum signal - to - noise ( s / n ) of 3 and a maximum s / n of 5 per smoothing beam . ] kev , which are corrected for both background and exposure . the best - fit two@xmath37 and @xmath37 models are shown with solid and dash lines , respectively . for a2204 only the @xmath37 model is plotted , which is sufficient to describe its sbp . ] as a function of turbulence length scales @xmath160 , which are plotted for the radius of 100 ( solid ) , 125 ( dash ) , 150 ( dash dot ) , 175 ( dot ) , and 200 ( thick solid ) @xmath1 kpc , respectively . the vertical lines mark the turnover points obtained in the wavelet spectra ( 4.1 and fig . 5 ) . ]

we present dynamical scaling relations for a homogeneous and representative sample of massive galaxies , selected only by stellar mass ( m ) and redshift ( ) as part of the ongoing galex arecibo sdss survey . we compare baryonic tully - fisher ( btf ) and faber - jackson ( bfj ) relations for this sample , and investigate how galaxies scatter around the best fits obtained for pruned subsets of disk - dominated and bulge - dominated systems . the bfj relation is significantly less scattered than the btf when the relations are applied to their maximum samples ( for the btf , only galaxies with hi detections ) , and is not affected by the inclination problems that plague the btf . disk - dominated , gas - rich galaxies systematically deviate from the bfj relation defined by the spheroids . we demonstrate that by applying a simple correction to the stellar velocity dispersions that depends only on the concentration index of the galaxy , we are able to bring disks and spheroids onto the same dynamical relation in other words , we obtain a generalized bfj relation that holds for all the galaxies in our sample , regardless of morphology , inclination or gas content , and has a scatter smaller than 0.1 dex . we compare the velocity - size relation for the three dynamical indicators used in this work , i.e. , rotational velocity , observed and concentration - corrected stellar dispersion . we find that disks and spheroids are offset in the stellar dispersion - size relation , and that the offset is removed when corrected dispersions are used instead . the generalized bfj relation represents a fundamental correlation between the global dark matter and baryonic content of galaxies , which is obeyed by all ( massive ) systems regardless of morphology . [ firstpage ] galaxies : kinematics and dynamics galaxies : evolution galaxies : fundamental parameters radio lines : galaxies

1110.4226

the observed global properties of galaxies obey a diverse set of scaling relations , which are fundamental tools to constrain models of galaxy formation and evolution . particularly interesting are the correlations between dynamics and luminosity / stellar mass or size , such as the tully - fisher ( tf ; * ? ? ? * ) relation for spirals , and the faber - jackson ( fj ; * ? ? ? * ) , @xmath4-@xmath5 @xcite , or fundamental plane ( fp ; * ? ? ? * ; * ? ? ? * ) relations for spheroids , because they link the luminous to the total mass of galaxies , thus providing insights into the interplay between their baryonic and dark matter components . historically , the importance of the tight tf and fp relations as secondary distance indicators has driven the community to assemble galaxy samples that meet strict selection criteria , in order to minimize systematic errors and scatter . while fp studies targeted elliptical and s0 galaxies , the samples used for tf applications have typically been restricted to late - type spirals with inclinations to the line - of - sight larger than 30 - 40 , preferably observed in red or infrared photometric bands to minimize extinction effects ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? these data sets are not ideal for characterizing the statistical properties of galaxies in general , because the excessive pruning means that they are not fair samples of the local universe . this issue is particularly problematic for the comparison with theoretical studies and numerical simulations of galaxy formation and evolution , which should be based on representative samples . for instance , tf studies of different classes of objects , such as s0s and early - type spirals ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , polar ring galaxies @xcite , barred spirals @xcite , and gas - rich dwarfs @xcite , have sometimes found disagreement with the tf relation of late - type spirals . most notably , gas - rich dwarfs lie systematically below the tf relation defined by bright galaxies . during the last decades , the interest in tf and fj - like scaling relations has shifted from cosmic flow applications to constraining galaxy formation models , and samples with broader morphological properties have been constructed specifically for this purpose ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? however , current large and homogeneous data sets include either spirals ( e.g. , in addition to the works mentioned above , @xcite and @xcite , hereafter ss11 ) or early - type galaxies ( e.g. , * ? ? ? * ; * ? ? ? . there have been attempts to move beyond the spiral / elliptical dichotomy , and uncover relations between stellar content and dynamics that hold for all galaxies , independent of morphology . @xcite found that all galaxies , from disks to spheroids and from dwarf spheroidals to giant ellipticals , lie on a two - dimensional surface defined by surface brightness , half - light radius , internal velocity and mass - to - light ratio . as a measure of internal velocity @xmath6 , they adopt either the rotational velocity @xmath7 for disks or the velocity dispersion @xmath5 for spheroids , thus the two types of systems are still treated separately ( especially because the sample is a large but heterogeneous collection of published data sets , for which either @xmath7 or @xmath5 is available ) . it is also important to point out that the search for scaling relations that are valid for all types of galaxies should make use of _ baryonic _ masses ( i.e. , the sum of stellar and gas masses ) instead of luminosities or stellar masses , because they could be more fundamental quantities . this is demonstrated by the fact that baryonic scaling relations hold for subsets of galaxies that do not follow the corresponding stellar relations . for example , the offset of the gas - rich dwarf galaxies from the stellar tf relation disappears when their gas mass is taken into account @xcite . the baryonic tf relation ( btf ; * ? ? ? * ) is linear over 5 orders of magnitude in ( stellar + gas ) mass , suggesting that the tf is fundamentally a relation between baryonic ( rather than luminous ) and total mass of the galaxy . intriguingly , although supported by limited statistics , there is some evidence that giant and dwarf ellipticals might lie on the same btf as the spirals @xcite . unfortunately , because they require estimates of both stellar and gas masses , btf samples ( e.g. , * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) are significantly smaller than tf ones . to summarize , it is still unclear if disk - dominated galaxies and spheroids obey the same dynamical scaling relations , mainly due to the lack of well - defined , representative samples of galaxies for which both rotation and stellar dispersion are measured . certainly , ellipticals might have no gas and no detectable rotation , and pure disks might have negligible stellar dispersions , but a significant fraction of local galaxies ( especially massive ones ) have a disk _ and _ a bulge ( e.g. * ? ? ? * ) , and the dynamical scaling relations should account for the smooth transition across galaxies with different bulge - to - disk ratios . as @xcite point out , a connection between the scaling relations of early - type and late - type galaxies is expected on the grounds that early - type systems are generally assumed to form through mergers of late - type ones . this is especially true at the high stellar mass end , where the blue sequence of star - forming disks merges onto the red sequence of passively - evolving , bulge - dominated galaxies ( e.g. * ? ? ? * ) , and the systems typically host a bulge and a disk . in this paper , we investigate dynamical scaling relations for a representative sample of @xmath0500 massive galaxies that are selected _ only _ by stellar mass ( @xmath8 m@xmath2 ) and redshift ( @xmath3 ) , as part of the ongoing galex arecibo sdss survey ( gass ; * ? ? ? * hereafter paper i ) . for these galaxies , we have homogeneous measurements of structural parameters and velocity dispersions from the sloan digital sky survey ( sdss ; * ? ? ? * ) , nuv@xmath9 colours from galex @xcite and sdss imaging , and hi masses and rotational velocities for the subset of objects detected at 21 cm with the arecibo radio telescope . this unique sample , which includes massive galaxies of all morphological types , allows us to investigate how objects that are typically not included in tf or fj / fp data sets scatter around those relations . in the spirit of works like those of @xcite and @xcite , we wish to establish if there is a _ fundamental _ correlation between baryonic mass and dynamics that is obeyed by the complete galaxy population , regardless of morphology . we show that , at least for the massive galaxies in our sample , such a relation does exist , and has a scatter smaller than 0.1 dex , comparable to that of the tf and fj relations applied to their respective pruned subsets . this paper is organized as follows . we summarize sample selection and measurements of relevant quantities in [ s_sample ] . we present the baryonic mass - velocity relations , starting with the tf and fj , in [ s_bvrel ] , and the velocity - size relations in [ s_vsize ] . we discuss our findings and conclude in [ s_disc ] . all the distance - dependent quantities in this work are computed assuming @xmath10 , @xmath11 and @xmath12 km s@xmath13 mpc@xmath13 . the sample used in this work is drawn from gass , an on - going survey which is gathering high - quality hi - line spectra for @xmath01000 massive galaxies , selected only by stellar mass ( greater than @xmath14 m@xmath2 ) and redshift ( @xmath15 ) . the gass targets are located within the intersection of the footprints of the sdss primary spectroscopic survey , the projected galex medium imaging survey and the on - going hi blind arecibo legacy fast alfa ( alfalfa ; * ? ? ? * ) survey . the galaxies are observed with the arecibo radio telescope until detected or until a gas fraction limit of @xmath16% is reached . for more details we refer the reader to paper i , where the first gass data release ( dr1 ) is presented . here , in addition to the dr1 data ( 176 galaxies ) , we use new arecibo observations of 240 galaxies that will be incorporated in the second gass data release ( catinella et al . , in preparation ) . as discussed in paper i , gass does not re - observe objects with good hi detections already available from the alfalfa survey or the cornell hi archive ( * ? ? ? * hereafter s05 ) . to correct the gass sample for its lack of hi - rich objects , we add galaxies from alfalfa and s05 in the correct proportions , following the procedure detailed in section 7.2 of paper i. the sample thus obtained , which is representative in terms of hi properties , includes 480 galaxies ( 296 detections and 184 non - detections ) . as explained below , we discard 44 galaxies for which the stellar velocity dispersion is not reliable . the final sample includes 436 galaxies ( 259 detections and 177 non - detections ) . as mentioned in paper i , the optical parameters are obtained from queries to the sdss dr7 @xcite data base server , unless otherwise noted . stellar masses are derived from sdss photometry using the spectral energy distribution ( sed ) fitting technique described in @xcite , assuming a @xcite initial mass function . a variety of model seds from the @xcite library are fitted to each galaxy , building a probability distribution for its stellar mass . the mean and the width of this distribution are used as measurements of the stellar mass and its formal error , respectively . over the interval probed by gass , stellar masses are believed to be accurate to better than 30 per cent . as a measure of galaxy size we adopt @xmath17 , i.e. half the 25 mag arcsec@xmath18 isophote diameter @xmath19 ( measured by us on the sdss _ g_-band images ) , in kpc . the baryonic mass is the sum of stars and gas ; the latter is computed from the hi mass , adopting the standard 1.4 correction factor to account for helium and metals , i.e. @xmath20 . this correction neglects the contribution of the molecular hydrogen . in our stellar mass regime , the amount of @xmath21 does not depend significantly on @xmath22 or concentration index and , on average , @xmath23 ( but with a large scatter , 0.41 dex ; @xcite ) . since the contribution of the hi to the baryonic mass is , for our sample , typically small ( see [ s_btf ] ) , it is safe to neglect the @xmath21 ( but we did check that including the molecular gas does not change our results ) . we set the hi masses of the non - detections to zero , but we note that using the upper limits ( which correspond to 1.5 - 5% of the stellar mass by survey design ) would make no difference to our plots . other parameters used in this work are discussed below . hi line widths from gass , alfalfa and the s05 archive are measured with the same technique , at the 50% of each peak level ( e.g. , * ? ? ? * ) . however , the corrections applied to the raw measurements , as originally published , are different . gass and s05 widths are corrected for both instrumental broadening and cosmological redshift , whereas alfalfa widths are corrected for instrumental broadening only , following equation 1 in @xcite . no turbulent motion or inclination corrections are applied . for convenience , we report here the corrections adopted : @xmath24 @xmath25 where @xmath26 is the measured velocity width , @xmath27 is the galaxy redshift , and @xmath28 is the instrumental broadening correction , which differs for the three sources . for gass paper i and alfalfa , @xmath28 is simply the final velocity resolution of the spectrum after smoothing ( i.e. , between 5 and 21 km s@xmath13 for gass spectra , which are hanning and boxcar smoothed , and @xmath010 km s@xmath13 for alfalfa spectra , which are hanning smoothed only ) . for s05 , @xmath28 equals @xmath29 , where @xmath30 is the channel separation in km s@xmath13 and @xmath31 is a complex function of signal - to - noise ratio ( snr ) and type of smoothing applied . for high snr , hanning - smoothed spectra with @xmath32 km s@xmath13 ( i.e. , the alfalfa case ) , @xmath33 , thus @xmath28 is of order of half the velocity resolution after smoothing . this is smaller than the correction adopted for gass in paper i , and larger than the alfalfa one , where the subtraction in quadrature makes the correction negligible ( i.e. , @xmath34 km s@xmath13 ) for velocity widths larger than 50 km s@xmath13 . the issue of homogenizing velocity widths obtained from different sources has been discussed in the past , and most recently by @xcite for the extragalactic distance database @xcite , a large compilation of data for galaxy distance and peculiar motion studies . we follow @xcite , who adopt a simplified version of the s05 solution , i.e. : @xmath35 , where @xmath36 and @xmath37 is the final velocity resolution of the spectrum after smoothing ( this is half the correction adopted in paper i ) . this is in better agreement with our own tests on high snr gass hi profiles , where the smoothing was gradually increased and the width remeasured . thus , we keep the s05 widths as published , we adopt the new @xmath28 correction for the gass data , and we uncorrect the alfalfa widths , @xmath38 , to obtain the raw @xmath26 measurements , then apply the same @xmath28 and @xmath39 corrections adopted for gass galaxies . lastly , we deproject all velocity widths to edge - on view . the inclination to the line - of - sight is computed from : @xmath40 where @xmath41 is the minor - to - major axis ratio from the _ r_-band exponential fit ( _ expab_r _ in the sdss database ) , and @xmath42 is the intrinsic axial ratio of a galaxy seen edge - on . we adopt @xmath43 ( @xcite ; see also , e.g. @xcite ) , and set the inclination to 90 for galaxies with @xmath44 . the value of @xmath43 applies to galaxies of morphological type earlier than sbc , whereas @xmath45 should be used for sbc and later spirals ( e.g. @xcite ; ss11 ) . however the difference is small , and our choice of @xmath42 seems appropriate for a sample of massive galaxies that was not selected for tf studies . thus , rotational velocities are computed as : @xmath46 where @xmath37 is the final velocity resolution of the spectrum after smoothing . as noted in paper i , the hi masses of galaxies in the alfalfa and s05 archives have been recomputed from the tabulated fluxes , in order to be made consistent with gass ones . we identified 31 galaxies for which hi confusion within the @xmath04 arecibo beam is certain , and therefore the measured hi parameters should not be trusted . specifically , we closely inspected the sdss images of all the galaxies in our sample , and flagged as `` confused '' those with at least one late - type , similar size companion ( based on sdss spectroscopy , i.e. with redshift available and within 0.002 of that of the target galaxy ) within the beam . these objects will be highlighted in our analysis when needed . galaxy velocity dispersions are measured by fitting stellar templates convolved with gaussian functions to sdss spectra , which are obtained through 3-diameter fiber apertures . because the fiber covers only a fraction of the galaxy light at the gass redshifts , these quantities ( catalogued as _ veldisp _ in the sdss spectroscopic data base , and here referred to as @xmath47 ) need to be corrected for aperture effects . as commonly done ( see , e.g. * ? ? ? * ; * ? ? ? * and references therein ) , fiber velocity dispersions are corrected to 1/8 effective radius as follows : @xmath48 where @xmath49 and @xmath50 is the circular galaxy radius in arcseconds , computed as @xmath51 ( @xmath52 and @xmath53 are the effective radius and axis ratio from the _ r_-band de vaucouleurs fit , respectively ) . this correction is small , @xmath03% on average , for the galaxies in our sample . only values of @xmath47 above the @xmath070 km s@xmath13 instrumental resolution of the sdss spectra are considered reliable . we thus discarded 44 galaxies that do not meet this requirement from our sample ; the effect of this restriction on the stellar mass distribution of the sample is shown in figure [ mstar ] . that are not included in our analysis.,width=264 ] the main goal of this work is to establish if there is a relation between baryonic mass and a measure of the dynamical mass ( estimated based on rotational velocity , stellar dispersion or a combination of the two ) that is obeyed by massive galaxies _ regardless of their morphology_. we begin by comparing the baryonic tf ( btf ) and baryonic fj ( bfj ) relations for our sample , in order to establish which quantity , the hi rotational velocity or the stellar dispersion , most reliably traces the baryonic mass of massive galaxies . thanks to its unique selection by stellar mass and redshift only , the gass sample includes massive galaxies of all morphological types , and it is thus ideal for this comparison . indeed , this is not a sample designed for either tf or fj studies , but we can prune it to separately study inclined , disk - dominated and spheroidal , bulge - dominated systems . most importantly , we can investigate outliers and residuals of the btf and bfj relations , which is essential in order to determine how to bring disks and spheroids onto the same relation . we conclude by illustrating two different ways of obtaining a baryonic relation that holds for all the massive galaxies in our sample , and whose dispersion is comparable to that of the btf and bfj relations . here we must restrict our sample to galaxies with hi detections , for which the rotational velocity can be measured . the btf relation for this sample is shown in figure [ btf]a . we note that , for our interval of stellar masses , the contribution of the gas to the baryonic mass is generally small if we plotted stellar instead of baryonic masses , most of the points would move downward by an amount comparable to the symbol size ( indeed , there are only 13 galaxies in our sample with gas fractions @xmath54/@xmath22@xmath55 ) . we plot the btf as usually done ( i.e. , baryonic mass versus rotational velocity ) , but we fit the inverse relation , since the scatter is clearly associated to the @xmath7-coordinate ( as can be seen by comparing figures [ btf]a and [ btf]b , explained below ) . the inverse fit to the data points , i.e.log @xmath7= a log ( @xmath22 + 1.4 @xmath54 ) + b , is shown as a dotted line and [ t2_fits ] list the expression of the fit , the scatter , and the number of galaxies contributing to the sample for all the main relations discussed in this work . ] . following ss11 , we compute the scatter in the @xmath56 variable around the best fit by applying tukey s bi - weight , which yields a robust estimate of the dispersion in presence of outliers ; the scatter ( in dex of velocity ) is noted at the bottom of the panel . also shown is the relation obtained by @xcite over nearly 5 orders of magnitude of baryonic mass ( solid line ) , which is in excellent agreement with our data . not surprisingly , the scatter around the best fit is large for our data set ( 0.127 dex ) , and discarding galaxies with confused hi spectra ( magenta symbols ; see [ s_hiwidths ] ) improves it only by a small amount ( 0.112 dex ) . however , we recover a significantly tighter relation when we prune the sample as usually done for tf studies . this is illustrated in figure [ btf]b , where a _ tf subset _ was obtained by selecting non - confused , disk - dominated , inclined galaxies ( i.e. , objects with concentration index @xmath57@xmath58 and inclination @xmath59 ; the scatter of the btf for this subset is 0.076 dex ) . we inspected the three low - velocity outliers that stand out on the left of the plot . two of them are galaxies for which the sdss inclination is clearly overestimated ( we measured 17 and 26 instead of 46 and 49 , respectively ) , and the third one has uncertain hi velocity width . removing these outliers , the scatter of the relation becomes 0.072 dex . how does this compare with the scatter of other tf samples ? there is a vast literature on the tf relation , showing that its scatter depends on the sample analyzed , velocity indicator , photometric band , and type of fit performed ( forward , inverse , bisector or orthogonal ) . @xcite studied the btf relation using a sample of 60 galaxies with extended hi rotation curves , and obtained a scatter of 0.191 dex in solar masses from a direct fit ( this is the average scatter for the relations marked as @xmath60 in his table 2 , where the stellar masses are computed using stellar population synthesis models as we do ) . for comparison , a direct fit to the data in figure [ btf]b ( excluding the three low - velocity outliers mentioned above ) yields a scatter of 0.221 dex in solar masses . a study that is directly comparable to ours is that of @xcite , who measured a scatter of 0.06 dex in velocity for the btf of a sample of 76 non - interacting disk galaxies , with inclinations @xmath61 and flat rotation curves ( see also , e.g. , * ? ? ? given the crude selection of our tf subset ( by concentration index ) , our sample is likely to include a broader range of disk galaxy types , thus the slightly larger scatter compared to these studies is not unexpected . figure [ btf]c demonstrates that the high - velocity outliers of the btf relation are all galaxies with small inclination to the line - of - sight . for these systems , the corrected rotational velocities become so unreliable that the correlation with baryonic mass effectively disappears . once galaxies with inclination smaller than 40 are removed from the sample , even bulge - dominated objects lie reasonably close to the relation obtained for disk - dominated ones , although with larger scatter ( empty circles in fig . [ btf]d ) . this agrees qualitatively with the findings of @xcite , based on a compilation of 792 galaxies with inclination larger than 30 and heterogeneous measurements from hyperleda @xcite . indeed , ho showed that a @xmath62-band tf relation exists for all hubble types , including elliptical and s0 galaxies ( although these represent a very small fraction of his sample ) . figure [ btf]c also shows a population of low - velocity outliers that can not be explained by small inclinations : these are galaxies with rotational velocities that are smaller than the stellar velocity dispersions ( red symbols ) . we note that @xcite also identifies a population of galaxies with unusually small @xmath63ratios , which are outliers in his tf relation . except for their small rotational velocities compared to their central stellar velocity dispersions , his outliers are otherwise normal luminous galaxies . these systems might be very interesting ho argues that a significant fraction of their hi gas must be dynamically unrelaxed , having been acquired through a minor merger episode or perhaps cold accretion . for our sample , the fraction of low @xmath63 outliers is smaller ( @xmath05% , against 17% for ho s sample ) . as shown in the appendix , where these galaxies are described in more detail , most of them have asymmetric hi profiles , suggesting that the hi distribution and/or kinematics might be disturbed ( although we can not prove that the gas was externally accreted ) . lastly , here and in other figures of this paper we highlight the positions of two interesting galaxies that were discussed in our previous works , gass 3505 and 35981 ( marked as a red and a blue star in fig . [ btf]a ) . these are both unusually hi - rich systems , which are outliers in the gas fraction plane discussed in paper i ( relating the hi mass fraction to stellar mass surface density @xmath64 and nuv@xmath9 color ) . however , gass 3505 is an early - type system , whereas gass 35981 ( ugc 8802 ) is a disk galaxy with a sharp metallicity drop in its outer disk , which suggests an external origin for the hi gas ( @xcite ; see also @xcite ) . as can be seen , from a dynamical point of view gass 35981 is not unusual , whereas gass 3505 is still an outlier . we now investigate how the residuals of the btf relation depend on a few representative galaxy parameters . the left column of figure [ residuals ] shows the btf residuals for the sample in figure [ btf]a with respect to the best inverse fit obtained for the _ tf subset _ in figure [ btf]b . in other words , residuals are computed as @xmath65 , where @xmath56 is the measured rotational velocity and @xmath66 is the value expected from our best btf relation for a galaxy with the same baryonic mass . cyan filled circles indicate objects with inclination smaller than 40 . from top to bottom , residuals are plotted as functions of concentration index , inclination , nuv@xmath9 color , stellar mass surface density , gas fraction , and distance from the gas fraction plane mentioned above and described in paper i ( galaxies more hi - rich than the average have positive distance ) , respectively . btf residuals do not exhibit strong dependence on structural and star - forming galaxy properties , except for a mild tendency toward increased scatter for more bulge - dominated , red galaxies , which largely disappears when the systems with low inclinations are removed from the sample . this is not surprising , as many attempts to identify a third parameter to minimize the scatter of the tf relation produced negative results ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . the results presented in this section show that the btf is not very promising for our purpose of determining a relation between baryonic and dynamical mass that holds for all massive galaxies . firstly , its application is restricted to galaxies with hidetections and inclinations larger than 40 . secondly , the bulge - dominated systems with hi detections are not simply offset from the relation defined by the disk - dominated galaxies , but are also more scattered . as demonstrated in the following two sections , the bfj relation does not suffer from these limitations , and can be generalized to include both disk - dominated and spheroidal systems . we now carry out a similar analysis for the bfj relation . since all the galaxies have a measurement of the stellar velocity dispersion from sdss , we can use here the full sample , but it is instructive to keep hi detections and non - detections separated . in figure [ bfj ] the baryonic mass is plotted as a function of the stellar velocity dispersion , and green upside - down triangles indicate galaxies that were not detected with arecibo . the relation plotted for the full gass sample ( panel a ) shows a clear segregation between hi detections and non - detections , the former being offset towards lower values of velocity dispersion . the gas - rich elliptical gass 3505 ( red star ) is not unusual in terms of its stellar dispersion ( gass 35981 is not plotted because it has @xmath67 km s@xmath13 ) . we recover a tighter relation when we restrict our sample to the _ fj subset _ ( panel b ) , which includes bulge - dominated galaxies ( i.e. , objects with @xmath57@xmath68 ) with inclinations smaller than 70 . we excluded from the subset highly flattened systems because these are likely to host a significant disk component , despite their high concentration index . the 26 bulge - dominated galaxies excluded by our inclination cut are shown as filled symbols in figure [ bfj]c . these are mostly hi detections , which already suggests that they are more likely to be disks than oblate spheroids . their sdss images ( figure [ bfj_edgeons ] ) confirm that they are inclined disks , often with dust lanes running along their major axis . as demonstrated by the bottom panels of figure [ bfj ] , disk - dominated galaxies are systematically offset from the best fit relation obtained for the data points in panel ( b ) , and they are mostly hi detections . these objects have low stellar velocity dispersions for fixed baryonic mass , thus they increase the scatter of the fj relation when they are plotted together with the bulge - dominated galaxies . a.,width=302 ] because fj studies usually do not include the gas component , we check our results by considering the _ stellar _ fj relation . in particular , we compared our fj relation with the one obtained for a large sample of sdss galaxies by @xcite , which is conveniently expressed in terms of stellar masses . the slope and scatter of our fj relation ( restricted to the _ fj subset _ ) are in remarkable agreement with those published by @xcite ( both relations have a scatter of 0.071 dex ) , whereas the zero points differ by 0.17 dex in solar masses . the @xcite sample includes galaxies with concentration index @xmath69 , redshift @xmath70 , and whose sdss spectra have a median signal - to - noise per pixel greater than 20 . removing our inclination cut does not bring the zero points of the two relations into agreement ( the offset is still 0.14 dex ) . however , @xcite note that their stellar masses are systematically larger than those derived by @xcite ( and adopted here ) by @xmath00.1 dex , thus our results are completely consistent . lastly , we notice that adding the gas to the stellar mass slightly increases the scatter of the relation ( from 0.071 dex for the fj to 0.074 dex for the bfj ) . this is because the most hi - rich galaxies in our sample ( see for instance the blue points in fig . [ bfj]d ) become even further displaced from the hi non - detections , which constitute the bulk of the _ fj subset_. a relation with similar scatter is obtained when , instead of restricting the sample to bulge - dominated systems , we consider only the bulge component itself . the mass of the bulge can be estimated from the correlation between sdss concentration index and bulge - to - total stellar mass ratio , b / t , presented by ( * ? ? ? * see their fig . 1 ) , which is based on the 2d multicomponent decomposition analysis of @xcite . our bilinear fit to their data yields b / t=(@[email protected])/2.276 . the relation between bulge mass and stellar velocity dispersion plotted in figure [ mbulge ] has a scatter of 0.075 dex in velocity , comparable to that of the fj and bfj relations applied to the fj subset ( see also * ? ? ? * ) . incidentally , it is somewhat surprising that we can recover such a tight relation given the uncertainties in our crude b / t estimate . we now focus again on the _ baryonic _ fj relation presented in figure [ bfj ] and study its residuals . the bfj residuals are plotted as functions of @xmath57 , inclination , nuv@xmath9 color , stellar mass surface density , gas fraction and distance from gas fraction plane on the right column of figure [ residuals ] . as for the btf , these are the residuals for the sample in figure [ bfj]a with respect to the best fit relation in figure [ bfj]b , computed as described in [ s_btf ] ( where @xmath56 is now stellar velocity dispersion ) . as in all the figures in this paper , green upside - down triangles indicate hi non - detections . interestingly , bfj residuals show clear trends with concentration index , nuv@xmath9 color , @xmath64 , and gas fraction , in the sense that more disky , gas - rich and star - forming galaxies have larger residuals . a similar trend is seen as a function of d@xmath724000 index , which is an indicator of the age of the stellar population sampled by the sdss 3-diameter fiber ( larger residuals are seen for smaller values of d@xmath724000 , not shown ) . it is well known that the fj relation for elliptical galaxies is a projection of a more general fundamental plane @xcite , which is usually parametrized in terms of stellar velocity dispersion , effective radius @xmath73 , and luminosity or surface brightness ( central , or average within @xmath73 ) . thus , one might expect to see a dependence of the fj residuals on a third parameter related to @xmath73 . however , the trends in figure [ residuals ] are _ not _ a consequence of the fundamental plane . firstly , our sample does not include only elliptical galaxies and spheroids . as can be seen by inspecting figure [ residuals ] , the trends are driven by the disk - dominated , star - forming galaxies . secondly , quantities such as @xmath57 , nuv@xmath9 color or gas fraction are not directly related to an effective radius . the comparison of the btf and bfj relations for our sample of massive galaxies illustrates several important points : ( a ) the bfj is significantly less scattered than the btf when the relations are applied to their maximum sample . when the two relations are applied to their respective `` good '' , morphologically - pruned subsets , the scatter in both is almost identical . ( b ) the bfj is insensitive to the inclination problems that plague the btf , which can be applied only to systems with inclination larger than 40 . furthermore , stellar dispersions are measured also for galaxies without hi detections . naturally , one could measure rotational velocities with other tracers and methods , e.g. , with @xmath74rotation curves . however , these have their own sets of problems ( for instance , @xmath74 emission is typically significantly less extended than hiemission , thus it may not trace the full rotational velocity . moreover , it is unclear at which spatial position the rotational velocity should be measured . see , e.g. , @xcite ) , and the inclination issue remains . ( c ) most importantly , and contrary to the btf case , the bfj residuals show systematic trends with other galaxy properties . as shown in figures [ residuals ] and [ bfj ] , disk - dominated galaxies do not form a scatter plot in the bfj plane , but they systematically deviate from the main relation defined by the bulge - dominated systems . one might wonder if , for disk - dominated , inclined galaxies , @xmath5 is strongly affected by rotation . in other words , the bfj for highly inclined disks ( figure [ bfj]c and [ bfj]d ) could just be a tf relation in disguise . from the average rotation curves of disk galaxies of @xcite , we estimated that the typical rotational velocity reached at 1.5 by gass objects is @xmath0100 km s@xmath13 ( but could be up to twice as large for the most luminous and/or most distant galaxies in our sample ) magnitudes into cousins @xmath75 magnitudes following @xcite , and adopted @xmath76 , the radius containing 90% of the petrosian flux in the sdss @xmath77-band , as a proxy for the optical radius @xmath78 , which is the radius encompassing 83% of the total _ i_-band light of the galaxy . the average _ i_-band absolute magnitude for our sample is @xmath7122.35 mag , and the average coverage of the sdss fiber radius is 14% @xmath78 . ] . however , this does not take inclination and bulge - to - disk ratio within the aperture into account . most importantly , the fiber measurements are intensity - weighted , thus most of the disk stars contributing to the dispersion are those closer to the galaxy center , where @xmath7 is negligible . thus we conclude that @xmath5 measurements are not significantly affected by contamination from disk stars in circular orbits . in summary , the sdss stellar velocity dispersions provide an estimate of dynamical mass that at least in the stellar mass and redshift regime probed by gass is not only more generally applicable ( i.e. , not limited to inclined galaxies with hi detections or with extended @xmath74 rotation curves ) , but also less affected by measurement problems ( which are largely responsible for the btf outliers ) compared to @xmath7 . moreover , the fact that the disk - dominated galaxies follow a bfj relation that is simply offset from that of the bulge - dominated systems suggests a simple way of bringing the two onto the same relation . this is accomplished in the next section . shown in figure [ residuals ] ( top right panel ) . as in other figures , hi non - detections are indicated as green triangles.,width=302 ] spheroids and disk - dominated galaxies can be brought onto the same bfj - like relation by correcting the velocity dispersions for the trends observed in figure [ residuals ] . the tightest dependencies of the bfj residuals are seen as a function of nuv@xmath9 color and concentration index . although the correlation with nuv@xmath9 is slightly tighter ( the dispersion of the linear fit is 0.073 dex ) , we decided to correct for the dependence on @xmath57 , which is a galaxy structural parameter . in fact , it is reasonable to expect larger corrections for more disk - dominated objects , where rotation is likely to be the main contributor to dynamical support . on the other hand , nuv@xmath9 color is linked to the star formation properties of the galaxy , thus the rationale for correcting velocity dispersions based on this quantity is much less evident . moreover , a correction based on nuv@xmath9 has the additional disadvantage of requiring uv photometry . ( b ) , and generalized bfj relation ( c).,width=302 ] the best linear fit to the data points in the top right panel of figure [ residuals ] is : @xmath79 where the residuals @xmath80 are computed with respect to the best fit bfj relation in figure [ bfj]b . we thus corrected the velocity dispersions @xmath81 of all the galaxies in our sample by subtracting the offset @xmath82 . the bfj relation obtained using the corrected velocity dispersions is plotted in figure [ bfj_corr ] . the inverse fit is almost indistinguishable from that of figure [ bfj]b , and has a scatter of 0.079 dex ( the scatter in solar masses , obtained from a direct fit , is 0.218 dex ) . this relation holds for all the galaxies in our sample , regardless of morphology , inclination or gas content , and is the main result of this work . we compared our results with the simulations performed by @xcite , which follow the formation of eight milky way - mass halos in a @xmath83-cold dark matter cosmology , including baryonic physics ( star formation , metal cooling , chemical enrichment , multiphase gas , thermal feedback from supernovae ) . for these simulated galaxies we can measure stellar and cold gas masses , and stellar velocity dispersions at a given radius , and calculate concentration indices ( from the ratio of the radii enclosing 90% and 50% of the total luminosity ; we used _ r_-band luminosities computed from the dust - free @xcite population synthesis models ) . we estimate total masses and luminosities within 10 kpc ( the mean value of @xmath84 for the galaxies in the gass sample ) , and obtain luminosity - weighted stellar velocity dispersions within 1.5 kpc ( the physical size subtended by a 1.5 radius at z=0.05 ) . using the concentration indices , we correct the stellar dispersions according to equation [ eq_modsigma ] . the results are illustrated in figure [ bfj_10kpc ] , where we compare the positions of the simulated galaxies ( red stars ) with those of the fj subset on the bfj plane ( a ) . the residuals from the bfj fit are plotted as a function of @xmath57 in ( b ) , and the generalized bfj relation is shown in ( c ) . the simulated galaxies are all disk - dominated according to our definition ( i.e. , @xmath85 ) except one , and their stellar dispersions are systematically offset from the fit in ( a ) towards smaller values , as the disk galaxies in the gass sample are . the correction applied to their stellar dispersions removes the offset , and the simulated galaxies lie on top of the relation in ( c ) . although it is based on eight halos only , the excellent agreement between simulations and data is encouraging . we have shown that disk - dominated galaxies are offset in the bfj plane , and that this is what allow us to obtain a tight baryonic mass - velocity relation that holds for all the galaxies in our sample . the reason why there is a _ disk _ bfj relation for massive galaxies is that @xmath5 is proportional to @xmath7 , and their ratio is a function of galaxy morphology . as demonstrated by @xcite , for a given @xmath7 , earlier type galaxies have higher @xmath5 . the brightest , bulge - dominated galaxies ( ellipticals , lenticulars and early - type spirals ) lie on the @xmath86 relation expected for isothermal stellar systems . later - type spiral and dwarf galaxies , on the other hand , depart from the isothermal relation by an amount that depends on morphology or total light concentration ( see their fig . 1 ) . we plot the relation between @xmath7 and @xmath5 for the hi - detected gass galaxies in figure [ vsigma ] . our sample shows two populations of outliers , characterized by too large or too low @xmath63 ratios compared to the rest of the data set . these are the same outliers of the btf relation ( see fig . [ btf ] ) ; in particular , galaxies with large @xmath63 ratios all have small inclinations ( @xmath87 , cyan ) , and objects with @xmath88 ( red ) were mentioned in [ s_btf ] and are discussed in the appendix . to avoid crowding , we show error bars for these galaxies only axis ratio in eq . 1 ) only . the errors on stellar velocity dispersions are from the sdss ( as noted in [ s_vdisp ] the aperture correction is very small , thus we do not propagate its uncertainty ) . ] . disregarding the outliers , one can see that the galaxies with larger stellar dispersions lie very close to the isothermal relation ( indicated by a dotted line ) , whereas there is a tail of galaxies with higher @xmath63 ratios at lower dispersions . the galaxies in this tail are all disk dominated . correcting the stellar dispersions according to equation [ eq_modsigma ] effectively translates into accounting for this departure from isothermality for disk - dominated galaxies . . dashed and dotted lines indicate the 1:1 correlation and the @xmath89 relation , respectively . error bars are plotted only for galaxies with inclination @xmath77 smaller than 40 ( cyan ) or with @xmath88 ( red ) that are not affected by beam confusion ( magenta diamonds ) . red symbols with cyan error bars are galaxies with both @xmath88 and @xmath90 . , width=302 ] in the previous section , we obtained a relation between baryonic mass and a measure of internal velocity that holds for our sample of massive galaxies , without any morphological pruning . we now ask whether we can find an even tighter relation by combining @xmath7 and @xmath5 measurements . both quantities measure the depth of the gravitational potential well of a galaxy , but they do so at different radii , namely those of the bulge and of the outer disk ( as traced by the hi gas , which typically extends well beyond the stellar disk ; however in very high - density environments hi could be severely stripped ; e.g. , @xcite ) . in principle , hi widths should provide the best measurement of the total mass ( including dark ) of a galaxy , regardless of the presence of a bulge component . however , as discussed above , reliable @xmath7 measurements are not possible for galaxies with small inclination to the line - of - sight in the limit of a perfectly face - on system , which has no measurable rotation , the velocity dispersion of the stellar component is a more useful quantity . thus , we have two methods of estimating the dynamical mass of a galaxy that are affected by different systematics and limitations . can we combine them in order to obtain a quantity that is , _ on average _ , more tightly correlated with the baryonic mass ? an interesting quantity in this context is the @xmath91 parameter ( e.g. , * ? ? ? * ) : @xmath92 this expression has been used in several studies , with different meanings for the dispersion component . @xcite studied the tf relation for an emission line - selected sample of 544 galaxies at @xmath93 , and demonstrated that a remarkably tighter relation is obtained when the @xmath91 parameter is adopted instead of the rotational velocity from the optical rotation curve . in their case , @xmath5 is the dispersion of the gas ( but this may be contaminated by velocity gradients , see their section 4.2 ) . interestingly , their sample includes early to late spirals , irregular galaxies and merging systems , and their @xmath91-stellar mass relation for the lowest redshift bin is in very good agreement with the fj relation of @xcite in terms of slope and zero point . more relevant for our work , @xcite used the @xmath91 parameter ( which they refer to as internal velocity , @xmath6 ) with @xmath5 measuring the stellar velocity dispersion of spheroidal galaxies . they showed that all classes of galaxies lie on a two - dimensional surface , which is expressed as a linear combination of the logarithms of the effective radius @xmath73 , the internal velocity squared , the surface brightness within @xmath73 , and the mass - to - light ratio within @xmath73 . their sample is a heterogeneous collection of 1925 spheroids and disk galaxies from existing data sets , which span the full range of galaxy types and luminosities . however , although they define @xmath94 , they use _ either _ the circular velocity @xmath95 for disk galaxies or the stellar velocity dispersion @xmath5 for spheroids , but never add the two contributions . another interesting analysis was carried out by @xcite , who studied the evolution of the stellar mass tf relation in a simulation of a disk merger . they argue that , since early - type galaxies are generally assumed to form through mergers of late - type systems , the scaling relations of early types should descend from those of late - type galaxies . thus , they simulate the evolution of a galaxy merger , mimic observations of emission lines , and study how the kinematics of the system changes with time . the progenitors are two identical sbc galaxies lying on the stellar mass tf relation , which merge and form a rotating elliptical galaxy . intriguingly , they show that , while rotation is converted into stellar dispersion , the @xmath91 parameter is approximately conserved suggesting that @xmath91 might be really tracing the mass distribution . the above studies motivated us to look at the relation between baryonic mass and @xmath91 parameter for the galaxies in our sample , with an important caveat . the @xmath91 parameter defined by equation [ eq_s05 ] has a physical meaning only if rotation and dispersion are associated to the _ same _ system . for instance , for a rotating isothermal sphere the dynamical support is provided by both ordered and thermal motions , and the above combination is a direct result of the virial theorem . for our galaxies , @xmath7 measures the rotational velocity of the gas , and @xmath5 is the velocity dispersion of the stars in the bulge , thus combining the two _ has no physical motivation_. however , we are not trying to partition the dynamical support of a galaxy between its bulge and disk components as we already pointed out , @xmath7 should account for the full dynamical support , regardless of the presence of a bulge . we wish instead to establish if we can average two measures of the potential well of a galaxy , which are based on different tracers , in order to obtain a quantity that , empirically , better correlates with the baryonic mass . after all , we expect the hi width to be a more reliable measurement for inclined disk galaxies , and @xmath5 to be better for elliptical or face - on , disk galaxies . thus , the combination of the two might work better _ on average_. the correlation between baryonic mass and @xmath91 parameter is plotted in figure [ s05 ] for the full sample ( a ) , for the galaxies with inclinations larger than 40 ( b ) , and for the tf and fj subsets ( c , d ) defined in [ s_btf ] and [ s_bfj ] . the @xmath91 relation for the full sample has a scatter of 0.094 dex , very close to that of the bfj relation for the full sample , but significantly larger than that of the generalized bfj plotted in figure [ bfj_corr ] . the @xmath91 relation still suffers from the inclination problems that affect the btf , although to a lesser degree . the strongest outliers seen in figure [ s05]a are removed by our 40 inclination cut ( b , c ) . notice that , when applied to disk - dominated galaxies with inclinations larger than 40 , the @xmath91 relation has a scatter of only 0.064 dex , i.e. , it is tighter than the btf shown in figure [ btf]b ( which has a scatter of 0.076 dex ) this is in fact the tightest relation presented in this work ( for comparison , the generalized bfj relation restricted to the same sample has a scatter of 0.077 dex ) . conversely , restricting the sample to bulge - dominated galaxies does not decrease the scatter of the @xmath91 relation . there is still a small offset between galaxies with and without hi detections . naturally , we do not have any information on the rotational velocity of the non - detections ( some of which are disks close to edge - on view ) thus , for a fraction of these , @xmath91 is certainly underestimated . overall , the baryonic @xmath91 relation is as good as our generalized bfj , but only for galaxies with inclinations larger than 40 . in this work we demonstrated that a simple correction applied to the stellar velocity dispersions removes the offset between disk - dominated galaxies and spheroids seen in the bfj relation , i.e. , to first order , it removes its dependence on the internal structure of the galaxy . we further investigate this by considering the size - velocity relation . the existence of such a correlation is well established for both disk galaxies ( e.g. , @xcite ; ss11 and references therein ) and ellipticals or bulges of early - type spirals ( the @xmath4-@xmath5 relation , where @xmath4 is the diameter of the galaxy at a given surface brightness level ; e.g. , @xcite ) . here , we compare this relation for the different velocity indicators and data subsets discussed in this work . specifically , we wish to test how the correction to the stellar velocity dispersion introduced in [ s_mbfj ] affects the size - velocity relation . the sizes of disk galaxies are commonly estimated from exponential scale lengths , @xmath96 . however , disk scale lengths are not only problematic to measure ( e.g. @xcite ; ss11 ) , but are also not meaningful for elliptical galaxies . @xcite and ss11 advocate the use of isophotal radii instead of disk scale lengths , as they yield scaling relations with significantly lower scatter . isophotal radii are also used as size indicators for early - type galaxies in the @xmath4-@xmath5 relation . thus , we adopt @xmath17 , which is half the 25 mag arcsec@xmath18 isophote diameter @xmath19 measured by us on the sdss _ g_-band images , as a measure of size for all the galaxies in our sample . for comparison , @xmath17 varies between 2 and 7 @xmath96 , where @xmath96 is the sdss exponential scale length in _ r_-band , for the disk - dominated galaxies in our sample . the relation between size and rotational velocity ( rv ) is shown in figure [ vsize]a for all the galaxies with hi detections . for this and the other relations presented in this section we performed bisector linear regressions instead of inverse fits , because it is less clear that the scatter is mostly confined to one variable . the scatter of @xmath17 about the best fit is computed as before , i.e. , by applying tukey s bi - weight , and is noted in the bottom right corner of the plot ( in dex of kpc ) . as in figures [ btf ] and [ vsigma ] , cyan and red circles indicate galaxies with inclinations smaller than 40 and with @xmath97 , respectively . some of the main outliers of this relation are also outliers for the btf , but there is not a 1-to-1 correspondence . for instance , the very hi - rich spiral gass 35981 ( marked as a blue star ) is an outlier on this plot , but not on the btf , whereas the gas - rich elliptical gass 3505 ( red star ) is an outlier for both relations . the rv relation is significantly weaker and more scattered than the btf , as known from other studies that use disk scale lengths as size indicators ( e.g. , @xcite report a pearson correlation coefficient of @xmath98 and a scatter of 0.33 in log @xmath99 ) . however , as mentioned above , ss11 obtain a very tight correlation using isophotal radii instead of disk scale lengths ( @xmath100 and a scatter of 0.11 in log r ; in the notation adopted by these authors , the scatter of courteau s relation becomes 0.165 dex ) . when restricted to the tf subset ( fig . [ vsize]b ) , the scatter of our relation ( 0.15 dex ) is intermediate between those of @xcite and ss11 samples . although we use a similar size indicator as ss11 , their sample includes only late - type spirals , thus the larger scatter of our relation is most likely due to the broader morphological mix of the galaxies in our data set . as for the btf and bfj relations , we plot the residuals of the rv relation as a function of several quantities in figure [ vsize_res ] ( left panels ) . there is a dependence of the rv residuals on inclination and , to a lesser extent , on @xmath64 . figure [ sigma_size ] shows the relation between @xmath17 and stellar velocity dispersion @xmath5 ( r@xmath5 ) for the full sample ( a ) and for the fj subset ( b ) . as for the bfj relation , galaxies with hi detections are displaced from the non - detections ( green triangles ) , and a clear trend is observed when the r@xmath5 residuals are plotted as a function of concentration index ( fig . [ vsize_res ] , top right panel ) . indeed , figure [ vsize_res ] shows that the r@xmath5 residuals behave similarly to the bfj ones ( see fig . [ residuals ] ) : disk - dominated galaxies systematically deviate from the best fit relation obtained for the fj subset , which is dominated by spheroids . aside from systematic trends , which are stronger for the r@xmath5 relation , the scatter of the rv and r@xmath5 relations is very similar , for both maximum samples and pruned subsets . a for the corrected velocity dispersion.,width=302 ] lastly , correcting the velocity dispersions according to equation [ eq_modsigma ] yields a slightly tighter size - velocity relation ( fig . [ sigmacorr_size ] ) . most importantly , our correction removes the offset between hi detections and non - detections present in figure [ sigma_size]a ( the remaining few hi detections lying above the relation are highly inclined galaxies , for which @xmath17 , which is not corrected for inclination , is likely to be overestimated ) . the main result of this work is the existence of a tight relation between baryonic mass and velocity that is independent of galaxy morphology . to first order , we can remove the dependence of dynamical scaling relations on the internal structure of the galaxy by applying a simple correction to the measured stellar velocity dispersions , which depends only on the concentration index @xmath57(eq . [ eq_modsigma ] ) . the correlation between baryonic mass and corrected @xmath5 thus obtained has a scatter of only 0.08 dex ( fig . [ bfj_corr ] and table [ t_fits ] ) . it is encouraging that our correction removes the offset between disks and spheroids also in the size - velocity relation ( expressed in terms of @xmath17 versus stellar dispersion ) . we tested if an even tighter baryonic mass - velocity relation could be obtained by using the @xmath91 parameter , which combines rotational velocity and stellar dispersion . we found no improvement , because the @xmath91 relation still suffers from the problem of low inclination galaxies that affects the btf , to a lesser extent . despite being measured at a smaller spatial scale than the hi width , the stellar velocity dispersion turns out to be a better tracer of mass , at least for the massive galaxies in our sample . this is surprising , as one would expect the rotational velocity to provide a more reliable measurement of mass for galaxies with hi detections , the vast majority of which are rotation dominated ( i.e. , 216/228 galaxies have @xmath101 ) , regardless of the presence of a bulge . and yet , the same result was obtained by @xcite for a sample of s0 galaxies with inclinations @xmath035@xmath7160 , which are also , overall , rotation - dominated : the central stellar velocity dispersion is a better predictor of _ luminosity than the circular speed at 2 - 3 exponential disk scale lengths , which they carefully measured from long - slit optical absorption - line spectra . as we already pointed out , the main limitation with rotational velocities is observational , not intrinsic . we are simply not able to reliably measure circular speeds from line - of - sight velocities , except for well - selected samples of undisturbed late - type , inclined spirals ( i.e. , the typical tf samples ) , for which the deprojection to edge - on view does not introduce very large uncertainties . as for the velocity dispersions , which are measured through 3-diameter fibers , there might be a concern about contamination from disk stars in circular orbits , especially for the most edge - on , disk - dominated galaxies . however , we have argued that this effect is negligible ( [ s_bfj ] ) . this agrees with the conclusion reached by @xcite , who reported that the contamination is small ( @xmath102 , based on simulations ) for milky way - type galaxies with pressure - supported bulges . we argued that the reason why disk - dominated galaxies are simply offset from the spheroids on the bfj plane ( which is the key point that allows us to bring all massive galaxies onto the same baryonic mass - velocity relation ) is that @xmath5 is proportional to @xmath7 , and their ratio is a function of galaxy morphology . although initial work indicated that spirals and ellipticals follow the same tight @xmath7-@xmath5 correlation for @xmath103 km s@xmath13(e.g . * ; * ? ? ? * ; * ? ? ? * ) , more recent analyses demonstrate that the relation between @xmath7 and @xmath5 is not universal , but depends on morphology . as mentioned in [ s_mbfj ] , @xcite show that the brightest , bulge - dominated galaxies lie on the @xmath86 relation expected for isothermal stellar systems , whereas later - type spirals and dwarfs are offset by an amount that depends on morphology or total light concentration . we have shown that correcting the stellar dispersions according to equation [ eq_modsigma ] effectively translates into accounting for this departure from isothermality for disk - dominated galaxies . @xcite noted that , despite the fact that a detailed understanding of what sets the relation between @xmath7 , @xmath5 and concentration index is still missing , one could use this relation to empirically reduce the scatter of scaling relations that involve dynamical parameters , such as the tf or fj . we showed that the existence of such relation allows us to do more than that we obtained a generalized bfj relation that holds for all the massive galaxies in our sample , with a scatter ( 0.079 dex ) that is as small as that of the btf and bfj relations applied to their pruned subsets ( i.e. 0.076 and 0.074 dex , respectively ) . for comparison , @xcite and @xcite report a scatter of 0.06 dex for the btf and 0.071 dex for the stellar fj relations , respectively . the implications of our generalized bfj relation for extragalactic studies are very promising . because it holds for all massive galaxies in our sample _ regardless of morphology _ , this relation appears to provide a more fundamental link between dark matter halo mass and baryonic content than that obtained using rotational velocities or velocity dispersions . as such , it gives more fundamental constraints to galaxy formation models than the tf or fj / fp relations . also this is the reference baryonic mass - internal velocity relation that higher redshift studies , which naturally target the most massive galaxies , should compare with . this relation is more resilient to systematic effects than the tf , and , contrary to both tf and fj / fp relations , does not require any morphological pruning a significant advantage since accurate morphological classifications are difficult to obtain for large samples beyond the very local universe . as pointed out throughout this work , our results are based on a representative sample of galaxies with stellar masses larger than @xmath14 m@xmath2 . it remains to be established how far down in stellar mass these results can be extrapolated . it would be beneficial to investigate whether a bfj relation for disk galaxies holds down to low baryonic masses ( where the gas contribution is more important ) , and with similarly low scatter , based on a representative and homogeneous sample such as gass . b.c . wishes to thank dennis zaritsky , susan kassin , thorsten naab and luca cortese for useful discussions . thanks also to simone weinmann for kindly providing the data on concentration index versus b / t ratio used in section 3.2 . the arecibo observatory is part of the national astronomy and ionosphere center , which is operated by cornell university under a cooperative agreement with the national science foundation . galex ( galaxy evolution explorer ) is a nasa small explorer , launched in april 2003 . we gratefully acknowledge nasa s support for construction , operation , and science analysis for the galex mission , developed in cooperation with the centre national detudes spatiales ( cnes ) of france and the korean ministry of science and technology . 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 , university of cambridge , 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 . we discuss here in more detail the 12 galaxies with hi rotational velocities smaller than their stellar velocity dispersions , which are outliers of both btf and @xmath7 versus @xmath5 relations . these objects are indicated by red symbols in figures [ btf ] and [ vsigma ] . while the high @xmath63 outliers are all galaxies with small inclinations , for which the rotational velocities are likely to be overestimated , the low @xmath63 outliers are potentially more interesting . as mentioned in [ s_btf ] , @xcite identifies a population of galaxies with unusually small @xmath63 ratios , which are outliers in his tf relation , and argues that these systems must have experienced gas accretion . figure [ vsigma_hi ] shows sdss images and hi - line profiles for 9 outliers detected by gass ( the other galaxies were either observed by alfalfa or included in the s05 archive ) . two of these galaxies , gass 30175 and 31156 , are outliers because their sdss inclinations are incorrect . they both have strong bars which dominate the sdss fits , thus overestimating the inclination ( 54 and 61 for gass 30175 and 31156 ; we measured 37 and 27 respectively ) . the same is true for gass 24236 ( agc 220363 , detected by alfalfa and not shown here ) , a nearly face - on barred galaxy with sdss inclination of 46 ( we measured 17 ) . of the remaining outliers , one is clearly morphologically disturbed ( gass 57099 , bottom row ) . another galaxy , gass 4030 , was observed as part of the coldgass survey ( * ? ? ? * see their figure a1 , row 6 ) . its ( uncorrected ) velocity width measured from the co(1 - 0 ) spectrum is 137 @xmath104 km s@xmath13 , which is exactly the same as the ( uncorrected ) hi width . since the co(1 - 0 ) emission typically originates from the central regions of a galaxy , perhaps the fact that co and hi widths are identical means that the hi gas in this galaxy is not extended enough to sample the flat part of the rotation curve this is not particularly unusual , as there are other examples of early - type spirals with truncated hi disks ( e.g. ngc 3623 , where the hi emission does not extend beyond the stellar disk ; @xcite ) or unexpectedly narrow hi profiles ( e.g. ngc 5854 ; @xcite ) . overall , even acknowledging that some of these profiles are not very high signal - to - noise detections , it is striking that most are asymmetric , suggesting disturbances in the distribution and/or kinematics of the hi gas . we will investigate asymmetries of hi profiles in more detail in a future work .

with the hubble space telescope we have obtained images of 9 of the most distant radio galaxies . the galaxies , which have redshifts between and , were observed with the wfpc2 camera in a broad band filter ( f606w or f707w , roughly equivalent to v or r - band ) , corresponding to the near ultraviolet emission in the rest frame of the radio galaxies . the total observing time was 2 orbits per object . in this paper we present the images overlayed on vla radio maps of comparable resolution . we also present previously unpublished images , taken from the hst archive , of two other high redshift radio galaxies , observed through similar broad band filters . we find that on the scale of the hst observations there is a wide variety of morphological structures of the hosting galaxies : most objects have a clumpy , irregular appearance , consisting of a bright nucleus and a number of smaller components , suggestive of merging systems . some observed structures could be due ( at least partly ) to the presence of dust distributed through the galaxies . the uv continuum emission is generally elongated and aligned with the axis of the radio sources , however the characteristics of the `` alignment effect '' differ from case to case , suggesting that the phenomenon can not be explained by a single physical mechanism . we compare the properties of our radio galaxies with those of the uv dropout galaxies and conclude that ( i ) the most massive radio galaxies may well evolve from an aggregate of uv dropout galaxies and ( ii ) high redshift radio galaxies probably evolve into present day brightest cluster galaxies .

astro-ph9809056

studying the optical morphology of high redshift ( @xmath2 ) radio galaxies ( hzrgs ) can contribute substantially to our understanding of galaxy formation and evolution in the early universe ( for a recent review see mccarthy 1993 ) . although the recent development of new techniques ( e.g. u and b band dropouts , steidel et al . 1996 ) has led to the discovery of a large population of high redshift galaxies , radio galaxies remain still of exceptional interest , because they pinpoint the most massive systems at high redshift and are potential signposts for finding high - redshift clusters of galaxies . + it has been shown that high luminosity radio sources associated with quasars and radio galaxies at redshift @xmath3 are located in rich clusters ( e.g. hill & lilly 1991 ) . at @xmath4 there are now several possible x - ray clusters that have been discovered around powerful radio galaxies , such as 3c324 at [email protected] ( dickinson et al . 1998 ) , 3c356 and 3c280 ( crawford & fabian 1996 ) . at @xmath6 the existence of clusters around hzrgs has not been established . however , there is an increasing number of important observational indications that hzrgs might be in clusters , including ( i ) the detection of possibly extended x - ray emission from the radio galaxy pks 1138 - 262 at z=2.156 , most probably coming from a hot cluster atmosphere ( carilli et al . 1998 ) ; ( ii ) strong faraday polarization and rotation of the radio emission of some hzrgs which might be due to dense gaseous halos ( carilli et al . 1997 ) ; ( iii ) possible excess of companion galaxies detected along the axes of the radio sources ( rttgering et al . 1996 ) ; ( iv ) possible excess of lyman break selected galaxies in the fields of several powerful radio sources ( e.g. lacy & rawlings , 1996 ) and ( v ) excess of candidate companion galaxies ( with two objects spectroscopically confirmed ) in the vicinity of mrc 0316 - 257 , at z=3.14 ( le fevre et al . + the hosts of powerful low redshift radio sources have long been identified with giant elliptical galaxies , containing old stellar population . the surprising continuity of the k z relation between the high redshift radio galaxies and the low redshift brightest cluster galaxies which shows little scatter up to redshift of @xmath7 ( although the scatter increases beyond redshift 2 , e.g. eales et al . 1997 ) , might indicate that the hosts of powerful radio sources are the most massive galaxies know at high redshifts . moreover , since hzrgs are probably located in forming clusters of galaxies , they could be the ancestors of brightest cluster galaxies . + we have previously presented and discussed hst images of the radio galaxy pks 1138 - 262 at z=2.156 , which shows the clumpiest optical morphology of all the hzrgs imaged with the hst ( pentericci et al . our conclusion was that pks 1138 - 262 is giant elliptical galaxy at the center of a protocluster in the late stages of its formation . + in this paper we present hst wfpc2 images for 9 powerful radio galaxies having redshifts between [email protected] and [email protected] . we also present deep hst archive images of 2 hzrgs observed with wfpc2 . we compare the hst images with vla maps of the associated radio sources having similar resolution . after discussing the sample selection ( sect . 2 ) , we describe the hst imaging and reduction procedures ( sect . 3 ) , the radio imaging and the problem of the relative astrometry between the radio and hst data ( sect . 4 ) . in sect . 5 we briefly discuss the most important characteristics of each object , also referring to previous results that are relevant to the interpretation of the new data . finally in sect . 6 we discuss some statistical trends of the properties of these high redshift radio sources , giving a qualitative interpretation . we then summarize our main results and present our conclusions . we also include the appendix new radio images of the radio galaxies tx 1707 + 105 and mrc 2104 - 242 . + throughout this paper we assume a hubble constant of h@xmath8 km s@xmath9 mpc@xmath9 and a deceleration parameter of q@xmath10 . the radio galaxies were initially selected from the more than 60 hzrgs which were known at the commencement of the project ( 1995 ) ( e.g. van ojik 1995 and references therein ) . most of these distant radio galaxies were found by observing ultra steep spectrum radio sources ( uss ) ( @xmath11 , where @xmath12 is the radio spectral index ) ( van ojik 1995 ) . + objects were selected according to the following criteria : ( i ) bright in the the r band ( r@xmath13 24 , i.e. sufficient to be mappable in a reasonable time with the hst ) ; ( ii ) amongst the brightest line emitters ( ly@xmath12 flux @xmath14 10@xmath15 erg s@xmath9@xmath16 ) . because of its high redshift , we also included the radio galaxy mg 2141 + 192 ( z=3.594 ) in the sample . + finally we obtained unpublished hst / wfpc2 images of the radio sources b2 0902 + 343 , at z=3.395 , and tx 0828 + 192 , at z=2.572 , from the hst archive . + for a statistical study of the properties of hzrgs it is important to enlarge the sample of objects with hst images : we therefore included in our analysis the other radio galaxies that have been imaged with the hst . these include the radio galaxy 4c 41.17 at z=3.8 , one of the the best studied hzrgs ( van breugel et al . 1998 ) ; the radio galaxy pks 1138 - 262 , at z=2.156 that was studied by our group ( pentericci et al . 1997 , 1998 ) ; mrc 0406 - 242 at z=2.44 that was object of a multi - frequency study , including wfpc imaging in different color bands , by rush et al . ( 1997 ) ; and 4c 23.58 at z=2.95 ( chambers et al . 1996a and 1996b ) . the first two objects were imaged with the wfpc2 camera , while the last two were imaged with the pre - refurbishment hst / wfpc . details of the observations can be found in the mentioned papers . in this way the final sample available for the statistical analysis of the properties of hzrgs consists of 15 galaxies . by including also radio galaxies that have been imaged with the pre - refurbishment hst and/or for which the total integration times are considerably different ( e.g. the observations of 4c 41.17 are much deeper than for the other objects ) , the quality of the images varies within the sample . however given the relatively small number of radio galaxies observed , it is important to increase the statistics . + in addition to the hst images , all the radio galaxies in the final sample have been imaged with the vla at several frequencies , to study their radio - polarimetric properties ( @xcite ) and have ly@xmath12 profiles taken with resolution of @xmath17 100 km s@xmath9 , thus allowing a detailed study of the morphology and kinematics of the ionized gas ( @xcite ) . for some of the radio galaxies , ground - based narrow band images of the ly@xmath12 emission gas , and broad band images in various color bands ( mostly r - band and k - band ) are also available ( see references for individual objects in sect . table 1 summarizes the observations . 9 radio galaxies were imaged with the planetary camera ( pc ) of wfpc2 during cycle 5 and/or cycle 6 . the pc utilizes an 800 @xmath18 800 pixel loral ccd as detector with pixel size of @xmath19 ( @xcite ) . the typical exposure time was 5300 sec ( 2 orbits ) for each galaxy . the total observing time was split between two exposures to facilitate removal of cosmic ray events . + the filters used for the observations were chosen to avoid contamination from the strong ly@xmath12 emission line at 1216and to have the rest frame wavelengths sampled as similar as possible throughout the sample . for the radio galaxies at redshift @xmath20 the filter used was the broad - band f707w filter ( centered at @xmath21 and with a fwhm of @xmath22 ) , similar to the cousins r band ; for the lower redshift galaxies we used the broad - band f606w filter ( @xmath23 and @xmath24 ) which is similar to the v band . + the radio galaxy tx 0828 + 193 was observed during cycle 4 with the wfpc2 by chambers et al . , using the filter f675w which is centered at @xmath25 and has a fwhm of @xmath26 . the observations were done in polarimetric mode . the galaxy was observed using the wf3 section of wfpc2 , which utilizes an 800 @xmath18 800 pixel loral ccd as detector with pixel size of @xmath27 ( @xcite ) . the total exposure time of 10000 s was split between ten observations . + the radio galaxy b2 0902 + 343 was observed during cycle 4 by eisenhardt and dickinson , using the pc of wfpc2 with the filter f622w , which is centered at @xmath28 and has a fwhm of @xmath29 . the total exposure time of 21600 s was split between nine observations . + in the redshift range observed the continuum emission may include contribution from the faint emission lines of heii , ciii ] and civ . for most of the radio galaxies we could estimate the total contamination using the line fluxes measured by low resolution spectra of the objects taken by van ojik ( 1995 ) . the detected lines are listed in table 1 , as well as the total contribution of the line emission to the measured flux , which ranges from 0 to 13.7% , with the highest contribution for the radio galaxy tx 0211 - 122 . we expect that for the 4 sources of which we do not have any such data available , the line contribution will be in the same range . therefore we can assume that the images represent to a good approximation the continuum emission from the galaxies . [ cols="<,^,^,^,^,^,^,^,^,^,^ " , ] \(1 ) catalog . ( 2 ) source name . ( 3 ) redshift . ( 4 ) date of hst observations . ( 5 ) filter used for hst observations . ( 6 ) rest - frame wavelength interval in angstroms . ( 7 ) total exposure time in second . ( 8) number of frames in which the total observing time was split . ( 9 ) wfpc2 magnitude within an aperture of radius 4@xmath30 . ( 10 ) emission line detected within the filter band . ( 11 ) percentual flux contamination from the detected lines . + @xmath31 see text for a detailed explanation + @xmath32 magnitude of galaxy 1707 + 105a + @xmath33 contribution to the total flux of 1707 + 105a and 1707 + 105b . + the data were reduced according to the standard space telescope science institute pipeline ( @xcite ) . further processing was performed using the noao image reduction and analysis facility ( iraf ) software package and the space telescope science data analysis system ( stsdas ) and involved cosmic ray removal and registering of the images . the shifts were measured from the peak positions of a non - saturated star present in both the pc images . the different frames were then added , background subtraction was performed using the average flux contained within 4 or more apertures placed on blank areas of the sky , as close as possible to the source , at different positions , to avoid introducing errors from residual gradients in the background flux . the resulting image was flux calibrated according to the precepts described in the `` hst data handbook '' ( 1995 edition ) , using the photometric parameters from the standard hst calibration and included in the file header . the images were then rotated to superimpose them to the vla radio maps ( see sect . 4.1 ) + lllllllllll source & ra & dec & f@xmath34 & @xmath12&cf@xmath34&size & rm & pa & ref . + & j2000&j2000 & mjy & & % & kpc & rad m@xmath35 & + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) + tx 0211@xmath36122&02:14:17.37&@xmath3611:58:46.7&54 & 1.5&3.8&134 & 160&73@xmath31&a + tx 1707@xmath37105&17:10:06.85&+10:31:09.0&64 & 1.2& & 173& & -58 & c + 4c 1410@xmath36001&14:13:15.13&@xmath3600:22:59.6&57 & 1.3&6.7&189&1510&44@xmath31 & a + mrc 2104@xmath36242&21:06:58.16&@xmath3624:05:11.3&68 & 1.3&1.6&177& & 12 & b + tx 0828@xmath37193&08:30:53.71&+19:13:18.5&22 & 1.6&21 & 98& & -42 & a + mrc 2025@xmath36218&20:27:59.45&@xmath3621:40:57.1&95 & 1.1&0.7 & 38&910 & -16 & a + 4c 1345@xmath37245&13:48:14.78&+24:15:50.0&115&1.4&0.7 & 17&750&-36 & a + mrc 0943@xmath36242&09:45:32.79&@xmath3624:28:49.8&55 & 1.8& & 29&&75 & a + b2 0902@xmath37343 & 09:05:30.10&+34:07:56.9&100&1.4&15 & 32&2500&37 & e + 4c 1243@xmath37036&12:45:38.43&+03:23:20.3&70 & 1.4&2.0 & 50&420 & 20 & d + mg 2141@xmath37192&21:44:07.50&+19:29:15.0&67 & 1.6& & 60& & 3 & a + + + ( 1 ) name of the source ( 2 ) and ( 3 ) coordinates of the radio core in the epoch j2000 . ( 4 ) total radio flux at 4.5 ghz . ( 5 ) radio spectral index between 4.5 ghz and 8.2 ghz , @xmath38 . ( 6 ) radio core flux /total flux at 4.5 ghz rest - frame . ( 7 ) maximum radio source size . ( 8) maximum rotation measure . ( 9 ) position angle of the inner radio axis relative to the direction north - south . ( 10 ) references : a. carilli et al . 1997 b. mccarthy et al . 1990 c. this paper d. van ojik et al . 1996 e. carilli 1995 + @xmath31 radio jets might be precessing . + the magnitudes were computed from the unrotated images ( which have less smoothing ) within a fixed aperture of diameter @xmath39 . in most cases this aperture is large enough to enclose all the light from the galaxies . the magnitudes were computed as : @xmath40 , where f is the measured flux and @xmath41 is the zero point for the hst magnitude scale normalized to vega . the results are presented in table 1 . + a number of different effects contributes to the errors in the photometric magnitudes ; ( i ) the poisson noise of the detected counts ; ( ii ) a @xmath422% uncertainty in the determination of the zero point ( burrows 1995 ) ; ( iii ) a @xmath424% systematic error due to the problem of charge transfer efficiency in the loral ccd ( holtzman et al . 1995 ) for which we did not correct ; ( iv ) accurate subtraction of the mean sky background ; ( v ) sky noise within the source aperture . the last two are usually the predominant effects . we estimate that the total uncertainty in the magnitudes is 0.1 or less for all galaxies . + a first order transformation from the f606w and f702w st magnitudes to the standard magnitude system was derived applying the precepts described by holtzman et al . the resulting transformation are @xmath43 , @xmath44 , @xmath45,and @xmath46 . + in table 1 we list for each galaxy the wfpc magnitude @xmath47 ; the emission lines that have been detected within the filter band and the total line contribution to the continuum flux . all the radio galaxies with the exception of b2 0902 + 343 , tx 1707 + 105 and mrc 2104 - 242 were imaged with the vla as part of a high resolution , multi - frequency radio polarimetric study carried out on a large sample of hzrgs by carilli et al . ( 1997 ) . a full description of the observations and the reduction procedure can be found in this paper . + the radio map of b2 0902 + 343 that we use in this paper is a high resolution ( 0.15@xmath30 ) radio continuum image of total intensity at 1.65ghz obtained by carilli by combining data from the vla and merlin ( see carilli 1995 for details ) . + the radio observations of tx 1707 + 105 and mrc 2104@xmath36214 were performed with the vla in b array . details of observations and reduction for both sources can be found in appendix a and b. the coordinate frame for the wfpc2 images determined from the image header information has uncertainties of the order of @xmath48 ( @xcite ) . since the optical galaxies are generally clumpy on a scale smaller than @xmath48 , it is important to get the better possible registration between the radio and the optical images , to allow a detailed inter - comparison between the emissions . + in overlaying the hst images with the radio vla images we made the following assumptions : for those sources showing a clear detection of the radio nucleus and for which good k - band ( or k@xmath49-band ) images existed ( mccarthy , private communication ) , we assumed that the peak position of the infra - red image would be a better indicator of the true location of the center of the host galaxy , rather than the peak of the hst image , since the uv continuum might be effected by dust extinction ( e.g de koff et al . we therefore identified the position of the radio core in the vla image with the peak position of the k - band image . finally we registered the hst frame and the infrared frame using the weighted positions of several stars which were present on both fields ; this can be achieved with an accuracy of 0.1@xmath30 which is then the total final uncertainty in the relative astrometry . this procedure was possible for the radio galaxies tx 0211 - 122 , 4c 1243 + 036 , mrc 2025@xmath36218 and mrc 2104@xmath36242 . + for those objects which had a clearly detected radio core but no k - band images , we associated the peak position of the hst image to the peak position of the radio emission . we followed this procedure for the radio galaxies 4c 1345 + 245 , 4c 1410 - 001 and tx 0828 + 128 ( for this last object see remarks made in the individual source description ) : these objects have a relatively simple morphology , hence it is reasonable to assume that the peak of the uv continuum represents the true nucleus of the galaxy ; the final uncertainty of the relative astrometry is then within a pixel i.e. @xmath50 . + for those objects which have no detected radio core ( mrc 0943@xmath36242 , tx 1707 + 105 and mg 2141 + 192 ) we used the hst absolute astrometry , and we then checked the peak position of several stars which were present on the wfpc2 frames , with the position given in the apm catalog ; with this method we achieved an accuracy of @xmath51 . finally for b2 0902 + 343 which has a radio core but no clear optical nucleus , we kept the natural hst astrometry : in this way the radio core falls in between the two optical peaks . this is consistent with what found e.g. by carilli ( 1995 ) . grey scale hst wfpc2 images ( smoothed with a gaussian function of fwhm equal to 2 pixels ) with vla radio contours superimposed are shown in figs . 1 - 10 . for every source we also show a contour map of the continuum emission to better delineate the morphology . we do not show such maps for b2 0902 + 343 and mg 2141 + 192 because they have very low surface brightness , and a contour map would add no information . for the very large radio galaxies ( namely tx 0211 - 122 , tx 1707 + 105 and 4c 1410 - 001 ) we also present a third image showing the complete field of the radio source . the objects are presented in order of increasing radio size , since it has been shown ( e.g van ojik 1995 ) that several properties of hzrgs tend to change with increasing radio size . + we shall now give brief descriptions of the ultraviolet morphology of each radio galaxy , with special emphasis on any peculiar characteristics ( such as distortions , jet - like features etc ) , and compare those with relevant previous results . + + * 4c 1345 + 245 * + this radio source at z=2.879 ( @xcite ) , is the smallest in the sample , being only @xmath52 in extent ( corresponding to 17 kpc in the adopted cosmology ) . the radio structure has been extensively studied with the vla at several frequencies by carilli et al . ( 1997 ) , who classified it as `` compact steep spectrum source '' ( css ) and by chambers et al . ( 1996b ) . the radio emission shows two lobes of roughly equivalent brightness , with a one sided feature extending from the core towards the eastern side , which has been identified as a jet . optical and infrared ground based observations show a compact object , with the emission extended along the radio axis , with one faint component or companion object along the radio axis to the southwest but beyond the radio lobe . ( @xcite ) . the new hst image shows that in uv continuum the emission has a bright compact nucleus . on the eastern side of this component there is a jet - like feature that follows remarkably well the small curvature of the radio jet : this suggests that we might be observing the optical counterpart of the radio - jet . however the radio - to - optical spectral index derived from the flux of the component ( 0.7 ) is completely different from the high - frequency radio spectral index ( -1.2 ) . such flattening of spectral indices into the optical is contrary to what is found for sources with observed optical synchrotron radiation ( e.g. meisenheimer et al . therefore we discard this possibility . a more likely interpretation is that star formation is taking place in that region triggered by the passage of the radio jet . other possible mechanisms to enhance the emission along the radio jet path have been proposed by bremer et al . ( 1997 ) . in sect . 6.1 we will discuss more extensively the alignment effect and how all the various models that have been proposed to explain it , apply to our sample of radio galaxies . + the object along the radio axis detected by chambers et al . ( see above ) is also detected in our hst image ( it is outside the field shown in fig . 1 ) ; its morphology indicates that it is most probably an edge - on spiral ( hence a foreground object ) . + + * mrc 0943 - 242 * + this radio source at z=2.923 ( @xcite ) is only 29 kpc in extent and has a simple double - morphology , with no nucleus detected in the present vla images ( @xcite ) . the hst image shows a bright elongated main component , plus a number of smaller clumps embedded in a halo of lower surface brightness emission with a peculiar overall curved morphology . the inner region of the uv emission shows a remarkably good alignment ( within @xmath53 ) with the radio axis . for comparison , the keck k - band image taken by van breugel et al . ( 1998 ) shows a somewhat rounder and more centrally concentrated morphology . + high resolution spectroscopy of the ly@xmath12 line shows spatially resolved absorption by associated neutral hydrogen , with the absorber covering the entire extended ly@xmath12 emission ( @xcite ) . similar deep high resolution observations have been performed on the civ and heii lines : the heii line does not show absorption , which is expected as this is a resonant line , while the civ line shows absorption due to the civ 1548/1551 doublet . the column density for the absorber is @xmath54 @xmath16 . combined with the measured column density for the hi absorber , this indicates that the spatially extended absorber is metal enriched ( @xcite ) . + + * b2 0902 + 343 * + this radio galaxy was identified by lilly ( 1988 ) and is one of the most extensively studied high redshift radio galaxies . it is 32 kpc in extent . + the radio emission has a bizarre structure showing a bright knotty jet with a sharp bend of almost 90@xmath55 at its northern end , and two southern components whose common orientation is perpendicular to the rest of the source ( carilli et al . 1994 ) . further multi - frequency radio studies lead carilli ( 1995 ) to conclude that most of the peculiarities of the radio galaxy can be explained by assuming that the source is oriented at a substantial angle ( between 45 and 60 degrees ) with respect to the plane of the sky , with the northern regions of the source approaching and that the central region of the galaxy is obscured by a substantial amount of dust . + from extensive studies eisenhardt & dickinson ( 1992 ) found that b2 0902 + 343 has a flat optical spectral energy distribution ( sed ) , and an unusually low surface brightness distribution at optical and ir wavelengths ; this lead to the suggestion that b2 0902 + 343 might be a proto - galaxy , undergoing a first major burst of star formation ( eales et al . 1993 , eisenhardt & dickinson 1992 ) . the presence of associated 21 cm neutral hydrogen in absorption against the radio continuum source was first detected by by uson et al . ( 1991 ) and confirmed by others ( @xcite ) . however no strong absorption in the ly@xmath12 emission line has been detected ( @xcite ) . + the optical morphology , as imaged by the hst , confirms the unusually low surface brightness distribution and shows that the galaxy consists of 2 regions , of approximately the same flux with a void in between , plus an extended fuzzy emission region to the north east of them . the source does not exhibit the radio - optical alignment effect ; the uv emission is almost perpendicular to the radio axis . with the present astrometry the radio core is situated in a valley between the optical peaks ; this morphology could be explained with the presence of large amounts of dusts . however the uncertainties in the astrometry are such that the radio core could be coincident with any of the two optical components . + + * mrc 2025 - 218 * + the galaxy associated with this uss radio source at z=2.630 ( 38 kpc in extent ) , was first identified by mccarthy et al . deep multi - frequency radio imaging show a double radio source with a jet on the southern side of the core , which has an extremely sharp bend towards the west , making an angle of @xmath56 ( @xcite ) . the northern lobe has a faint extension in the direction of the core which could be a counter jet . ground based near infrared imaging show a compact object ( van breugel et al . 1998 ) , while the ly@xmath12 emission extends for more than 5@xmath30 along the radio axis and is distributed bimodally . the total sed of the galaxy is well fit by a main stellar population aged 1.5 gyrs , combined with a young star - burst contributing 20% of the total light at 5000 (@xcite ) . cimatti et al . ( 1993 ) find that that the rest frame uv continuum emission is linearly polarized ( @xmath57 ) , with the electric vector oriented perpendicular to the uv emission axis . + the hst image shows that the host galaxy has a compact morphology , consisting of a bright nucleus , two smaller components and extended low surface brightness emission , which is elongated and well aligned with the radio axis . the angle between the inner radio axis and the extended uv emission is only @xmath58 . there is no direct one - to - one relation between the radio components and the uv emission , unlike 4c 1345 + 245 ; however if we draw a cone of opening angle @xmath59 along the radio axis , all the uv emission on both sides of the radio core is then constrained within this cone . such a morphology , reminiscent of an ionization cone , is expected in models where the aligned uv continuum emission is scattered light of a buried quasar , and is supported by the polarization measurements by cimatti et al . + the present hst image reveals little uv emission near the bend : however high resolution spectroscopic observations of the ly@xmath12 emission line show that the galaxy is embedded in a very large halo of ionized gas , extended well beyond the radio source ( more than 60 kpc i.e. double the size of the radio source ) ; therefore the most likely explanation for the bend is that interaction between the radio plasma and the surrounding gas deflects the jet , as observed in other cases ( e.g. pentericci et al . 1997 ) . + + * 4c 1243 + 036 * + this radio galaxy at z=3.570 which has an extension of 50 kpc , was identified and extensively studied by van ojik et al . the radio source is double with a sharp bent structure on the southern side . strong depolarization of the radio emission indicates that the source is embedded in a magneto - ionic medium . high resolution spectroscopy and narrow band imaging of the ly@xmath12 emission line have detected the presence of a giant ( 100 kpc ) halo of ionized gas showing ordered motion , possibly due to rotation of a proto - galactic gas disk , out of which the galaxy associated with 4c 1243 + 036 is forming . furthermore the ly@xmath12 emission shows a secondary peak at the location of the bending of the radio jet , consistent with a gas cloud being responsible for the deflection of the radio jet ( ojik et al . + the morphology of the galaxy as imaged by the hst consists of a nucleus from where a narrow and elongated structure departs , which then bends to the south . there is also a smaller component , about 1@xmath30 beyond the northern radio hot - spot , which could belong to the system , since narrow band ly@xmath12 imaging shows that there is ly@xmath12 emission at this location ( van ojik et al . + the most remarkable characteristic of 4c 1243 + 036 is that the uv light follows closely the direction of the radio source , both in the inner 2@xmath30 region where the light is aligned with the radio axis to within 15 degrees , but especially at the location of the bend : here both the uv emission and the radio jet bend rather sharply to the south , suggesting a direct relation between the radio jet and the uv component . this is similar to the case of the radio galaxy 4c 1345 + 245 . note that recent k - band keck imaging of 4c 1243 + 036 by van breugel et al . ( 1998 ) , although at a different resolution , indicate that also the k - band continuum emission is elongated and follows the bend of the radio jet . + + * mg 2141 + 192 * + this galaxy at z=3.594 ( 60 kpc in extent ) was identified by spinrad et al . ( 1992 ) and since then has been extensively studied by various groups . the radio source has a simple double morphology , with no nucleus detected in the present images . eales & rawlings ( 1996 ) who imaged this object in the infrared , report the detection of a relatively brighter component half way between the radio hot - spots and a second fainter one , @xmath39 north , approximately coincident with the northern radio hot - spot . recently van breugel et al . ( 1998 ) re - imaged the object in the near - infrared with the keck telescope , finding additional extended low surface brightness emission southern of the nucleus . armus et al . ( 1998 ) imaged the [ oiii ] emission line nebula associated with the galaxy , which has an extent of more than 70 kpc ( equal to the separation between the radio lobes ) , is extremely narrow and aligned with the radio axis . by comparing fluxes of the different emission lines they also find indications for the presence of large amounts of dust . finally maxfield et al . ( 1997 ) find that the emission nebulae of ly@xmath12 , civ and heii are not only spatially extended but also have remarkable velocity structure with multi - components velocity displacement up to 1900 km s@xmath9 , which are most consistent with a shock ionization picture . + the hst shows that the host galaxy is very faint in the uv restframe , and consists of a nucleus with a faint filamentary extension and a small clump to the west . in the hst image some fuzzy emission ( at a 3@xmath60 level ) is present near the position of the radio hot - spot , where the second infrared component is located . we also detect similar emission very close to the position of the southern radio component . overall the uv rest - frame emission is extremely faint , consistent with the presence of large amounts of dust . deeper images are needed to delineate the morphology of this galaxy in more detail . + + * tx 0828 + 193 * + this large radio source ( 98 kpc in extent ) at z=2.572 ( van ojik 1995 ) , has a double morphology with a jet extending from the core towards the northern hot - spot ( most probably the approaching side ) . the end of the jet contains multiple hot - spots and has a 90 degrees bent . the southern part of the radio source consists only of a single hot - spot . + the hst image shows a small galaxy consisting of several clumps arranged in a triangular shape . we choose to identify the radio core with the brightest optical component ( the same procedure we followed for other galaxies , see sect . 4 ) ; however another possible registration would be with the radio core at the vertex of the triangulum . at this position , an ionization cone on both sides of the radio core , would encompass all the uv emission . the morphology of tx 0828 + 193 , like that of mrc 2025@xmath36218 , strongly suggest that a large fraction of the uv light might be scattered light from a buried agn . the axis of this `` scattering cone '' is aligned with the radio axis to within a few degrees ( @xmath61 ) . + there is another object located along the radio axis which could be associated with the radio source ( a companion galaxy ) : it is bright in the uv continuum but shows no line emission , so it could as well be an intervening system at a different redshift ( @xcite ) . + the ly@xmath12 emission from this radio galaxy has a spectacular shape , with the entire blue wing of the emission line profile absorbed by neutral gas associated with the galaxy ( van ojik et al . 1997 ) . if the companion object is at the same redshift as tx 0828 + 193 , then it is possible that a neutral gaseous halo associated with it is the responsible for the absorption . since the absorption is very steep and broad , it is probably due to a combination of absorbing systems each at a slightly different velocity with respect to the ly@xmath12 peak . also in the red wing of the ly@xmath12 profile a broad shoulder is observed that maybe be due to multiple hi absorption systems or to intrinsic velocity structure in the ionized gas . + + * tx 0211 - 122 * + this large radio source ( 134 kpc ) at z=2.336 ( van ojik et al . , 1994 ) has a simple double morphology . a jet feature extends from the core towards south , curves and reaches the eastern lobe ; this structure suggests that the radio axis might be precessing . + the galaxy , as shown from the hst image , consists of a bright nucleus and a much smaller clump , both embedded by lower surface brightness emission , distributed in an irregular way . the contour image of the central component shows that it consists of two `` tails '' , one of which points in the direction of the inner radio jet . + the optical spectrum of this source is peculiar with the ly@xmath12 emission being anomalously weak when compared to higher ionization lines : the flux ratio of ly@xmath12 to nv a factor of 30 smaller than that of typical hzrgs while the large nv / civ ratio indicates that the line - emitting gas is over - abundant in nitrogen ( @xcite ) . van ojik et al . consider various mechanism that could produce these features , and conclude that the galaxy is likely to be undergoing a massive star - burst in the central region , possibly as the result of the passage of the radio jet . the star - burst would produce large amounts of dust , which when mixed through the emission line gas partly absorbs the ly@xmath12 emission , giving it a very patchy morphology , while the enhancement of nitrogen emission could be produced either by shocks or photo - ionization . + + * tx 1707 + 105 * + this radio source at z=2.349 ( van ojik 1995 ) which is 173 kpc in extent is one of the most peculiar systems in our sample : it consists of two galaxies ( labeled a and b in fig . 9 ) both showing strong and extended ly@xmath12 emission at the same redshift . the 2 objects lie almost exactly along the radio axis and they are both clumpy and elongated in a direction with is almost perpendicular to it . in particular galaxy 1707a ( the brightest one ) is comprised of a series of knots of approximately the same brightness , which form a sort of string , while galaxy 1707b consists of only two clumps . there is a further emission component , which in fig . 9 is indicated as c , that lies in between the two galaxies and could be part of the system . it does not show line emission , although probably when the high resolution spectrum was taken this object fell outside our @xmath52 wide slit . + with the present data it is not possible to determine exactly which galaxy is associated with the radio emission . given the large extension of the source , we expect , for symmetry reasons , that the radio source is associated to the galaxy closest to the center , i.e. galaxy 1707a . if this is the case then 1707b , and possibly 1707c , would be companion galaxies located along the radio axis . there are many cases of companion galaxies of high and low redshift radio galaxies . the best known case is minkowsky s object : the location of this dwarf galaxy is at the end of the radio jet emanating from the radio galaxy pks 0123@xmath36016 at @xmath62 , suggesting that its origin is due to jet - induced star formation ( @xcite ) . a similar star forming region associated with the nearby powerful radio galaxy 3c285 has also been reported ( @xcite ) . the most recent example is the radio source 3c34 ( at z=0.69 ) , which shows a clumpy emission feature along the radio axis and and oriented towards a radio hot - spot . also in this case , the emission has been associated with a region of massive star formation triggered by the passage of the radio jet ( @xcite ) . finally , rttgering et al . ( 1996 ) find that companion galaxies of radio sources tend to be distributed along the direction of the radio axis , which , in their interpretation , could be due to the luminosity of merging dwarf galaxies being enhanced by scattering and/or jet - induced star formation . + + * mrc 2104 - 242 * + this radio galaxy at z=2.491 is 177 kpc in extent and was first identified by mccarthy et al . it has a simple double morphology and a relatively bright nucleus . + the ly@xmath12 emission is spectacular : narrow band imaging show two large gas clumps , extending for more than 12@xmath30 along the radio axis . spectroscopy of the line showed that both components have very large velocity distribution ( @xmath631000 - 1500 km s@xmath9 ) , large equivalent width and have a net velocity difference of about 500 km s@xmath9 . each component contains multiple velocity peaks and kinematic data at various position angles indicate that there is no overall ordered motion ( @xcite ) . a detailed study of the ly@xmath12 emission line showed that a model based on shocks from direct interaction between the radio plasma and the gas can explain both the kinematics and the morphology of the gas ( @xcite ) . + the hst image is remarkable : the host galaxy is one of the clumpiest of our sample , consisting of a number of knots of similar brightness and size , located around the radio core . unfortunately some of the components are confused with the residuals from a spike of an extremely bright nearby star . furthermore there is a filamentary component that is more than 2@xmath30 long and extremely narrow . this last component is aligned with the radio axis to within a few degrees . the overall extension of the host galaxy is almost 7@xmath30 , making it the largest optical galaxy in our sample . + + -1 cm * 4c 1410 - 001 * + this radio galaxy at z=2.363 ( van ojik , 1995 ) is , with 189 kpc , the largest radio source in the sample . the host galaxy is highly elongated ( @xmath64 ) . -1 cm -1 cm it consists of a compact nucleus , a second bright component and extended lower surface brightness emission which is clumpy . the galaxy and the radio source are strongly misaligned : the angle between the optical and radio axis is nearly 45 degrees . however , the northern component of the radio source is curved , suggesting the radio axis might be precessing , in which case the elongated optical emission could be located along the previous path of the radio jet . the galaxy has extended ( @xmath65 kpc ) bright ly@xmath12 emission , exhibiting a velocity shear that could be due to rotation of the gas . the amplitude of this shear is almost equal to the overall velocity width of the line ( van ojik et al . the uv - optical continuum emission from hzrgs is generally aligned with the main axis of the radio emission ; several models have been proposed to explain the nature of the optical continuum emission and of this alignment effect ( for a review see mccarthy 1993 and references therein ) . the most viable ones are : ( i ) star - formation stimulated by the radio jets as it propagates outward from the nucleus ( @xcite ) ; ( ii ) scattering of light from an obscured nucleus by dust or free electrons ( di serego alighieri et al 1989 ; scarrott et al 1990 ; tadhunter et al 1992 ; cimatti et al 1993 ; di serego alighieri et al 1994 ) ; ( iii ) nebular continuum emission from warm line emitting clouds excited by the obscured nucleus ( @xcite ) . + so far the only hzrg for which there is direct spectroscopic evidence that the uv continuum clumps are star forming regions , not dominated by scattered light , is 4c41.17 : the spectrum of this galaxy shows absorption lines and p - cygni profiles similar to those found in the spectra of high redshift star forming galaxies ( dey et al . 1997 ) . + until recently , polarization measurements were possible only for z@xmath66 radio galaxies and showed that in most cases a large fraction of the uv continuum emission could be explained as scattered light . recently though , observations of @xmath67 radio galaxies have led to quite contradictory results : while some objects show considerable amounts of polarization ( e.g cimatti et al . 1997 and references therein ) , others such as 4c41.17 have upper limits consistent with the complete absence of polarization ( @xcite ) . + the hst data with their high resolution provide informations about the inner regions of hzrgs , and confirm that the radio / optical alignment is still present at scales of less than an arcsecond . in fig . 12 we plot the distribution of position angle difference ( @xmath68 ) for our sample . to determine the optical position angle ( pa ) , for the galaxies with a more regular morphology we smoothed the hst images with a gaussian function having a fwhm of 1@xmath30 and then we fit the inner 3@xmath30 region with ellipses , using the iraf package isophote , which also gives the orientation of the major axis of the ellipse . for the galaxies with irregular morphologies , the fits gave meaningless results , so we selected as optical axis the line passing through the 2 brightest peaks on the images . -1 cm -1.5 cm -1 cm the position angle of the radio emission is given by the line joining the radio core to the nearest hot - spots ( or the line joining the hotspots , if the core is not detected ) . + despite the fact that 13 out of 15 radio galaxies in our sample have @xmath69 , we notice that the properties of the alignment effect vary considerably from object to objects . + we can distinguish various groups : + ( i ) radio galaxies that show a remarkable one - to one relation between radio emission and uv continuum light : this includes 4c 1345 + 245 , that has an optical jet - like feature , and 4c 1243 + 036 where the uv light follows the bending of the radio jet . these structures can be easily explained by the jet - induced star formation models ( see references above ) . alternatively bremer et al . ( 1997 ) proposed a mechanism by which , when the radio jet passes through the gas clouds , it breaks them apart thus increasing the surface area of cool gas exposed to the ionizing beam . consequently the material along the jet path becomes a far more efficient scatterer of nuclear radiation and the uv emission is enhanced in a very narrow region . + ( ii ) radio galaxies in which the uv continuum emission has a triangular - shaped morphology , reminiscent of an ionization cone . this category includes tx 0828 + 193 , mrc 2025@xmath36218 and mrc 0406 - 242 . such morphologies are expected in models that consider the aligned optical continuum as being scattered light of a central buried quasar . + ( iii ) radio galaxies where the alignment between the optical morphology and the radio axis is good but there no one - to one relation between radio and uv components . this group includes mrc 0943@xmath36242 , mrc 2104@xmath36242 , 4c 1410@xmath36001 , tx 0211@xmath36122 , mg 2141 + 192 , pks 1138@xmath36262 , 4c28.58 and 4c41.17 . the degree to which the two components are aligned varies strongly even within this group : for example in the radio galaxy 4c 1410 - 001 the difference in position angles between the radio and the uv emission is 45@xmath55 , however the radio map indicates that the radio jets probably had a different direction in the past , corresponding to the direction along which the uv light is elongated . + ( iv ) finally , galaxies that show total misalignment between radio and optical emission . there are 2 such cases in our sample . first the radio galaxy b2 0902 + 343 ( fig . 3 ) , where dust could play an important role in obscuring the central regions ( eales et al . 1993 , eisenhardt and dickinson 1992 ) , thus `` masking '' the alignment effect . second , the extremely peculiar and complex system tx 1707 + 105 ( fig . 9 ) , which is comprised of 2 ( possibly 3 ) separate galaxies , with similarly strong ly@xmath12 emission : the galaxies are located along the radio axis , but they are clumpy and extended almost perpendicularly to the radio axis . this unusual morphology would be hard to explain just by invoking the presence of dust , since the dust should have an extremely complicated distribution in multiple lanes parallel to the radio axis . + in summary the new data confirm that there is no single model that can satisfactorily explain the optical morphology of all hzrgs and the nature of the aligned optical continuum emission . at the same time , none of the proposed models can be ruled out by the present data . therefore it seems likely that all three mechanism contribute to the aligned light , but their relative importance varies greatly from object to object . + a striking feature of the hst images of the radio galaxies is the widespread clumpiness of the optical continuum emission . most galaxies are comprised of several components , regardless of the fact whether they are aligned or not with the radio axis ; the clumps are resolved and their typical sizes are in the range 2 - 10 kpc . to give a consistent definition of `` clumpiness '' we proceeded in the following way : since the size of our sample is small , and for the faintest galaxies it is difficult to delineate the structures , we first normalized the total observed flux of each galaxy ( within a fixed aperture ) taking the faintest and most distant galaxy mg 2141 + 192 as reference . we then defined the parameter n as the number of components which have at least one contour at a flux level of @xmath70 erg @xmath16 sec@xmath9@xmath9 . this value was chosen so that the radio galaxy mg 2141 + 192 had 3 clumps . note that , despite the difference in restframe frequencies sampled by the observations ( see table 1 ) , this is a good approximation because the spectral energy distribution of hzrgs in the uv wavelength range ( 1300 - 2000 ) is generally flat . 13 we present a plot showing how n , our measure of clumpiness , varies with radio size , for all the radio galaxies in the sample . clearly there is a tendency for the larger radio sources to have a clumpier optical continuum : the sources with radio sizes greater than @xmath65kpc have on average more than twice as many clumps as the smaller radio galaxies . a spearman rank correlation test gives a significance level of 95@xmath71 for this correlation . + a possible explanation for this trend is that the medium around the hosts of powerful agn is dense and clumpy on a scale of more than 100 kpc ; as the radio sources expand through the gas , they light up more and more material either by triggering star formation in the gas clouds , or by enhancing the scattering properties of the material in the vicinity of the jets . this result is contrary to that found by best et al . ( 1996 ) for a complete sample of @xmath72 3cr radio sources , which have been imaged with the hst : they found that smaller radio sources tend to be comprised of a string of several knots , while larger radio galaxies are made generally of only 2 optical components . however note that the range of radio sizes of the @xmath72 3cr sample is 3 times as large as that of our sample . our sample covers a redshift range from z@xmath52 to [email protected] , which correspond to look - back times from 80% to 90% of the total age of the universe ( for @xmath73 ) . this epoch is close to the epoch of the formation of these hzrgs , therefore it is interesting to search for any evolution in the properties of the radio galaxies with increasing cosmic time . we follow a similar approach to that used by van breugel et al . ( 1998 ) for a sample of powerful hzrgs , observed with keck in the near infrared , which corresponds to the restframe optical emission ( @xmath74 ) . their sample is similar to ours , being comprised or a similar number of sources , with the same radio power , but has a higher average redshift ( @xmath75 versus @xmath76 ) and more galaxies having @xmath77 . there are 6 radio galaxies common to both samples . 14 we present the results for our sample of hzrgs : the left plot shows how the radio /optical size ratio vary with redshift ; the radio sizes are measured as the distances between the most distant hot - spots , on either side of the nucleus , while the optical lengths are defined to be the maximum extension of the optical emission in the direction of the radio source . in cases of multiple systems , such as pks 1138 - 262 and tx 1707 + 105 , all the optical components where considered , so that the radio / optical size ratio gives an indication of how much emission there is within the radio source extension . the plot indicates that there is no significant evolution in the ratio of radio to optical size . if we divide the sample in two redshift ranges , then the average radio / optical size ratio is 3.2 for the highest redshift bin ( @xmath78 ) and 3.4 for the lowest redshift radio galaxies , so the difference is negligible . this is different from the result of an breugel et al . for the infrared emission : they present marginal evidence that the host of @xmath77 radio galaxies are comparable in size with the radio sources , while the @xmath79 radio sources appear systematically larger that the hosts galaxies . + in the right plot of fig . 14 we show how the strength of the radio - optical alignment , represented by the difference in position angle between the optical and the radio emission @xmath68 ( see sect . above for definition ) varies with z. again there is no significant difference between the lowest redshift radio galaxies , which have an average pa difference of 20@xmath80 ) , is extremely different from the pa of whole system ( 3 galaxies , pa@xmath81 ) ] and the highest redshift sources which have an average of 18@xmath82 . on the contrary van breugel et al . find a strong evolution in the alignment of the host galaxies from @xmath77 to @xmath83 : specifically the infrared morphologies become smoother and less elongated at @xmath83 and the infrared / radio alignment strength decreases . the best interpretation is that , while for the lower redshift sources in their sample the near ir emission is dominated by the most evolved stellar population , ( which is less effected by the presence of the radio jets ) , for the very high redshift galaxies the observed near ir emission starts to be dominated by young stars , probably formed following the passage of the radio jets . + on the other hand , our hst observations sample the uv restframe emission which is though to be dominated by the younger stellar populations in all cases , regardless of redshift . these young hot stars are formed in subsequent small bursts , induced either by the interaction of the jets with the medium or by mergers of smaller subunits . such events may involve only little amounts of mass , but can still produce remarkable uv morphologies ( e.g 1138 - 262 pentericci et al . 1998 ) , and their frequency is not expected to change from redshift 4 to 2 . therefore we do nt expect any strong evolution in the uv restframe properties of the radio galaxies in this redshift interval . + + it is interesting to compare the morphologies of our high redshift radio galaxies with those of the high redshift galaxies , which have been recently discovered with uv dropout techniques and extensively studied by various groups , also with the hst ( see for example steidel et al . 1996 , giavalisco et al . 1996 , williams et al . 1996 ) . + as pointed out in sect . 6.1 , a fraction of the uv continuum emission of hzrgs is directly connected , through various possible mechanisms , to the presence of the agn . also some of the features that we see in the galaxies can be easily explained by a direct correlation with the radio jets , for example the narrow elongated structures seen in 4c 1243 + 036 and mrc 2104@xmath36242 and the jet - like feature observed in 4c 1345 + 245 . however in other cases there is a striking similarity between the _ individual components _ of the radio galaxies and the population of uv dropout galaxies , which clearly favors a stellar origin for the emission coming from those clumps . + particularly in some of the clumpiest and most extended radio galaxies , such as tx 1707 + 105 and mrc 2104@xmath36242 , there are components that have a compact and regular morphology , with sizes of the order of few kpc resembling that of the high redshift radio - quiet galaxies . in a previous paper we made a detailed inter - comparison between the clumps that are observed around the radio galaxy 1138@xmath36262 to the uv dropout galaxies : the conclusion was that those components had characteristics similar to the uv dropout galaxies such as absolute magnitudes , surface brightness profiles , half - light radii ( @xmath84 kpc ) and inferred star formation rates ( 5 - 10 @xmath85yr@xmath9 per clump ; pentericci et al . also 4c41.17 has a similar very clumpy morphology with compelling evidence that the clumps are star forming regions . however we must note that it is not yet possible to determine the masses of neither of the 2 classes of objects ( uv dropouts and radio galaxies clumps ) ; so it could as well be that they are intrinsically different objects , with a similar amount of star - bursting activity that make them look similar in the uv continuum emission . + it seems that at least some of the high redshift radio galaxies consist of a central large galaxy , that hosts the agn and a number of small star forming subunits , resembling the uv dropout galaxies , which are located in a region as large as @xmath86 kpc around the radio source . powerful radio sources would then pin - point to regions in which the density of star forming units is higher than average . the central host galaxies of radio sources may well have formed trough merging of these small sub - galactic stellar systems . note that the mergers of these gas - rich subunits with the host galaxies could have triggered ( or re - triggered ) the radio emission by providing fuel for the central engine of the agns , as it seen in many cases at low z ( e.g osterbrock 1993 ) . + our observations provide some qualitative support for hierarchical galaxy evolution models , which predict that the morphological appearance of galaxies during their formation period should be highly irregular and clumpy ( e.g baron & white 1989 ) . in particular semi - analytical models predict that one of the forms in which massive elliptical galaxies accrete their mass is from multiple merging of smaller subunits ( aragon - salamanca et al . 1998 , and references therein . a possible problem arises from the fact that in standard hierarchical cold dark matter models such massive systems are thought to form relatively late ( @xcite ) , i.e. at much lower redshift , and in the majority of galaxies the main population of stars is formed more recently ( after @xmath87 ) @xcite . however , white & frenk ( 1991 ) argue that a mechanism that could explain the formation of massive elliptical galaxies at an earlier epoch is over - merging of star - burst galaxies and indeed , as we have reviewed in the introduction , there is now increasing evidence that high redshift radio galaxies are probably located in the over - dense regions of the early universe . + therefore we conclude that high redshift radio galaxies may be formed from a aggregates of sub - galactic units , similar to the uv dropout galaxies , and will probably evolve into present day brightest cluster galaxies . in this paper we have presented new hst / wfpc2 images of 11 high redshift radio galaxies , all complemented with vla radio maps of comparable resolution . the images reveal a wide variety in the morphology of the host galaxies of these high redshift radio sources : in particular most objects have a clumpy , irregular appearance , consisting of a bright nucleus and a number of smaller components . the number of clumps seems to increase with increasing radio size . the uv continuum emission is generally elongated and aligned with the axis of the radio sources , however the characteristics of the `` alignment effect '' differ greatly from case to case . the new data confirm that none of the proposed models can satisfactorily explain the phenomenon and that most probably the aligned continuum emission is a mixture of star light , scattered light , and nebular continuum emission . our data show no significant evolution in the morphological properties over the redshift interval . finally , we compare the properties of our radio galaxies with those of the uv dropout galaxies and conclude that high redshift radio galaxies might be forming from aggregates of sub - clumps similar to the uv dropout galaxies and that they will probably evolve into present day brightest cluster galaxies . + in a future paper we will present complementary hst / nicmos data of an enlarged sample of high redshift radio galaxies . the new infrared observations will provide constrains to the age of the older stellar population of the host galaxies . with the high resolution we will be able to determine if also the older stellar population shows significant clumpy sub - structures and to what extent are the forming brightest cluster ellipticals already assembled and relaxed . this work is based on observations with the nasa / esa hubble space telescope , obtained at the space telescope science institute , which is operated by aura inc . under contract with nasa . we thank c. carilli for doing the reduction of the radio data for tx 1707 + 105 . hjar acknowledges support from an eu twinning project , a programme subsidy granted by the netherlands organization for scientific research ( nwo ) and a nato research grant . the work by wvb at igpp / llnl was performed under the auspices of the us department of energy under contract w-7405-eng-48 . lcccccccccc component & s@xmath88&s@xmath89&i@xmath88&i@xmath89&@xmath12&s@xmath90 & s@xmath91 & fp@xmath88 & fp@xmath89 & rm + & mjy & mjy & mjy / beam & mjy / beam & & mjy & mjy & % & % & rad m@xmath35 + + nw & 2.85 & 7.47 & 1.07 & 4.69 & 2.6 & & & & + se & 27.3 & 55.8 & 23.4 & 53.1 & 1.5 & 4.85&4.12 & 18&7.4 & 47 + lcccccccc component & s@xmath92&s@xmath89&s@xmath93&i@xmath92&i@xmath89&i@xmath93 & @xmath94 & @xmath95 + & mjy & mjy & mjy & mjy / beam & mjy / beam & mjy / beam & + + north lobe & 22.9 & 55.2 & 294 & 10.4 & 49.8 & 278 & 1.45&2.9 + core & 0.55 & & & 0.53 & & & + south lobe & 1.77 & 7.14 & 34.8 & 1.27 & 4.66 & 19.2 & 1.2 & 2.5 + we present here multi - frequency maps of the radio galaxy tx 1707 + 105 obtained with the vla in b array . + observations were made at 4.5 and 8.2 ghz , using two frequency channel each having a 50 mhz bandwidth , for a total integration time of 700 s and 1020 s respectively . data processing was done performed using the astronomical image processing system ( aips ) in the standard way . the system gains were calibrated with respect to the standard sources 3c286 . phase calibration was performed using the nearby calibrator 1658 + 076 . the antenna polarization response terms were determined using multiple scans of the calibrator 1850 + 284 over a large range in parallactic angle . absolute linear polarization position angles were measured using a scan of 3c286 . the calibrated data were then edited and self - calibrated using standard procedures to improve the dynamic range . images of the three stokes polarization parameters , i , q and u were synthesized and all images were cleaned down to a level of approximately 3 times the theoretical rms noise using the aips task imagr . the observations at the different frequencies were added in the image plane to produce the the final maps of total and polarized flux . 1 we show the maps at 4.5 ghz and 8.2 ghz ( with a resolution respectively of 1.2@xmath30 and 0.7@xmath30 ) of the total flux ( left panels ) and polarized flux ( right panels ) . in all images contours are spaced in a geometric progression with a factor of 2 , with the first contour level equal to 3@xmath60 , where @xmath60 the off - source rms which is 0.12 mjy for the 4.5 ghz map , 0.1 mjy for the 8.2 ghz map and 0.17 mjy for the polarized flux maps . + the radio galaxy has a simple double morphology with no radio core detected in the present images . the two lobes are nearly symmetric in total radio brightness , but the northern hot - spot is totally depolarized at both frequencies , while the southern one is polarized . in fig . 2 we present maps of the radio galaxy mrc 2104@xmath36242 obtained with the vla in b array at 3 different frequencies : 1.4 ghz , 4.5 ghz , and 8.2 ghz , with a resolution , respectively of 3.9@xmath30 , 1.2@xmath30 and 0.7@xmath30 . in all images contours are spaced in a geometric progression with a factor of 2 , with the first contour lever equal to 3@xmath60 , where @xmath60 the off - source rms , and is respectively at 1.74 mjy for the 1.5 ghz map , 0.19 mjy for the 4.5ghz map and 0.05 mjy for the 8.2 ghz map . + the radio source is a double showing fainter diffuse emission between the hot - spots and the core . the northern hot - spot is elongated in a direction the is different from the radio axis . osterbrock , d.e . : `` astrophysics of gaseous nebulae and active galactic nuclei '' , university science books , 1989 . pentericci l. , rottgering h. , miley g. , carilli c. , mccarthy p. : 1997 , _ a&a _ * 500 * , 580

near - field and resonance effects have a strong influence on the nanoscale electromagnetic energy transfer , and detailed understanding of these effects is required for the design of new , optimized nano - optical devices . we provide a comprehensive microscopic view of electromagnetic energy transfer phenomena by introducing quantum langevin heat baths as local noise sources in the equations of motion for the thermally fluctuating electric dipoles forming dielectric bodies . the theory is , in a sense , the microscopic generalization of the well - known fluctuational electrodynamics theory and thereby provides an alternative and conceptually simple way to calculate the local emission and absorption rates from the local langevin bath currents . we apply the model to study energy transfer between silicon carbide nanoparticles located in a microcavity formed of two mirrors and next to a surface supporting propagating surface modes . the results show that the heat current between the dipoles placed in a cavity oscillates as a function of their position and distance and can be enhanced by several orders of magnitude as compared to the free space heat current with a similar interparticle distance . the predicted enhancement can be viewed as a many - body generalization of the well - known cavity purcell effect . similar effects are also observed in the interparticle heat transfer between dipoles located next to a surface of a polar material supporting surface phonon polaritons .

1310.4677

electromagnetic energy transfer between dielectric bodies at different temperatures is commonly described using the fluctuational electrodynamics ( fed ) approach @xcite developed by rytov @xcite and first applied to condensed matter physics by lifshitz @xcite . according to fed , thermal motion of charged particles in a body creates random currents , which induce electromagnetic fields . outside the body , the field is then either radiated to free space or absorbed in the near or far - field regime by another body . proximity effects involving the evanescent waves in the near - field were first observed by hargreaves @xcite , and fed was consequently applied to theoretically predict strong near - field enhancement of heat transfer in various geometries @xcite . the predictions have been explored in more detail also experimentally @xcite . near - field effects are expected to have numerous applications in , e.g. , thermal microscopy @xcite , infrared thermophotovoltaics @xcite and narrow - band infrared antennas @xcite . as a statistical model , fed is closely related to langevin dynamics commonly applied to describe the random thermal motion of non - charged bodies @xcite . in langevin dynamics the particle is assumed to be coupled to a bath of harmonic oscillators , whose effect on the particle can effectively be described by a random force and deterministic friction @xcite . huttner and barnett @xcite essentially applied langevin dynamics to study the quantization of the electromagnetic field in absorbing dielectrics by coupling the polarization field to a bath of harmonic oscillators . the explicit inclusion of the microscopic degrees of freedom responsible for absorption solves the problem of temporally decaying field commutators arising if one blindly applies the standard electromagnetic field quantization methods to dielectrics . relation of the langevin dynamics to fed was recently highlighted by rosa , dalvit and milonni @xcite , who used langevin dynamics to derive the fluctuation - dissipation theorem ( fdt ) @xcite for the fluctuating polarization field from the microscopic motion of the oscillating dipoles . the goal of this paper is to show that the microscopic dipole oscillator model combined with quantum langevin dynamics can be used to transparently and microscopically treat also the problem of electromagnetic heat transfer between dielectric bodies held at different temperatures . the primary motivation for studying heat transfer using the quantum langevin equation instead of fed is that we can derive heat transfer rates in full analogy with phononic @xcite and electronic @xcite heat transfer and , by following the mathematical manipulations presented in ref . @xcite , we are able to arrive at an identical landauer - bttiker - like formula for the energy transmission function . the theory presented here enables , therefore , a unification of phononic , electronic and photonic heat transfer under the common langevin theory . when written in terms of particle polarizabilities and the electromagnetic green s function , the transmission function reduces to the form derived earlier from fed @xcite . in contrast to these works , we ( i ) consistently include the electromagnetic self - interaction produced by the local electromagnetic green s dyadic by following the discrete dipole approximation @xcite , ( ii ) include the inhomogeneous environment enabling , e.g. , accounting for cavity resonance effects , and ( iii ) present an alternative and conceptually simple way to derive electromagnetic energy transfer rates starting from the microscopic equations of motion . for presentational simplicity , we initially assume the particles to be small enough for the dipole approximation to hold . however , overcoming this assumption by following the well - established discrete dipole approximation mentioned above is also discussed . as an application of the formalism , we study the enhancement of heat transfer rates between sic particles placed in a microcavity and close to a polar surface supporting surface phonon polaritons ( spps ) . in a microcavity , the heat transfer rate between particles is shown to oscillate as a function of their distance and the cavity enhancement can be several orders of magnitude as compared to the free space heat current with a similar interparticle distance . enhanced heat current is predicted also for particles close to a sic surface , where the spps transport electromagnetic energy between the particles . the paper is organized as follows . in sec . [ sec : theory ] , we present the formulation to calculate thermal energy transfer between dielectric particles and the environment . we ( i ) represent the polarization fields inside the particles by their total dipole moments , ( ii ) solve the quantum langevin equations of motion for coupled dipole moment dynamics in terms of the dipole displacement green s function , ( iii ) calculate the thermal averages of heat currents using the fluctuation - dissipation theorem for the bath noises and the thermal field of the environment , and ( iv ) express the heat currents in terms of particle polarizabilities and the electromagnetic green s dyadic . in secs . [ sec : results_cavity ] and [ sec : results_surface ] , we investigate heat transfer between sic nanoparticles in a microcavity and above a sic surface , respectively . in this section we formulate the many - particle electromagnetic heat transfer problem in terms of the dipole approximation @xcite and langevin dynamics ( ld ) . our final results for the energy transfer rates are essentially equivalent to those obtained using fluctuational electrodynamics ( fed ) @xcite . therefore readers only interested in the numerical results regarding the energy transfer rates in inhomogeneous environments are not required to go through the rigorous derivations of this section to understand the results . however , our derivation based on ld provides an alternative view to the fed approach , offering additional insight to the physics involved . in contrast to the fed approach , where the electric fields and the induced dipole moments arising from the fluctuating dipole moments are solved to calculate the energy dissipated by the induced currents , ld is based on describing fluctuations by stochastic forces in the microscopic dipole equations of motion @xcite . with the help of poynting s theorem , we show that the locally absorbed power can be calculated in ld simply from the steady - state energy current to the local langevin heat bath . other advantages of the langevin approach compared to fed are discussed below at the end of sec . [ sec : transmission_polarizability ] . in our system setup , the studied dielectric particles with electric susceptibilities @xmath0 are located in an environment defined through its relative permittivity @xmath1 , so that the overall relative permittivity is given @xmath2 outside the particles and @xmath3 at particle coordinate @xmath4 as illustrated in fig . [ fig : sud1](a ) . consequently the environment , consisting in the example shown in fig . [ fig : sud1](a ) of the two cavity walls , is described as a single object whose temperature is constant but the permittivity can be inhomogeneous . if the particles are located in pure vacuum environment so that @xmath5 @xmath6 , the environment only acts as a source of black - body thermal radiation . if the environment is inhomogeneous , it not only generates background thermal radiation but also scatters the radiation emitted by the dipoles . for simplicity , we assume that all particles are non - magnetic such that the relative magnetic permeability equals unity everywhere . small dielectric particles with positions @xmath7 , electric susceptibilities @xmath8 and temperatures @xmath9 is located in an inhomogeneous environment , consisting in the shown case of two dielectric ( or metallic ) bodies occupying the left and right half - spaces . the overall relative permittivity is given by @xmath2 outside the particles and @xmath3 at each particle coordinate @xmath4 . the part described by the environment dielectric constant , consisting in this example of the two cavity walls , is assumed to act as a source of thermal radiation at temperature @xmath10 . the polarization field inside each particle @xmath11 is treated as an oscillating point dipole moment @xmath12 coupled to a local langevin bath at temperature @xmath9 as shown in ( b ) . the total local field @xmath13 driving each dipole moment @xmath12 is the sum of the stochastic background field @xmath14 and the fields @xmath15 created by each dipole @xmath16.,title="fig : " ] small dielectric particles with positions @xmath7 , electric susceptibilities @xmath8 and temperatures @xmath9 is located in an inhomogeneous environment , consisting in the shown case of two dielectric ( or metallic ) bodies occupying the left and right half - spaces . the overall relative permittivity is given by @xmath2 outside the particles and @xmath3 at each particle coordinate @xmath4 . the part described by the environment dielectric constant , consisting in this example of the two cavity walls , is assumed to act as a source of thermal radiation at temperature @xmath10 . the polarization field inside each particle @xmath11 is treated as an oscillating point dipole moment @xmath12 coupled to a local langevin bath at temperature @xmath9 as shown in ( b ) . the total local field @xmath13 driving each dipole moment @xmath12 is the sum of the stochastic background field @xmath14 and the fields @xmath15 created by each dipole @xmath16.,title="fig : " ] following the dipole approximation , the internal polarization field of each particle with susceptibility @xmath8 is modeled as a dipole located at the central coordinate @xmath4 of the particle @xmath11 as illustrated in fig . [ fig : sud1](b ) . the microscopic dipole moments , which represent the fluctuating electric polarization inside each particle , are then coupled to ( i ) local heat baths describing thermal fluctuations and dissipation , ( ii ) to the electromagnetic field arising from other dipoles , and ( iii ) to the thermal field originating from the environment . the dipole formulation results in equations of motion for the dipole displacements , which are solved by using their fourier transform in sec . [ sec : equationofmotion ] . the electromagnetic green s dyadic coupling the dipoles is defined in sec . [ sec : greens_dyadics ] . combined with the microscopic expressions for local absorption and emission derived from the poynting theorem in sec . [ sec : poynting ] , the solutions of the equations of motion and the correlation functions of the langevin noise and the environment thermal field described in sec . [ sec : noise ] are then used in sec . [ sec : lb_formula ] to derive concise expressions for the electromagnetic energy transfer between dipoles as well as the environment field . finally , in secs . [ sec : polarizabilities ] and [ sec : transmission_polarizability ] we relate the local oscillator parameters to the particle polarizabilities and express the electromagnetic energy transfer rates in terms of these quantities and the full electromagnetic green s dyadic . the dipole approximation is strictly valid if the particles are much smaller than the dominant wavelength and the electric field amplitude is constant inside each particle @xcite . however , these restrictions can be easily lifted by dividing the particles into sufficiently small dipolar subvolumes by following the well - established discrete dipole approximation @xcite . the formulas presented in this section are then simply written for a larger number of dipoles constituting the particles . the general forms of the green s dyadic and the langevin equations of motion remain unchanged , and consequently the mathematical derivation leading to the expressions , and for locally absorbed power is then identical with the derivation given here . we model the dipole dynamics of the polarization field in each particle @xmath11 by the classical oscillator model @xcite accompanied by quantum langevin dynamics @xcite describing local thermal fluctuations . the equation of motion and its fourier transform for the displacement coordinate @xmath17 corresponding to dipole moment @xmath12 of dipole @xmath11 located at @xmath4 are then @xmath18 and @xmath19 respectively . the local electric field @xmath20 , which acts as a driving term in the equations of motion , is discussed below . the dipole mass @xmath21 , resonance frequency @xmath22 , langevin friction constant @xmath23 and charge @xmath24 are later incorporated into the definition of particle polarizability as discussed in sec . [ sec : polarizabilities ] . the friction term @xmath25 represents the damping of dipole oscillations due to the coupling to a local heat bath and is responsible for dissipation , which appears as a non - zero imaginary part in the polarizability . dissipation to the local heat bath is accompanied by thermal fluctuations described by the random force @xmath26 , which turns the equation of motion into a stochastic differential equation . the relative magnitude of fluctuations and dissipation at each dipole site @xmath11 depends on the bath temperature @xmath9 through the fluctuation - dissipation relation presented in sec . [ sec : noise ] . for notational simplicity , we have assumed the friction to be memoryless and proportional to the instantaneous velocity with friction constant @xmath27 , representing ohmic damping @xcite . for non - ohmic bath , the derivation of the heat transfer rates proceeds similarly and the final results for heat transfer rates , derived in sec . [ sec : lb_formula ] , are given in a form valid for arbitrary damping . the fourier transform in eq . is defined , as usual , by @xmath28 with the corresponding inverse transform @xmath29 e^{-i\omega t } \hat{f}(\omega)$ ] . the fourier - transformed local electric field @xmath30 appearing in eq . can be written following the discrete dipole approximation @xcite as @xmath31 here @xmath32 is the stochastic thermal field originating from the environment and @xmath33 is the electric field due to dipole moment @xmath34 , given in frequency domain by @xmath35 the electromagnetic green s dyadic @xmath36 appearing in eq . is defined in sec . [ sec : greens_dyadics ] , where also the definition of the local green s dyadic @xmath37 accounting for the polarization field due to near - neighborhood and the radiation damping force is discussed . for green s functions , only the frequency - domain representations are used in this paper so we omit their hats for brevity . equation is semiclassical in the sense that we treat the displacements @xmath38 , noise @xmath26 and field @xmath39 as classical commuting variables and only include quantum effects by imposing quantum fluctuation - dissipation relations for the symmetrized correlators of @xmath26 and @xmath39 @xcite . we expect that a full quantum treatment of the mechanical degrees of freedom would give , after proper symmetrization of observables , identical results for heat transfer rates . this follows from the linearity of the equations of motion and is illustrated in refs . @xcite for phonon heat transfer . the substitution of eqs . and to eq . gives @xmath40 equation can be rearranged as @xmath41 by defining an inverse propagator @xmath42\delta_{ij}{\mathbf}{i}_{3\times 3}+ q^2\omega^2\mu_0 { \mathbb{g}}_{ij},\ ] ] where @xmath43 is the @xmath44 unit matrix . the solution to eq . can be written compactly in matrix form as @xmath45 , \label{eq : usol}\ ] ] where the dipole displacement green s function @xmath46 is @xmath47+i\gamma^{\textrm{bath}}(\omega)/2+i\gamma^{\textrm{rad}}(\omega)/2}. \label{eq : g_expression1}\ ] ] here we adopt a matrix notation where the dipole indices @xmath48 and the spatial components @xmath49 are combined into a composite index resulting in matrices and vectors of size @xmath50 and @xmath51 , respectively . in the following , we will use an index notation where the subscript @xmath52 ( @xmath11 ) always refers to the @xmath44 matrix ( 3-component vector ) corresponding to the notation used before eq . . in eq we have additionally defined the block - diagonal resonance frequency matrix as @xmath53 , the block - diagonal bath coupling matrix as @xmath54 , and the radiation coupling function defined through the imaginary part of the electromagnetic interaction by@xmath55 . \label{eq : gammarad_def}\ ] ] in the denominator of eq . and in all matrix - valued expressions appearing below , the scalar terms should be interpreted as being proportional to the unit matrix of size @xmath56 . equation shows that both the coupling to heat baths and the coupling to the radiation field produce broadening in the green s function via the coupling functions @xmath57 and @xmath58 . this broadening in the green s function reflects dissipation that , along with the accompanying thermal fluctuations , enables energy transfer between dipoles with different bath temperatures . the electric green s dyadic @xmath60 , which appears in eq . through the definition @xmath61 , is the green s function of the nonhomogeneous helmholtz equation for the electric field and defined by the equation @xcite @xmath62 here the wavenumber is @xmath63 defined in terms of frequency @xmath64 and the speed of light @xmath65 in vacuum . by definition , the green s dyadic @xmath59 accounts for the scattering of the electromagnetic field from the inhomogeneities in the environment but not from the dipoles themselves . this effect is , however , automatically included in the formalism by the coupled dipole equations of motion . since magnetic effects are neglected in this paper , we use the terms electric green s dyadic and electromagnetic green s dyadic interchangeably . the electric green s dyadic can be separated into two parts as @xmath66 where the free - space green s dyadic @xmath67 is the particular solution of the inhomogeneous equation with @xmath68 , which in our case corresponds to @xmath5 , @xcite @xmath69 \frac{e^{ik_0|{\mathbf}{r}-{\mathbf}{r}'|}}{4\pi|{\mathbf}{r}-{\mathbf}{r}'|}.\ ] ] the scattering part @xmath70 is then the solution to the homogeneous equation that satisfies the proper boundary conditions due to inhomogeneities in @xmath1 . the separation is necessary since the real part of the green s dyadic @xmath71 diverges for @xmath72 , so the local green s dyadic @xmath73 appearing in eq . is defined for each particle @xmath11 as @xcite @xmath74 { { \mathbf}{i}_{3 \times 3}}+ { \mathbb{g}}^s({{\mathbf}{r}}_i,{{\mathbf}{r}}_i;\omega ) \label{eq : mi_expr } \ ] ] where we have assumed each particle to be a sphere with radius @xmath75 and volume @xmath76 . we have also only included the lowest - order terms in @xmath77 . the imaginary part of the term in braces in eq . is the imaginary part of the local free - space green s dyadic , @xmath78=i\omega/(6\pi c ) { { \mathbf}{i}_{3 \times 3}}$ ] . the electric field produced by this local term is proportional to @xmath79 and produces the abraham - lorentz radiation damping @xcite proportional to the third time - derivative of the dipole moment in the equation of motion . the fluctuations accompanying the radiation damping are responsible for generating thermal radiation @xcite . the most straightforward approach to evaluate the energy transfer rates between the dipoles is to first calculate the electric and magnetic fields @xmath80 and @xmath81 due to the dipoles and to directly integrate the electromagnetic power flux obtained from the poynting vector @xmath82 over a surface enclosing the dipole under study . however , the calculation of the surface integrals is complicated and , in the present approach , one can simplify the problem by studying the heat transfer in terms of the dissipation and emission by the discrete dipoles . to calculate the power absorbed by the dipole @xmath11 , we shift momentarily back to time domain and apply maxwell s equations @xmath83 and @xmath84 , for non - magnetic media with polarization density @xmath85 and no free charges , on the surface integral of the poynting vector over the boundary @xmath86 enclosing the particle volume @xmath87 . by also applying the gauss divergence theorem , we then get @xcite @xmath88 d{\mathbf}{r } - \int_{v_i } { \mathbf}{e } \cdot \frac{\partial { \mathbf}{p}}{\partial t } d{\mathbf}{r } \\ & \approx - \int_{v_i } \frac{1}{2 } \frac{\partial}{\partial t } \left[\varepsilon_0 { \mathbf}{e}^2 + \mu_0 { \mathbf}{h}^2 \right ] d{\mathbf}{r } - \delta v_i { \mathbf}{e}_i \cdot \frac{\dot{{\mathbf}{p}}_i}{\delta v_i } \label{eq : poynting_4 } \\ & = - \frac{d } { d t } \int_{v_i } \frac{1}{2 } \left[\varepsilon_0 { \mathbf}{e}^2 + \mu_0 { \mathbf}{h}^2 \right ] d{\mathbf}{r } - \left[m\ddot{{{\mathbf}{u}}}_i+m\omega_i^2{{\mathbf}{u}}_i-\xi_i+m\gamma\dot{{{\mathbf}{u}}}_i \right]\cdot \dot{{{\mathbf}{u}}}_i . \label{eq : poynting_3 } \ ] ] in eq . , we approximated the electric field inside the volume @xmath87 by the constant @xmath20 and similarly the polarization density by @xmath89 . in moving to eq . , we applied the equation of motion to solve for @xmath20 . the expression is interpreted as the energy conservation law @xmath90 the first term on the right - hand side is the time derivative of the electromagnetic energy @xmath91d{\mathbf}{r } $ ] stored in @xmath87 , the second term is the rate of change of the mechanical energy @xmath92 stored in the oscillator @xmath11 , and the third term @xmath93 is the energy current to the langevin heat bath . by thermal averaging ( defined below in sec . [ sec : noise ] ) and assuming steady state so that the total time - derivatives in eq . vanish , one gets @xmath94 , implying that the energy absorbed from the bath is equal to the emitted electromagnetic radiation . in order to calculate the thermal average of the bath heat current , we need to specify the statistical properties of the stochastic langevin noise @xmath95 and the thermal background field @xmath96 appearing in the solution to the equation of motion . the langevin noise has zero average @xmath97 and its ( symmetrized ) autocorrelation function satisfies the quantum fluctuation - dissipation theorem @xcite @xmath98 \delta_{ij } , \label{eq : xiixij}\end{aligned}\ ] ] where @xmath9 is the local bath temperature and the bose - einstein function is @xmath99^{-1}$ ] . here @xmath100 is the reduced planck s constant and @xmath101 is the boltzmann constant . the delta function @xmath102 reflects translational invariance in time and the kronecker delta @xmath103 ensures that the heat baths at sites @xmath11 and @xmath16 are uncorrelated , i.e. , the baths are local . for the memoryless friction assumed in eq . , the bath coupling function is @xmath104 as defined in sec . [ sec : equationofmotion ] . the thermal background field @xmath96 has zero average @xmath105 and its symmetrized autocorrelation function satisfies the fluctuation - dissipation relation @xcite @xmath106}\label{eq : ebgebg1},\end{aligned}\ ] ] where @xmath58 was defined in eq . . the simultaneous presence of @xmath107 $ ] both in eq . and as a source of dissipation in the dipole displacement green s function ensures the onset of thermal equilibrium when the dipole and environment bath temperatures are equal as discussed in sec . [ sec : lb_formula ] . we also note that the bath noises @xmath26 and the thermal background field @xmath39 are uncorrelated , @xmath108 having the solution for the dipole displacements and noise correlations and available , we are equipped to calculate the thermal average of the energy flow to each local heat bath . as shown in app . [ sec : appendix1 ] , the thermal average of the energy current flowing to the local heat bath at site @xmath11 can be written in the landauer - bttiker form @xmath109 \notag \\ & \quad + \int_0^{\infty } \frac{d\omega}{2\pi } \hbar \omega { \mathcal}{t}_{i,\textrm{rad}}{(\omega)}[f_b(\omega,{t_{\textrm{env}}})-f_b(\omega , t_i ) ] , \label{eq : landauer_formula}\end{aligned}\ ] ] where the energy transmission function between dipoles @xmath11 and @xmath16 is @xmath110 \label{eq : tij}\ ] ] and between dipole @xmath11 and the radiation field @xmath111_{ii } \right\}. \label{eq : tirad}\ ] ] note that also here we use the index notation where @xmath112_{ii}$ ] is a @xmath113 matrix describing the dyadic elements related to dipole @xmath11 . the first term of eq . accounts for the heat transfer between dipoles . the transmission function is of the well - known caroli form @xcite , giving the energy - dependent transmission function in terms of the green s function @xmath114 and bath coupling functions @xmath115 . in the context of electron @xcite and phonon @xcite transport , the bath coupling functions typically represent the coupling of the scattering region to semi - infinite leads , although they may also model coupling to local self - consistent voltage probes @xcite or heat baths @xcite to account for inelastic scattering of electrons and phonons , respectively . the second term of eq . consists of absorption due to the environment field at temperature @xmath10 and the radiation emitted by the dipole . the environment field compensates the energy loss due to radiation , ensuring that in the uniform temperature configuration , @xmath116 , energy transfer to all local heat baths vanishes and the system is in thermal equilibrium . equation naturally implies that the total energy transfer to the radiation field is @xmath117.\ ] ] as the transmission functions and are symmetric , it is then easy to show that @xmath118 , showing that total energy is conserved . in this subsection we relate the parameters appearing in the dipole equation of motion and the transmission functions and to the polarizabilities of the particles by studying the induced dipole moment @xmath119 when a single particle @xmath11 is placed in an external field @xmath120 . for generality , we treat the particle polarizabilities as tensors , since the final expressions for the transmission functions are valid also for anisotropic media . anisotropy would simply appear in the equation of motion through direction - dependent parameters and , in the case of a non - principal coordinate system , as coupling between the different cartesian components of the dipole displacement vector . for completeness and due to the several widely used definitions for the local polarizability tensor , we specifically define the bare , clausius - mossotti and effective polarizabilities through relations @xmath121 , \\ \langle\hat{{{\mathbf}{p}}}_i(\omega ) \rangle & = \varepsilon_0 \alpha_{\textrm{eff}}^i(\omega ) \hat{{\mathbf}{e}}_0({{\mathbf}{r}}_i,\omega),\end{aligned}\ ] ] where the bare polarization @xmath122 relates the dipole moment and the total macroscopic electric field @xmath123 , the clausius - mossotti polarizability @xmath124 relates the dipole moment to the sum of @xmath125 and the polarization field @xmath126 , the latter modifying the local field seen by the dipole , and the effective polarizability @xmath127 relates the dipole moment directly to the external field . by solving for @xmath128 from the equation of motion for a single dipole in an external field @xmath129 , one gets @xmath130 it is easy to see that the polarizabilities are related by @xmath131 note that the definition of the effective polarizability depends on the environment through the local green s dyadic @xmath73 [ eq . ] . definition based on the effective polarizability in free - space is obtained by replacing @xmath73 in eq . by the free - space green s dyadic @xmath132 [ eq . ] . the polarizabilities are directly related to the known bulk dielectric constants @xmath133 of the particles through @xmath134 and @xmath135 . using the definition of the bare polarizability , we see that @xmath136 . using also eqs . - , one gets @xmath137{{\mathbf}{i}_{3 \times 3}}\\ \alpha_{\textrm{cm}}^i(\omega ) & = 3\delta v_i \frac{\varepsilon_i(\omega)-1}{\varepsilon_i(\omega)+2 } { { \mathbf}{i}_{3 \times 3 } } , \label{eq : alphacm_epsilon } \\ \alpha_{\textrm{eff}}^i(\omega ) & =3 \delta v_i \frac{\varepsilon_i(\omega)-1}{\varepsilon_i(\omega)+2 } \times \left\ { { { \mathbf}{i}_{3\times 3}- \frac{3{k_0}^2[\varepsilon_i(\omega)-1]}{\varepsilon_i(\omega)+2 } } \mathbb{m}_i \right\}^{-1}. \label{eq : alphaeff_final}\end{aligned}\ ] ] equation is the well - known clausius - mossotti relation ( or lorentz - lorenz relation in optics ) connecting the local polarizability to the dielectric constant @xcite . in eq . , we have defined the dyadic @xmath138 that excludes the polarization dyadic from the local green s function @xmath139 [ eq . ] as @xmath140 { { \mathbf}{i}_{3 \times 3}}+ \delta v_i { \mathbb{g}}^s({{\mathbf}{r}}_i,{{\mathbf}{r}}_i;\omega).\end{aligned}\ ] ] definition arises in the discrete dipole approximation as the unique definition of the local polarizability that ensures the equivalence of the method of coupled dipoles ( which solves the dipole moments ) with the method of moments ( which solves the electric fields ) @xcite . as a side note , we note that our formalism inherently includes the modified fluctuation - dissipation theorem derived by manjavacas and garca de abajo @xcite . the fluctuation - dissipation theorem for dipole moments can be derived from our formalism by noting that for a single dipole coupled to a thermal bath at temperature @xmath9 and located in empty space with no environment field , the expectation value of the dipole moment correlation function can be written as @xmath141 \delta(\omega+\omega ' ) \label{eq : pp_corr},\ ] ] where the reduced susceptibility tensor @xmath142 ( note the difference to the bulk electric susceptibility @xmath0 ) is defined in terms of the effective polarizability @xmath143 as @xmath144 - \frac{{k_0}^3}{6\pi } \alpha_{\textrm{eff}}^i(\omega ) \alpha_{\textrm{eff}}^i(\omega)^{\dagger}. \label{eq : gammabath_chi}\ ] ] we do not , however , need to impose eq . for the dipole moment fluctuations , since fluctuations are naturally included in the formalism by the coupling to langevin baths . since all three definitions , and of the polarizability can be directly related to the dielectric constant , the choice of which polarizability to use is a matter of taste . we choose to express the transmission functions and in terms of the clausius - mossotti polarizability . we first note that the term @xmath145 , reflecting the strength of coupling to the local heat bath , can be solved from the definition of the clausius - mossotti polarizability to get the bath coupling function @xmath146^{-1 } \textrm{im}[\alpha^i_{\textrm{cm}}(\omega)][\alpha^i_{\textrm{cm}}(\omega)^{\dagger}]^{-1}. \label{eq : gammabath_cm}\end{aligned}\ ] ] similarly we can express the dipole displacement green s function in terms of the block - diagonal polarizability matrix @xmath147 $ ] as @xmath148 note that the unknown microscopic parameters @xmath21 , @xmath22 and @xmath23 appearing in the dipole equation of motion @xmath149 and in the dipole displacement green s function have now been fully absorbed into the particle polarizabilities , which can in turn be inferred from the dielectric constants using eq . . substituting eq . , its complex conjugate and eq . to eq . for @xmath150 , one gets after straightforward algebraic manipulations @xmath151 { { { \mathbb{g}}}^{\textrm{cm}}}_{ij}{(\omega)}\textrm{im}[\alpha^j_{\textrm{cm}}(\omega ) ] { { { \mathbb{g}}}^{\textrm{cm}}}_{ji}(\omega)^{\dagger}\right ] . \label{eq : tij_final}\ ] ] here the electromagnetic green s dyadic @xmath152 expressed in terms of the clausius - mossotti polarizabilities of the particles has been defined as @xmath153 \alpha_{\textrm{cm}}(\omega)^{-1 } \\ & \equiv \frac{1}{{k_0}^2 } \alpha_{\textrm{cm}}(\omega)^{-1 } + \underbrace { \left[\frac{1}{1-{k_0}^2 { \mathbb{g}}(\omega)\alpha_{\textrm{cm}}(\omega ) } \right ] { \mathbb{g}}(\omega)}_{{{{\mathbb{g}}}^{\textrm{full}}}}. \label{eq : gemmb_cm_app}\end{aligned}\ ] ] the first term of eq . is local and does not contribute to dipole - dipole energy transfer . the second term of eq . , denoted by @xmath154 , is non - local and therefore responsible for dipole - dipole energy transfer . it is easy to see that this part satisfies the equation @xmath155 comparing with the well - known dyson equation satisfied by the electromagnetic green s dyadic in an inhomogeneous environment @xcite and noting that @xmath59 is the green s dyadic unperturbed by the dipoles , one notices that @xmath154 can be interpreted as the electromagnetic green s dyadic that fully incorporates the scattering caused by the dipoles . equation arises also from the combination of eqs . and , if one divides the dipole moment into the induced and fluctuating parts and calculates the electric field due to the fluctuating parts @xcite . here eq . arises as a convenient definition that allows us to express the transmission function in terms of purely optical quantities . the radiation transmission function can be similarly written in terms of the polarizabilities and the clausius - mossotti green s dyadic as @xmath156 { { { \mathbb{g}}}^{\textrm{cm}}}_{ij}(\omega ) \alpha_{\textrm{cm}}^j(\omega ) \textrm{im}[{\mathbb{g}}_{jk}(\omega ) ] \alpha^k_{\textrm{cm}}(\omega)^ * { { { \mathbb{g}}}^{\textrm{cm}}}_{ki}(\omega)^{\dagger } \right\}. \label{eq : tirad_final}\end{aligned}\ ] ] note that also the first , local term of eq . contributes to the sum through the local terms @xmath157 . it is straightforward to derive similar expressions for @xmath150 and @xmath158 also in terms of the bare and effective polarizabilities , but since the final results are anyway equivalent , only eqs . and are presented here . equations and for the transmission functions have been derived also from fluctuational electrodynamics ( fed ) @xcite . the main difference between fed and langevin dynamics ( ld ) is that in fed , thermal dipole fluctuations are described by the dipole fluctuation - dissipation theorem . accompanied with self - consistent equations for the electric fields and the induced parts of the dipole moments , the locally dissipated energy at each dipole site can be calculated from the product of the local electric field and the induced dipole current . in langevin dynamics , on the other hand , dipolar fluctuations are described by stochastic forces in the microscopic dipole equation of motion , which separately describes all driving and dissipative forces acting on the dipole displacements . with the help of poynting s theorem , we showed that the calculation of the locally absorbed power from ld does not require solving for the electric fields : only the dipole moments and the fdts and are required [ eq . ] . langevin dynamics has also other advantages compared to fed . first , we were able to show that the dipole - dipole energy transmission function has exactly the same caroli form @xcite as in electronic @xcite and phononic @xcite transport . our formalism therefore presents an appealing unification of photonic energy transfer with the electronic and phononic theories , promoting the development of coupled energy transfer models covering all three carriers . second , our methodology offers a simple , alternative derivation of the modified fluctuation - dissipation theorem , in which the abraham - lorentz radiation damping [ eq . ] is subtracted from the imaginary part of the effective polarizability in the fluctuation - dissipation theorem to avoid non - physical fluctuations in non - absorbing particles @xcite . in fed , this subtraction is not inherently built in the theory but has to be derived separately . finally , langevin formalism does not require artificial separation between the fluctuating and induced parts of the dipole moment as in fed : only the total dipole moment appears in our formalism . the well - known purcell effect @xcite symbolizes the strong position dependence of the spontaneous emission rate of an atom placed in an optical cavity . the same effect has also been reported for dipole - dipole interactions in a resonant cavity @xcite , but the details of the thermal energy transfer and the perturbations caused by the dipoles themselves have not previously been studied in cavities including several dipoles . as an example of the developed formalism , we investigate in sec . [ sec : results_cavity ] how cavity modes modify the energy exchange rate between sic nanoparticles in a microcavity . in sec . [ sec : results_surface ] , we also investigate the modification of the interparticle heat transfer rate above a sic surface , which can support coupled surface modes of phonons and photons known as surface phonon polaritons @xcite . in both configurations , the nanoparticle radii are set to @xmath159 nm and the dielectric constant of sic is modeled by the lorentz model @xcite @xmath160 where @xmath161 , @xmath162 mev , @xmath163 mev and @xmath164 @xmath165ev . the imaginary parts of the particle polarizabilities @xmath166 [ eq . ] peak strongly at the resonance energy @xmath167 mev corresponding to the condition @xmath168=-2 $ ] . the strong absorptance peak at this energy dominates the heat transfer rates at room temperature , and the associated resonant mode is sometimes called the frhlich mode @xcite . . particle @xmath169 is located at the center of a planar cavity ( @xmath170 ) and the location of the particle @xmath171 is varied.,width=325 ] \}/(\pi c^2)$ ] in an empty cavity of length @xmath172 @xmath165 m as a function of distance @xmath173 from the cavity wall and energy @xmath174 , with arbitrary in - plane positions @xmath175 and @xmath176 . the confined modes appear as position - dependent densities of states above threshold energies @xmath177 . ( b ) the in - plane dispersion of the confined electromagnetic modes in the cavity . the resonance energy @xmath167 mev of a sic nanoparticle , shown as the dashed line , can couple to the third mode at @xmath178 mev [ crossing denoted by ( 3 ) ] , to the second mode at @xmath179 mev ( 2 ) and to the first mode at @xmath180 mev ( 1 ) . the corresponding in - plane wavelengths @xmath181 are @xmath182 @xmath165 m , @xmath183 @xmath165 m and @xmath184 @xmath165 m for ( 1 ) , ( 2 ) and ( 3 ) , respectively.,title="fig : " ] \}/(\pi c^2)$ ] in an empty cavity of length @xmath172 @xmath165 m as a function of distance @xmath173 from the cavity wall and energy @xmath174 , with arbitrary in - plane positions @xmath175 and @xmath176 . the confined modes appear as position - dependent densities of states above threshold energies @xmath177 . ( b ) the in - plane dispersion of the confined electromagnetic modes in the cavity . the resonance energy @xmath167 mev of a sic nanoparticle , shown as the dashed line , can couple to the third mode at @xmath178 mev [ crossing denoted by ( 3 ) ] , to the second mode at @xmath179 mev ( 2 ) and to the first mode at @xmath180 mev ( 1 ) . the corresponding in - plane wavelengths @xmath181 are @xmath182 @xmath165 m , @xmath183 @xmath165 m and @xmath184 @xmath165 m for ( 1 ) , ( 2 ) and ( 3 ) , respectively.,title="fig : " ] in this section , the nanoparticles are assumed to be located in a planar cavity of length @xmath185 , illustrated in fig . [ fig : cavity_geom ] . the cavity green s dyadic @xmath186 is calculated in the presence of the cavity walls occupying the half - spaces @xmath187 and @xmath188 @xcite . the space between the walls is vacuum corresponding to the refractive index @xmath189 . the refractive index in the two half - spaces is chosen to be @xmath190 , which corresponds to the reflection coefficient @xmath191 and phase shift @xmath192 at normal incidence . the environment dielectric function is then @xmath193 for @xmath187 and @xmath188 and unity otherwise . we decompose the wavevector @xmath194 into longitudinal and in - plane parts as @xmath195 . as the cavity walls are very reflective , modes satisfying the constructive interference condition @xmath196 show up in the local electromagnetic density of states in fig . [ fig : rho](a ) , where the electric density of states @xmath197\}/(\pi c^2 ) $ ] @xcite in a cavity of length @xmath172 @xmath165 m is shown . the threshold energies of the cavity modes are @xmath177 , the smallest three of which are visible at energies @xmath198 mev , @xmath199 mev , and @xmath200 mev . for frequency @xmath64 , the corresponding in - plane wavevectors are @xmath201 , whose dispersion is shown in fig . [ fig : rho](b ) . the frhlich resonance energy of a sic nanoparticle , shown as the dashed line , can couple to each of the three modes at different in - plane wavevectors @xmath202 . ( solid blue or dark gray , left axis ) and the interparticle heat current @xmath203 ( solid red or light gray , right axis ) as a function of sic nanoparticle separation @xmath204 with in - plane separation @xmath205 . one particle is fixed at the center of the cavity , @xmath170 , as illustrated in the inset . the vertical axes are logarithmic and the free space transmission functions and heat currents are shown as dashed lines . in the transmission function , the frequency @xmath64 is chosen to correspond to the nanoparticle mode resonance @xmath167 mev , where the particles can strongly couple to the third standing mode [ point ( 3 ) in fig . [ fig : rho](b ) ] . for the heat current , the particle temperatures are @xmath206 k and @xmath207 k. the cavity length is @xmath172 @xmath165 m and the nanoparticle radius @xmath159 nm . ( b ) transmission function @xmath208 and the heat current @xmath209 as a function of @xmath204 for cavity ( solid ) and free space ( dashed ) . for the heat current , the environment temperature is @xmath210 k and the particle temperature is @xmath207 k. in contrast to ( a ) , the vertical axes are linear.,title="fig : " ] ( solid blue or dark gray , left axis ) and the interparticle heat current @xmath203 ( solid red or light gray , right axis ) as a function of sic nanoparticle separation @xmath204 with in - plane separation @xmath205 . one particle is fixed at the center of the cavity , @xmath170 , as illustrated in the inset . the vertical axes are logarithmic and the free space transmission functions and heat currents are shown as dashed lines . in the transmission function , the frequency @xmath64 is chosen to correspond to the nanoparticle mode resonance @xmath167 mev , where the particles can strongly couple to the third standing mode [ point ( 3 ) in fig . [ fig : rho](b ) ] . for the heat current , the particle temperatures are @xmath206 k and @xmath207 k. the cavity length is @xmath172 @xmath165 m and the nanoparticle radius @xmath159 nm . ( b ) transmission function @xmath208 and the heat current @xmath209 as a function of @xmath204 for cavity ( solid ) and free space ( dashed ) . for the heat current , the environment temperature is @xmath210 k and the particle temperature is @xmath207 k. in contrast to ( a ) , the vertical axes are linear.,title="fig : " ] figure [ fig : t12_a](a ) shows the transmission functions @xmath211 and the interparticle heat current @xmath203 as a function of particle distance @xmath204 for two sic nanoparticles in a cavity . one of the particles is fixed at the center of the cavity at @xmath170 and the in - plane separation is set to @xmath205 as shown in the inset . the cavity length is @xmath172 @xmath165 m and the frequency in the transmission function is set to correspond to the nanoparticle frhlich resonance @xmath212 mev . in the heat current calculation , the particle temperatures are set to @xmath206 k and @xmath207 k. the heat current closely follows the trends of the transmission function due to the strong absorptance peak at the frhlich mode frequency @xmath213 . for small @xmath204 ( but @xmath214 ) the interparticle heat current @xmath215 diverges as @xmath216 due to the near - field divergence of the free - space green s dyadic @xcite . as the separation @xmath204 increases , the interparticle heat current quickly starts to decrease . due to the coupling to the cavity modes , the transmission function and the heat current have , however , another local maximum at @xmath217 @xmath218 , corresponding to the amplitude maximum of the third standing mode at @xmath219 in fig . [ fig : rho](a ) and depicted by point ( 3 ) in fig . [ fig : rho](b ) . at the amplitude maximum , the interparticle transmission is enhanced ten - fold compared to the free - space value ( dashed line ) . the transmission function @xmath220 and the particle - environment heat current @xmath209 are plotted in fig . [ fig : t12_a](b ) . in the heat current calculation , we set the environment temperature to @xmath210 k. coupling to the radiation field depends on the particle position due to the position - dependent coupling to the third cavity mode and results in the well - known purcell effect . at @xmath217 @xmath165 m , where the interparticle transmission @xmath221 has a local maximum as a function of particle @xmath171 position , the transmission @xmath208 also has a local maximum and the value exceeds the interparticle transmission nearly by two orders of magnitude . this implies that if the temperature of the nanoparticle 1 was increased above @xmath10 , the steady - state temperature @xmath222 of the particle @xmath171 would be dominated by the environment temperature @xmath10 , unless @xmath223 . the latter condition could be achieved either by strong heating of particle 1 or by the cooling of the environment . and @xmath220 as a function of in - plane separation @xmath224 for @xmath225 , with cavity length @xmath172 @xmath165 m and frequency @xmath212 mev . the vertical axis is logarithmic and the geometry is illustrated in the inset . the dashed lines show the transmission functions in free space . ] figure [ fig : t12_b ] shows the transmission functions @xmath211 and @xmath220 as a function of in - plane separation @xmath224 of the sic nanoparticles . apart from the near - field , the interparticle transmission function @xmath221 is again visibly larger ( on the logarithmic scale ) in the cavity than in free space . the transmission function oscillates as a function of @xmath224 due to the interference of three propagating cavity modes . the oscillation can be shown to be an interference effect by evaluating @xmath221 in a narrow cavity , where only one of cavity modes can be excited at @xmath226 mev , and noting that the oscillations vanish ( not shown ) . the enhanced interparticle energy transfer demonstrated in figs . [ fig : t12_a](a ) and [ fig : t12_b ] could have applications , for example , in enhancing the spatial and temporal control of plasmonic heating in metal nanoparticle collections@xcite . to investigate how the surface phonon polaritons affect interparticle heat transfer close to a sic surface , we plot in fig . [ fig : t12_sic](a ) the transmission functions for two nanoparticles at a distance of @xmath169 @xmath165 m from a sic surface , as shown in the inset . the transmission functions are again evaluated at the energy @xmath212 mev . for @xmath227 @xmath165 m , the interparticle transmission function @xmath221 for particles close to a surface ( solid line ) is clearly enhanced compared to the free space value ( dashed line ) by the coupling to the surface phonon polaritons . similarly to the case of nanoparticles in a cavity , the enhanced interparticle transmission is in most cases negligible compared to the strong coupling @xmath208 to the environment radiation which also increases due to the coupling to surface phonon polaritons . ( blue or dark gray ) and @xmath220 ( red or light gray ) at @xmath212 mev as a function of in - plane nanoparticle separation @xmath224 , with ( solid ) and without ( dashed ) a sic surface . the distance from the surface is @xmath169 @xmath165 m . both interparticle transmission @xmath221 and the radiation coupling are strongly enhanced compared to the free - space value by the coupling to surface phonon polaritons . the geometry is illustrated in the inset . ( b ) the real and imaginary parts and the absolute value of the @xmath228-component of the interparticle green s dyadic @xmath229 as a function of @xmath224 . both real and imaginary parts oscillate at the surface phonon polariton wavelength , but the absolute value decays without oscillations . inset : the in - plane spatial fourier transform of the real and imaginary parts of the green s dyadic . the imaginary part has a peak at the surface phonon polariton wavevector @xmath230}$ ] ( vertical dashed line).,title="fig : " ] ( blue or dark gray ) and @xmath220 ( red or light gray ) at @xmath212 mev as a function of in - plane nanoparticle separation @xmath224 , with ( solid ) and without ( dashed ) a sic surface . the distance from the surface is @xmath169 @xmath165 m . both interparticle transmission @xmath221 and the radiation coupling are strongly enhanced compared to the free - space value by the coupling to surface phonon polaritons . the geometry is illustrated in the inset . ( b ) the real and imaginary parts and the absolute value of the @xmath228-component of the interparticle green s dyadic @xmath229 as a function of @xmath224 . both real and imaginary parts oscillate at the surface phonon polariton wavelength , but the absolute value decays without oscillations . inset : the in - plane spatial fourier transform of the real and imaginary parts of the green s dyadic . the imaginary part has a peak at the surface phonon polariton wavevector @xmath230}$ ] ( vertical dashed line).,title="fig : " ] in contrast to fig . [ fig : t12_b ] , the interparticle transmission does not oscillate as a function of particle separation . this can be understood by noting that since the phonon polaritons carrying the heat are excited at a single wavevector @xmath202 , no interference oscillations can be expected . to show explicitly that the interparticle heat transfer enhancement is mainly due to propagating surface phonon polaritons , the @xmath228-component of the scattering part of the interparticle green s dyadic @xmath229 is plotted in fig . [ fig : t12_sic](b ) . the @xmath231- and @xmath232-components of the green s dyadics are similar , and the @xmath228-component is chosen only for illustration . the phase of the green s dyadic can be seen to rotate at the wavelength @xmath233 @xmath165 m corresponding to the wavevector @xmath234 @xmath165m@xmath235 , which agrees with the phonon polariton in - plane wavevector @xmath236}\approx ( 0.83 + 0.033i)$ ] @xmath165m@xmath235 . the @xmath237 phase difference of the real and imaginary parts of green s dyadic physically means that the phase of the electric field at the location of the particle @xmath169 rotates as a function of separation @xmath224 from dipole @xmath171 . since the polaritons propagate freely on the surface , no interference oscillations can be seen either in the absolute value of the green s dyadic or in the transmission function . interference effects could be observed by suitably structuring the surface such that phonon polaritons could be reflected by the obstructions or by replacing the surface by a thin film supporting two branches of surface phonon polaritons @xcite . we developed a microscopic dipole oscillator model to describe electromagnetic heat transfer between an arbitrary number of objects each described as a single oscillating dipole or more generally as a collection of dipoles in the spirit of the discrete dipole approximation . the dynamics of each dipole was described by the langevin theory accounting for the dissipation and noise induced by the thermal fluctuations in the dipole moment . the formulation enables a physically insightful approach to model the energy transfer and results in heat transfer rates that are fully analogous with the landauer - bttiker formulas of electron and phonon transfer . we also expressed the energy transmission functions in terms of purely optical properties , i.e. the polarizabilities of the dipoles and the electromagnetic green s dyadic . we applied the formalism to calculate heat transfer rates between silicon carbide nanoparticles placed in a microcavity , showing that the interparticle heat transfer rate can be enhanced by orders of magnitude as compared to a similar nanoparticle configuration in free space . the coupling of the background thermal field is , however , also enhanced by the microcavity and typically dominates the net energy transfer rate of a nanoparticle with its surroundings . as a function of in - plane separation of two nanoparticles in a microcavity , the interparticle heat current also oscillates due to the interference of the excited cavity modes . the quantum langevin approach presented in this paper opens up new possibilities for modeling energy transfer and the coupling of different energy transfer mechanisms in nanostructures . for instance , coupling the presently studied system to phonon heat transfer model described by similar langevin equations of motion will enable a detailed description of energy transfer by a coupled photon and phonon system . further coupling of phonon and electron baths would capture inelastic energy transfer between the lattice vibrations and electrons , thereby providing a very thorough picture of energy transport and conversion processes in nanostructures , with numerous applications in the modeling of future energy technological devices . we thank o. heikkil for useful discussions especially on the calculation of the cavity green s dyadic . we acknowledge the aalto science - it project and finnish it center for science for computer time . the work is in part funded by the academy of finland and the aef research program of aalto university and by the graduate school in electronics , telecommunications and automation ( geta ) . in this appendix , we calculate the thermal average of the langevin bath current @xmath238 defined in eq . using the solution of the dipole displacements and the correlation functions and . for convenience , we define the reduced correlator @xmath239 , valid for any functions @xmath240 , @xmath241 , through the relation @xmath242 we separate the energy current @xmath238 into two terms as @xmath243 , where @xmath244 and @xmath245 . the rate of work done by the stochastic force @xmath26 on the dipole at site @xmath11 is @xmath246 { { \mathbf}{g}}_{ji}^t \hat\xi_i \right\rangle_{\omega } , \end{aligned}\ ] ] where we substituted the solution for the dipole dynamics . we also introduced the trace , which allows us to cycle the position of the term inside braces : @xmath247\\ & = i \int \frac{d\omega}{2\pi } \hbar \omega \textrm{tr } \left[{{\mathbf}{g}}_{ii } { ( \omega)}\gamma^{\textrm{bath}}_i{(\omega)}\right ] { \left[f_b(\omega , t_{i})+\frac{1}{2 } \right]}.\end{aligned}\ ] ] here we used eqs . and . since the green s function is a real function in time - domain , the real part of @xmath248 is an even function of frequency and since @xmath249 and @xmath250 $ ] are both odd functions of frequency , only the imaginary part of @xmath248 survives the integration . the same result would follow directly by considering the symmetrized function @xmath251 instead of @xmath252 , which would be necessary for heisenberg operators @xcite . we get @xmath253\gamma^{\textrm{bath}}_i{(\omega)}\right\ } { \left[f_b(\omega , t_{i})+\frac{1}{2 } \right]}.\end{aligned}\ ] ] now we apply the identity @xmath254 = -\frac{1}{2 } { \mathbf}{g}{(\omega)}\left[\gamma^{\textrm{bath}}{(\omega)}+\gamma^{\textrm{rad}}(\omega ) \right]{\mathbf}{g}{(\omega)}^{\dagger } , \label{eq : gammaidentity}\ ] ] where the block - diagonal matrices @xmath255 $ ] and @xmath256 $ ] were defined in sec . [ sec : equationofmotion ] . we get @xmath257 { \left[f_b(\omega , t_{i})+\frac{1}{2 } \right]}\notag \\ & \quad + \int_0^{\infty } \frac{d\omega}{2\pi } \hbar \omega \textrm{tr } \left\ { \left[{{\mathbf}{g}}{(\omega)}\gamma^{\textrm{rad } } { ( \omega)}{{\mathbf}{g}}(\omega)^{\dagger } \right]_{ii } \gamma^{\textrm{bath}}_i{(\omega)}\right\ } { \left[f_b(\omega , t_{i})+\frac{1}{2 } \right]}.\end{aligned}\ ] ] here we noted that the integrands are even functions in @xmath64 . the power absorbed by the bath due to the friction force is @xmath258 { { \mathbf}{g}}_{ji}^t { { \mathbf}{g}}_{ik } \left[\hat\xi_k+q{\hat{{\mathbf}{e}}_{\textrm{env}}}({{\mathbf}{r}}_k)\right ] \right\rangle_{\omega } \\ & = \int \frac{d\omega}{2\pi } m\gamma\omega^2 \sum_{j , k=1}^n \textrm{tr}\left\ { { { \mathbf}{g}}_{ji}{(\omega)}^t { { \mathbf}{g}}_{ik}{(\omega)}^ * \left\langle \left[\hat\xi_k+q{\hat{{\mathbf}{e}}_{\textrm{env}}}({{\mathbf}{r}}_k ) \right ] \left[\hat\xi_j^t+q{\hat{{\mathbf}{e}}_{\textrm{env}}}({{\mathbf}{r}}_j)^t\right ] \right\rangle_{\omega } \right\ } \\ & = \int \frac{d\omega}{2\pi } m\gamma \hbar \omega^2 \sum_{j=1}^n \textrm{tr}\left[{{\mathbf}{g}}_{ji}{(\omega)}^t { { \mathbf}{g}}_{ij}{(\omega)}^ * \gamma_j^{\textrm{bath}}{(\omega)}\right ] { \left[f_b(\omega , t_{j})+\frac{1}{2 } \right]}\notag \\ & \quad + \int \frac{d\omega}{2\pi } m\gamma \hbar \omega^2 \sum_{j , k=1}^n \textrm{tr}\left[{{\mathbf}{g}}_{ji}{(\omega)}^t { { \mathbf}{g}}_{ik}{(\omega)}^ * \gamma_{kj}^{\textrm{rad}}{(\omega)}\right ] { \left[f_b(\omega,{t_{\textrm{env}}})+\frac{1}{2 } \right]}. \\ & = \frac{1}{2 } \int \frac{d\omega}{2\pi } \hbar \omega \sum_{j=1}^n \textrm{tr}\left[\gamma_i^{\textrm{bath}}{(\omega)}{{\mathbf}{g}}_{ij}{(\omega)}^ * \gamma_j^{\textrm{bath}}{(\omega)}{{\mathbf}{g}}_{ji}{(\omega)}^t\right ] { \left[f_b(\omega , t_{j})+\frac{1}{2 } \right]}\notag \\ & \quad + \frac{1}{2 } \int \frac{d\omega}{2\pi } \hbar \omega \sum_{j , k=1}^n \textrm{tr}\left[\gamma_i^{\textrm{bath}}{(\omega)}{{\mathbf}{g}}_{ik}{(\omega)}^ * \gamma_{kj}^{\textrm{rad}}{(\omega)}{{\mathbf}{g}}_{ji}{(\omega)}^t \right ] { \left[f_b(\omega,{t_{\textrm{env}}})+\frac{1}{2 } \right]}. \end{aligned}\ ] ] taking the complex conjugate gives finally @xmath259 { \left[f_b(\omega , t_{j})+\frac{1}{2 } \right]}\notag \\ & \quad + \int_0^{\infty } \frac{d\omega}{2\pi } \hbar \omega \textrm{tr}\left\{\gamma_i^{\textrm{bath}}{(\omega)}[{{\mathbf}{g}}{(\omega)}\gamma^{\textrm{rad}}{(\omega)}{{\mathbf}{g}}{(\omega)}^{\dagger}]_{ii } \right\ } { \left[f_b(\omega,{t_{\textrm{env}}})+\frac{1}{2 } \right]}. \end{aligned}\ ] ] the second term is absent , if the environment field is absent . the total heat current flowing to the bath is @xmath260 , so we get eq . .

this paper investigates how multiuser dimensions can effectively be exploited for target degrees of freedom ( dof ) in interfering broadcast channels ( ibc ) consisting of-transmitters and their user groups . first , each transmitter is assumed to have a single antenna and serve a singe user in its user group where each user has receive antennas less than . in this case , a-transmitter single - input multiple - output ( simo ) interference channel ( ic ) is constituted after user selection . without help of multiuser diversity , interfering signals can not be perfectly removed at each user since the number of receive antennas is smaller than or equal to the number of interferers . only with proper user selection , non - zero dof per transmitter is achievable as the number of users increases . through geometric interpretation of interfering channels , we show that the multiuser dimensions have to be used first for reducing the dof loss caused by the interfering signals , and then have to be used for increasing the dof gain from its own signal . the sufficient number of users for the target dof is derived . we also discuss how the optimal strategy of exploiting multiuser diversity can be realized by practical user selection schemes . finally , the single transmit antenna case is extended to the multiple - input multiple - output ( mimo ) ibc where each transmitter with multiple antennas serves multiple users . multiuser diversity , degrees of freedom , interference alignment measure , interfering broadcast channel

1111.1051

interference is a major performance - limiting factor in modern wireless communication systems . many interference mitigation strategies have been proposed to improve network spectral efficiency . by allowing partial or full cooperation among interfering base stations , interference can effectively be managed and spectral efficiency can be improved . joint beamforming @xcite and network mimo ( or multicell processing ) @xcite among base stations have been shown to be effective interference mitigation techniques . however , if cooperation among transmitters is not allowed , orthogonal multiple access has been a traditional solution to interference . in a @xmath0-user single - input single - output ( siso ) interference channel ( ic ) , for example , each user can achieve @xmath2 degrees of freedom ( dof ) by time division multiple access . in recent years , interference alignment ( ia ) techniques have received much attention @xcite . the basic concept of ia is to align the interfering signals in a small dimensional subspace . in a @xmath0-user siso ic , @xmath3 dof have been shown to be achievable using ia @xcite . although ia provides a substantial asymptotic capacity gain in interference channels , there are many practical challenges for implementation . ia requires global channel state information at the transmitter ( csit ) , and imperfect channel knowledge severely degrades the gain of ia . in some channel configurations , symbols should be extended in the time / frequency domain to align interfering signals . the high computational complexity is also considered as a major challenge . to ameliorate these difficulties , many ia algorithms have been proposed such as iterative ia @xcite and a subspace ia @xcite . for interference suppression , multiuser diversity can also be exploited by opportunistic user selection for minimizing interference . the interference reduction by multiuser diversity can be enjoyed without heavy burden on global channel knowledge because user selection in general requires only a small amount of feedback @xcite . in this context , opportunistic interference alignment ( oia ) has been recently proposed in @xcite and has attracted much attention . in a 3-transmitter @xmath4 mimo interfering broadcast channel ( ibc ) , the authors of @xcite proved that @xmath5 ( where @xmath6 $ ] ) dof per transmitter is achievable when the number of users scales as @xmath7 . in @xcite and @xcite , a @xmath0-transmitter @xmath8 simo ibc and a @xmath0-transmitter @xmath9 simo ibc have been studied , respectively . for simo interfering multiple access channel ( imac ) constituted by @xmath0-cell uplink channels with @xmath10 transmit antennas and single antenna users , the authors of @xcite showed that @xmath11 dof are achievable when the number of users scales as @xmath12 . in these schemes , user dimensions are used to align the interfering signals ; each transmitter opportunistically selects a user whose interfering signals are most aligned among the users associated with the transmitter . contrary to the conventional opportunistic user selection techniques @xcite , the oia scheme exploits the multiuser dimensions for interference alignment . in this paper , we investigate the optimal role of multiuser diversity for the target dof in the ibc with @xmath0-transmitters and generalize the results of @xcite@xcite . for the @xmath0-transmitter simo ibc , each transmitter selects and serves a single user in its user group consisting of @xmath13 users . once after @xmath0-transmitters select their serving users , a @xmath0-user ic is constructed . each user has @xmath14 antennas less than or equal to the number of interferers , i.e. , @xmath15 . thus , without help of multiuser diversity , interference at each user can not be perfectly removed so that the achievable rate of each transmitter goes to zero as signal - to - noise ratio ( snr ) increases . consequently , the achievable dof per transmitter becomes zero . however , non - zero dof per transmitter is achievable by exploiting multiuser diversity as the number of users increases . since opportunistic user selection can focus on either enhancing the desired signal or decreasing interference , non - zero dof can be obtained by properly enhancing the desired signal strength and reducing interference via user selection . that is , the non - zero dof @xmath16 comprises a dof gain term @xmath17 from the desired signal and a dof loss term @xmath18 caused by interference such that @xmath19 , and the target dof @xmath16 can be obtained by a proper combination of @xmath20 and @xmath21 . however , many questions remain unsolved ; what is the feasible and optimal combination of @xmath22 for the target dof @xmath23 and what is the sufficient number of users for the target dof achieving strategy . we answer these fundamental questions and analytically investigate how the multiuser dimensions can be optimally exploited for the target dof in the ibc . specifically , from geometric interpretation of interfering channels , we define an interference alignment measure that indicates how well interference signals are aligned at each user . using the interference alignment measure , we first consider the @xmath0- transmitters simo ibc and show that the dof gain term @xmath20 can be achieved if the number of users scales in terms of transmit power @xmath24 as @xmath25 and the dof loss term can be reduced to @xmath21 if the number of users scales as @xmath26 . from these results , we find the optimal strategy of exploiting multiuser diversity for the target dof @xmath16 in terms of the required number of users ; the optimal target dof achieving strategies @xmath27 are @xmath28 and @xmath29 for the target dof @xmath30 $ ] and @xmath31 , respectively . we also investigate how the optimal target dof achieving strategy @xmath27 can be realized by practical user selection schemes . then , we extend our results to the @xmath0-transmitter mimo ibc where each transmitter has @xmath32 multiple antennas and serves multiple users with @xmath14 receive antenna each . our generalized key findings are summarized as follows : * for the target dof @xmath33 $ ] , @xmath34 is the optimal target dof achieving strategy that minimizes the required number of users . that is , the multiuser dimensions should be exploited to make the dof loss @xmath35 . the sufficient number of users for this strategy scales like @xmath36 . * for the target dof @xmath37 , @xmath38 is the optimal target dof achieving strategy which minimizes the required number of users . that is , the multiuser dimensions should be exploited to make the dof loss term zero as well as to make the dof gain term @xmath16 . the sufficient number of users for this strategy scales like @xmath39 . the rest of this paper is organized as follows . in section ii , we describe the system model . in section iii , a geometric interpretation of interfering channels is provided , and the interference alignment measure is defined . section iv derives the optimal strategies of achieving the target dof in terms of the required number of users . in section v , we show how various practical user selection schemes exploit multiuser diversity for the target dof and discuss their optimality to achieve the target dof . the system model is extended for the mimo ibc in section vi . numerical results are shown in section vii , and we conclude our paper in section viii . _ notations _ throughout the paper , we use boldface to denote vectors and matrices . the notations @xmath40 , @xmath41 , and @xmath42 denote the conjugate transpose , the @xmath43th largest eigenvalue , and the eigenvector of matrix @xmath44 corresponding to the @xmath43th largest eigenvalue . for convenience , the smallest eigenvalue , the largest eigenvalue , and the eigenvectors corresponding eigenvectors of @xmath44 are denoted as @xmath45 , @xmath46 , @xmath47 , and @xmath48 , respectively . also , @xmath49 , @xmath50 , and @xmath51 indicate the @xmath52 identity matrix , the @xmath53-dimensional complex space , and the set of @xmath54 complex matrices , respectively . our system model is depicted in fig . [ fig : system_model ] . the system corresponds to the interfering broadcast channel ( ibc ) of which capacity is unknown . there are @xmath0 transmitters with @xmath32 transmit antennas each , and each transmitter has its own user group consisting of @xmath13 users with @xmath14 antennas each . first , each transmitter is assumed to have a single antenna , i.e. , @xmath55 , and serves a single user selected in its user group so that @xmath0-transmitter simo ic is opportunistically constituted . the system model with multiple transmit antennas ( i.e. , @xmath56 ) becomes statistically identical with the single transmit antenna model if each transmitter uses a random precoding vector . in section [ sec : extension ] , we extend our system model to the @xmath0-transmitter mimo ibc where each transmitter with multiple antennas serves the multiple users through orthonormal random beams . in this paper , we focus on the cases that the number of receive antennas is smaller than the number of transmitters , i.e. , @xmath57 . otherwise ( i.e. , if @xmath58 ) , each user can suppress all interfering signals through zero - forcing like schemes so that dof one is trivially guaranteed at each transmitter . we also assume that collaboration or information sharing among the transmitters is not allowed . since the user selection at each transmitter is independent of the other transmitters , we only consider the achievable rate of the first transmitter without loss of generality . note that the average achievable rate per transmitter will be same if the configurations of the transmitters are identical . at the first transmitter , the received signal at the @xmath53th user denoted by @xmath59 is given by @xmath60 where @xmath61 is the vector channel from the @xmath62th transmitter to the @xmath53th user whose elements are independent and identically distributed ( i.i.d . ) circularly symmetric complex gaussian random variables with zero means and unit variance . also , @xmath63 is the transmitted signal using random gaussian codebook from the @xmath62th transmitter such that @xmath64 , where @xmath24 is the power budget at each transmitter . also , @xmath65 is a circularly symmetric complex gaussian noise with zero mean and an identity covariance matrix , i.e. , @xmath66 . assuming perfect channel estimation at each receiver , the channel state information @xmath67 is available at the @xmath53th user . the received signal is postprocessed at each user using multiple receive antennas . let @xmath68 be the postprocessing vector of the @xmath53th user such that @xmath69 . then , the received signal after postprocessing becomes @xmath70 to aid user selection at the transmitter , each user feeds one scalar value back to the transmitter . various user selection criteria and corresponding feedback information will be discussed in the following sections . since no information is shared among the transmitters , each transmitter independently selects a single user based on the collected information . let @xmath71 be the index of the selected user at the first transmitter . then , the average achievable rate of the first transmitter is given by @xmath72 we decompose @xmath73 into two terms @xmath74 and @xmath75 such that @xmath76 , which are given , respectively , by @xmath77 then , the achievable dof of the first transmitter becomes @xmath78 we call @xmath79 and @xmath80 as _ dof gain term _ and _ dof loss term _ , respectively . the achievable rate of each transmitter depends on the number of users because multiuser dimensions are exploited for a rate increase . when there are fixed number of users , the achievable rate of each transmitter will be saturated in the high snr region due to interferences because the number of receive antennas at each user is smaller than the number of total transmitters . consequently , the first transmitter can not obtain any dof , i.e. , @xmath81 in this case , both the dof gain term and the dof loss term become one , i.e. , @xmath82 on the other hand , when the transmit power is fixed , the achievable rate of the selected user can increase to infinity as the number of users increases , i.e. , @xmath83 then , how much dof can be achieved when both the number of users and the transmit power increase ? obviously , non - zero dof can be obtained by exploiting multiuser dimensions , and the achievable dof @xmath84 \label{eqn : opportunistic_dof}\end{aligned}\]]will depend on the increasing speeds of @xmath13 and @xmath24 . in this case , dof @xmath85 at the first transmitter comprises the dof gain term @xmath86 and the dof loss term @xmath87 such that @xmath88 , i.e. , @xmath89 we call @xmath90 as a _ target dof achieving strategy _ if @xmath91 for the target dof @xmath16 . since each strategy requires different user scaling , we need to find the optimal dof achieving strategy that exploits multiuser diversity most efficiently , i.e. , which requires the minimum user scaling . for the _ target dof _ per transmitter @xmath92 , we find the optimal target dof achieving strategy @xmath27 satisfying @xmath93 and derive the required user scaling . note that the definition of dof in this paper is extended from the conventional definition of dof in order to properly capture multiuser diversity gain in terms of achievable rate . achievable dof defined in depends on increasing speeds of @xmath13 and @xmath24 ; and can have non - zero values even larger than one if the number of users properly scales with the transmit power . from the definitions of the rate gain term and the rate loss term given in and , respectively , the strategies which achieve the target dof @xmath16 are given by @xmath94 the following lemma shows that we do not need to consider all of the candidate strategies in but take into account only a subset of to find the optimal target dof achieving strategy . [ lemma : dof_terms_range ] for any non - negative target dof , the optimal dof achieving strategy is in the set @xmath95,~d_1-d_2=d \big\}. \label{eqn : dof_term_range}\end{aligned}\ ] ] at each channel realization , the achievable dof has the form of @xmath96 where @xmath97 and @xmath98 are its own signal power and the interfering signal power at the selected user , respectively . since the function is an increasing function of @xmath99 and a decreasing function of @xmath100 , for an increase of , the multiuser dimension should be used for increasing @xmath99 , for decreasing @xmath100 , or mixture of them . this fact results in . lemma 1 provides a basic guideline of using the multiuser dimension ; multiuser diversity should not be used for either decreasing dof gain term or increasing dof loss term . since the optimal target dof achieving strategy is obtained in the reduced set of candidate strategies , we consider the dof gain term larger than one and dof loss term smaller than one , i.e. , @xmath101 and @xmath102 $ ] , in the latter parts of this paper . in our system model , each user suffers from @xmath1 interfering channels which is larger than or equal to the number of receive antennas , i.e. , @xmath103 . since the interfering channels are isotropic and independent of each other , they span @xmath14-dimensional space . thus , the whole signal space at the receiver is corrupted by interfering signals , and hence the dof loss term becomes one if no effort is made to align interfering signals . on the other hand , the dof loss can be reduced by aligning interfering signals in smaller dimensional subspace . for example , if the interfering signals are perfectly aligned in @xmath104-dimensional subspace , they can be nullified by postprocessing so that we can make the dof loss zero . the transmitter can exploit the multiuser dimensions to align interfering signals by simply selecting a user whose interfering channels are most aligned . thus , each user needs to measure how much the interfering channels are aligned in @xmath104-dimensional subspace at the receiver . we call this measure as the _ interference alignment measure_. in this section , we geometrically interpret the interfering channels and define the interference alignment measure at each user . the interference alignment measure will be used for computing the reducible dof loss via multiuser diversity in section iv . let @xmath105 be the surface of the @xmath14-dimensional unit hypersphere centered at the origin , i.e. , @xmath106 for an arbitrary unit vector @xmath107 and an arbitrary non - negative real number @xmath108 , we can divide @xmath105 into two parts , @xmath109 and @xmath110 , given by @xmath111 when @xmath112 , two parts @xmath113 and @xmath114 are represented in fig . [ fig : sphere_s1s2 ] . let @xmath115 be the surface area of @xmath116 for @xmath117 . the surface area of an @xmath14-dimensional complex unit hypersphere is given by @xmath118 , and it was shown that @xcite @xmath119 which is invariant with @xmath120 . therefore , we obtain @xmath121 from the relationship @xmath122 . from this fact , we obtain the following lemma . [ lemma : p_n_lambda ] let @xmath123 be independent and isotropic unit vectors in @xmath124 . for an arbitrary unit vector @xmath125 and @xmath126 $ ] , the probability that @xmath127 contains @xmath128 becomes @xmath129 = \left(1-(1-\lambda)^{n_r-1}\right)^m , \label{eqn : prob_m_lambda}\end{aligned}\ ] ] which is invariant with @xmath120 . from the ratio of @xmath130 and @xmath131 , we obtain @xmath132 = \frac{a(s_2(\mathbf{c } , \lambda))}{a(s_0 ) } = 1-(1-\lambda)^{n_r-1 } , \quad \forall i.\end{aligned}\ ] ] since @xmath123 are independent of each other , it is satisfied that @xmath133 & = \pr[\mathbf{g}_1 \in s_2 ( \mathbf{c } , \lambda)]^m , { \nonumber}\end{aligned}\ ] ] which is given in . in this subsection , we define the interference alignment measure at each user . the dof loss is determined by how much the interfering channels are closely aligned in @xmath104-dimensional subspace . only if interfering channels are perfectly aligned in @xmath104-dimensional subspace , we can have zero dof loss . the interference alignment measure is used for computing the dof loss at each user . let @xmath134 be the @xmath135 normalized interfering channels at a user and @xmath136 be the interference alignment measure among them . consider the following optimization problem : @xmath137.{\nonumber}\end{aligned}\ ] ] from the definition of @xmath138 given in , this problem is equivalent to @xmath139,{\nonumber}\end{aligned}\ ] ] which can be solved by linear programming @xcite . let @xmath140 be the solution of the above problem . then , @xmath141 has the smallest surface area among all @xmath142 containing @xmath134 . using @xmath143 , we can divide an @xmath14-dimensional space into two subspaces which are the one - dimensional subspace spanned by @xmath143 and the @xmath104-dimensional complementary subspace denoted by @xmath144 . if there exists @xmath143 such that @xmath145 , it is satisfied that @xmath146 and @xmath147 , and hence @xmath148 becomes zero . in this case , we can say that the interfering channels are perfectly aligned in @xmath104-dimensional subspace in @xmath124 . note that @xmath149 is an @xmath104-dimensional subspace orthogonal to @xmath150 , and @xmath151 is the @xmath14-dimensional complex hypersphere , @xmath105 . when @xmath148 is the smaller , the vectors are the more aligned in the @xmath104-dimensional subspace , @xmath144 . thus , we will use @xmath152 as an _ interference alignment measure _ to quantify how much the interfering channels are closely aligned in an @xmath104-dimensional subspace , i.e. , @xmath153).\label{eqn : lambda*}\end{aligned}\ ] ] in other words , we use the mini - max distance of the interfering channels from an @xmath104-dimensional subspace . in fig . [ fig : iam_concept ] , the interference alignment measure is geometrically represented . the more the interfering channels are aligned , the smaller the interference alignment measure becomes . since the interference alignment measure is obtained from the optimization problem , the exact distribution is difficult to find . instead , we obtain the lower bound for the cumulative distribution function ( cdf ) of the interference alignment measure in the following lemma . [ lemma : prob_bar_d_m ] when @xmath154 , the probability that the interference alignment measure @xmath136 is smaller than @xmath155 $ ] is lower bounded on @xmath156 } \ge \left(1-(1-\lambda)^{n_r-1}\right)^{k - n_r}. \label{eqn : am1_prob}\end{aligned}\ ] ] we consider two events : @xmath157 where @xmath158 is the @xmath14-dimensional unit vector such that @xmath159 . by the definition of the interference alignment measure given in , @xmath160 is true whenever @xmath161 is true , equivalently , @xmath162 } \ge { \mathrm{pr}\left[\mathrm{(e2)}\right]}$ ] . the probability of @xmath161 is obtained by @xmath163 } & = { \mathrm{pr}\left[s_2(\bar{\mathbf{c } } , \lambda ) \supset \{\mathbf{g}_1 , \ldots , \mathbf{g}_{k-1}\}\right]}{\nonumber\\}&\stackrel{(a)}{= } { \mathrm{pr}\left[s_2(\bar{\mathbf{c } } , \lambda ) \supset \{\mathbf{g}_{n_r } , \ldots , \mathbf{g}_{k-1}\}\right]}{\nonumber\\}&\stackrel{(b)}{= } \left(1-(1-\lambda)^{n_r-1}\right)^{k - n_r},\end{aligned}\ ] ] where the equality @xmath164 is from the definition of @xmath158 such that @xmath165 . also , the equality @xmath166 holds from lemma [ lemma : p_n_lambda ] and from the fact that @xmath158 is independent of @xmath167 . thus , we obtain @xmath168 } \ge \left(1-(1-\lambda)^{n_r-1}\right)^{k - n_r}.\end{aligned}\ ] ] the remaining question is how much we can reduce the interference alignment measure via user selection . in the first user group , the @xmath53th user has @xmath1 interfering channels , @xmath169 . the interference alignment measure at the @xmath53th user can be written by @xmath170 where @xmath171 is the normalized interfering channel , i.e. , @xmath172 . thus , the achievable smallest interference alignment measure via user selection is given by @xmath173 obviously , the smallest interference alignment measure will decrease as the number of users increases . in the following lemma , we find the relationship between and the number of total users ( i.e. , @xmath13 ) . [ lemma : lambda_n1 ] when there are @xmath13 users , the expectation of the smallest interference alignment measure is upper bounded on @xmath174 < n^{-\frac{1}{k - n_r}}.\label{eqn : am1_bound}\end{aligned}\ ] ] the complementary cdf of is bounded on @xmath175 } & \qquad={\mathrm{pr}\left[\mathfrak{q } \big ( \tilde{\mathbf{h } } _ { n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k}\big ) \ge \lambda \textrm{~for all~ } n\right]}{\nonumber\\}&\qquad=\prod_{n=1}^n{\mathrm{pr}\left [ \mathfrak{q } \big ( \tilde{\mathbf{h } } _ { n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k}\big ) \ge \lambda\right]}{\nonumber\\}&\qquad=\left(1-{\mathrm{pr}\left [ \mathfrak{q } \big ( \tilde{\mathbf{h } } _ { n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k}\big ) \le \lambda\right]}\right)^n{\nonumber\\}&\qquad\stackrel{(a ) } { < } \big[1- ( 1-(1 - \lambda)^{n_r-1})^{k - n_r}\big]^n , \label{eqn : prob_bound}\end{aligned}\ ] ] where @xmath176 $ ] , and the inequality @xmath164 holds from lemma [ lemma : prob_bar_d_m ] . using this bound , we obtain as @xmath177 & = \int_0 ^ 1 { \mathrm{pr}\left[\min_n \mathfrak{q } \big ( \tilde{\mathbf{h } } _ { n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k}\big)\ge \lambda\right ] } d \lambda { \nonumber\\}&\le \int_0 ^ 1 \big[1- ( 1 - ( 1- \lambda ) ^{n_r-1})^{k - n_r}\big]^n d \lambda{\nonumber\\}&\stackrel{(a)}{\le } \int_0 ^ 1 \big[1- ( 1 - ( 1- \lambda))^{k - n_r}\big]^n d \lambda{\nonumber\\}&\stackrel{(b)}{= } \frac{1}{k - n_r}\beta\left(\frac{1}{k - n_r } , n+1\right){\nonumber\\}&\stackrel{(c)}{= } \frac{\gamma\left(1+\frac{1}{k - n_r}\right)\gamma(n+1 ) } { \gamma\left(n+ 1 + \frac{1}{k - n_r}\right ) } { \nonumber\\}&\stackrel{(d ) } { < } n^{-\frac{1}{k - n_r}},{\nonumber}\end{aligned}\ ] ] where the inequality @xmath164 is due to @xmath178 for @xmath108 , and the equality @xmath166 holds from the representation of beta function @xcite @xmath179 the equality @xmath180 comes from the definition of the beta function @xmath181 and the property of the gamma function @xmath182 . in the right - hand - side of the equality @xmath180 , it holds @xmath183 because @xmath184 for @xmath185 . also , it is satisfied that @xmath186 where @xmath187 is from the gautschi s inequality @xcite given by @xmath188 with @xmath189 and @xmath190 . thus , the inequality @xmath191 holds . in this section , we derive the optimal strategies of exploiting multiuser diversity for the target dof @xmath16 . we first decompose the target dof @xmath16 into the dof gain term @xmath20 and the dof loss term @xmath21 such that @xmath192 , and find the required user scalings for @xmath20 and @xmath21 , respectively . then , the optimal target dof achieving strategy is derived by determining the optimal combination @xmath193 which requires the minimum user scaling for the target dof @xmath16 . in this subsection , we find the required user scaling to reduce the dof loss . via user selection , the rate loss term given in can be minimized by @xmath194 . \label{eqn : rloss_min}\end{aligned}\ ] ] this value is upper bounded on @xmath194 & \stackrel{(a)}{= } \mathbb{e}_{\vert\mathbf{h}\vert , \tilde{\mathbf{h } } } \left[\min_{n,\mathbf{v}_n } ~\log_2\left(1 + p \sum_{k=2}^{k } \vert\mathbf{h}_{n , k}\vert^2\vert \mathbf{v}_{n}^\dagger \tilde{\mathbf{h}}_{n , k } \vert^2\right ) \right]{\nonumber\\}&\stackrel{(b)}{\le } \mathbb{e}_{\tilde{\mathbf{h } } } \left[\min_{n,\mathbf{v}_n } ~\mathbb{e}_{\vert\mathbf{h}\vert}\log_2\left(1 + p \sum_{k=2}^{k } \vert\mathbf{h}_{n , k}\vert^2\vert \mathbf{v}_{n}^\dagger \tilde{\mathbf{h}}_{n , k } \vert^2\right ) \right]{\nonumber\\}&\stackrel{(c)}{\le } \mathbb{e}_{\tilde{\mathbf{h } } } \left[\min_{n,\mathbf{v}_n } ~\log_2\left(1 + n_rp \sum_{k=2}^{k } \vert \mathbf{v}_{n}^\dagger \tilde{\mathbf{h}}_{n , k } \vert^2\right ) \right]{\nonumber\\}&\stackrel{(d)}{\le } \mathbb{e}_{\tilde{\mathbf{h } } } \left[\min_{n } ~\log_2\left(1 + n_rp ( k-1)\mathfrak{q}(\tilde{\mathbf{h}}_{n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k } ) \right)\right]{\nonumber\\}&\stackrel{(e)}{\le } ~\log_2\left(1 + n_rp ( k-1 ) \mathbb{e}_{\tilde{\mathbf{h}}}\left [ \min_{n } \mathfrak{q}(\tilde{\mathbf{h}}_{n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k } ) \right]\right){\nonumber\\}&\stackrel{(f)}{\le } ~\log_2\left(1 + n_rp ( k-1 ) n^{-\frac{1}{k - n_r}}\right ) , \label{eqn : rloss_bound}\end{aligned}\ ] ] where the equality @xmath164 is obtained by decomposing the channel vector into direction and magnitude independent of each other such that @xmath195 . the inequality @xmath166 holds because the minimum of the average is larger than the average of the minimum . the inequality @xmath180 is from the jensen s inequality and @xmath196 . also , the inequality @xmath191 holds from the fact that @xmath197 & \le \min_{\mathbf{v}_n } \left[(k-1 ) \max_{2\le k\le k } \vert \mathbf{v}_n^\dagger \tilde{\mathbf{h}}_{n , k}\vert^2 \right]{\nonumber\\}&= ( k-1 ) \mathfrak{q } ( \tilde{\mathbf{h}}_{n,2 } , \ldots , \tilde{\mathbf{h}}_{n , k}),\end{aligned}\ ] ] where @xmath198 is the interference alignment measure at the user @xmath53 given in . the inequality @xmath187 is from the jensen s inequality , and the inequality @xmath199 holds from lemma [ lemma : lambda_n1 ] . we obtain the following theorem . [ theorem : n_scaling_dof_loss ] we can obtain the dof loss term @xmath200 $ ] when the number of users in each group scales as @xmath201 to obtain the dof loss term @xmath21 , it is enough to make satisfying @xmath202 which is achieved if @xmath203 . a tighter upper bound of the rate loss term than could exist , but the derived upper bound in enables us to compare the increasing speeds of the transmit power and the required number of users , which is the crucial factor of dof calculation . the scaling law of the required number of users obtained from , which is derived in theorem [ theorem : n_scaling_dof_loss ] , is enough to find the optimal target dof achieving strategy as shown in section [ sec : dof_achieving_strategy ] . we also find the required user scaling to increase the dof gain term . from the definition of the rate gain term given in , the maximum rate gain term obtained by user selection is @xmath204 . \label{eqn : rgain_max}\end{aligned}\ ] ] this value is lower bounded on @xmath205 & \ge\mathbb{e}\left [ \max_{n , \mathbf{v}_n } \log_2\left(1 + p \vert \mathbf{v}_{n}^\dagger { \mathbf{h}_{n,1 } } \vert^2\right)\right]{\nonumber\\}&= \mathbb{e}\left [ \max_{n } \log_2\left(1 + p \vert { \mathbf{h}_{n,1 } } \vert^2\right)\right],\label{eqn : rgain_lower_bound}\end{aligned}\ ] ] and upper bounded on @xmath205 & \le \mathbb{e}\left [ \max_{n } \log_2\left(1 + p \sum_{k=1}^{k } \vert { \mathbf{h}_{n , k } } \vert^2\right)\right]{\nonumber\\}&\le \mathbb{e}\left [ \max_{n } \log_2\left(1 + pk \max_k \vert { \mathbf{h}_{n , k } } \vert^2\right)\right].\label{eqn : rgain_upper_bound}\end{aligned}\ ] ] since all @xmath206 are i.i.d . @xmath207 random variables , for sufficiently large @xmath13 , the bounds and acts like @xcite @xmath208 & \sim \log_2(1+p\log n ) { \nonumber\\}\mathbb{e}\left [ \max_{n , k } \log_2\left(1 + pk \vert { \mathbf{h}_{n , k } } \vert^2\right)\right ] & \sim \log_2(1+pk\log ( kn)).{\nonumber}\end{aligned}\ ] ] thus , when both @xmath13 and @xmath24 are large enough , act like @xmath209 , i.e , @xmath210 \sim \log_2(p\log n ) . \label{eqn : rgain_asymptotic}\end{aligned}\ ] ] therefore , we establish the following theorem . [ theorem : n_scaling_dof_gain ] the dof gain term @xmath211 is achievable when the number of users in each group scales as @xmath212 we use . by setting @xmath213 we obtain the required user scaling for the dof gain term @xmath20 given by @xmath25 . in theorem [ theorem : n_scaling_dof_loss ] and theorem [ theorem : n_scaling_dof_gain ] , we found the required user scalings for the dof loss term @xmath21 and the dof gain term @xmath20 , respectively . in this subsection , we find the optimal target dof achieving strategy which requires the minimum user scaling . we start with the following theorem . [ theorem : scaling1 ] for the target dof up to one , the whole multiuser dimensions should be devoted to minimizing the dof loss caused by interfering signals . the optimal dof achieving strategy for the target dof @xmath214 $ ] is @xmath215 , and the corresponding sufficient user scaling is @xmath216 in theorem [ theorem : n_scaling_dof_loss ] , we have shown that the target dof @xmath217 $ ] is achievable by reducing the dof loss term with the user scaling @xmath218 . on the other hand , this user scaling can not increase the dof gain term . substituting @xmath218 into the dof gain term @xmath20 becomes @xmath219 which is the same as when there is a fixed number of users as described in . that is , any other combinations @xmath220 which achieve the target dof @xmath16 requires larger user scaling than @xmath218 , where @xmath221 since @xmath222 . therefore , the optimal target dof achieving strategy is given by @xmath215 , and the sufficient user scaling is @xmath218 . now , we derive the target dof achieving strategy when the target dof @xmath16 is greater than one . to find the optimal dof achieving strategy , we firstly find the sufficient user scaling for an arbitrary strategy @xmath90 achieving dof @xmath223 . then , we show that the optimal target dof achieving strategy for the target dof @xmath224 is @xmath225 . [ lemma : general_scaling_d>1 ] for the target dof @xmath226 , the sufficient user scaling for an arbitrary strategy @xmath90 achieving dof @xmath227 is given by @xmath228 where @xmath222 and @xmath102 $ ] . as a target dof achieving scheme , we consider a two - stage user selection scheme ; the first stage is to increase the dof gain term , and the second stage is to decrease the dof loss term . the considered two stage user selection strategy is illustrated in fig . [ fig : user_selection_2stage ] . we randomly divide total @xmath13 users into @xmath229 subgroups having @xmath230 users each such that @xmath231 . then , the user selection in each stage is performed as follows . * stage 1 : in each subgroup , a single user having the largest channel gain is selected among @xmath230 users . as a result , we have @xmath229 selected users after stage 1 . * stage 2 : among the @xmath229 users , the transmitter selects a single user to minimize the dof loss term . in stage 1 , the dof gain term @xmath20 is obtained at each selected user when @xmath232 as stated in theorem [ theorem : n_scaling_dof_gain ] . in stage 2 , we can make the dof loss term @xmath21 when @xmath233 as shown in theorem [ theorem : n_scaling_dof_loss ] . thus , the target dof @xmath226 with the strategy @xmath90 such that @xmath192 can be obtained by the user scaling @xmath234 , which is given in . from lemma [ lemma : general_scaling_d>1 ] , we obtain the optimal dof achieving strategy for the target dof @xmath235 in following theorem . [ theorem : scaling2 ] the optimal target dof achieving strategy for @xmath236 is to increase the dof gain term to @xmath16 and to perfectly eliminate the dof loss , i.e. , @xmath237 . consequently , the sufficient user scaling for target dof @xmath226 becomes @xmath238 the proof is similar to that of theorem [ theorem : scaling1 ] . from lemma [ lemma : general_scaling_d>1 ] , we can obtain the target dof @xmath226 by the strategy @xmath239 with the sufficient user scaling given in . however , this scaling can not increase the dof gain term larger than @xmath16 even when the user scaling is only used to increase the dof gain term . substituting into , we still have @xmath240 this implicates that the user scaling given in is sufficient for the strategy @xmath239 but not enough for other strategies @xmath241 as well as @xmath242 which requires the higher user scaling than that of @xmath241 , where @xmath243 $ ] . therefore , the optimal strategy for the target dof @xmath226 becomes @xmath244 . in fig . [ fig : dof_achieving_strategy ] , the optimal dof achieving strategy @xmath27 is plotted according to the target dof @xmath23 . in this section , we discuss how the optimal target dof achieving strategy can be realized by practical user selection schemes . the practical schemes considered in this section require no cooperation and no information exchange among the transmitters . for practical scenarios , we assume that each user has knowledge of channel state information ( csi ) of the direct channel and the covariance matrix of the received signal without explicit knowledge of csi of the interfering channels . that is , the @xmath53th user knows csi of its own desired channel @xmath245 and the covariance matrix of the received signal @xmath246 = \mathbf{i}_{n_r } + p\sum_{k=1}^k \mathbf{h}_{n , k } \mathbf{h}_{n , k } ^\dagger$ ] . from these values , the user @xmath53 easily obtains the interference covariance matrix denoted by @xmath247 such as @xmath248 - p\mathbf{h}_{n,1}\mathbf{h}_{n,1}^\dagger -\mathbf{i}_{n_r}.\end{aligned}\ ] ] therefore , the achievable rate at the first transmitter given in can be rewritten by @xmath249 to increase @xmath73 , various user selection schemes can be considered , but we focus on several popular techniques in the following subsections to maximize the postprocessed snr @xmath250 , to minimize the postprocessed inr @xmath251 , and to maximize the postprocessed sinr @xmath252 . in the max - snr user selection scheme , each user maximizes the postprocessed snr , and the transmitter selects the user having the maximum postprocessed snr . consequently , the postprocessed snr at the selected user becomes @xmath253 \stackrel{(a)}{= } \max_n p\vert { \mathbf{h}_{n,1}}\vert^2,\end{aligned}\ ] ] where the equality @xmath164 holds when the @xmath53th user adopts the postprocessing vector @xmath254 . thus , the selected user denoted by @xmath255 becomes @xmath256 and the desired channel gain at each user ( @xmath257 for the user @xmath53 ) should be informed to the transmitter . in the min - inr user selection scheme , each user minimizes the postprocessed inr , and the transmitter selects the user having the minimum postprocessed inr . thus , the postprocessed inr at the selected user becomes @xmath263 \stackrel{(a)}{= } \min_n \left[\lambda _ { \min } \left(\mathbf{r}_{n}\right)\right],\end{aligned}\ ] ] where the equality @xmath164 is obtained by the postprocessing vector of the @xmath53th user @xmath264 the required feedback information from the @xmath53th user is @xmath265 , and index of the selected user denoted by @xmath266 becomes @xmath267.\end{aligned}\ ] ] note that this scheme minimizes the rate loss term defined in . using the min - inr scheme , the transmitter can decrease the dof loss term while the dof gain term remains to be one . therefore , the min - inr scheme realizes the optimal target dof achieving strategy @xmath268 for the target dof @xmath30 $ ] . the required number of users by the min - inr scheme for the target dof @xmath30 $ ] scales like @xmath269 which is the required user scaling of the optimal target dof achieving strategy when the target dof is @xmath270 $ ] as shown in theorem [ theorem : scaling1 ] . the max - sinr user selection scheme is known to maximize the achievable rate at the transmitter although it requires additional complexity for postprocessing at the receivers . the achievable rate by the max - sinr scheme denoted by @xmath271 becomes @xmath272 . \label{eqn : r_sinr}\end{aligned}\ ] ] at each channel realization , the postprocessed sinr at the selected user is given by @xmath273 . \label{eqn : max_sinr}\end{aligned}\ ] ] to maximize the postprocessed sinr , the @xmath53th user adopts the postprocessing vector given by @xmath274 which is identical with the mmse - irc in @xcite . the corresponding postprocessed sinr at user @xmath53 becomes @xmath275 @xcite , and hence the selected user at the transmitter denoted by @xmath276 is given by @xmath277 [ lemma : scaling1 ] to obtain the target dof @xmath30 $ ] , the required user scaling of the max - sinr scheme is exactly the same as that of the min - inr scheme . from the fact that @xmath278 we obtain @xmath279 where the equality @xmath164 is because @xmath280 as shown in the proof of theorem [ theorem : scaling1 ] . therefore , the required user scaling for @xmath281 is exactly the same as the required user scaling for @xmath282 , equivalently , for @xmath283 . lemma [ lemma : scaling1 ] indicates that the max - sinr scheme realizes the optimal dof achieving strategy @xmath284 for the target dof @xmath30 $ ] . [ lemma : scaling2 ] the max - sinr scheme realizes the dof achieving strategy @xmath285 whenever the target dof @xmath16 is greater than 1 . since the max - sinr scheme is the optimal user selection scheme , it achieves dof @xmath226 with the user scaling @xmath286 as stated in theorem [ theorem : scaling2 ] . from the definition of , we obtain @xmath287 , and hence we have @xmath288 . as shown in , the sufficient user scaling for the max - sinr scheme to obtain the target dof @xmath16 can not increase the dof gain term larger than @xmath16 even if the whole user scaling is only devoted to increasing the dof gain term . this implicates that when we obtain the target dof @xmath226 by the max - sinr scheme with the user scaling @xmath286 , we obtain @xmath289 and have the dof gain @xmath16 at most ( i.e. , @xmath290 ) . therefore , the max - sinr scheme can only have @xmath291 if @xmath292 . for the target dof @xmath226 , the two - stage user selection scheme described in the proof of lemma [ lemma : general_scaling_d>1 ] can be adopted . more specifically , the transmitter selects the users by the max - snr scheme in the first stage . then , in the second stage , the transmitter selects a single user by the min - inr scheme or the max - sinr scheme . as shown in the proof of lemma [ lemma : general_scaling_d>1 ] , the two - stage user selection scheme can realize the optimal target dof achieving strategy for the target dof @xmath226 . in this section , we extend our system to interfering mimo bc cases . more specifically , each transmitter with @xmath32 antennas sends @xmath32 independent streams over @xmath32 orthonormal random beams using equal power allocation . similar to the user selection procedure in @xcite , each transmitter broadcasts @xmath32 orthonormal random beams , and each user feeds @xmath32 scalar values corresponding to all beams back to the transmitter . the feedback information corresponding to each stream such as snr , inr , and sinr can be easily found in a similar way to the simo case . a single user is selected for each beam , but the same user can be selected for different beams . however , it rarely occurs that multiple streams are transmitted for a single user as the number of users increases . when multiple streams are transmitted for a single user , the user is assumed to decode each stream treating the other streams as interferences . we denote the orthonormal random beams by @xmath293 which satisfies that @xmath294 and @xmath295 for all @xmath296 . we start from the following remark . let @xmath297 be the channel matrix whose elements are i.i.d . gaussian random variables . then , for an arbitrary unitary matrix @xmath298 ( i.e. , @xmath299 ) , the distributions of @xmath300 and @xmath301 are identical . since the @xmath14 columns of @xmath300 are independent and isotropic random vectors in @xmath124 , so are the @xmath14 columns of @xmath302 . owing to remark 1 , the @xmath0-transmitter mimo interfering bc is statistically identical with the @xmath303-transmitter simo interfering bc . let @xmath304 be the channel matrix from the @xmath62th transmitter to the @xmath53th user in the first user group . since the random beams satisfy that @xmath305^\dagger [ \mathbf{u}_1 , \ldots , \mathbf{u}_{n_t } ] = [ \mathbf{u}_1 , \ldots , \mathbf{u}_{n_t } ] [ \mathbf{u}_1 , \ldots , \mathbf{u}_{n_t}]^\dagger = \mathbf{i}_{n_t}$ ] , the user @xmath53 in the first user group has @xmath303 independent and isotropic channel vectors @xmath306 ~~ i\in[n_t],{\nonumber}\end{aligned}\ ] ] formed by the random beams and channel matrices from all transmitters . if the @xmath53th user in the first group is served by the @xmath43th random beam , the user has desired channel @xmath307 and the @xmath308 interfering channels , which correspond to @xmath309 inter - stream interfering channels @xmath310 and @xmath311 inter - transmitter interfering channels @xmath312 } \{\mathbf{h}_{n , k}\mathbf{u}_j\}_{k=2}^k.{\nonumber}\end{aligned}\ ] ] consequently , each random beam can be regarded as a single antenna transmitter with the transmit power @xmath313 . this fact leads to the following theorems . [ theorem : mimo_bc ] in a mimo bc where a transmitter with @xmath32 antennas supports @xmath32 users among @xmath13 users with @xmath314 antennas each , the optimal dof achieving strategy for the target dof @xmath315)$ ] is @xmath316 and requires the number of users to scale as @xmath317 . for the target dof @xmath318 , the optimal dof achieving strategy is @xmath239 and requires the number of users to scale as @xmath319 . the dof gain term @xmath320 is obtained when each stream achieves dof gain term @xmath321 . thus , with the same procedure given in section [ sec : dof_gain_scaling ] , we can easily show that the dof gain term @xmath320 is obtained when @xmath322 . on the other hand , the dof loss term @xmath323 is obtained when each stream achieves dof loss term per stream @xmath324 . as stated earlier , using @xmath32 orthonormal random beams at the transmitter each of which independently supports a single user , the mimo bc can be translated into an @xmath32-transmitter simo ic where each transmitter supports one of @xmath13 users with @xmath14 antennas . in this case , dof loss term @xmath323 , i.e. , dof loss @xmath325 per stream , is obtained when @xmath326 . therefore , we can conclude that the optimal dof achieving strategy for the target dof @xmath315)$ ] is @xmath327 and requires the number of users to scale as @xmath328 . also , for the target dof @xmath318 , the optimal dof achieving strategy is @xmath239 and requires the number of users to scale as @xmath329 . [ theorem : mimo_ibc ] consider a @xmath0-transmitter interfering mimo bc where the @xmath62th transmitter with @xmath330 antennas supports @xmath330 users among @xmath331 users with @xmath332 antennas each . at the @xmath62th transmitter , the optimal dof achieving strategy for the target dof @xmath333)$ ] is @xmath334 and requires the number of users to scale as @xmath335 . for the target dof @xmath37 , the optimal dof achieving strategy becomes @xmath239 and requires the number of users to scale as @xmath336 . the proof is similar to that of theorem [ theorem : mimo_bc ] . the @xmath62th transmitter obtains dof gain term @xmath337 when each stream obtains dof gain term @xmath338 , and the required user scaling is exactly given by @xmath339 . on the other hand , the @xmath62th transmitter obtains @xmath340 when the dof loss term per stream becomes @xmath341 . using @xmath330 orthonormal random beams at each transmitter each of which independently supports a single user , the interfering mimo bc can be translated into an @xmath342-transmitter simo ic where each transmitter supports a single user among @xmath331 users with @xmath343 antennas . thus , the @xmath62th transmitter obtains the dof loss term @xmath340 when @xmath344 . therefore , we can conclude that the optimal dof achieving strategy of the @xmath62th transmitter for the target dof @xmath333)$ ] is @xmath334 and obtained when the number of users scales as @xmath345 . also , for the target dof @xmath346 , the optimal dof achieving strategy is @xmath239 and requires the number of users to scale as @xmath336 . in this section , we first compare achievable rates of the practical user selection schemes for given number of users . then , we check if the target dof can be achievable with increasing number of users by showing achievable rates per transmitter for the practical user selection schemes . we have also considered two time division multiple access ( tdma ) schemes . in the first tdma scheme ( tdma1 ) , a single transmitter operates at each time so that @xmath2 dof is achieved at each transmitter . in the second tdma scheme ( tdma2 ) , only @xmath14 of @xmath0 transmitters operate at each time so that @xmath347 dof is achieved at each transmitter . [ fig : oia_k4_a10_n10 ] shows the achievable rates of each transmitter for various user selection schemes in ibc when there are 4 transmitters and each transmitter has 10 users with three receive antennas each . it is confirmed that the achievable rates are saturated in the high snr region and the achievable dof per transmitter becomes zero for the fixed number of users . now , we show that the target dof can be achievable if the number of users properly scales . in fig . [ fig : oia_k4_a10_kn ] , the number of users scales as @xmath348 , i.e. , @xmath349 for the target dof one . specifically , two user scaling @xmath350 and @xmath351 are considered , and other configurations except the number of users are the same as those in fig . [ fig : oia_k4_a10_n10 ] . fig . [ fig : oia_k4_a10_kn ] verifies that the min - inr and the max - sinr schemes achieve dof one per transmitter as predicted in theorem [ theorem : scaling1 ] and lemma [ lemma : scaling1 ] . in fig . [ fig : oia_k4_a10_n_powerp ] , we consider two different user scaling @xmath352 and @xmath353 from those in fig . [ fig : oia_k4_a10_kn ] . according to theorem [ theorem : scaling1 ] and lemma [ lemma : scaling1 ] , the achievable dof at each transmitter by either the max - sinr scheme or the min - inr scheme is @xmath16 when the number of users scales as @xmath354 . as predicted , fig . [ fig : oia_k4_a10_n_powerp ] shows that the achieved dof per transmitter is 0.5 and 1 when @xmath352 and @xmath353 , respectively , by either the min - inr scheme or the max - sinr scheme . we first studied the optimal way of exploiting multiuser diversity in the @xmath0-transmitter simo ibc where each transmitter with a single antenna selects a user and the number of transmitters is larger than the number of receive antennas at each user . we proved that the multiuser dimensions should be used first for decreasing the dof loss caused by interfering signals ; the whole multiuser dimensions should be exploited to reduce the dof loss term to @xmath355 for the target dof @xmath356 $ ] , while the multiuser dimensions should be devoted to making the dof loss zero and then to increasing the dof gain term to @xmath16 for the target dof @xmath357 . we also derived the sufficient user scaling for the target dof . the dof per transmitter @xmath33 $ ] is obtained when the number of users scales as @xmath358 , and the dof per transmitter @xmath357 is achieved when the number of users scales as @xmath359 . also , we extended the results to the @xmath0-transmitter mimo ibc where each transmitter having the multiple antennas supports the multiple users . k. gomadam , v. r. cadambe , s. a. jafar , `` approaching the capacity of wireless networks through distributed interference alignment , '' in _ proc . of ieee global telecommunications conference _ , pp . 16 , dec . 2008 . y. huang and b. rao , random beamforming with heterogeneous users and selective feedback : individual sum rate and individual scaling laws " , _ ieee trans . wireless commun . 5 , pp . 20802090 , may 2013 . j. h. lee and w. choi , opportunistic interference aligned user selection in multi - user mimo interference channels , " in _ proc . of ieee global telecommunications conference _ , miami , fl , usa , dec . j. h. lee and w. choi , interference alignment by opportunistic user selection in 3-user mimo interference channels , " in _ proc . of ieee international conference on communications _ , kyoto , japan , june , 2011 . j. h. lee and w. choi , on the achievable dof and user scaling law of opportunistic interference alignment in 3-transmitter mimo interference channels , " _ ieee trans . wireless commun . 12 , no . 6 , pp . 2743 - 2753 , june 2013 . j. h. lee and w. choi , opportunistic interference alignment by receiver selection in a k - user 1 x 3 simo interference channel , " in _ proc . of ieee global telecommunications conference _ , houston , tx , usa , dec . 2011 . t. kim , d. j. love , and b. clerckx , instantaneous degrees of freedom of downlink interference channels with multiuser diversity , " in _ proc . of asilomar conference on signals , systems , and computers _ , pp.12931297 , nov . b. c. jung , d. park , and w .- y . shin , opportunistic interference mitigation achieves optimal degrees - of - freedom in wireless multi - cell uplink networks , " _ ieee trans . 60 , no . 7 , july 2012 . j. joung and a. h. sayed , user selection methods for multiuser two - way relay communications using space division multiple access , " _ ieee trans . wireless commun . _ , vol . 9 , no . 7 , pp 21302136 , july 2010 . y. ohwatari , n. miki , t. asai , t. abe , and h. taoka , performance of advanced receiver employing interference rejection combining to suppress inter - cell interference in lte - advanced downlink " , in _ proc . ieee vehicular technology conference ( vtc - fall ) _ , sep . y. tokgoz , b. d. rao , m. wengler , and b. judson , performance analysis of optimum combining in antenna array systems with multiple interferers in flat rayleigh fading , " _ ieee trans . 10471050 , july 2004 .

the axes of solar active regions are inclined relative to the east west direction , with the tilt angle tending to increase with latitude ( `` joy s law '' ) . observational determinations of joy s law have been based either on white - light images of sunspot groups or on magnetograms , where the latter have the advantage of measuring directly the physically relevant quantity ( the photospheric field ) , but the disadvantage of having been recorded routinely only since the mid-1960s . white - light studies employing the historical mount wilson ( mw ) database have yielded tilt angles that are smaller and that increase less steeply with latitude than those obtained from magnetic data . we confirm this effect by comparing sunspot - group tilt angles from the debrecen photoheliographic database with measurements made by li and ulrich using mw magnetograms taken during cycles 2123 . whether white - light or magnetic data are employed , the median tilt angles significantly exceed the mean values , and provide a better characterization of the observed distributions . the discrepancy between the white - light and magnetic results is found to have two main sources . first , a substantial fraction of the white - light `` tilt angles '' refer to sunspots of the same polarity . of greater physical significance is that the magnetograph measurements include the contribution of plage areas , which are invisible in white - light images but tend to have greater axial inclinations than the adjacent sunspots . given the large uncertainties inherent in both the white - light and the magnetic measurements , it remains unclear whether any systematic relationship exists between tilt angle and cycle amplitude during cycles 1623 .

1412.2329

as toroidal magnetic flux rises buoyantly through the convection zone to form sunspots and active regions , the coriolis force acts to twist the flux tubes so that they emerge slightly inclined relative to the east west direction ( see , e.g. , caligari et al . 1995 ; dsilva & choudhuri 1993 ; fan et al . 1994 ; fisher et al . 1995 ; wang & sheeley 1991 ) . this axial tilt plays a central part in flux - transport dynamo models , where it is an essential ingredient in the formation and evolution of the polar fields ( see , e.g. , wang & sheeley 1991 ; dikpati & charbonneau 1999 ; mackay et al . 2002 ; baumann et al . 2004 ; cameron et al . 2010 ; muoz - jaramillo et al . 2010 , 2013 ; jiang et al . 2011a , 2011b , 2013 , 2014 ; kitchatinov & olemskoy 2011 ; cameron & schssler 2012 ; hathaway & upton 2014 ; upton & hathaway 2014a , 2014b ) . the tendency for the leading / westward parts of sunspot groups to be located equatorward of their trailing / eastward parts was first studied by joy , using the sunspot drawings of carrington and sprer made between 1856 and 1893 ( cycles 1013 ) . from the data listed in table ii of hale et al . ( 1919 ) , we infer that the 2633 sunspot groups measured by joy had an area - weighted average inclination of 61 ; a linear fit to the tilt angles as a function of latitude @xmath0 gives @xmath1 over the range @xmath2 . by definition , @xmath3 where @xmath4 is the latitudinal separation between the leading and trailing poles of the sunspot group or active region , @xmath5 is their longitudinal separation , and @xmath6 is taken to be positive if the leading pole lies equatorward of the trailing pole ( in either hemisphere ) , negative otherwise . this definition restricts the tilt angles to the range @xmath7 , and does not distinguish between different polarity orientations , so that it is suited for white - light measurements in which the polarities are unknown . hale et al . ( 1919 ) noted that `` the angle of inclination was found to depend entirely on the latitude of the group , without reference to the number of the cycle or the time within the cycle . a knowledge of the polarities of the spots would have aided greatly in determining the position of the axes of the groups . '' joy s results , as reported in hale et al . ( 1919 ) , were subsequently confirmed by brunner ( 1930 ) , using drawings of 1981 sunspot groups made at the zrich observatory between 1894 and 1928 ( cycles 1316 ) . brunner obtained a value of @xmath8 for the mean tilt angle of his entire sample ; a linear fit to the latitudinally binned data in his table i yields @xmath9 the slope thus being considerably steeper than that found by joy . it should be noted that the derived slopes depend sensitively on the relatively few measurements made at very low and very high latitudes : thus , for the 0@xmath105@xmath10 ( 25@xmath1030@xmath10 ) latitude bin , brunner obtained a mean tilt angle of 06 ( 144 ) , whereas joy s measurements gave 37 ( 93 ) , leading to a much flatter slope . according to brunner , the average axial tilts of sunspot groups decrease systematically from @xmath1178 to @xmath1146 , as they evolve from small spots / pores without penumbrae to large , widely separated spots . in contrast , hale et al . ( 1919 ) stated that `` there is little change in this angle during the life of the group . '' joy s law was confirmed using magnetograph measurements by wang & sheeley ( 1989 , 1991 ) , who analyzed the statistical properties of 2710 bipolar magnetic regions ( bmrs ) that appeared in daily full - disk magnetograms recorded by the national solar observatory at kitt peak ( nso / kp ) during cycle 21 ( 19761986 ) . here , the centroidal positions of the positive- and negative - polarity flux were estimated visually by sheeley from photographic prints , with the help of stonyhurst overlay grids . the bmr tilt angles were found to increase with latitude with a slope of approximately 0.5 , close to the value obtained by brunner ( 1930 ) ; averaged over the entire data set , @xmath12 was as large as 9@xmath10 . an important difference from most other tilt - angle studies is that each bmr was measured only once , at or as close as possible to the time when its flux reached its peak ( the original purpose having been to provide a source term for surface flux - transport modeling ) . by tracking the day - to - day evolution of the flux by eye , it was usually possible to distinguish between closely spaced active regions , which is often not the case when automated methods are employed . howard ( 1989 , 1991a ) applied a field - strength thresholding algorithm to identify active regions in coarse - format versions of mount wilson ( mw ) observatory daily magnetograms recorded during 19671990 , and then derived their axial tilts from the flux - weighted positions of their leading and trailing sectors . he obtained a less regular increase in the tilt angle with latitude than wang & sheeley ( 1989 ) , and suggested that the difference might be due to the inclusion of older , decayed regions in his sample . from an analysis of 517 bmrs appearing on magnetograms taken at the huairou solar observing station during 19882001 , tian et al . ( 2003 ) derived a joy s law slope of 0.5 , in good agreement with the results of wang & sheeley ( 1989 , 1991 ) . the active regions in this study were selected for measurement when they were `` almost mature '' and located within 15@xmath10 of central meridian . more recently , stenflo & kosovichev ( 2012 ) and li & ulrich ( 2012 ) developed automated procedures to derive active - region tilt angles from magnetograms recorded with the michelson doppler imager ( mdi ) on the _ solar and heliospheric observatory _ ( _ soho _ ) , finding slopes of @xmath110.56 and 0.36 , respectively , for joy s law . li & ulrich ( 2012 ) also applied their algorithm to mw magnetograms taken during 19742012 , obtaining a slope of 0.45 and a mean tilt angle of 62 . howard ( 1991b ) derived tilt angles for sunspot groups appearing in mw daily white - light photographs taken between 1917 and 1985 . here , the umbral positions and areas were measured and grouped by proximity as described in howard et al . ( 1984 ) ; the umbrae located westward ( eastward ) of the area - weighted centroid of each group were then defined as belonging to the `` leading '' ( `` following '' ) sector ( despite the absence of polarity information ) . as may be seen from figure 3 in howard ( 1991b ) , the white - light tilt angles show a fairly regular increase up to latitude @xmath13 , but the average slope is only about one - half that shown in figure 5 of howard ( 1991a ) for the tilts derived from mw magnetograms . reviewing his white - light and magnetic results , howard ( 1996 ) noted that the average tilt angle based on the 24,308 mw sunspot - group measurements over the period 19171985 was @xmath14 , whereas the corresponding value for the 15,692 `` plages '' in mw magnetograms taken during 19671995 was @xmath15 . employing the mw sunspot - group tilt - angle measurements and the analogous kodaikanal ( kk ) database covering the period 19061987 , dasi - espuig et al . ( 2010 ) derived a slope of 0.26/0.28 ( mw / kk ) for joy s law and a mean tilt angle of 425/45 ( mw / kk ) . in addition , they found a tendency for @xmath16 , the mean tilt angle normalized by the mean latitude of the sunspot groups , calculated for each of cycles 1521 , to decrease with increasing cycle strength . from the pulkovo database of sunspot group measurements during 19481991 , ivanov ( 2012 ) obtained a joy s law slope of 0.38 and a mean tilt angle of 61 . he also showed that the anticorrelation between @xmath16 and cycle strength in the mw data is almost entirely due to the unusually low value of @xmath12 during cycle 19 ( see also the corrigendum of dasi - espuig et al . 2013 ; a strong anticorrelation remains present in the normalized kk tilt angles for cycles 1521 ) . in their study of hemispheric asymmetries in the mw sunspot - group tilt angles , mcclintock & norton ( 2013 ) found that the mean value of @xmath6 during cycle 19 was 46 in the northern hemisphere , but only 18 in the southern hemisphere . in line with the results of dasi - espuig et al . ( 2010 ) and the flux transport simulations of schssler & baumann ( 2006 ) , they also advocated a downward revision in the slope of joy s law . from this brief review , we infer that the active - region tilt angles derived from magnetograms have tended to be systematically larger than those deduced from white - light observations . as an apparent exception , the sunspot group measurements of brunner ( 1930 ) are in accord with the magnetograph results . however , an inspection of brunner s table iii suggests that roughly one - third of his measurements were made at a very early , pre - penumbral stage in the evolution of the sunspot groups , when their mean tilt was as large as 78 . such emerging regions , which consist of small , faint sunspot pores , are likely to constitute a much smaller fraction of the sunspot groups in the mw and kk databases , which are presumably dominated by fully developed regions whose tilt angles have an average value of 46 , according to table iii of brunner ( 1930 ) . since the statistical studies of magnetic tilt angles are also heavily weighted toward fully emerged active regions , the good agreement between brunner s results and those of wang & sheeley ( 1989 , 1991 ) , tian et al . ( 2003 ) , li & ulrich ( 2012 ) , and stenflo & kosovichev ( 2012 ) could be partly fortuitous . the main objective of this study is to understand why white - light observations of sunspot groups tend to yield smaller tilt angles than magnetograph measurements of active regions . for this purpose , we compare the tilt angles given in the debrecen photoheliographic data ( dpd ) sunspot catalog with the corresponding values derived by li & ulrich ( 2012 ) from mw magnetograms taken during cycles 2123 ( section 2 ) . section 3 places these recent magnetic and white - light tilt - angle measurements in the context of the mw white - light results for cycles 1621 . in section 4 , we examine specific cases in which the white - light and magnetic measurements give very different tilt angles , and discuss how such systematic differences arise . our results are summarized in section 5 . the online dpd catalog contains daily measurements of sunspot areas and positions from 1974 to the present , based on full - disk white - light images taken mainly at the debrecen observatory and its gyula observing station ( see gyri et al . the database also provides sunspot - group tilt angles derived in two different ways : ( 1 ) from the area - weighted umbral positions alone ( `` u '' ) , as in the case of the mw and kk sunspot - group tilt angles ; and ( 2 ) from the area - weighted positions of all visible components of the sunspot group , including umbrae , penumbrae , and pores ( `` upp '' ) . again , the `` leading '' and `` following / trailing '' portions of a sunspot group are defined relative to the area - weighted centroid of the group , and do not necessarily coincide with the actual leading- and trailing - polarity sectors of the active region . we also remark that the number of dpd / upp tilt - angle measurements considerably exceeds ( by @xmath1170% ) the number of dpd / u measurements , because many sunspot groups contain at least a few faint pores , but only one umbra ( usually in the leading sector ) or , in a roughly equal number of cases , no umbrae at all . in addition to dpd , the debrecen observatory team has constructed a database that employs mdi continuum images and magnetograms recorded during 19962011 . the _ soho_/mdi debrecen data ( sdd ) sunspot catalog provides four kinds of tilt angle measurements , two of which use only the continuum intensity images and are exactly analogous to dpd / u and dpd / upp . the remaining two types of tilt angle estimates exploit the mdi magnetograms to assign polarities to the u and upp continuum measurements , so that the leading and following sectors are distinguished by polarity instead of being defined by the area - weighted centroid of the sunspot group . because the statistical properties of the sdd and dpd tilt angles turn out to be rather similar , however , this study will focus on the dpd database , which covers three complete sunspot cycles . a detailed analysis of the sdd and dpd measurements and comparison with mw / kk may be found in baranyi ( 2014 ) . employing daily full - disk mw magnetograms recorded between 1974 and 2012 , li & ulrich ( 2012 ) determined the tilt angles of active regions based on the flux - weighted centroids of the leading- and trailing - polarity sectors . since the period that they studied nearly coincides with that covered by the dpd database , it is interesting to compare their magnetic tilt angles ( `` mw / mag '' ) with the dpd white - light results . let the tilt angles derived from the dpd / u , dpd / upp , and mw / mag measurements be designated by @xmath17 , @xmath18 , and @xmath19 , respectively . throughout this study , we include only measurements made within 45@xmath10 of longitude from central meridian , so as to avoid distortions due to foreshortening near the solar limb . to determine the latitudinal dependence of the tilt angles , we bin the measurements into 5@xmath10-wide intervals of unsigned latitude @xmath20 , and calculate either the arithmetic mean or the median of the data within each bin . when performing averages , we follow the convention of treating multiple measurements of a single active region as if they represented measurements of different active regions ( instead of combining them into one value ) , even though this procedure gives greater weight to longer - lived regions . we begin by examining the average behavior of the tilt angles over the entire interval from 1976 june through 2008 december , comprising cycles 2123 . for this 32 yr period , li & ulrich ( 2012 ) made a total of 25,322 measurements of @xmath19 based on 7958 active regions ; the dpd database provides 23,204 measurements of @xmath17 for 5973 noaa sunspot groups , and 38,975 measurements of @xmath18 for 8575 noaa sunspot groups . ( although the dpd catalog also contains sunspot groups without noaa designations , we omit them in order to maximize the overlap with the active region sample of li and ulrich . ) figure 1 shows @xmath19 , @xmath17 , and @xmath18 as a function of @xmath20 . in the top panel , arithmetic means of the tilt angles are plotted for each latitude bin between 0@xmath105@xmath10 and 30@xmath1035@xmath10 , while median values are plotted in the bottom panel . the standard error of the mean / median is indicated by the vertical bar through each point ; the errors are largest for the highest- and lowest - latitude bins because they contain the smallest number of measurements . the straight lines represent least - squares fits to the plotted points up to latitude @xmath21 ; the point at 30@xmath1035@xmath10 is omitted both because it is based on relatively few measurements , and because it may be significantly affected by differential rotation , which acts to decrease the tilt angles with time and has its steepest gradients at midlatitudes . the slope @xmath22 and @xmath23-intercept @xmath24 of the regression line then define joy s law for the given set of tilt angle measurements : @xmath25 from figure 1 , the magnetic tilt angles ( @xmath19 ) are seen to increase more steeply than those derived from white - light observations ( @xmath17 , @xmath18 ) . in addition , regardless of whether magnetic or white - light data are used , the median values yield systematically larger slopes @xmath22 than the mean values . if the tilt angle measurements for all latitudes are now combined , we find that the mean values of @xmath19 , @xmath17 , and @xmath18 during cycles 2123 are @xmath26 , @xmath27 , and @xmath28 , respectively , while their median values are @xmath29 , @xmath30 , and @xmath31 ( all quoted errors are 1@xmath32 ) . thus the magnetic tilt angles are systematically larger than their white - light counterparts , and the median values are systematically larger than the mean values for both magnetic and white - light tilts . the white - light tilt angles derived from umbral , penumbral , and pore positions are slightly larger than those derived from umbral positions only . both magnetic and white - light tilt angles exhibit an extremely wide scatter , due both to measurement errors and to the randomizing effect of the ambient supergranular convection ; in general , the root - mean - square ( rms ) tilt angles are much larger than their arithmetic means , and increase with decreasing bmr size or total flux ( see howard 1989 , 1991a , 1991b , 1996 ; wang & sheeley 1989 , 1991 ) . figure 2 shows histograms of @xmath19 , @xmath17 , and @xmath18 , again including all measurements taken during cycles 2123 . the bin width is 5@xmath10 , and each histogram ( plotted as a smooth curve connecting the discrete bin values ) has been normalized by dividing by the total number of measurements and multiplying by a constant scaling factor . all three tilt - angle distributions exhibit a gaussian - like profile , with peak located in the 5@xmath1010@xmath10 bin and a half width at half - maximum ( hwhm ) ranging from 154 ( dpd / u ) to 175 ( mw / mag ) . ( despite their gaussian appearance , however , the distributions are not well fitted by single gaussian functions . ) because the peak is shifted to the right of @xmath33 , the wings of the profile are not symmetric over the range @xmath34 , and the `` outliers '' near the left / negative end of the distribution act to decrease the mean value of @xmath6 relative to the median value . to represent the centroid of such skewed distributions , it is preferable to employ the median ( as was done by li & ulrich 2012 ) rather than the mean . from a physical viewpoint , we note that a nonnegligible fraction of the negative outliers are likely to represent bmrs with reversed ( `` anti - hale '' ) east west polarity orientations , which would be assigned tilt angles @xmath35 if a polarity - dependent definition were adopted , in which @xmath6 ranges between @xmath36180@xmath10 and @xmath37180@xmath10 . in the absence of the artificial cutoff at @xmath3890@xmath10 , the mean of the distribution would then increase and converge toward the median . it should be remarked that the great majority of extreme outliers beyond @xmath39 are associated with small active regions having leading following separations of less than 25 . although the distributions of @xmath19 , @xmath17 , and @xmath18 have remarkably similar , quasi - gaussian profiles , the curves of figure 2 deviate significantly from each other around their peaks and for relatively large , positive tilt angles . in particular , mw / mag shows a `` deficit '' of tilt angles in the range @xmath110@xmath1015@xmath10 but a `` surplus '' in the range @xmath1125@xmath1040@xmath10 , as compared to dpd / upp and especially dpd / u . from the fact that tilt angles of @xmath1125@xmath1040@xmath10 are characteristic of active regions located at mid to high latitudes , we infer that such active regions ( typically associated with the rising phase of the cycle ) may be an important source of the differences among the three distributions , as is also suggested by figure 1 . we now consider the behavior of the tilt angles for the individual cycles . in the top panel of figure 3 , we plot the median values of @xmath19 as a function of @xmath20 separately for cycles 21 , 22 , and 23 ; the middle and bottom panels display the corresponding results for @xmath17 and @xmath18 , respectively . neither the magnetic nor the white - light tilt angles show any clear evidence for systematic variations among these three cycles ( which , it should be remarked , have rather similar sunspot - number amplitudes ) . in the case of the tilt angles derived from mw magnetograms , the slope @xmath40 ranges from @xmath41 ( cycle 22 ) to @xmath42 ( cycle 21 ) ; the median tilt angle based on all latitudes , @xmath43 , ranges from @xmath44 ( cycle 23 ) to @xmath45 ( cycle 22 ) . in the case of the tilt angles derived from umbral data alone , @xmath40 varies from @xmath46 ( cycle 21 ) to @xmath47 ( cycle 22 ) , while @xmath43 varies from @xmath48 ( cycle 22 ) to @xmath49 ( cycle 23 ) . when penumbral areas and pores are included , @xmath40 varies from @xmath50 ( cycle 21 ) to @xmath51 ( cycle 23 ) , and @xmath43 varies from @xmath52 ( cycle 22 ) to @xmath53 ( cycle 23 ) . thus @xmath40 has its largest value in cycle 21 according to the magnetic measurements , in cycle 22 when umbral positions alone are used , but in cycle 23 when umbral , penumbral , and pore data are employed ; on the other hand , @xmath43 has its largest value in cycle 22 according to the magnetic observations , but in cycle 23 according to both sets of white - light observations . these seemingly random differences give some indication of the uncertainties involved in attempting to deduce systematic long - term trends from tilt angle measurements . the tilt angle statistics described here and in the next section are summarized in tables 1 and 2 . the tilt angles derived from the sunspot - group measurements of howard et al . ( 1984 ) , using daily white - light photographs taken at mw during 19171985 , have been analyzed previously by howard ( 1991b , 1996 ) , dasi - espuig et al . ( 2010 ) , ivanov ( 2012 ) , and mcclintock & norton ( 2013 ) . because howard et al . ( 1984 ) measured the positions and areas of umbrae only , this data set , referred to here as `` mw / wl , '' may be considered analogous to dpd / u . it is clearly of interest to compare the mw / wl tilt angles with those derived from the more recent databases . for this purpose , we divide the mw / wl time series into six cycles , starting with cycle 16 ( 1923 august1933 august ) and ending with cycle 21 ( 1976 june1986 august ) ; we discard the observations for cycle 15 , since they do not include its rising phase , and we assume that the data gap during the final months of cycle 21 does not significantly affect the overall statistics for that cycle . we omit all measurements taken more than 45@xmath10 from central meridian longitude , as well as those for which the centroids of the leading and following parts are separated by an angle @xmath54^{1/2}$ ] exceeding 20@xmath10 , which is the maximum pole separation among the bmrs analyzed by wang & sheeley ( 1989 ; see their figure 2 , and also the discussion of dasi - espuig et al . 2010 ) . to derive joy s law for each cycle , we again compute the mean or median value of @xmath6 within 5@xmath10-wide bins in @xmath20 and apply a linear least - squares fit over the latitude range 0@xmath1030@xmath10 . figure 4 ( top panel ) shows the joy s law slope , @xmath22 , derived for each of cycles 1621 by fitting the mean values of @xmath6 as a function of @xmath20 . also plotted for comparison are the corresponding slopes obtained for cycles 2123 from the dpd / u , dpd / upp , and mw / mag measurements . as a reminder of the relative strengths of the different cycles , the asterisks indicate the ( arbitrarily scaled ) maximum yearly sunspot number @xmath55 for each cycle . the bottom panel of figure 4 displays the same quantities , but with @xmath22 determined from the median rather than the mean value of the tilt angle as a function of latitude . from figure 4 , we see that the joy s law slopes obtained from the mw / wl sunspot - group data tend to be substantially smaller than those determined from both the dpd and the magnetic measurements . averaged over cycles 1621 , the mean - based ( median - based ) mw / wl slope is @xmath56 ( [email protected] ) , derived by dasi - espuig et al . ( 2010 ) from the mw / wl data ( 0.26 ) is higher than ours ( 0.20 ) because they forced their least - squares fit to pass through zero , setting @xmath57 in equation ( 4 ) ( see also the discussion of mcclintock & norton 2013 ) . ] whereas the corresponding value for dpd / u , averaged over cycles 2123 , is @xmath58 ( [email protected] ) . however , for cycle 21 , the only period when the databases overlap , mw / wl actually gives a larger slope than dpd / u ; the significance of this `` convergence '' is unclear to us . figure 5 ( top panel ) displays the mean values of @xmath6 , averaged over each cycle and over all latitudes , for the different data sets ; the median values are plotted in the bottom panel . when averaged over cycles 1621 , the mw / wl tilt angles have a mean ( median ) value of 44@xmath3802 ( 58@xmath3802 ) , roughly @xmath59 smaller than the corresponding dpd values averaged over cycles 2123 , and @xmath112@xmath10 smaller than the mw / mag values . again , the different data sets are in relatively good agreement for cycle 21 , when the mw / wl tilt angles attain their highest cycle - averaged values . we note that the dpd tilt angles for cycle 21 remain almost unchanged even if the 1986 measurements are omitted when averaging over this cycle ; thus it is unlikely that the agreement can be attributed to the absence of mw / wl data in 1986 . as pointed out by dasi - espuig et al . ( 2010 ) , the cycle - averaged values of @xmath6 depend on the latitudinal distribution of the measured active regions , which may vary from cycle to cycle and from one database to another . to correct for this effect ( following dasi - espuig et al . ) , we now divide the cycle - averaged tilt angles by their corresponding cycle - averaged , unsigned latitudes . figure 6 shows the resulting normalized quantities , @xmath16 ( top panel ) and @xmath60 ( bottom panel ) . the latitude normalization considerably reduces the cycle - to - cycle variations and the differences between the various data sets . this partial convergence reflects the fact that , averaged over cycles 1621 , the mw / wl measurements were made at a mean ( median ) latitude of 151 ( 144 ) , as compared to 159 ( 153 ) for the dpd measurements , averaged over cycles 2123 . even after the latitude normalization is applied , it is clear from figure 6 that cycle 19 remains an outlier characterized by unusually small values of @xmath16 and @xmath60 , according to the mw / wl measurements . for comparison , as indicated by figure 4(c ) in ivanov ( 2012 ) , the pulkovo measurements for this cycle give @xmath61 , twice the mw / wl value . however , it should also be noted that the pulkovo data coverage of cycle 19 is incomplete , with figure 3 of ivanov ( 2012 ) showing a gap that extends through the maximum phase 19571960 . as pointed out by mcclintock & norton ( 2013 ) , the unusually low mw / wl value for cycle 19 is mainly attributable to the southern - hemisphere sunspot groups , which have a latitude - normalized mean tilt angle of 0.15 , as compared to 0.26 for the northern hemisphere . in flux transport models , the evolution of the polar fields depends sensitively on the tilt angles of active regions and on the meridional flow speed . the magnetograph observations of babcock ( 1959 ) indicate that the south polar field reversed as early as mid-1957 , whereas the north polar field reversed in late 1958 ( see also the polar faculae counts of sheeley 2008 ) . this timing seems difficult to reconcile with the very small tilt angles that the mw / wl database assigns to the southern hemisphere , unless large variations in the meridional flow speed are invoked . in figure 7 , we have plotted a histogram of the 22,533 mw / wl tilt - angle measurements made during cycles 1621 ; for comparison , we again display the histogram of the 23,204 dpd / u tilt angles measured during cycles 2123 . the mw / wl tilt angles show a significantly larger scatter , with the gaussian - like distribution having a hwhm of 179 , as compared to 154 for the dpd / u measurements . from a sample of mw white - light images and sunspot - group drawings with polarity labeling for the period 19611967 , baranyi ( 2014 ) found that @xmath1184% of the sunspot groups whose leading and following parts had separations @xmath62 consisted entirely of spots of the same polarity . in the great majority of these `` unipolar '' groups , the actual opposite - polarity ( usually trailing ) sector contained no visible umbrae ; in the remaining cases , the two sectors of a larger sunspot group were incorrectly taken to be separate groups . based on all of the mw / wl measurements made within 45@xmath10 of central meridian during cycles 1621 , the median angular separation of the leading and following sectors is only 29 , and 43% of the sunspot groups have @xmath62 . as may be seen from the red dotted curve in figure 8 , the tilt angle distribution of these `` compact '' groups is extremely broad ( hwhm = 257 ) ; furthermore , the mean ( median ) value of @xmath6 is only 27 ( 40 ) , as compared to 44 ( 58 ) for all sunspot groups . when this component is separated out and only sunspot groups having @xmath63 are retained , the hwhm of the original tilt - angle distribution decreases from 179 to 166 , and the outliers in the far wings @xmath64 almost disappear ( compare the black solid and blue dashed curves in figure 8) . after filtering out sunspot groups with @xmath62 from the mw / wl database , we find that @xmath12 ( @xmath43 ) increases from 44 ( 58 ) to 58 ( 64 ) , while @xmath16 ( @xmath60 ) increases from 0.29 ( 0.40 ) to 0.38 ( 0.45 ) ; likewise , @xmath65 ( @xmath66 ) increases from 0.20 ( 0.27 ) to 0.32 ( 0.34 ) . when the same procedure is applied to the dpd / u measurements for cycles 2123 , for which 33% of the sunspot groups have @xmath62 , @xmath12 ( @xmath43 ) increases from 53 ( 66 ) to 64 ( 70 ) , @xmath16 ( @xmath60 ) increases from 0.33 ( 0.43 ) to 0.40 ( 0.45 ) , and @xmath65 ( @xmath66 ) increases from 0.28 ( 0.40 ) to 0.40 ( 0.45 ) . the filtering has a smaller effect on the tilt angles derived from the dpd / upp measurements , for which only 23% of the sunspot groups have @xmath62 . in all cases , the median tilt angles undergo considerably less change than the corresponding means , again suggesting that the median provides a more robust statistical measure than the mean . figure 9 shows the overall effect of omitting sunspot groups with @xmath62 on the joy s law slopes derived from the different databases ( compare figure 4 ) . we note that only 11% of the mw / mag measurements have the leading- and trailing - polarity centroids separated by less than 25 , consistent with the conclusion of baranyi ( 2014 ) that most of the mw / wl measurements having @xmath62 refer to `` unipolar '' sunspot groups . although filtering out the unipolar groups brings the different data sets into better agreement with each other , the magnetic measurements still appear to provide an upper bound on the tilt angles ( see tables 1 and 2 ) . the physical basis of the remaining discrepancy is discussed in section 4 . even when new active regions are at their peak development , only @xmath1150% of their magnetic flux is contained in sunspots ( schrijver 1987 ; sheeley 1966 ) , with the rest being in the form of plages / faculae , which are invisible in white - light images except near the solar limb . plages , best seen in k - line spectroheliograms , represent very strong fields that fall below the @xmath111500 g threshold for sunspots . we now consider the possibility that the magnetic tilt angles tend to be larger than their white - light counterparts because the magnetograph measurements include the contribution of plage areas . for this purpose , we examine a number of specific cases in which magnetic and white - light measurements have been carried out for the same active region , but with very different results . these examples are all taken from the rising phase of the activity cycle , when the tilt angles and the discrepancies between the measurements tend to be greatest . figure 10 shows a white - light image of active region noaa 807 , as recorded by the debrecen observatory on 1977 april 18 ; also displayed are longitudinal magnetograms from mw ( spatial resolution 37/pixel ) and nso / kp ( @xmath111@xmath67/pixel ) taken on the same day . the region is centered at latitude @xmath68 , and is dominated by a pair of leading sunspots separated by 45 . for this sunspot group , the tilt angles given by the dpd / u and dpd / upp databases are 25 and 18 , respectively ; in contrast , li & ulrich ( 2012 ) obtained a value of @xmath6 as large as 30@xmath10 from the mw magnetogram , in agreement with the measurement by sheeley from the kp magnetogram ( see wang & sheeley 1989 ) . to show why the white - light and magnetic results are so different , we have overplotted on each of the images in figure 10 the axes connecting the centroids of the leading and following sectors , according to dpd / u , li & ulrich ( mw / mag ) , and sheeley ( kp / mag ) . it is evident that the white - light tilt angles are so small because the easternmost of the two sunspots is incorrectly assigned to the following sector . from the magnetograms , we see that both spots have leading ( negative ) polarity , while the trailing - polarity sector , centered to the southeast of the sunspot pair , consists mainly of plage and is almost invisible in the white - light image . figure 11 displays a debrecen white - light image of active region noaa 882 , together with the corresponding mw and kp magnetograms . the region is centered at @xmath69 , and the observations were all made on 1977 september 5 . a prominent sunspot is seen only in the leading sector . the tilt angles measured by dpd / u , dpd / upp , mw / mag , and kp / mag are 12@xmath10 , 19@xmath10 , 29@xmath10 , and 39@xmath10 , respectively . the dpd / u tilt - angle measurement ( not shown ) can be rejected because it takes the following sector to be centered on a small satellite umbra within the leading sunspot , so that @xmath70 . the dpd / upp measurement takes into account the small pores in the trailing - polarity sector , but not the extensive plage areas ; the inferred axis ( red line in figure 11 ) is thus less steeply inclined than the axes derived from the mw and kp magnetograms ( indicated by the blue and green lines , respectively ) . figure 12 shows a debrecen image of noaa 1203 on 1978 july 14 , along with the corresponding mw and kp magnetograms . for this active region , centered at @xmath71 , the tilt angles given by dpd / u , dpd / upp ( not shown ) , mw / mag , and kp / mag are 6@xmath10 , 5@xmath10 , 21@xmath10 , and 19@xmath10 , respectively . here , it is again clear that the axial inclination of the bmr can not be deduced from the white - light image in the absence of polarity information . in this case , the large structure dominating the eastern part of the sunspot group encloses spots of both polarities , so that the basic assumption underlying the white - light measurements that sunspots of opposite polarity do not overlap in longitude breaks down . the dpd tilt angles are too small because all of the spots located eastward of the area - weighted centroid of the sunspot group have automatically been assigned to the following - polarity sector . as an example from the rising phase of cycle 23 , figure 13 shows a continuum image and magnetogram of active region noaa 8086 , located at @xmath72 . both images were recorded by _ soho_/mdi on 1997 september 17 . the sunspot group again consists of a prominent leading spot trailed by a smattering of smaller spots / pores . the dpd and sdd measurements all give values of @xmath6 in the range @xmath1119@xmath1023@xmath10 ; the sdd / u tilt angle is indicated by the red line connecting the leading spot to the pores located to its northeast . in contrast , based on the mdi magnetogram , li & ulrich ( 2012 ) derived a tilt angle of 41@xmath10 ( indicated by the blue line ) . this steeper inclination reflects the presence of ( positive ) leading - polarity plage to the east of the large leading spot and ( negative ) trailing - polarity plage to the west of the small trailing spots . from these case studies , it is clear that the use of white - light observations to derive tilt angles is subject to many pitfalls , and that the results often do not reflect the actual axial inclinations of active regions , as defined by the distribution of magnetic field . one of the main sources of error is the well - known asymmetry between the leading- and following - polarity sectors , with the latter being characterized by smaller spots and less concentrated flux ( more extensive plage areas ) ( see fan et al . 1993 , and references therein ; murakzy et al . if only the darkest features ( umbrae ) are used to determine the centroids of the two sectors , as is the case for the mw / wl and dpd / u data sets , spots belonging to the leading - polarity sector may be assigned to the following sector , particularly when there are no visible umbrae of following polarity ( as in figure 10 ) . moreover , the absence of tilt angle measurements when only a single ( or no ) umbra is observed reduces the size of the sample by as much as @xmath1140% , introducing an unknown bias into the statistics . another major source of error is associated with active regions where sunspots of opposite polarity overlap in longitude , as in figure 12 . in such cases , in the absence of polarity information , the white - light measurements will tend to underestimate the actual tilt angles . as illustrated by the examples of figures 10 , 11 , and 13 , a more fundamental shortcoming of the white - light measurements is that they do not include the contribution of plage areas . that the plage component of active regions tends to have a greater axial inclination than the sunspot component was pointed out earlier by howard ( 1996 ) , who compared a histogram of tilt angles derived from mw magnetograms taken during 19671995 with the tilt angle distribution derived from the 19171985 mw white - light database ( see his figure 1 ) . the daily magnetograms used by howard were smoothed to a resolution of @xmath7334/pixel , so that sunspots were no longer identifiable and active regions effectively consisted entirely of plage fields . in general , as illustrated by some of the cases presented here , the trailing - polarity sectors of active regions tend to be dominated by plages , which occupy a larger area and extend to higher latitudes than the sunspots . it is widely accepted that joy s law has its physical origin in the coriolis force , which acts to twist toroidally oriented flux tubes in the north south direction as they rise buoyantly to the surface ( for a dissenting view , see kosovichev & stenflo 2008 ; stenflo & kosovichev 2012 ) . the tendency for the sunspot component of active regions to have a smaller axial inclination than the plage areas may then be understood as follows . as it approaches the surface , an @xmath74-loop expands rapidly in order to maintain pressure equilibrium with its surroundings . the expansion ( or diverging flow ) in the longitudinal direction , with speed @xmath75 , gives rise to a coriolis force ( @xmath76 ) that acts over a timescale @xmath77 to produce an axial tilt @xmath78 . since @xmath75 ( like the buoyant velocity ) scales roughly as the alfvn speed @xmath79 , it follows that the tilt angle @xmath6 should scale inversely as the field strength @xmath80 . similar results are obtained if the limiting effects of magnetic tension and drag forces are included ( see fan et al . 1994 ; fisher et al . 1995 ) . on physical grounds , then , we would expect the tilt angles associated with sunspots to be smaller than those associated with the weaker plage fields . it should be emphasized that the axial tilt is here regarded to be a function of the local flux density , not of the area - integrated flux . indeed , there is no clear evidence that the tilt angles of bmrs vary systematically with their total flux or areal size ( wang & sheeley 1989 , 1991 ; kosovichev & stenflo 2008 ; stenflo & kosovichev 2012 ) , even though their spread ( rms values ) increase with decreasing size , an effect attributable to random displacements by supergranular convective motions ( see , e.g. , howard 1989 , 1996 ; wang & sheeley 1989 , 1991 ) . our main objective has been to clarify the differences between magnetic and white - light determinations of active - region tilt angles , and in particular to understand why magnetograms ( used by wang & sheeley 1989 ; howard 1991a ; tian et al . 2003 ; stenflo & kosovichev 2012 ; li & ulrich 2012 ) tend to yield larger inclinations than white - light images of sunspot groups ( used by howard 1991b ; dasi - espuig et al . 2010 ; ivanov 2012 ; mcclintock & norton 2013 ) . because the previous white - light studies have depended heavily on the historical mw and kk databases , we have employed the more recent dpd measurements to confirm this systematic effect and to diagnose its causes . for comparison with the white - light results , we have adopted the tilt angle measurements of li & ulrich ( 2012 ) as representative of the results of magnetogram - based studies . we now summarize our conclusions ( see also tables 1 and 2 ) . when averaged over cycles 2123 ( 19762008 ) , the mw magnetic measurements yield larger tilt angles and joy s law slopes than the dpd white - light observations . the mean ( median ) tilt angle ranges from 62@xmath3802 ( 75@xmath3802 ) for mw / mag , to 59@xmath3801 ( 69@xmath3802 ) for dpd / upp , to 53@xmath3802 ( 66@xmath3802 ) for dpd / u . the mean - based ( median - based ) values of @xmath22 are found to be [email protected] ( [email protected] ) for mw / mag , 0.32@xmath81 ( [email protected] ) for dpd / upp , and [email protected] ( [email protected] ) for dpd / u . the historical mw white - light database gives systematically smaller tilt angles and joy s law slopes than the dpd measurements . when averaged over cycles 1621 ( 19231985 ) , mw / wl yields a mean ( median ) tilt angle of 44@xmath3802 ( 58@xmath3802 ) and a mean - based ( median - based ) joy s law slope of [email protected] ( [email protected] ) . however , the mw and dpd data sets show surprisingly good agreement where they overlap in cycle 21 ( see figures 46 ) . \3 . in 43% of the mw sunspot - group measurements for cycles 1621 , the angular separation between the leading and following parts is less than 25 . in the great majority ( @xmath1184% ) of these cases , as shown by baranyi ( 2014 ) , the two sectors are likely to have the same polarity . filtering out all regions with @xmath62 from the mw / wl and dpd / u data sets substantially increases both the average tilt angles and the joy s law slopes . that a large fraction of the umbral - based measurements refer to `` unipolar '' groups reflects the tendency for well - defined sunspots to be found preferentially in the leading - polarity sectors of active regions . \4 . in view of the skewed nature of the measured distributions , which are centered on positive values of @xmath6 and are thus asymmetric with respect to the imposed endpoints at @xmath82 , medians provide a better characterization of the tilt angle properties than simple arithmetic means ( as was recognized earlier by li & ulrich 2012 ) . as evident from a comparison of tables 1 and 2 , the median - based values of @xmath22 , @xmath6 , and @xmath83 tend to be substantially larger than the corresponding mean - based values . it is also noteworthy that the medians change less than the means when sunspot groups with @xmath62 are omitted . \5 . that the tilt angles derived from magnetograms remain somewhat larger than those deduced from white - light images , even after filtering out unipolar sunspot groups , is due to the contribution of plages , which are seen in magnetograms but not in white light . the plage areas of an active region tend to have a larger axial inclination than the sunspots . this property was first noted by howard ( 1996 ) , although he did not correct the white - light tilt angles for the contribution of unipolar groups . the difference between the sunspot and plage tilt angles can be attributed to the action of the coriolis force , which imparts to each buoyant flux tube a twist proportional to its rise / expansion time . since @xmath84 scales inversely as the local field strength , the weaker , more slowly rising / expanding fields that form plages end up with a greater net twist than the stronger sunspot fields . the same holds if the twist due to the coriolis force is limited by magnetic tension ( see , e.g. , fan et al . 1994 ) . given the large uncertainties involved in measuring tilt angles , whether white - light images or magnetograms are employed , it remains unclear whether joy s law undergoes physically significant variations between different cycles . however , figure 6 suggests that the cycle - averaged , latitude - normalized parameters @xmath16 ( @xmath110.30.4 ) and @xmath60 ( @xmath110.40.5 ) remain relatively constant from cycle to cycle . our main conclusion is that white - light measurements generally underestimate the axial inclinations of active regions , even when the leading- and following - polarity sectors are correctly identified , because the sunspot component is characterized by a smaller tilt than the plage component . by the same token , measurements employing magnetograms that underestimate the flux in sunspots ( due to low spatial resolution or saturation effects ) may slightly overestimate the axial tilts of active regions . it is apparent that the tilt angles extracted from the mw white - light database do not provide a reliable basis for long - term flux - transport simulations , or for deducing cycle - to - cycle relationships among tilt angles , cycle amplitudes / total sunspot areas , and the polar fields ( see dasi - espuig et al . 2010 ; muoz - jaramillo et al . however , these measurements could be improved by exploiting the polarity information available in the daily mw sunspot - group drawings . in addition , use should be made of the digitized mw k - line images spanning the period 19151985 , which show the plage areas associated with the sunspot groups , while at the same time outlining more clearly the boundaries of the active regions ( see , e.g. , figures 1 to 12 in sheeley et al . it would be particularly instructive to compare the tilt angles defined by the emission contours with those derived from the white - light sunspots . from an analysis of white - light facular data compiled by the royal greenwich observatory , foukal ( 1993 ) concluded that the ratio of facular to sunspot area decreased during the highest - amplitude cycles ( 18 and 19 ) . since faculae are the white - light counterparts ( visible near the limb ) of plages and strong magnetic network , such a decrease would imply that the axial tilts of active regions should be somewhat smaller during very active cycles than in weaker cycles . however , recent photometric measurements by chapman et al . ( 2011 ) indicate that the facular / plage area in fact increases linearly with sunspot area , without flattening out for large total sunspot areas . an important question not addressed in this study is the effect of evolutionary processes on the measured tilt angles . that the values obtained depend on when the measurements are made was already emphasized by brunner ( 1930 ) , who noted that the axial inclinations of emerging sunspot groups decrease with time due to the westward proper motions of their leading spots ( see also kosovichev & stenflo 2008 ) . the slope derived for joy s law depends sensitively on the relatively few active regions at latitudes @xmath85 , which are also those most strongly affected by differential rotation . for a bmr centered at @xmath86 with a latitudinal pole separation of 2@xmath10 , for example , @xmath5 increases at a rate of @xmath11009 day@xmath87 , so that rotational shearing may have a noticeable effect on the inclinations of regions older than @xmath1110 days . tilt angle measurements should also be performed before supergranular diffusion has converted the sunspot fields into network / plage . it is clear from this study that reliable and physically meaningful tilt angles can not be derived from white - light images alone , which not only lack the essential polarity information , but also omit the plage component of active regions . even when magnetograms are employed , care is needed when applying automated techniques . ideally , each active region should be tracked from day to day and its axial inclination measured at the time of peak development , after the flux has fully emerged but before it begins to decay through transport processes . such tracking , which also makes it easier to identify the boundaries of closely spaced active regions , is probably best done by human eye ; a computer algorithm can then be employed to determine the centroids of the leading- and trailing - polarity flux within the visually checked / adjusted active - region boundaries . it will be especially interesting to compare the magnetic tilt angles determined for the current weak cycle with those derived for cycles 2123 . accurate measurements using a combination of visual and automated techniques may help to resolve the question of whether cycle - to - cycle variations in the polar fields are related to systematic variations in the axial tilts of active regions . we thank the referee for comments , and g. chintzoglou , j. s. morrill , k. muglach , n. r. sheeley , jr . , d. g. socker , r. k. ulrich , and the participants of the international space science institute workshop on `` the solar activity cycle : physical causes and consequences '' for helpful discussions . we are also indebted to the debrecen heliophysical observatory , mw / ucla , and nso / kp for providing the data used in this investigation , which was funded by the office of naval research and by the european community s seventh framework programme ( project eheroes ) . @xmath65 & mw / mag & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & mw / wl & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & ( @xmath63 ) & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & dpd / u & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & dpd / upp & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + @xmath12 & mw / mag & & & & & & 58@xmath3803 & 68@xmath3803 & 61@xmath3803 & & 62@xmath3802 + & ( @xmath63 ) & & & & & & 60@xmath3803 & 70@xmath3803 & 63@xmath3803 & & 64@xmath3802 + & mw / wl & 44@xmath3806 & 47@xmath3805 & 48@xmath3805 & 36@xmath3804 & 41@xmath3805 & 53@xmath3805 & & & 44@xmath3802 & + & ( @xmath63 ) & 66@xmath3805 & 55@xmath3805 & 64@xmath3804 & 46@xmath3804 & 55@xmath3804 & 64@xmath3805 & & & 58@xmath3802 & + & dpd / u & & & & & & 52@xmath3803 & 51@xmath3803 & 56@xmath3803 & & 53@xmath3802 + & ( @xmath63 ) & & & & & & 66@xmath3803 & 62@xmath3802 & 64@xmath3802 & & 64@xmath3801 + & dpd / upp & & & & & & 55@xmath3802 & 61@xmath3802 & 63@xmath3802 & & 59@xmath3801 + & ( @xmath63 ) & & & & & & 61@xmath3802 & 67@xmath3802 & 70@xmath3802 & & 66@xmath3801 + @xmath16 & mw / mag & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & mw / wl & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & ( @xmath63 ) & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & dpd / u & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & dpd / upp & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + @xmath66 & mw / mag & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & mw / wl & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & ( @xmath63 ) & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & dpd / u & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & dpd / upp & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + @xmath43 & mw / mag & & & & & & 75@xmath3804 & 78@xmath3804 & 71@xmath3803 & & 75@xmath3802 + & ( @xmath63 ) & & & & & & 755@xmath3804 & 78@xmath3804 & 71@xmath3803 & & 75@xmath3802 + & mw / wl & 61@xmath3807 & 53@xmath3807 & 64@xmath3806 & 47@xmath3805 & 55@xmath3806 & 74@xmath3806 & & & 58@xmath3802 & + & ( @xmath63 ) & 74@xmath3807 & 585@xmath3806 & 74@xmath3805 & 51@xmath3805 & 58@xmath3805 & 78@xmath3806 & & & 64@xmath3802 & + & dpd / u & & & & & & 67@xmath3804 & 60@xmath3804 & 70@xmath3804 & & 66@xmath3802 + & ( @xmath63 ) & & & & & & 72@xmath3803 & 64@xmath3803 & 725@xmath3803 & & 70@xmath3802 + & dpd / upp & & & & & & 68@xmath3803 & 65@xmath3803 & 74@xmath3803 & & 69@xmath3802 + & ( @xmath63 ) & & & & & & 72@xmath3803 & 69@xmath3803 & 78@xmath3803 & & 73@xmath3802 + @xmath60 & mw / mag & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & mw / wl & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & ( @xmath63 ) & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & & [email protected] & + & dpd / u & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & dpd / upp & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + & ( @xmath63 ) & & & & & & [email protected] & [email protected] & [email protected] & & [email protected] + -wide bins in unsigned latitude @xmath20 , and a least - squares linear fit applied to the six points spanning the range 0@xmath1030@xmath10 . vertical bars indicate one standard error of the mean . ( b ) median - based joy s law : here , median values of the measured tilt angles have been calculated for each latitude bin and a least - squares fit applied over the range 0@xmath1030@xmath10 . vertical bars represent one standard error of the median . ] ( the discrete bin points are connected by a smooth curve rather than being plotted in steplike fashion ) . vertical dotted ( dash - dotted ) lines mark the location of @xmath88 ( @xmath89 ) . significant differences between the three gaussian - like distributions occur over the interval @xmath90 , with dpd / upp and especially mw / mag showing a greater relative number of these large tilt angles than dpd / u , which makes up for its deficit with a surplus at small tilt angles . the restriction of @xmath6 to the range between @xmath91 and @xmath92 causes the distribution functions to be skewed leftward relative to their peaks , which in turn explains why their means are smaller than their medians . ] ) have been normalized by dividing by the total number of measurements in each data set and multiplying by an arbitrary constant . the mw / wl distribution has hwhm = 179 , and shows a larger frequency of both positive and negative outliers than dpd / u ( hwhm = 154 ) . ] ( blue dashed curve ) , and a component consisting of the remaining 9756 sunspot groups with @xmath62 ( red dotted curve ) . all three histograms have bin width 5@xmath10 and have been normalized by dividing by the total number of mw / wl measurements ( 22,533 ) and multiplying by a constant scaling factor . the tilt angle distribution of sunspot groups with @xmath62 , the great majority of which have leading and following sectors of the same polarity , is extremely broad ( hwhm = 257 ) . filtering out this component reduces the hwhm of the original distribution from 179 to 166 , and removes most of the outliers in the far wings . ] . ( a ) mean values . ( b ) median values . compare figure 4 , where all active regions are included in the averages . the filtering condition removes most of the `` unipolar '' sunspot groups from the mw / wl and dpd data sets . ] . ( a ) white - light image taken by the debrecen observatory at 06:59 ut on 1977 april 18 . ( b ) mw magnetogram recorded at 22:24 ut . ( c ) nso / kp magnetogram recorded at 15:33 ut . colored lines overplotted on each image connect the measured centroids ( corrected for differential rotation ) of the leading and following sectors . red : dpd / u ( @xmath93 ) . blue : mw / mag ( @xmath94 ; see li & ulrich 2012 ) . green : kp / mag ( @xmath95 ; see wang & sheeley 1989 ) . the white - light image is dominated by a pair of sunspots separated by 45 ; dpd / u assigns the easternmost spot to the following sector , even though the magnetograms show that it has the leading polarity . the resulting @xmath6 is much smaller than that obtained from the magnetic measurements . this active region also provides an example of longitudinal overlap between leading- and trailing - polarity sectors , another major source of error in white - light tilt - angle determinations ( see also figure 12 ) . ] . ( a ) white - light image taken by the debrecen observatory at 10:16 ut on 1977 september 5 . ( b ) mw magnetogram recorded at 17:44 ut . ( c ) nso / kp magnetogram recorded at 14:10 ut . red line : dpd / upp ( @xmath96 ) . blue line : mw / mag ( @xmath97 ) . green line : kp / mag ( @xmath98 ) . the magnetic measurements give larger tilt angles than dpd / upp because the axis defined by the plage areas is steeper than that defined by the sunspots . the dpd / u measurement ( not plotted ) has the centroid of the following sector coinciding with a very small satellite umbra within the leading sunspot , giving a tilt angle of 123 . ] . ( a ) white - light image taken by the debrecen observatory at 09:57 ut on 1978 july 14 . ( b ) mw magnetogram recorded at 18:02 ut . ( c ) nso / kp magnetogram recorded at 14:33 ut . red line : dpd / u ( @xmath99 ) . blue line : mw / mag ( @xmath100 ) . green line : kp / mag ( @xmath101 ) . in this case , the white - light measurements have the following sector centered on the large , eastern sunspot structure , whereas the magnetograms show that the equatorward half of this structure actually has leading polarity . the magnetic axis thus has a larger inclination than suggested by the white - light image . ] . ( a ) mdi continuum image recorded at 00:00 ut on 1997 september 17 . ( b ) simultaneous mdi magnetogram . red line : sdd / u ( @xmath102 ) . blue line : mdi / mag ( @xmath103 ; li & ulrich 2012 ) . again , the magnetic tilt angle is larger because the plage areas visible in the magnetogram have a more steeply inclined axis than the sunspots seen in the continuum image . ]

an element of a coxeter group is fully commutative if any two of its reduced decompositions are related by a series of transpositions of adjacent commuting generators . an element of a coxeter group is cyclically fully commutative if any of its cyclic shifts remains fully commutative . these elements were studied by boothby _ et al_. in particular the authors enumerated cyclically fully commutative elements in all coxeter groups having a finite number of them . in this work we characterize and enumerate cyclically fully commutative elements according to their coxeter length in all finite or affine coxeter groups by using a new operation on heaps , the cylindric transformation . in finite types , this refines the work of boothby _ et al . _ , by adding a new parameter . in affine type , all the results are new . in particular , we prove that there is a finite number of cyclically fully commutative logarithmic elements in all affine coxeter groups . we study afterwards the cyclically fully commutative involutions and prove that their number is finite in all coxeter groups . [ multiblock footnote omitted ]

1403.1130

let @xmath0 be a coxeter group . an element @xmath1 is said to be fully commutative ( fc ) if any reduced word for @xmath2 can be obtained from any other one by only short braid relations . these elements were introduced and studied independently by fan in @xcite , graham in @xcite and stembridge in @xcite , where among other things , he classified the coxeter groups with a finite number of fully commutative elements and enumerated them in each case . fully commutative elements appear naturally in the context of ( generalized ) temperley - lieb algebras , as they index a linear basis of the latter . recently , in @xcite , biagioli , jouhet and nadeau refined stembridge s enumeration by counting fully commutative elements according to their coxeter length in any finite or affine coxeter group . + in this paper , we focus on a certain subset of fully commutative elements , the cyclically fully commutative elements ( which we denote cfc from now on ) . these are elements for which every cyclic shift of any reduced expression is a reduced expression of a fc element . they were introduced by boothby et al . in @xcite , where they classified the coxeter groups with a finite number of cfc elements ( they showed that the latter are the groups with a finite number of fc elements ) and enumerated them . the main goal of @xcite is to establish necessary and sufficient conditions for a cfc element @xmath1 to be logarithmic , i.e to satisfy @xmath3 for any positive integer @xmath4 . this is the first step towards studying a cyclic version of matsumoto s theorem for cfc elements . here we will focus on the combinatorics of cfc elements . we introduce a new operation on heaps , which will allow us to give a new characterization of cfc elements in all coxeter groups ( see theorem [ propcfc ] ) . for finite or affine coxeter groups , this characterization can be reformulated in terms of words , by using the work on fc elements from @xcite . it allows us to enumerate the cfc elements by taking into account their coxeter length . we will also prove that the number of cfc involutions is finite in all coxeter groups . this paper is organized as follows . we recall in section 1 some definitions and properties on coxeter groups , we introduce a new operation on heaps , the _ cylindric transformation _ , in two different ways , and we deduce a new characterization of cfc elements in terms of pattern - avoidance for cylindric transformed heaps ( see theorem [ propcfc ] ) . in section 2 , we use this characterization in order to obtain a complete classification ( in terms of words ) of cfc elements in the affine group @xmath5 . we also deduce a classification of cfc elements in the group @xmath6 , and use this to enumerate cfc elements in both groups , according to their coxeter length . the same work is done for the groups @xmath7 , and @xmath8 in section 3 . in section 4 , we will focus on cfc involutions . the main result is that there is a finite number of cfc involutions in all coxeter groups . we also give a characterization of cfc involutions for all coxeter groups and enumerate them in finite and affine types . [ sec : intro ] + let @xmath0 be a coxeter group with a finite generating set @xmath9 and coxeter matrix @xmath10 . recall ( see @xcite ) that this notation means that the relations between generators are of the form @xmath11 for @xmath12 , @xmath13 is a symmetric matrix with @xmath14 and @xmath15 . the pair ( @xmath16 ) is called a coxeter system . we can write the relations as @xmath17 , each side having length @xmath18 ; these are usually called braid relations . when @xmath19 , this is a commutation relation or a short braid relation . we define the coxeter diagram @xmath20 associated to ( @xmath21 ) as the graph with vertex set @xmath9 , and for each pair ( @xmath22 ) with @xmath23 , an edge between @xmath24 and @xmath25 labelled by @xmath18 . if @xmath26 , we usually omit the labelling . + for @xmath1 , we define the coxeter length @xmath27 as the minimum length of any expression @xmath28 with @xmath29 . an expression is called reduced if it has a minimal length . the set of all reduced expressions of @xmath2 is denoted by @xmath30 . a classical result in coxeter group theory , known as matsumoto s theorem , is that any expression in @xmath30 can be obtained from any other one using only braid relations . an element @xmath2 is said to be fully commutative if any expression in @xmath30 can be obtained from any other one using only commutation relations . for clarity , we will write @xmath2 for an element in @xmath0 , and @xmath31 for one of its words ( or expressions ) . [ defcfc ] ( ( * ? ? ? * definition 3.4 ) ) given an expression @xmath32 , we define the left ( _ resp . _ right ) cyclic shift of @xmath31 as @xmath33 ( _ resp . _ @xmath34 ) . a cyclic shift of @xmath31 is @xmath35 for @xmath36 . an element @xmath1 is said to be cyclically fully commutative if every cyclic shift of any expression in @xmath30 is a reduced expression for a fully commutative element . we denote the set of cfc elements of @xmath0 by @xmath37 . + + we follow stembridge @xcite in this section . fix a word @xmath38 in @xmath39 , the free monoid generated by @xmath9 . we define a partial ordering of the indices @xmath40 by @xmath41 if @xmath42 and @xmath43 and extend by transitivity on a partial ordering . we denote by @xmath44 this poset together with the `` labeling '' @xmath45 : this is the _ heap _ of @xmath46 . we will often call the elements of @xmath47 points . in the hasse diagram of @xmath48 , elements with the same labels will be drawn in the same column . the size @xmath49 of a heap @xmath50 is its cardinality . given any subset @xmath51 , we will note @xmath52 the subposet induced by all elements of @xmath50 with labels in i. in particular @xmath53 for @xmath54 is the chain @xmath55 where @xmath56 . we also denote by @xmath57 the number @xmath58 ( note that this also counts the number of occurences of s in * w * ) . in @xmath50 , we say that there is a _ chain covering relation _ between @xmath59 and @xmath60 , denoted by @xmath61 , if @xmath62 and one of the two following conditions is satisfied : * @xmath63 and there is no element @xmath64 with the same label as @xmath59 or @xmath60 such that @xmath65 , * @xmath66 and there is no element @xmath64 such that @xmath65 . note that the set of chain covering relations corresponds to the set of edges in the corresponding heap . in other words , this set is equal to the set of covering relations of all subposets @xmath67 , with @xmath23 . in figure [ exheap ] , we fix a coxeter diagram on the left , and we give two examples of words with the corresponding heaps . two different heaps . ] if we consider heaps up to poset isomorphisms which preserve the labels , then heaps encode precisely _ commutativity classes _ , that is , if the word @xmath68 is obtainable from @xmath31 by transposing commuting generators then there exists a poset isomorphism between @xmath44 and @xmath69 . in particular , if @xmath2 is fully commutative , the heaps of the reduced words are all isomorphic , and thus we can define the heap of @xmath2 , denoted @xmath70 . let @xmath71 be a word . a _ linear extension _ of the poset @xmath72 is a linear ordering @xmath73 of @xmath74 such that @xmath75 implies @xmath41 . now let @xmath76 be the set of words @xmath77 where @xmath73 ranges over all linear extensions of @xmath44 . let @xmath2 be a fully commutative element . then @xmath78 is equal to @xmath79 for some ( equivalently every ) @xmath80 . we say that a chain @xmath81 in a poset @xmath50 is convex if the only elements @xmath82 satisfying @xmath83 are the elements @xmath84 of the chain . the next result characterizes _ fc heaps _ , namely the heaps representing the commutativity classes of fc elements . [ propfc ] a heap @xmath50 is the heap of some fc element if and only if @xmath85 there is no convex chain @xmath86 in @xmath50 such that @xmath87 and @xmath88 where @xmath89 , and @xmath90 there is no chain covering relation @xmath91 in @xmath50 such that @xmath92 . now , we will focus on cfc elements . before this , we need to define a new operation on heaps . [ def ] let @xmath93 be a heap of a word @xmath94 . the cylindric transformation @xmath95 of @xmath50 is the labelling @xmath45 and a relation on the indices @xmath96 , made of the chain covering relations @xmath97 of @xmath50 , together with some new relations defined as follows : * for each generator @xmath24 , consider the minimal point a and the maximal point b in the chain @xmath98 ( for the partial order @xmath99 ) . if a is minimal and b is maximal in the poset @xmath50 , we add a new relation @xmath100 . * for each pair of generators ( s , t ) such that @xmath101 , consider the minimal point a and the maximal point b in the chain @xmath67 ( for the partial order @xmath99 ) . if these points have different labellings ( one has label s and the other has label t ) , we add a new relation @xmath100 . _ example : _ for the simply laced linear coxeter diagram with 7 generators , consider the word @xmath102 . its heap and its cylindric transformation are represented in figure [ excylindric ] . a heap and its cylindric transformation . ] @xmath103 is not a poset , even if we extend the relation @xmath104 by transitivity . indeed , each point in @xmath103 would otherwise be in relation with itself . as @xmath103 is not a poset , we can not draw its hasse diagram . we define the diagram of @xmath103 as the hasse diagram of @xmath50 , in which we add oriented edges representing the new relations described in definition [ def ] . let us explain the name cylindric transformation " that we decided to use . consider the coxeter system ( @xmath16 ) corresponding to the linear coxeter diagram @xmath105 of figure [ linear ] . the linear coxeter diagram . ] in this case , when we draw the diagram of @xmath103 , it should be seen as drawn on a cylinder instead of a plane as for hasse diagrams of heaps ( in our cylinder , each chain @xmath106 for a generator s can be seen as a circle ) . as for heaps , if @xmath103 is a cylindric transformation of a heap , the size @xmath107 is its cardinality . given any subset @xmath51 , we will write @xmath108 for the set of points in @xmath103 with labels in @xmath109 . unlike in the definition of @xmath52 , this definition does not include the relations @xmath97 between those points . we will now give the construction of the cylindric transformation of a heap in terms of words . [ propositioncfc ] if @xmath110 is a word and h its heap , let @xmath111 be the shortest prefix of @xmath31 such that all generators which appear in @xmath31 also appear in @xmath112 . we write @xmath113 where we denote by @xmath114 and @xmath115 . consider the heap @xmath116 of the word * w*. in @xmath116 we keep only the points and the chain covering relations , we identify the points @xmath59 and @xmath117 for @xmath118 , and we denote the resulting diagram by @xmath119 . then we have @xmath120 . note that @xmath119 define in proposition [ propositioncfc ] is not a poset , although @xmath121 is . for @xmath122 . ] we have @xmath123 . we will first prove that the chain covering relations @xmath97 in @xmath119 are included in those of @xmath103 . there are three types of chain covering relations in @xmath124 : * the ones between two points @xmath59 and @xmath60 such that @xmath125 . they correspond to chain covering relations which also appear in @xmath50 , and are therefore present in @xmath103 . * the ones between two points @xmath59 and @xmath60 such that @xmath126 . after the identification between @xmath59 ( _ resp . _ @xmath60 ) and @xmath127 ( _ resp . _ @xmath128 ) , these relations also appear in @xmath50 , and so in @xmath103 . * the ones between two points @xmath59 and @xmath60 such that @xmath129 and @xmath130 . by construction of @xmath119 , the point @xmath60 will be identified with the point @xmath131 which is minimal in @xmath132 ( and even minimal in @xmath50 if @xmath133 ) . likewise , the point i is maximal in @xmath134 ( and even maximal in @xmath50 if @xmath133 ) . so , when we construct @xmath103 , we will add a chain covering relation @xmath135 . thus the chain covering relations of @xmath119 are included in those of @xmath103 . the reverse inclusion can be proved by similar arguments . [ injectif ] the canonical mapping @xmath136 is not injective . more precisely , if the word @xmath137 is commutation equivalent to some cylic shift of the word @xmath138 , then @xmath139 . for @xmath24 a generator , @xmath140 a word , define @xmath141 and @xmath142 . we will prove that the chain covering relations of @xmath103 are included in those of @xmath143 . by applying this to the successive right cyclic shifts of @xmath144 , we may conclude that @xmath145 , which is sufficient to prove the proposition . + we denote by @xmath59 ( _ resp . _ @xmath146 ) the point of @xmath50 ( _ resp . _ @xmath119 ) corresponding to the first ( _ resp . _ last ) occurrence of s in @xmath147 ( _ resp . _ @xmath144 ) . there are three types of chain covering relations in @xmath103 : * the ones between two points different from @xmath59 ; they are also present in @xmath143 . * the ones between @xmath59 and another point @xmath60 . two cases can occur : @xmath61 in @xmath148 or not . in the first case , @xmath60 must be minimal among all points with the same label . as @xmath146 is maximal in @xmath119 , we add a relation @xmath149 in @xmath143 when we construct @xmath143 . the second case occurs when we add a relation between @xmath59 and @xmath60 during the construction of @xmath103 . in this case , @xmath60 is maximal in @xmath50 among all points with the same label and so there is a relation @xmath150 in @xmath119 . this relation stays naturally present in @xmath143 . * the ones between @xmath59 and itself . it is clear that @xmath146 is also in relation with itself in @xmath143 . now we characterize the cfc elements in terms of heaps . set @xmath50 a heap and @xmath24 , @xmath25 two generators such that @xmath23 . consider a chain @xmath151 in @xmath103 with @xmath152 , all elements @xmath153 distinct and such that @xmath154 and @xmath155 . + such a chain is called cylindric convex if the only elements @xmath156 , satisfying @xmath157 with all elements involved in this second chain distinct , are the elements @xmath84 of the first chain . [ remark1 ] assume there exists a convex chain @xmath151 of length @xmath158 in a heap @xmath50 , such that all elements @xmath153 are distinct and @xmath154 and @xmath155 . then this chain is also cylindric convex in the cylindric transformation @xmath103 . let @xmath151 be such a convex chain in @xmath50 and assume there is a chain @xmath157 with all elements distinct in @xmath103 , and @xmath159 . two cases can occur : * @xmath160 : as @xmath153 is not a maximal element in @xmath161 , the relation @xmath162 holds in @xmath50 . consequently , the relation @xmath163 holds in @xmath50 . so @xmath151 is not a convex chain in @xmath50 , which is a contradiction , * @xmath164 : as @xmath165 is not a minimal element in @xmath166 , the relation @xmath167 holds in @xmath50 . consequently , the relation @xmath168 holds in @xmath50 . so , as @xmath169 , @xmath151 is not a convex chain in @xmath50 , which is a contradiction . [ propcfc ] a heap @xmath50 is the heap of some word @xmath31 corresponding to a cfc element w if and only if both the following conditions are satisfied : * there is no cylindric convex chain @xmath170 in the cylindric transformation @xmath103 such that @xmath154 and @xmath155 , where @xmath89 ; * there is no chain covering relation @xmath91 in the cylindric transformation @xmath103 such that @xmath171 . let @xmath31 be a word and @xmath172 its heap . assume that @xmath31 is a reduced word of a non cfc element . there exists by definition [ defcfc ] a cyclic shift of @xmath31 which is commutation equivalent to @xmath173 or @xmath174 . let @xmath119 be the heap of @xmath175 . by proposition [ propfc ] , in the first case , @xmath119 contains a chain covering relation @xmath91 such that @xmath176 , and in the second case , @xmath119 contains a convex chain @xmath86 such that @xmath154 and @xmath155 , where @xmath89 . by lemma [ remark1 ] , these two cases give a chain covering relation and a cylindric convex chain in @xmath143 , respectively . but @xmath177 is commutation equivalent to a cyclic shift of @xmath31 , so @xmath178 , as seen in proposition [ injectif ] . + conversely , if @xmath103 contains the cylindric convex chain @xmath179 , let @xmath175 be the cyclic shift of @xmath31 beginning by @xmath180 , and let @xmath119 be its heap . as all elements in the cylindric convex chain are distinct , @xmath119 contains the convex chain @xmath181 . therefore @xmath119 is not fc by proposition [ propfc ] and @xmath50 is not cfc . the same argument also holds if @xmath103 contains a relation @xmath91 such that @xmath171 , by letting this time @xmath175 be the cyclic shift of @xmath31 beginning by @xmath182 . for example , this theorem ensures that the heap in figure [ excylindric ] does not correspond to a cfc element , as it contains a chain covering relation @xmath91 in the cylindric transformation @xmath103 such that @xmath183 . the example in figure [ excfc ] corresponds to a cfc element , according to theorem [ propcfc ] . a cfc element in @xmath184 . ] before ending this section , let us recall the set of alternating heaps which was defined in @xcite , and will be useful later . [ alternating ] in a coxeter group with linear diagram @xmath105 , a reduced word is alternating if for @xmath185 , the occurrences of @xmath186 alternate with those of @xmath187 . a heap is called alternating if it is the heap of an alternating word . if the coxeter group is @xmath5 ( see figure [ dynkina ] for the coxeter diagram ) , the diagram is not linear but we define the alternating word in the same way by setting @xmath188 . [ generatoronce ] the subset of alternating words defined as the set of words in which each generator occurs at most once is particular , in the sense that each element of this subset is a reduced expression for a cfc element ( we can not use any braid relation of length at least 3 ) . this fact justifies why this subset will be treated separately from other alternating words in the rest of the article . in this section , we will give a characterization and the enumeration of cfc elements in both types @xmath6 and @xmath191 . the corresponding coxeter diagrams are given in figure [ dynkina ] . and @xmath5.,title="fig : " ] and @xmath5.,title="fig : " ] + the elements of @xmath192 were enumerated in @xcite by using recurrence relations . our characterization in terms of heaps allows us to take into account the coxeter length . actually , we can compute the generating functions @xmath193 for @xmath194 and @xmath5 . in particular , when @xmath195 and @xmath194 , we get back the enumeration from @xcite ( recall that the number of cfc elements in type @xmath5 is infinite ) . our strategy is the following : first , we obtain a characterization of cfc elements in type @xmath5 , we deduce from it a characterization of cfc elements in type @xmath6 , then we derive the enumeration of cfc elements in type @xmath6 and deduce from it the enumeration in type @xmath5 . note that , in this type , the diagram of a cylindric transformation of a heap can be seen as drawn on a torus . [ thmantilde ] an element @xmath197 is cfc if and only if one ( equivalently , any ) of its reduced expressions @xmath31 verifies one of these conditions : * each generator occurs at most once in @xmath31 , * @xmath31 is an alternating word and @xmath198 . as said in remark [ generatoronce ] , if each generator occurs at most once in @xmath31 , then @xmath2 is a cfc element . so let @xmath31 be a reduced expression of a cfc element @xmath2 having a generator occuring at least twice in @xmath31 . recall that according to @xcite , @xmath199 is fully commutative if and only if @xmath31 is an alternating word . let @xmath200 be a generator which occurs at least twice in @xmath31 and such that for all @xmath201 . we will prove that @xmath202 where we set @xmath203 , which is sufficient to show that each generator occurs the same number of times . as @xmath31 is alternating , there are three possibilities : * @xmath204 . it contradicts the maximality of @xmath205 . * @xmath206 . by maximality of @xmath205 , there are two possibilities for @xmath207 : @xmath208 or @xmath209 . we obtain either a cylindric convex chain @xmath210 where @xmath211 and @xmath64 have label @xmath200 and @xmath212 has label @xmath213 , or a chain covering relation between two indices @xmath214 and @xmath215 with label @xmath200 , which allows us to conclude that @xmath2 is not a cfc element by theorem [ propcfc ] ( see the figure below , where we have circled the points @xmath211 , @xmath212 , @xmath64 , and @xmath214 , @xmath215 respectively ) . + * @xmath202 which is the needed condition . conversely , let @xmath31 be an alternating word such that @xmath198 . @xmath216 can not contain a cylindric convex chain @xmath217 of length 3 such that @xmath218 and @xmath219 ( _ resp . _ @xmath220 ) : indeed , the required condition on @xmath31 implies that there exists an index @xmath221 such that @xmath222 with @xmath223 ( _ resp . _ @xmath224 ) , which is a contradiction with the cylindric convexity of the chain . the same argument holds for chain covering relations involving indices with the same labellings . in this case , we will both characterize the cfc elements and compute the generating function @xmath225 we begin with a lemma which is a consequence of corollary 5.6 in @xcite . nevertheless , we will give a shorter proof using theorem [ thmantilde ] . [ lemmean]the cfc elements in type @xmath226 are those having reduced expressions in which each generator occurs at most once . let @xmath31 be a reduced word of a cfc element in type @xmath6 . by definition of the coxeter diagram ( cf figure [ dynkina ] ) , it is a reduced word of a cfc element in type @xmath5 in which the generator @xmath227 does not occur . according to theorem [ thmantilde ] , the only such cfc elements in type @xmath5 are those in which each generator occurs at most once . conversely , if all generators occur at most once in @xmath31 , remark [ generatoronce ] ensures that @xmath31 is a reduced expression of a cfc element . [ thman ] we have @xmath228 , @xmath229 and for @xmath230 , @xmath231 equivalently , we have the generating function : @xmath232 according to ( * ? ? ? * proposition 2.7 ) , fc elements in @xmath6 are in bijection with motzkin type paths of length @xmath233 , with starting and ending points at height 0 , where the horizontal steps are labeled either l or r ( and horizontal steps at height 0 are always labeled r ) . we recall the bijection , which is defined as follows : let @xmath2 be a fc element in @xmath6 , and @xmath50 its heap . to each @xmath234 , we associate a point @xmath235 . as @xmath2 is alternating , three cases can occur : * @xmath236 , corresponding to an ascending step , * @xmath237 , corresponding to a descending step , * @xmath238 , corresponding to an horizontal step , labelled by r if @xmath186 occurs before @xmath187 in @xmath2 , l otherwise . according to lemma [ lemmean ] , the restriction of this bijection to cfc elements is a bijection between cfc elements and the previous motzkin paths , having length @xmath233 , whose height does not exceed 1 . by taking into account the first return to the @xmath211-axis , we obtain the following recurrence relation for @xmath230 : @xmath239 where we write @xmath240 ( which fits with @xmath228 , @xmath229 and the expected recurrence relation ) . rewriting ( [ equationa ] ) with @xmath233 replaced by @xmath241 and combining with ( [ equationa ] ) allows us to eliminate the sum over @xmath242 , and leads to the recurrence relation ( [ equationa2 ] ) . classical techniques in generating function theory and the values @xmath228 , @xmath229 enable us to compute the desired generating function . notice that , as expected , if @xmath243 , we find the odd - index fibonacci numbers generating function of @xcite . this @xmath215-analog was already known , see sequence a105306 in @xcite . we have for @xmath230 @xmath244 where @xmath245 is a polynomial of degree n satisfying for n@xmath246 @xmath247 with @xmath248 , @xmath249 , and @xmath250 . moreover , we can compute the generating function : @xmath251 therefore the coefficients of @xmath252 are ultimately periodic of exact period n , and the periodicity starts at length n. in the same way as for the finite type @xmath6 ( see @xcite ) , fc elements in @xmath5 are in bijection with motzkin type paths of length n satisfying the following conditions : * the starting point @xmath253 and the ending point @xmath254 have the same height , * horizontal steps at height 0 are always labeled r , * if the path contains only horizontal steps at height @xmath255 1 , the steps must have not all the same labelling . the construction of the path corresponding to a fc element is the same as in type @xmath190 if we set @xmath203 . in theorem [ thmantilde ] , the alternating cfc elements in which all generators occur at least twice and in the same number correspond to motzkin type paths which have only horizontal steps . therefore there are @xmath256 such paths for all fixed starting height @xmath257 . it leads to the generating function @xmath258 which can be summed to obtain the second term of the right hand side of ( [ equationatilde ] ) . in theorem [ thmantilde ] , the cfc elements which correspond to reduced expressions with at most one occurrence of each generator correspond to motzkin type paths which stay at height @xmath259 . we denote by @xmath245 the generating function of such elements . let @xmath59 ( _ resp . _ @xmath60 ) be the first ( _ resp . _ last ) return to the x - axis ( see figure [ motzkinpathsatilde ] for an example ) . . ] we get @xmath260 where the first term counts paths which stay at height 1 , the second term counts paths such that @xmath261 , and the last term counts paths such that @xmath262 and @xmath263 . rewriting ( [ equationatilde2 ] ) with @xmath241 replaced by @xmath264 , combining with ( [ equationatilde2 ] ) and using the relation in theorem [ thman ] allow us to eliminate the double sum , and leads to the following relation : @xmath265 from this and , it is easy to derive the expected relation and the generating function p. as @xmath245 is a polynomial in q , the periodicity is clear , and its beginning comes from the fact that the leading coefficient of @xmath245 is @xmath266 and @xmath267p_{n-1}(q ) \neq 0 $ ] , where @xmath267p_{n-1}(q)$ ] is the coefficient of @xmath268 in @xmath245 . the situation in type @xmath5 is very different from all other types that we study , as we will see later : this is the only case where the generating series @xmath269 of cfc elements which correspond to reduced expression with at most one occurrence of each generator do not satisfy the recurrence relation @xmath270 . there are three other infinite families of affine coxeter groups , they correspond to types @xmath273 all these groups contain an infinite number of cfc elements . in any of these cases , using theorem [ propcfc ] , we are able to characterize the cfc elements and to compute the generating function @xmath274 . we also show that this generating function is always ultimately periodic . as a corollary , we deduce a characterization and the enumeration of cfc elements in finite types @xmath275 and @xmath272 . the coxeter diagram of type @xmath278 is represented below . . ] the situation is more complicated than in type @xmath196 , but we obtain an explicit characterization of cfc elements . [ thmctilde]an element @xmath279 is cfc if and only if one ( equivalently any ) of its reduced expressions @xmath31 verifies one of the three conditions : * each generator occurs at most once in @xmath31 , * @xmath31 is an alternating word and @xmath280 , * @xmath31 is a subword of the infinite periodic word @xmath281 , where @xmath282 and @xmath283 ( i.e we have @xmath284 or @xmath285 , where @xmath286 and @xmath287 ) . let @xmath2 be a cfc element in @xmath288 and @xmath31 be one of its reduced expressions . we denote by @xmath50 the heap of * w * and @xmath103 its cylindric transformation . in ( * theorem 3.4 ) , fc elements are classified in five families , the first corresponding to alternating elements . as before , we distinguish two cases for elements in this first family : * each generator occurs at most once in @xmath31 . according to remark [ generatoronce ] , @xmath2 is cfc as no long braid relations can be applied . these elements satisfy ( a ) . * @xmath31 is an alternating word in which a generator occurs at least twice . in this case , the proof will essentially be the same as for alternating words in type @xmath5 . we write @xmath289 and @xmath290 . let @xmath291\ ! ] $ ] be a maximal interval such that @xmath292 and @xmath293 @xmath294 . assume @xmath295 , by maximality of @xmath296 and the fact that @xmath31 is alternating , we have @xmath297 and there are two possibilities in @xmath103 : @xmath298 or @xmath299 . we obtain either a cylindric convex chain @xmath300 of length 4 where @xmath301 , @xmath212 both have label @xmath302 and @xmath211 , @xmath64 both have label @xmath303 , or a chain covering relation between two indices @xmath214 and @xmath215 with label @xmath302 , which allows us to conclude that @xmath2 is not cfc by theorem [ propcfc ] ( see figure below , in each case , we have circled the points @xmath301 , @xmath211 , @xmath212 , @xmath64 and @xmath215 , @xmath304 respectively ) . + + so , @xmath305 has to be equal to @xmath233 if @xmath2 is cfc . the same argument for @xmath153 will lead to @xmath306 , and so @xmath307 . these elements satisfy ( b ) . next , we will show that among the four remaining families from @xcite , the cfc elements must satisfy ( c ) . there are two cases : * @xmath2 belongs to the family called zigzag , i.e @xmath31 is a subword of the infinite periodic word @xmath308 with at least one generator which occurs more than three times ( actually , this condition will not be used in this proof ) . denote by @xmath186 ( _ resp . _ @xmath200 ) the first ( _ resp . _ last ) letter of @xmath31 . we assume that the second letter is @xmath309 ( the case where the second letter is @xmath187 is symmetric and can be treated similarly ) . + if @xmath310 , @xmath103 contains a cylindric convex chain of length 3 involving points in @xmath311 or a chain covering relation between two points with label @xmath186 . + if @xmath200 = @xmath309 , then the last but one letter in @xmath31 is either @xmath186 or @xmath312 , so @xmath103 takes one of the two forms in figure [ patternctilde4 ] . + the two possible heaps for @xmath313 . ] + + in the first case , @xmath103 contains a cylindric convex chain involving points in @xmath311 of length 4 , given by the two first and the two last letters of @xmath31 ( the corresponding points are circled in figure [ patternctilde4 ] , left ) . by theorem [ propcfc ] , @xmath31 is not cfc . in the second case , @xmath103 contains a cylindric convex chain involving points in @xmath311 of length 3 , given by the two first and the last letters of @xmath31 ( they are circled points in figure [ patternctilde4 ] , right ) . by theorem [ propcfc ] , @xmath31 is not cfc , unless @xmath314 , and therefore @xmath31 satisfies ( c ) . + if @xmath200 = @xmath187 , then the last but one letter in @xmath31 is either @xmath186 or @xmath315 , so @xmath103 takes one of the two forms in figure [ patternctilde5 ] . + the two possible heaps for @xmath316 . ] + + in the first case , @xmath103 contains a cylindric convex chain involving points in @xmath317 , of length 3 , given by the first and the two last letters of @xmath31 ( the corresponding points are circled in figure [ patternctilde5 ] , left ) . by theorem [ propcfc ] , @xmath31 is not cfc , unless @xmath318 , which is not possible because the second letter of @xmath31 is @xmath309 , or unless @xmath319 , and therefore @xmath31 satisfies ( c ) . in the second case , @xmath31 satisfies ( c ) . * @xmath2 belongs to one of the three remaining families , called right peak , left peak and left - right peak . in this case , it is easy to find either a chain covering relation @xmath320 such that @xmath214 and @xmath215 have the same label or a cylindric convex chain in @xmath103 , except if @xmath2 satisfies ( c ) ( and has length @xmath321 ) . conversely , any @xmath2 satisfying one of the three conditions ( a ) , ( b ) , or ( c ) is a cfc element . indeed , @xmath103 contains neither a relation @xmath91 such that i and j have the same label nor a cylindric convex chain of length 3 , except involving points in @xmath322 or @xmath323 which do not contain any cylindric convex chain of length 4 : for @xmath2 satisfying ( a ) or ( c ) , this is trivial , and for @xmath2 satisfying ( b ) , the same proof as in type @xmath189 holds ( see theorem [ thmantilde ] ) . we have the generating function @xmath324 the coefficients of @xmath325 are ultimately periodic of exact period n(n+1 ) if n is odd , and 2n(n+1 ) if n is even . moreover , periodicity starts at length n. notice that the sets of elements satisfying condition ( a ) , ( b ) or ( c ) of theorem [ thmctilde ] are disjoint . the first term in ( [ equationctilde ] ) corresponds to the set of cfc elements satisfying ( a ) . this set is clearly in bijection with @xmath326 . + now , we show that the second term in ( [ equationctilde ] ) corresponds to cfc elements satisfying ( b ) . if we fix an integer @xmath257 , there are @xmath327 such elements satisfying @xmath328 . indeed , for each generator @xmath329 , there are two possibilities : @xmath186 appears before or after @xmath187 ( and @xmath25 appears before or after @xmath330 ) . the generating function of elements satisfying ( b ) is then equal to @xmath331 . + with the condition on the number of occurrences of each generator , we can see that the length of elements satisfying ( c ) must be a multiple of @xmath321 , and there are @xmath321 such elements for each possible length ( the first two letters in @xmath31 can be @xmath332 or @xmath333 for all @xmath334\!],$ ] or @xmath335 or @xmath336 . this gives the term @xmath337 . + @xmath338 is a polynomial in q , and the two other terms in ( [ equationctilde ] ) have periodic coefficients , so @xmath325 is ultimately periodic . the period is the least common multiple of @xmath339 and @xmath321 , which is @xmath340 if @xmath233 is odd , and @xmath341 if @xmath233 is even . as the leading coefficient of @xmath338 is @xmath342 and @xmath343 a_{n+1}^{cfc}(q ) \neq 0 $ ] , the beginning of the periodicity is @xmath233 . we can deduce the characterization in type @xmath275 , whose corresponding coxeter diagram is recalled below . . ] the cfc elements in type @xmath344 are those having reduced expressions in which each generator occurs at most once . moreover , @xmath345 let @xmath2 be a cfc element in type @xmath346 , and @xmath31 one of its reduced expression . consequently , @xmath2 is a cfc element in type @xmath347 in which the generator @xmath82 does not occur . but , according to theorem [ thmantilde ] , the only such cfc elements are those satisfying ( a ) ( and having no u ) : if @xmath2 satisfies ( b ) ( _ resp . _ ( c ) ) , all generators appear the same number of times ( _ resp . _ u must appear in @xmath31 ) . conversely , if all generators occur at most once in @xmath31 , we already saw in remark [ generatoronce ] that @xmath2 is a cfc element . we will obtain once again a characterization of cfc elements in type @xmath350 , deduce a characterization and the enumeration in type @xmath349 , and finally deduce the enumeration in type @xmath350 . in what follows , by @xmath31 is an alternating word " , we mean that if we replace @xmath351 and @xmath352 by @xmath227 , and @xmath82 by @xmath353 , @xmath31 is an alternating word in the sense of definition [ alternating ] , and @xmath351 and @xmath352 alternate in @xmath31 . and @xmath349.,title="fig : " ] and @xmath349.,title="fig : " ] [ thmbtilde ] an element @xmath354 is cfc if and only if one ( equivalently any ) of its reduced expressions @xmath31 verifies one of these conditions : * each generator occurs at most once in @xmath31 , * @xmath31 is an alternating word , @xmath355 , and @xmath356 ( in particular , @xmath357 is even ) , * @xmath31 is a subword of @xmath358 , which is an infinite periodic word , where @xmath351 and @xmath352 are allowed to commute , such that @xmath359 , and @xmath360 ( i.e @xmath31 takes one of the five forms : @xmath361 or @xmath285 or @xmath362 or @xmath363 or @xmath364 ) . the steps of this proof are the same as in theorem [ thmctilde ] . let @xmath2 be a cfc element in @xmath365 and let @xmath31 be one of its reduced expressions . we denote by @xmath50 the heap of @xmath31 and @xmath103 its cylindric transformation . in ( * theorem 3.10 ) , fc elements are classified in five families . the first corresponds to alternating elements . as before , we distinguish two cases : * each generator occurs at most once in @xmath31 . these elements satisfy ( a ) . * @xmath31 comes from an alternating word of type @xmath366 , where we applied the replacements @xmath367 or @xmath368 . the same proof as for type @xmath369 gives @xmath370 . if @xmath371 is odd , after replacement we have the subword of @xmath372 where we keep only the generators @xmath373 and @xmath24 is @xmath374 ( or @xmath375 ) , which prevents @xmath2 from being cfc . so @xmath371 is even , and as @xmath31 is alternating , @xmath376 . these elements satisfy ( b ) . next , we will show that among the four remaining families from @xcite , the cfc elements must satisfy ( c ) . there are two possibilities : * @xmath31 is a subword of @xmath358 , which is an infinite periodic word , where @xmath351 and @xmath352 are allowed to commute , such that a generator occurs more than twice . the same case distinction as in type @xmath369 leads us to the required condition ( c ) . * @xmath47 is a heap among special cases analoguous to the ones in type @xmath366 which are trivially non cfc , except for those which satisfy ( c ) ( and are of length @xmath377 ) . conversely , all elements @xmath2 satisfying one of the three conditions ( a ) , ( b ) or ( c ) are cfc elements . indeed , @xmath103 contains neither relations @xmath41 such that @xmath59 and @xmath60 have the same label nor cylindric convex chains of length 3 , except involving points in @xmath378 , and these points can not form a cylindric convex chain of length 4 . the cfc elements in type @xmath379 are those having reduced expressions in which each generator occurs at most once . moreover , we have @xmath380 , @xmath381 , @xmath382 and for @xmath230 : @xmath383 in other words , we have @xmath384 let @xmath385 be a cfc element of type @xmath349 and let @xmath31 be one of its reduced expressions . consequently , @xmath2 is a cfc element of type @xmath350 in which the generator @xmath82 does not occur . but , according to theorem [ thmbtilde ] , the only such cfc elements are those satisfying condition ( a ) ( and having no u ) : if @xmath31 satisfies ( b ) or ( c ) , u must appear in @xmath31 . conversely , if all generators occur at most once in @xmath31 , we already saw that @xmath2 is cfc . + so , any cfc element of type @xmath349 can be otained from a cfc element @xmath386 in @xmath6 by adding to @xmath386 nothing , or one occurrence of @xmath351 ( with two choices : before or after @xmath330 ) , or one occurrence of @xmath352 ( with two choices ) , or one occurrence of @xmath351 and @xmath352 ( with four choices ) . if @xmath330 does not occur in @xmath386 , we have no choice for adding @xmath351 or / and @xmath352 as @xmath351 and @xmath352 commute with all other generators . this leads to the following recurrence relation : @xmath387 with this relation and ( [ equationa2 ] ) in theorem [ thman ] , it is easy to derive ( [ equationd ] ) . the generating function @xmath388 is computed through classical methods . we have the generating function @xmath389 furthermore , the coefficients of @xmath390 are ultimately periodic of exact period + 2(n+1)(2n+1 ) and the periodicity starts at length @xmath391 . notice that the sets of elements satisfying condition ( a ) , ( b ) or ( c ) of theorem [ thmbtilde ] are disjoint . the first term in ( [ equationbtilde ] ) corresponds to the set of cfc elements satisfying ( a ) . this set is clearly in bijection with @xmath392 . + the second term in ( [ equationbtilde ] ) corresponds to cfc elements satisfying ( b ) , which have length a multiple of @xmath393 , and there are @xmath394 such elements for each possible length . + with the condition on the number of occurrences of each generator , we can see that the length of elements satisfying ( c ) must be a multiple of @xmath395 , and there are @xmath396 such elements for each possible length ( the first two letters can be @xmath397 , @xmath398 , @xmath399 , @xmath400 , @xmath401 or @xmath402 ) . this gives the third term of . + moreover , @xmath403 is a polynomial in q , and the two other terms of ( [ equationbtilde ] ) are periodic , so @xmath390 is ultimately periodic . the period is the least common multiple of @xmath393 and @xmath395 , which is @xmath404 . the beginning of the periodicity is @xmath391 , as the leading coefficient of @xmath403 is @xmath405 and @xmath406d_{n+2}^{cfc}(q ) \neq 0 $ ] . the situation is very similar in type @xmath407 for the characterization , but the generating function takes a slightly different form , due to the specificity of the coxeter diagram . . ] in what follows , by @xmath31 is an alternating word " , we mean that if we replace @xmath351 and @xmath352 by @xmath227 , and @xmath408 and @xmath409 by @xmath353 , the image of @xmath372 is an alternating word in the sense of definition [ alternating ] , @xmath351 and @xmath352 alternate in @xmath2 , and @xmath408 and @xmath409 alternate in @xmath2 . [ thmdtilde ] an element @xmath410 is cfc if and only if one ( equivalently any ) of its reduced expressions * w * verifies one of these conditions : * each generator occurs at most once in @xmath140 , * @xmath31 is an alternating word , @xmath282 , and @xmath411 ( in particular , @xmath357 is even ) , * @xmath31 is a subword of @xmath412 , which is an infinite periodic word , where @xmath351 and @xmath352 , @xmath408 and @xmath409 are allowed to commute , such that @xmath359 , and @xmath413 . the same proof as for theorem [ thmbtilde ] holds , one only needs to add the replacements @xmath414 @xmath415 or @xmath416 . again , we can compute the corresponding generating function . we have the generating function @xmath417 where @xmath418 is a polynomial in q of degree @xmath419 such that @xmath420 , @xmath421 , and for @xmath422 : @xmath423 moreover , the coefficients of @xmath424 are ultimately periodic of exact period 2(n+1 ) , and the periodicity starts at length @xmath391 . note that the generating function @xmath425 of the polynomials @xmath426 is computable through classical techniques . however it does not have a nice expression . notice that the sets of elements satisfying condition ( a ) , ( b ) or ( c ) of theorem [ thmdtilde ] are disjoint . the first term in ( [ equationdtilde ] ) corresponds to the set of cfc elements satisfying ( a ) . any such element can be otained from a cfc element @xmath2 of type @xmath349 by adding nothing , or one occurrence of @xmath408 ( with two choices : before or after @xmath427 ) , or one occurrence of @xmath409 ( with two choices ) , or one occurrence of @xmath408 and @xmath409 ( with four choices ) . if @xmath427 does not occur in @xmath2 , we have no choice for adding @xmath408 or / and @xmath409 , as @xmath408 and @xmath409 commute with all other generators . this leads to the following recurrence relation : @xmath428 using this and , we find the expected recurrence relation for the polynomials @xmath429 . elements satisfying ( b ) have length a multiple of @xmath393 and there are @xmath430 of them . elements satisfying ( c ) have also length a multiple of @xmath393 , and there are @xmath377 of them ( by inspecting the first two letters ) . the property of ultimate periodicity is clear . exceptional finite types are @xmath431 @xmath432 . enumerating cfc elements according to the length in these two last ones is trivial , while other groups are special cases of the families @xmath433 ( @xmath434 ) , @xmath435 ( @xmath436 ) , @xmath437 ( @xmath230 ) . it is shown in @xcite that these families contain a finite number of cfc elements . it is possible to apply our method to characterize them in terms of cylindric heaps and obtain recurrence relations for their generating functions according to the length . however , we leave this to the interested reader . for exceptional affine types having a finite number of cfc elements ( @xmath438 and @xmath439 ) , generating functions of cfc elements are polynomials recursively computable as in type @xmath190 , because @xmath440 and @xmath441 . it remains to study three exceptional affine types having an infinite number of cfc elements . their coxeter diagrams are represented below . for these types , we only give the result without detailing the proof . for the interested readers , we just sketch it : as for other affine types , we look at the classification of fc elements in ( * ? ? ? * lemmas 5.25.4 ) , and see that there exists cfc elements only for some explicit length , which must be either bounded or an integer multiple of a constant depending on the considered type . , @xmath442 and @xmath443.,title="fig : " ] , @xmath442 and @xmath443.,title="fig : " ] , @xmath442 and @xmath443.,title="fig : " ] let @xmath444 ( _ resp . _ @xmath445 , _ resp . _ @xmath446 ) be an exceptional affine coxeter group . we define in @xmath0 the element @xmath447 as @xmath448 ( _ resp . _ @xmath449 , _ resp . _ @xmath450 ) . an element @xmath451 is cfc if and only if one ( equivalently any ) of its reduced expressions @xmath31 verifies one of these conditions : * @xmath31 belongs to a finite set depending on @xmath0 , * @xmath452 , where n is a non - negative integer , @xmath453 and @xmath454 . consequently , the following generating functions hold : @xmath455 @xmath456 @xmath457 where @xmath458 and @xmath459 are polynomials . here we study some particular elements of a general coxeter group @xmath0 . recall that for @xmath460 , the support @xmath461 of @xmath2 is the set of generators that appear in a ( equivalently , any ) reduced expression of @xmath2 . it is well known that if @xmath462 and @xmath4 is a positive integer , then @xmath463 . if equality holds for all @xmath464 ( @xmath465 @xmath466 is reduced for all k ) , we say that @xmath385 is logarithmic ( see @xcite for more information about logarithmic elements ) . for example , it is trivial that there is no logarithmic element in finite coxeter groups , because there is a finite number of reduced expressions . by using our characterizations for affine types in theorems [ thmantilde ] , [ thmbtilde ] , [ thmctilde ] , [ thmdtilde ] , we derive the following consequence . a generalization of this result was proved for all coxeter groups in ( * ? ? ? * corollary e ) , by using geometric group theoretic methods . nevertheless , as our approach is different and more combinatorial , we find interesting to give our proof , although it only works for affine types . for @xmath0=@xmath467 or @xmath468 , if @xmath2 is a cfc element , @xmath2 is logarithmic if and only if a ( equivalently , any ) reduced expression @xmath372 of @xmath2 has full support ( i.e all generators occur in @xmath372 ) . in particular , there is a finite number of cfc elements which are not logarithmic . let @xmath2 be in @xmath37 and let @xmath31 be one of its reduced expressions . assume that @xmath372 has not full support . in this case , @xmath2 belongs to a proper parabolic subgroup of @xmath0 , which is a finite coxeter group by a classical property of affine coxeter groups ( see @xcite ) . so @xmath2 is not logarithmic , and there is a finite number of such elements , which correspond to elements satisfying ( a ) in theorems [ thmantilde ] , [ thmbtilde ] , [ thmctilde ] , [ thmdtilde ] and such that at least one generator does not occur in @xmath31 . + conversely , assume that @xmath2 is a cfc element with full support . according to theorems [ thmantilde ] , [ thmbtilde ] , [ thmctilde ] , [ thmdtilde ] , @xmath372 must satisfy ( b ) or ( c ) or has to be a coxeter element ( @xmath465 each generator occurs exactly once in * w * ) . if @xmath2 is a coxeter element , @xmath2 is logarithmic ( see @xcite ) . if @xmath31 satisfies ( b ) ( _ resp . _ ( c ) ) , it is easy to verify that @xmath469 also satisfies ( b ) ( _ resp . _ ( c ) ) and is therefore reduced . it follows that @xmath2 is logarithmic . we can also notice that the powers of coxeter elements are always cfc in affine types @xmath5 and @xmath470 because they satisfy the alternating word condition , but are never cfc elements in types @xmath350 and @xmath8 ( because they satisfy neither the alternating word condition nor condition ( c ) ) . a natural question that arises in the study of fc elements is to compute the number of fc involutions in finite and affine coxeter groups ( as this number is , for some groups , the sum of the dimensions of irreducible representations of a natural quotient of the iwahori - hecke algebra associated to the group , see @xcite ) . similarly , we now focus on cfc elements which are involutions . the main result is that there is always a finite number of such elements in all coxeter groups , and we are able to characterize them in terms of words . we also use the characterization of cfc elements to enumerate cfc involutions in finite and affine coxeter groups . [ finiteness ] let @xmath0 be a coxeter group and let i(@xmath0 ) be its subset of involutions . the set @xmath471i(@xmath0 ) is finite . moreover an element @xmath2 belongs to @xmath471i(@xmath0 ) if and only if for any generator s in @xmath461 , @xmath472 and for all t such that @xmath23 , @xmath473 ( i.e two non commuting generators can not occur in @xmath2 ) . let @xmath2 be a cfc involution , let s be a generator in @xmath461 and @xmath31 be a reduced expression of @xmath2 . assume that @xmath474 . consider a cyclic shift @xmath475 of @xmath31 which begins with s. as @xmath2 and s are involutions , @xmath175 is an expression of an involution . moreover as @xmath2 is cfc , @xmath175 corresponds to a fc element @xmath386 . according to @xcite , a fc element @xmath2 is an involution if and only if r(@xmath2 ) is palindromic , i.e r(@xmath2 ) includes the reverse of all of its members . applying this to @xmath175 allows us to say that @xmath175 is commutation equivalent to a word @xmath476 . so a cyclic shift of @xmath31 is commutation equivalent to @xmath477 , which is in contradiction with the cfc property of @xmath2 . therefore we have @xmath478 . we consider @xmath175 as before . as r(@xmath386 ) is palindromic , we can conclude that all generators in @xmath479=@xmath461 commute with s. [ remarkinv ] the number of cfc involutions in a coxeter group @xmath0 depends only on the edges of the coxeter diagram , without taking into account the values @xmath18 . let us enumerate the cfc involutions in classical types , according to their coxeter length . if @xmath0 is a coxeter group , we define @xmath480 . [ ainvo ] in types @xmath190 , @xmath275 , and @xmath481 the following relation holds for all @xmath482 : @xmath483 and @xmath484 , @xmath485 . moreover , we can compute the generating function @xmath486 the equality @xmath487 comes from remark [ remarkinv ] . let @xmath2 be a cfc involution in @xmath488 . if @xmath489 belongs to @xmath461 , by theorem [ finiteness ] , @xmath427 does not belong to @xmath490(@xmath2 ) , and @xmath2 is equal to @xmath489 concatenated to a cfc involution of @xmath491 . if @xmath489 does not belong to @xmath490(@xmath2 ) , @xmath2 is a cfc involution of @xmath6 . this yields the expected relation . the generating function is computed through classical techniques . in particular , if @xmath492 , we obtain the number of cfc involutions in type @xmath6 , which is the @xmath493 fibonacci number . in types d and @xmath494 , the following relations hold for all @xmath495 : @xmath496 @xmath497 and @xmath498 , @xmath499 , @xmath500 . moreover , we can compute the generating function @xmath501 to prove , let @xmath2 be a cfc involution in @xmath349 . if @xmath502 belongs to @xmath490(@xmath2 ) , by theorem [ finiteness ] , @xmath503 do not belong to @xmath490(@xmath2 ) , and @xmath2 is equal to @xmath502 concatenated to a cfc involution of @xmath504 . if @xmath502 does not belong to @xmath490(@xmath2 ) , @xmath2 is a cfc involution of @xmath6 concatenated to @xmath351 , @xmath352 , @xmath400 or nothing . this shows . equation is simply a consequence of , and remark [ remarkinv ] . the generating function comes from classical techniques . if @xmath492 , we obtain the number of cfc involutions in type @xmath349 , which leads to a fibonacci - type sequence with starting numbers 1 and 4 . in type @xmath189 , the following relations hold for @xmath505 : @xmath506 where @xmath507 , @xmath508 . therefore , we can compute the generating function @xmath509 we also have @xmath510 where @xmath511 is the @xmath512 lucas polynomials , defined explicitely ( see sequence a114525 in @xcite ) by @xmath513 $ ] . this equality can be proved by using generating functions . let @xmath2 be a cfc involution in @xmath5 . if @xmath514 belongs to @xmath490(@xmath2 ) , by theorem [ finiteness ] , @xmath427 and @xmath330 do not belong to @xmath490(@xmath2 ) , and @xmath2 is equal to @xmath514 concatenated to a cfc involution of @xmath504 . if @xmath514 does not belong to @xmath490(@xmath2 ) , @xmath2 is a cfc involution of @xmath6 . this leads to the relation:@xmath515 using this and ( [ equationinva ] ) , it is easy to deduce ( [ equationinvatilde ] ) . the generating function comes from classical techniques . in type @xmath516 , the following relation holds for @xmath422 : @xmath517 with @xmath518 and @xmath519 . we can therefore compute the generating function : @xmath520 let @xmath2 be a cfc involution in @xmath8 . if @xmath502 belongs to @xmath490(@xmath2 ) , by theorem [ finiteness ] , @xmath521 , @xmath351 and @xmath352 do not belong to @xmath490(@xmath2 ) , and @xmath2 is equal to @xmath502 concatenated to a cfc involution of @xmath522 . if @xmath502 does not belong to @xmath490(@xmath2 ) , @xmath2 is a cfc involution @xmath523 concatenated to @xmath351 , @xmath352 , @xmath400 or nothing . this leads to the relation:@xmath524 use this and ( [ equationinvd ] ) to derive ( [ equationinvdtilde ] ) . if @xmath492 , we obtain the number of involutions in type @xmath8 , which is a fibonacci - type sequence with starting numbers 10 and 7 . in sections [ finite ] and [ affine ] , we obtained two @xmath215-analogs of the number of cfc elements in finite types . the first , in type @xmath190 , corresponds to permutations avoiding 321 and 3412 , taking into account their coxeter length ( see @xcite ) . the second , in type @xmath272 , is apparently new . one may wonder if it corresponds to another combinatorial object . + we can notice that if @xmath0 is a finite or affine coxeter group , the generating function of cfc elements is rational . this is in fact true for all coxeter groups , as will be proved in the forthcoming work @xcite . we would like to thank m. macauley for pointing out reference @xcite on logarithmic elements , and t. marquis for valuable remarks on the infinite number of non logarithmic cfc elements . finally , we also thank p. nadeau for interesting comments and discussions on an earlier version of this manuscript .

the aim of this note is to give a geometric proof for classical local rigidity of lattices in semisimple lie groups . we are reproving well known results in a more geometric ( and hopefully clearer ) way .

1702.00342

let @xmath0 be a semisimple lie group , and let @xmath1 be an irreducible lattice . we denote by @xmath2 the space of deformations of @xmath3 in @xmath0 , i.e. the space of all homomorphisms @xmath4 with the topology of pointwise convergence . recall that @xmath3 is called _ locally rigid _ if there is a neighborhood @xmath5 of the inclusion map @xmath6 in @xmath2 such that any @xmath7 is conjugate to @xmath8 . in other words , this means that if @xmath3 is generated by the finite set @xmath9 , then there is an identity neighborhood @xmath10 such that if @xmath11 is a homomorphism such that @xmath12 for any @xmath13 , then there is some @xmath14 for which @xmath15 for any @xmath16 ( such @xmath17 is called a trivial deformation ) . local rigidity was first proved by selberg @xcite for uniform lattices in the case @xmath18 , and by calabi @xcite for uniform lattices in the case @xmath19 , @xmath20 . then , weil @xcite generalized these results to any uniform irreducible lattice in any @xmath0 , assuming that @xmath0 is not locally isomorphic to @xmath21 ( in which case lattices have many non - trivial deformations - the teichmuller spaces ) . later , garland and raghunathan @xcite proved local rigidity for non - uniform lattices @xmath1 when @xmath22 will always mean the real rank of @xmath0 . ] under the necessary assumption that @xmath0 is neither locally isomorphic to @xmath21 nor to @xmath23 . for non - uniform irreducible lattices in higher rank semisimple lie groups , the local rigidity is a consequence of the much stronger property : super - rigidity , which was proved by margulis @xcite . our aim here , is to present a simple proof of the following theorem : [ thm1 ] under the necessary conditions on @xmath0 and @xmath3 , if @xmath17 is a deformation , close enough to the identity @xmath8 , then @xmath24 is a lattice and @xmath17 is an isomorphism . in the higher rank case , we need the lattice to be arithmetic , which is always true by margulis theorem . but this is of course cheating . anyway our method provides , for example , a simple geometric proof of the local rigidity of @xmath25 in @xmath26 for @xmath20 . note also that margulis arithmeticity theorem is easier in the non compact case @xcite ( it does not rely on super - rigidity ) which is the only case we need here . in fact , we prove a `` stronger '' result than the statement of theorem [ thm1 ] . namely , if @xmath27 is the associated symmetric space , then for a small deformation @xmath17 , @xmath28 is homeomorphic to @xmath29 and @xmath30 is an isomorphism . with theorem [ thm1 ] , local rigidity follows from mostow rigidity . recall that mostow s rigidity theorem @xcite says that if @xmath0 is a connected semisimple lie group not locally isomorphic to @xmath21 , and @xmath31 are isomorphic irreducible lattices in @xmath0 , then they are conjugate in @xmath0 , i.e. @xmath32 for some @xmath14 . in selberg s paper @xcite a main difficulty was to prove a weaker version of [ thm1 ] in his special case . selberg also indicated that his proof ( which is quite elementary and very elegant ) would work for more general higher rank cases if the general version ( [ thm1 ] above ) of his lemma 9 could be proved . this supports the general feeling that it should be possible to give a complete elementary proof of local rigidity which does not rely on mostow s rigidity . however , the authors decided not to strain too much , in trying to avoid mostow rigidity , since in the rank one case there are by now very elegant and straightforward proofs of mostow rigidity ( such as the proof of gromov and thurston ( see @xcite ) for compact hyperbolic manifolds , and of besson , courtois and gallot @xcite for the general rank one case , note also that mostow s original proof is much simpler in the rank one case ) while for higher rank spaces the much stronger margulis supper - rigidity holds . our proof of theorem [ thm1 ] relies on an old principle implicit in ehresmann s work and to our knowledge first made explicit in thurston s notes . this principle implies in particular that any sufficiently small deformation of a cocompact lattice in a lie group stays a lattice . it seems to have remained unknown for many years , as both selberg and weil spent some effort in proving partial cases of it . it is by now fairly well known , but the authors could not find a complete written proof of it in the literature although everyone knows that it is implicit in ehresmann s work . in the second section of this paper , we thus decided to write a complete proof , using ehresmann s beautiful viewpoint , of a slight generalization of this principle to the non - compact case needed for our purpose . in the third section we prove theorem [ thm1 ] in the rank one case . in the last section we prove theorem [ thm1 ] for arithmetic lattices of higher rank . non - uniform lattices in the group @xmath33 are not locally rigid , although mostow rigidity is valid for @xmath34 . this shows that local rigidity is not a straightforward consequence of mostow rigidity . moreover , it was shown by thurston that if @xmath35 is a non - uniform lattice then there is an infinite family of small deformations @xmath36 of @xmath3 such that @xmath37 is again a lattice , but not isomorphic to @xmath3 , and in fact many of the @xmath37 s may be constructed to be uniform . the proof which we present in this note is in a sense a product of our attempt to understand and formulate why similar phenomenon can not happen in higher dimensions . in fact , we shall show that some weak version of local rigidity is valid also for non - uniform lattices in @xmath34 . this would allow us to obtain a direct proof of the fact that for any ( non - uniform ) lattice in @xmath38 ( as well as in any connected @xmath0 which is defined over @xmath39 as an algebraic group and is locally isomorphic to @xmath23 ) , there is an algebraic number field @xmath40 , and an element @xmath14 , such that @xmath41 . this fact ( which is well known in the general case for any lattice in any @xmath0 which is not locally isomorphic to @xmath21 ) can be proved by a few line argument when @xmath3 is locally rigid in @xmath0 . when @xmath0 is locally isomorphic to @xmath23 , and @xmath1 is non - uniform ( and hence not locally rigid ) , this was proved in @xcite , section 8 , by a special and longer argument . our method allows to unify this special case to the general case . in section [ ehresmann - thurston ] we shall formulate and prove the ehresmann - thurston principle and use it in order to prove local rigidity for uniform lattices ( weil s theorem ) . in section [ rank-1 ] , which is in a sense the main contribution of this note , we shall present a complete and elementary proof for local rigidity of non - uniform lattices in groups of rank one ( originally due to garland and raghunathan ) . then , in section [ higher - rank ] , we shall apply our method to higher - rank non - uniform lattices . in some cases , our proof remains completely elementary . however , for the general case we shall rely on some deep results . in this section @xmath42 is a model manifold and @xmath0 a transitive group of real analytic homeomorphisms of @xmath42 . given such data one can construct all possible manifolds by choosing open subsets of @xmath42 and pasting them together using restrictions of homeomorphisms from the group @xmath0 . following ehresmann and thurston we will call such a manifold a @xmath43-manifold . more precisely , a _ @xmath43-manifold _ is a manifold @xmath44 together with an atlas @xmath45 such that the changes of charts are restrictions of elements of @xmath0 . such a structure enables to develop the manifold @xmath44 along its paths ( by pasting the open subsets of x ) into the model manifold @xmath42 . we can do this in particular for closed paths representing generators of the fundamental group @xmath46 . starting from one of the open sets @xmath47 of the given atlas , the development produces a covering space @xmath48 of @xmath44 , a representation @xmath49 called the holonomy and a structure preserving immersion , the developing map @xmath50 , which is equivariant with respect to the action of @xmath46 on @xmath48 and of @xmath51 on @xmath42 . there is an ambiguity in the choice of one chart @xmath52 about a base point . however , deferent charts around a point are always identified on a common sub - neighborhood modulo the action of @xmath0 . ehresmann s way of doing mathematics was to find neat definitions for his concepts so that theorems would then follow easily from definitions . we shall try to follow his way . recall first his fiber bundle picture of a @xmath43-structure and his neat definition of the developing map . to make it simple , let us describe first the picture for one chart @xmath53 which is a topological embedding . we can associate a trivial fiber bundle @xmath54 by assigning to any @xmath55 the model space @xmath56 . the manifold @xmath57 is embedded as the diagonal cross section @xmath58 . its points are the points of tangency of fibers and base manifolds . for the global picture of a @xmath43-structure on a manifold @xmath44 , we will restrict ourselves to compatible charts that are topological embeddings @xmath53 for open sets @xmath59 . a point of the fiber bundle space @xmath60 over @xmath44 is , by definition , a triple @xmath61 where @xmath62 , @xmath52 is a compatible chart and @xmath63 , modulo the equivalence relation given by the action of @xmath0 , i.e. @xmath64 iff there is @xmath14 such that @xmath65 and @xmath66 on some sub - neighborhood @xmath67 . in @xmath60 , the manifold @xmath44 is embedded as the diagonal cross section @xmath68 , whose points are represented by triples @xmath69 . the horizontal leaves represented by the triples @xmath70 give a foliation @xmath71 of the total space @xmath60 , which induces an @xmath72-plane field @xmath73 in @xmath60 transversal to the fibers and transversal to the cross section @xmath68 . given such data , ehresmann has defined the classical notion of holonomy . the holonomy is obtained by lifting a closed curve starting and ending at @xmath74 , into all curves tangent to @xmath73 , and by this getting an element of @xmath0 acting on @xmath75 . for contractible closed curves in the base space @xmath44 , the holonomy is of course the identity map of the fiber @xmath75 . this leads us , following ehresmann , to consider the general notion of _ flat cartan connections_. in fact @xmath43-manifolds are the flat cases of manifolds @xmath44 with a general @xmath43-connection . @xmath43-connection _ is defined in @xcite as follows : 1 . a fiber bundle @xmath76 with fiber @xmath42 over @xmath44 , 2 . a fixed cross section @xmath68 , 3 . an @xmath72-plane field @xmath73 in @xmath60 transversal to the fibers and transversal to the fixed cross section , such that 4 . the holonomy of each closed curve starting and ending at @xmath74 is obtained by lifting it to all curves tangent to @xmath73 , and is a homeomorphism of @xmath75 which is induced by an element of @xmath0 . a @xmath43-connection is said to be _ flat _ if contractible closed curves have trivial holonomy . we have just seen that a @xmath43-structure on a manifold @xmath44 yields a flat @xmath43-connection on @xmath44 . let us describe how we can recover the holonomy and the developing map of a given @xmath43-structure from the induced flat @xmath43-connection . let @xmath44 be a manifold with a flat @xmath43-connection . for general closed curves , starting and ending at @xmath77 , the holonomy gives a representation of @xmath78 into the group @xmath0 acting on @xmath75 . when the @xmath43-connection is induced by a @xmath43-structure on @xmath44 , this map is the usual holonomy map of the @xmath43-structure . @xmath60 can be described as the quotient @xmath79 where the action is the diagonal one , with @xmath46 acting on @xmath48 by covering transformations and on @xmath42 via the holonomy map through the action of @xmath0 on @xmath42 . the development of a curve ending at @xmath80 , is obtained by dragging along @xmath73 the corresponding points until they arrive in the fiber @xmath75 . as soon as the connection @xmath73 is flat , homotopic curves with common initial and end points give the same image of the initial point in the end fiber and the development map @xmath81 is well defined . we thus conclude that any flat @xmath43-connection on a manifold @xmath44 yields a @xmath43-structure . it is then an easy matter to prove the following useful result , which to our knowledge first appeared in a slightly different setup in thurston s notes @xcite , but which , as we will see , and seems to be a common knowledge , is elementary using ehresmann s viewpoint of @xmath43-structures . for a slightly different proof , from which we have borrowed some ideas , see @xcite which is a very nice reference for all basic material on @xmath43-structures . [ prin ] let @xmath82 be a smooth compact manifold possibly with boundary . let @xmath83 be the reunion of @xmath82 with a small collar @xmath84 of its boundary . assume @xmath83 is equipped with a @xmath43-structure @xmath44 whose holonomy is @xmath85 . then , for any sufficiently small deformation @xmath17 of @xmath8 in @xmath86 , there is a @xmath43-structure on the interior of @xmath82 whose holonomy is @xmath17 . let @xmath87 be a deformation of @xmath8 . we define the following two fiber bundles over @xmath83 with fibers @xmath42 @xmath88 and @xmath89 where the action of @xmath90 on @xmath42 is respectively induced via @xmath8 and @xmath17 by the natural action of @xmath0 on @xmath42 . we denote by @xmath91 and @xmath92 the two diagonal actions of @xmath90 on @xmath93 considered above . note that since @xmath82 is compact , @xmath90 is finitely generated and hence @xmath94 has the structure of an analytic manifold and , in particular , it is locally arcwise connected . theorem [ prin ] is a consequence of the following easy claim . [ diffeo ] if @xmath17 is a deformation of @xmath8 in the same connected component of @xmath94 then there exists a fiber bundle map @xmath95 such that 1 . @xmath96 restricts to a diffeomorphism @xmath97 above @xmath82 , 2 . as @xmath17 gets closer to @xmath8 , the lifted diffeomorphism @xmath98 gets closer to the identity in the compact - open topology . we temporarily postpone the proof of the claim and conclude the proof of theorem [ prin ] . when @xmath17 is in the same connected component as @xmath8 , we have , by claim [ diffeo ] , a diffeomorphism @xmath97 between the compact fiber bundles @xmath99 and @xmath100 which are defined over @xmath82 . thus , if @xmath71 denotes the horizontal foliation in @xmath101 , the restriction of @xmath71 to @xmath99 induces via @xmath97 an @xmath72-plane field @xmath73 on @xmath100 . when @xmath17 is close enough to @xmath8 , the lifted diffeomorphism @xmath102 is close to the identity so that the @xmath72-plane field @xmath73 is transversal to both the fibers and the image by @xmath97 of the diagonal cross section in @xmath99 . moreover , the holonomy which is obtained by lifting a closed curve starting and ending at @xmath103 , to all curves tangent to @xmath73 , is equal to the image by @xmath17 in @xmath0 of the homotopy class of the curve . thus the restriction of the bundle @xmath100 to a bundle over the interior of @xmath82 yields a @xmath43-connection . this connection is obviously flat and hence gives a @xmath43-structure on the interior of @xmath82 whose holonomy is @xmath17 . this concludes the proof of theorem [ prin ] modulo claim [ diffeo ] . let us now prove claim [ diffeo ] . let @xmath104 be a finite open covering of @xmath83 such that * @xmath105 , * for each @xmath106 , @xmath107 is simply connected and the fiber bundles @xmath108 and @xmath101 are both trivial above @xmath107 . for each integer @xmath109 , we fix a trivialization @xmath110 ( resp . @xmath111 ) of @xmath108 ( resp . @xmath101 ) above @xmath107 . for each pair of integers @xmath112 , we then denote by @xmath113 ( resp . @xmath114 ) the diffeomorphism , which is the product of the identity on the first factor with the corresponding change of charts on the second factor , between the trivializations @xmath115 and @xmath116 ( resp . @xmath117 and @xmath118 ) . let @xmath119 @xmath120 and for each integer @xmath121 , let @xmath122 be a shrinking of @xmath123 ( i.e. for each integer @xmath124 , @xmath125 is an open set whose closure is included in @xmath126 ) such that @xmath127 is still a covering of @xmath83 . as both @xmath108 and @xmath101 are trivial above @xmath128 , the identity map @xmath129 induces a diffeomorphism @xmath130 . we will define by induction on @xmath131 , a diffeomorphism @xmath132 between @xmath101 and @xmath108 above @xmath133 which is equal to @xmath134 above @xmath135 . but for the sake of clarity let us first define the diffeomorphism @xmath136 . let s first describe the @xmath137 step . first note that , as @xmath17 is a deformation of @xmath8 in the same connected component , the diffeomorphism @xmath138 of @xmath139 is isotopic to the identity , by an isotopy which preserves each of the fiber @xmath140 . we denote by @xmath141 , @xmath142 $ ] , this isotopy . we want to define @xmath136 above @xmath143 , recall it will be equal to @xmath144 above @xmath145 . let @xmath146 be an open set such that @xmath147 we define @xmath136 above @xmath143 as follows : * above @xmath148 , both @xmath101 and @xmath108 are trivial and the identity map @xmath149 induces a diffeomorphism @xmath150 . * we define @xmath151 above @xmath152 . ( in the coordinates @xmath153 induced by the trivializations above @xmath154 , the diffeomorphism @xmath155 is equal to the restriction of @xmath138 . ) * we extend @xmath155 above a small collar neighborhood @xmath156 of the boundary @xmath157 of @xmath152 in @xmath158 by the map from @xmath159 to itself , given by @xmath160 where @xmath161\to [ 0,1]$ ] is the classical smooth bump function . * finally we can extend @xmath155 ( by the identity ) above all @xmath162 to get a smooth diffeomorphism . * we then define @xmath136 above @xmath143 to be the restriction of @xmath155 to that set . if we let @xmath136 to be equal to @xmath144 above @xmath163 , we get a well defined diffeomorphism @xmath136 from @xmath101 to @xmath108 above @xmath164 . to follow the induction we need to define @xmath165 above @xmath166 . first note that above @xmath167 , the fiber bundles @xmath101 and @xmath108 are both trivial . let @xmath146 be an open set such that @xmath168 we define @xmath165 above @xmath166 as follows : * above @xmath169 , both @xmath101 and @xmath108 are trivial and we define a function @xmath155 to be the canonical diffeomorphism between their trivializations . * above @xmath170 we define @xmath155 to be equal to @xmath132 . * we extend @xmath155 as in the first inductive step , by using an isotopy between changes of charts composed with a smooth bump function , to get a smooth diffeomorphism between @xmath101 and @xmath108 above all @xmath171 . * we define @xmath165 above @xmath166 to be equal to the restriction of @xmath155 above @xmath166 . if we let @xmath165 to be equal to @xmath132 above @xmath172 , we get a well defined diffeomorphism @xmath165 from @xmath101 to @xmath108 above @xmath173 . the local trivialization of the fiber bundle @xmath108 depends continuously ( in the @xmath174-topology ) on @xmath17 , and hence @xmath165 depends continuously on @xmath132 and @xmath17 . we continue the induction until @xmath175 and get a diffeomorphism between @xmath101 and @xmath108 above @xmath176 which restricts to a fiber bundle diffeomorphism between the corresponding bundles above @xmath82 . moreover we have seen that the restriction of @xmath96 above @xmath82 depends continuously ( in the @xmath174 topology ) on @xmath17 . this concludes the proof of the claim . recall now ehresmann s definition @xcite of a complete @xmath43-structure on a manifold @xmath44 . let @xmath44 be a @xmath43-manifold . any curve @xmath177 starting from a point @xmath77 in @xmath44 can be developed to a curve @xmath178 in @xmath42 . the @xmath43-manifold is said to be _ complete _ if , conversely , any curve @xmath179 extending @xmath178 in @xmath42 is a development of a curve in @xmath44 extending @xmath177 . when @xmath42 is a riemannian homogeneous @xmath0-space , for a lie group @xmath0 of isometries of @xmath42 , it is a classical theorem of hopf and rinow that this notion of completeness coincides with the usual notion of metric completeness . the ehresmann - thurston principle above is especially useful when combined with the following . [ complet ] if @xmath44 is a complete @xmath43-manifold , then the developing map induces an isomorphism between its universal cover and the universal cover of @xmath42 . as a corollary of theorem [ prin ] and theorem [ complet ] we get the following classical result of weil @xcite ( see also @xcite ) . [ compact rigidity ] let @xmath180 be a cocompact lattice in a connected center free semi - simple lie group without compact factors @xmath0 . if @xmath17 is a deformation of @xmath180 in @xmath0 , close enough to the identity , then @xmath181 is still a cocompact lattice and is isomorphic to @xmath180 . the group @xmath3 is finitely generated and by selberg s lemma it has a torsion - free finite index subgroup . it is easy to see that if [ compact rigidity ] holds for a finite index subgroup of @xmath3 then it also holds for @xmath3 . hence , we shall assume that @xmath3 itself is torsion - free . let @xmath42 be the symmetric space associated to @xmath0 . it is simply connected . let @xmath44 be the @xmath43-manifold @xmath182 . it is compact and its fundamental group is @xmath180 . according to theorem [ prin ] , if @xmath17 is a sufficiently small deformation of the inclusion @xmath183 , then there exists a new @xmath43-structure @xmath48 on the manifold @xmath44 , whose holonomy is @xmath184 . but as @xmath44 is compact and @xmath42 is riemannian , this new @xmath43-structure is complete and , by theorem [ complet ] , the developing map is a @xmath180-equivariant isomorphism between its universal cover and @xmath42 . this implies that @xmath185 acts properly discontinuously on @xmath42 and that @xmath186 . as @xmath48 is homeomorphic to @xmath44 this concludes the proof . we shall conclude this section by stating the following natural generalization of corollary [ compact rigidity ] which is also due to weil @xcite . [ gwt ] let @xmath0 be a connected lie group and @xmath1 a uniform lattice . then for any sufficiently small deformation @xmath17 of @xmath3 in @xmath0 , @xmath24 is again a uniform lattice and @xmath30 is an isomorphism . assume first that @xmath0 is simply connected . then we can chose a left invariant riemannian metric @xmath187 on @xmath0 , and argue verbatim as in the proof corollary [ compact rigidity ] letting @xmath188 to stand for @xmath42 , and @xmath0 to act isometrically by left multiplication . now consider the general case . let @xmath189 be the universal covering of @xmath0 . let @xmath1 be a uniform lattice and let @xmath190 be its pre - image in @xmath189 . then @xmath190 is a uniform lattice in @xmath189 , and @xmath191 is homeomorphic to @xmath192 . since @xmath3 is finitely generated , the deformation space @xmath193 has the structure of a manifold and in particular it is locally arcwise connected . therefore if @xmath17 is sufficiently small deformation of @xmath3 in @xmath0 then there is a curve of deformations @xmath194 with @xmath195 the identity , and @xmath196 . now for any @xmath16 and any @xmath197 above @xmath198 , the curve @xmath199 lifts uniquely to a curve @xmath200 . the uniqueness guaranties that @xmath201 is again a homomorphism for any @xmath202 . hence @xmath203 lifts to a curve @xmath204 . in particular @xmath205 lifts to a deformation @xmath206 of @xmath207 in @xmath189 . it is also easy to see that the lifting @xmath208 on a neighborhood of the inclusion @xmath8 is continuous as a map from an open set in @xmath193 to @xmath209 . thus , the general case follows from the simply connected one . the proof of theorem [ thm1 ] will follow the same lines , we just need to understand the non - compact parts , namely the ends . let @xmath0 be a connected center free simple lie group of rank one , with associated symmetric space @xmath27 . let @xmath3 be a non - uniform lattice in @xmath0 . we are going to investigate the possible small deformations of @xmath3 , and to prove local rigidity when @xmath210 , and some weak version of it for dimensions @xmath211 and @xmath212 . by selberg s lemma @xmath3 is almost torsion free . note that a small deformation which stabilizes a finite index subgroup must be trivial . in the sequel we shall assume that @xmath3 is torsion free . we keep the notation @xmath6 for the inclusion , and @xmath17 for a small deformation of @xmath8 . we denote by @xmath213 the corresponding locally symmetric manifold . we shall show that under some conditions on @xmath17 , @xmath24 is again a lattice in @xmath0 and @xmath214 is homeomorphic to @xmath44 , and that these conditions are fulfilled when @xmath215 . our strategy is to use ehresmann - thurston s principle . as @xmath44 is not compact , we shall decompose it to a compact part @xmath216 which `` exhausts most of @xmath44 '' and to cusps . more precisely , let @xmath217 be a fixed compact submanifold with boundary , which is obtained by cutting all the cusps of @xmath44 along horospheres . each such cusp is contained in a connected component of the thin part of the thick - thin decomposition of @xmath44 corresponding to the constant @xmath218 of the margulis lemma ( see @xcite or @xcite ) , and is homeomorphic to @xmath219 for some @xmath220-dimensional compact manifold @xmath221 . we shall call such cusps canonical . more precisely , a _ canonical cusp _ is a quotient of a horoball by a discrete group of parabolic isometries which preserves the horoball and acts freely and cocompactly on its boundary horosphere . notice that a canonical cusp has finite volume . to see this , one can look at the iwasawa decomposition @xmath222 which corresponds to a lifting of the given cusp , express the haar measure of @xmath0 in terms of the haar measures of @xmath223 and @xmath82 , and estimate the volume of the cusp in the same way as one shows that a siegel set has a finite volume . in the argument below we shall assume , for the sake of clearness , that @xmath44 has only one cusp , instead of finitely many . it is easy to see that the proof works equally well in the general case . we choose another horosphere which is contained in @xmath216 and parallel to the boundary horosphere , and denote the collar between them by @xmath224 $ ] ( see figure 1 ) . by the ehresmann - thurston principle , there is a @xmath43-structure @xmath225 on @xmath216 whose holonomy is @xmath17 . we will show that under some conditions , the @xmath43-structure on @xmath224 $ ] which is induced from @xmath225 coincides with a @xmath43-structure which is induced from some canonical cusp @xmath226 ( whose holonomy coincides with @xmath227 ) to a subset of it , which is homeomorphic to @xmath224 $ ] . in such case we can glue @xmath216 with @xmath226 along @xmath224 $ ] and obtain a complete @xmath43-manifold @xmath48 homeomorphic to @xmath44 whose fundamental group is @xmath24 . additionally , @xmath48 has a finite volume , being a union of the compact set @xmath225 with @xmath226 which is a canonical cusp and hence has a finite volume . recall the following basic facts about isometries of rank one symmetric spaces : * maximal unipotent groups correspond to points in the ideal boundary @xmath228 . a maximal unipotent group acts simply transitively on each horosphere around its fixed point at @xmath229 , its dimension is @xmath230 and it is the unipotent radical of the parabolic group which is the stabilizer of this point . the maximal unipotent groups are metabelian . * there are only 3 types of isometries : elliptic , hyperbolic and parabolic . an elliptic isometry fixes a point in @xmath42 and is contained in some compact subgroup of @xmath0 , it is semisimple and all its eigenvalues have absolute value @xmath231 . + a hyperbolic isometry @xmath198 has exactly two fixed points at the ideal boundary @xmath228 and it acts by translations on the geodesic connecting them , called the axis of @xmath198 , which is also characterized as the set of points where the displacement function @xmath232 attains its minimum . a hyperbolic element is semisimple and has an eigenvalue of absolute value @xmath233 . + a parabolic element is one which has no fixed points in @xmath42 and has a unique fixed point in the ideal boundary @xmath228 . if @xmath234 is parabolic then @xmath235 and the infimum is not attained . an element is parabolic iff it is not semisimple . in particular , unipotents are paraboloics . in general , all the eigenvalues of a parabolic element have absolute value @xmath231 . * the quotient map @xmath236 is open , and the action map @xmath237 is continuous . let @xmath238 be a subgroup of @xmath3 which corresponds to the fundamental group of the cusp of @xmath44 . then @xmath239 is group of parabolic elements , and by @xcite corollary 8.25 , it contains a subgroup @xmath240 of finite index which is unipotent , i.e. a lattice in the unipotent radical of the corresponding parabolic group . we want to show that @xmath241 is the fundamental group of some canonical cusp . so , in particular , we need to show that @xmath241 is again a group of parabolic elements . this fact is true only when @xmath242 ( but the rest of the argument applies also for dimensions @xmath211 and @xmath212 ) . we shall prove it by showing that otherwise , @xmath241 is contained in the stabilizer group of some geodesic or some point , and then derive a contradiction since @xmath241 is `` too large '' and there is `` no room '' for it in such a subgroup . we shall divide this argument into few claims . [ index ] there exists an integer @xmath243 such that any nilpotent subgroup @xmath244 of @xmath245 contains a subgroup @xmath246 such that * @xmath247 , * @xmath157 is contained in a connected abelian group . dividing @xmath245 by the factor @xmath248 , we can prove the analogous statement for @xmath249 . let @xmath250 be a small symmetric identity neighborhood with the following property : a subset @xmath251 generates a nilpotent group if and only if @xmath252 generates a nilpotent lie sub - algebra ( c.f . @xcite lemma 8.20 ) . let @xmath253 be a haar measure on @xmath249 , and take @xmath254 let now @xmath244 be a nilpotent subgroup of @xmath249 . take @xmath255 . then @xmath157 is contained in the connected nilpotent group which corresponds to the nilpotent lie subalgebra generated by @xmath256 . the closure of this connected nilpotent group is compact , connected and nilpotent , hence it is abelian , and in particular so is @xmath157 . now let @xmath257 be a set of coset representatives for @xmath258 . the sets @xmath259 must be disjoint , for otherwise we will have @xmath260 for some @xmath261 . therefore @xmath262 [ rmk ] the proof of lemma [ index ] shows that the analogous statement holds for the group @xmath263 as well . below , we shall assume that the index @xmath124 is good for both groups @xmath245 and @xmath263 . the maximal dimension of a connected abelian subgroup of @xmath249 is @xmath264 $ ] - the absolute rank of @xmath265 , and hence the maximal dimension of a connected abelian subgroup of @xmath245 is @xmath264 + 1 $ ] . moreover , the stabilizer group of a geodesic @xmath266 embeds naturally in @xmath245 . this implies : [ smalldim ] assume @xmath267 . let @xmath268 be a connected abelian subgroup . if @xmath157 is not unipotent then @xmath269 . by lie s theorem , @xmath157 is triangulable over @xmath270 . thus , since @xmath157 is not a unipotent group , there is a semisimple element @xmath271 . the isometry @xmath272 is either elliptic or hyperbolic . if @xmath272 is hyperbolic , then as @xmath157 is abelian , any @xmath273 preserves the axis of @xmath272 . by the above paragraph we get @xmath274 + 1 < n-1\ ] ] as @xmath267 . if @xmath272 is elliptic then its set of fixed points @xmath275 is a totally geodesic sub - symmetric space of dimension @xmath276 which is @xmath157-invariant . hence @xmath157 embeds in @xmath277 and the result follows by an inductive argument on the dimension , keeping in mind that that any connected unipotent subgroup of @xmath275 has dimension @xmath278 . let @xmath279 be the fixed point of @xmath239 , let @xmath280 be the parabolic group which corresponds to @xmath281 and let @xmath282 be the langland s decomposition of @xmath283 . then @xmath82 is the maximal unipotent group which corresponds to @xmath281 , and @xmath284 can be characterized as the subgroup of @xmath0 which preserve the horospheres around @xmath281 , and we have @xmath285 , and @xmath286 . additionally , @xmath239 ( resp . @xmath240 ) is a cocompact lattice in @xmath284 ( resp . in @xmath82 ) because @xmath287 is homeomorphic to @xmath221 ( where @xmath288 is an arbitrary point and @xmath289 the corresponding horosphere . in particular any subgroup of finite index @xmath290 is zariski - dense in the unipotent group @xmath82 , and thus , such subgroup @xmath157 is not contained in a connected proper subgroup of @xmath82 . as @xmath82 acts simply transitively on each horosphere , its dimension is @xmath291 . so we see that @xmath292 is @xmath291-dimensional and conclude : [ bigdim ] assume @xmath267 . no subgroup of finite index of @xmath239 is contained in a connected abelian subgroup of @xmath0 of dimension @xmath293 . the following claim is obvious . [ 12 ] for a finitely generated group @xmath294 , and a convergent sequence of homomorphisms @xmath295 , the property : `` @xmath296 is contained in a connected abelian group of dimension @xmath297 '' is preserved by taking the limit . we can now prove the following main proposition : assume @xmath267 . if @xmath17 is a sufficiently small deformation of @xmath8 , then the group @xmath241 is contained in a parabolic group . moreover , @xmath241 contains no hyperbolic elements . let @xmath298 be a central element . if @xmath299 is parabolic then @xmath300 is contained in the parabolic group corresponding to the unique fixed point of @xmath299 at @xmath228 . assume that @xmath299 is not parabolic . then it is either elliptic or hyperbolic . but if @xmath301 is hyperbolic then , as it is central , @xmath302 is contained in the stabilizer group of the axis of @xmath299 . however , by lemma [ index ] we obtain that @xmath300 contains a subgroup of index @xmath303 which is contained in a connected abelian group , which by proposition [ smalldim ] has dimension @xmath304 . since @xmath305 is finitely generated ( being a lattice in @xmath82 ) it has only finitely many subgroups of index @xmath303 . therefore the property `` to have a subgroup of index @xmath303 which is contained in a connected abelian group of dimension @xmath306 '' is preserved by taking a limit . this , however , contradicts claim [ bigdim ] . suppose therefore that @xmath301 is elliptic . then its set of fixed points is a sub - symmetric space @xmath307 of smaller dimension , on which @xmath300 acts nilpotently , and we continue by taking @xmath308 which is central with respect to this action . if @xmath309 is parabolic we conclude that @xmath300 is contained in the parabolic group corresponding to the unique fixed point of @xmath310 at @xmath311 . if @xmath310 is not parabolic then it can not act parabolically on @xmath307 . as above @xmath310 can not acts hyperbolically . if @xmath312 acts elliptically , we continue by induction , defining @xmath313 . by this way we conclude that @xmath300 is contained in a parabolic group . now if all the @xmath314 s constructed above act elliptically , then @xmath300 is contained in a compact group , and ( letting @xmath315 ) we derive a contradiction to remark [ rmk ] , proposition [ smalldim ] and claim [ bigdim ] as above . therefore one of the @xmath314 s , say @xmath316 , must act as a parabolic . this implies that @xmath300 has a unique fixed point at infinity , and since @xmath240 is normal in @xmath239 it follows that @xmath241 fix this point and hence contained in the corresponding parabolic group . now if @xmath241 would contain a hyperbolic element then so would @xmath300 since a power of hyperbolic is hyperbolic . however hyperbolic element can not commute with parabolic , while the action of any element on the corresponding subsymmetric space @xmath317 commute with the action of the parabolic element @xmath318 . therefore @xmath241 does not contain hyperbolic elements . since the action of @xmath0 at @xmath228 is continuous , the unique fixed point @xmath319 of @xmath241 is close to the fixed point @xmath320 of @xmath239 , and since the orbit map @xmath236 is open , there is some @xmath14 near @xmath231 which takes @xmath320 to @xmath319 . conjugating @xmath17 by @xmath321 we get @xmath322 which is again a small deformation . moreover , @xmath323 is a deformation inside the parabolic subgroup @xmath283 which corresponds to the cusp . the group @xmath324 is contained in @xmath284 . this follows immediately from the previous proposition since @xmath284 is exactly the set of elements in @xmath283 which are not hyperbolic . we shall now use ehresmann - thurston s principle ( proposition [ gwt ] ) to deduce : the group @xmath325 is a cocompact lattice in @xmath284 , and @xmath326 is an isomorphism . since @xmath325 ( and @xmath241 ) is discrete and torsion free , it acts properly and freely on each horosphere @xmath289 , and hence @xmath327 ( and hence also @xmath328 ) is homeomorphic to @xmath287 - the boundary of our cusp . this implies that @xmath241 is the fundamental group of some canonical cusp homeomorphic to @xmath329 . this completes the proof of the following theorem and thus completes the rank one case : [ rk-1thm ] let @xmath0 be a connected center free simple lie group of rank one neither isomorphic to @xmath330 nor to @xmath34 , and let @xmath1 be a non - uniform lattice . then there is a neighborhood @xmath5 of the inclusion @xmath6 in the deformation space @xmath2 such that for any @xmath7 , @xmath331 is a lattice in @xmath0 , @xmath30 is an isomorphism , and @xmath28 is homeomorphic to @xmath29 . from theorem [ rk-1thm ] and mostow rigidity , we conclude local rigidity in the rank one case . it is well known that if @xmath3 is a locally rigid lattice in the group of real points @xmath332 of some @xmath39-algebraic group @xmath333 , then it can be conjugated by an element @xmath14 into @xmath334 for some number field @xmath40 this follows from the fact that @xmath3 is finitely generated , and hence , the deformation space @xmath2 has the structure of a real algebraic variety @xmath335 which is defined over @xmath39 . if we let @xmath336 denote the algebraic closure of @xmath39 in @xmath270 , then this implies that @xmath337 is dense in @xmath338 in the real topology . thus we can find a point @xmath339 arbitrarily close to the inclusion @xmath8 . as @xmath3 is locally rigid , @xmath17 is given by conjugation , and as @xmath3 is finitely generated , @xmath24 lies in some algebraic number field . let now @xmath340 , and let @xmath1 be a ( torsion free ) non - uniform lattice . then @xmath3 is not locally rigid . for this reason garland and raghunathan had to deal with this case separately in @xcite , section 8 . using the above presented proof , one can treat the @xmath23 case in the same manner as the general case , since , as explained above , if @xmath17 is a small deformation of @xmath3 which sends unipotents to unipotents , then @xmath24 is a lattice and @xmath30 is an isomorphism , and hence by mostow rigidity , it is given by conjugation . in fact , the fundamental group of each cusp is finitely generated , and it is enough to choose such finite generating set for each cusp , and to require that @xmath17 sends all these finitely many elements to unipotents . then any unipotent in @xmath3 , being conjugate to some element of the chosen fundamental group of some cusp , must also be sent to a unipotent . we can choose a large finite generating set @xmath9 for @xmath3 which contains a generating set for a chosen fundamental group for each of the finitely many cusps . then , we add to the set of relations defining @xmath2 the conditions that every unipotent in @xmath9 must be sent to unipotent , and call this subspace @xmath341 . since , the unipotents are the zeros of the @xmath39-polynomial @xmath342 we see that the space @xmath343 has the structure of a real algebraic variety defined over @xmath39 . we conclude that : [ sl_2(c ) ] let @xmath344 be a @xmath39-algebraic linear group with @xmath332 isomorphic to @xmath34 , and let @xmath3 be a non - uniform lattice in @xmath0 . then @xmath3 is conjugate to a subgroup of @xmath334 for some number field @xmath345 . in the next section , we shall proceed by induction on @xmath346 . the rank one case , thus , would be the base step in this induction argument . we shall need the following weak version of local rigidity , which holds in the general case ( also for @xmath347 ) , and is proved by the same argument as above . let @xmath0 be a connected rank one simple lie group with associated symmetric space @xmath42 , and let @xmath1 be a torsion free lattice . let @xmath17 be a small deformation of @xmath3 which takes all the unipotents in @xmath3 to unipotents . then @xmath24 is a lattice , @xmath348 is homeomorphic to @xmath349 , and @xmath30 is an isomorphism . we shall now consider the case where @xmath0 is a connected center - free semisimple lie group of rank @xmath350 without compact factors , with associated symmetric space @xmath27 , and @xmath1 a non - uniform ( torsion free ) irreducible lattice , corresponding to the locally symmetric manifold @xmath213 . as in the rank one case , we are going to apply ehresmann - thurston s principle . so we need to work with a compact submanifold with boundary @xmath217 which exhausts most of @xmath44 . we want to obtain @xmath216 from @xmath44 by cutting out its ends , and to have enough information on the structure of the ends , so that we can argue as we did in the rank one case . of course , it would be nicer to give a simple proof of local rigidity for general lattices , without assuming arithmeticity . however , we shall use the description of leuzinger @xcite for the structure of @xmath351 , and hence assume , as in @xcite , that @xmath3 is arithmetic . we shall prove the following result , which originally is a consequence of margulis super - rigidity theorem . let @xmath1 be a non - uniform arithmetic irreducible lattice , and let @xmath17 be a small deformation of the inclusion @xmath6 . then @xmath185 is a lattice , @xmath352 is homeomorphic to @xmath44 , and @xmath30 is an isomorphism . the case of @xmath39-@xmath353 is very similar to the @xmath248-@xmath354 one case . however , the picture is much more sophisticated when @xmath39-@xmath355 . then @xmath44 has only one end , i.e. if @xmath216 is defined appropriately then @xmath351 is connected . moreover , `` from infinity '' @xmath44 `` looks like '' a simplicial complex of dimension @xmath39-@xmath356 . our argument will use an induction on the @xmath248-@xmath354 . let @xmath157 be a parabolic subgroup of @xmath0 corresponding to a cusp of @xmath180 , and @xmath357 be its langland s decomposition over @xmath39 . let @xmath358 , @xmath359 , and @xmath226 be the center of @xmath360 . by arithmeticity , @xmath361 is a lattice in the unipotent group @xmath82 , and in particular it is finitely generated . let @xmath294 be a finitely generated group . an element @xmath363 is called a _ @xmath57-element _ if the length of @xmath364 , with respect to the word metric on @xmath294 corresponding to a fixed finite generating set , grows like @xmath365 . if @xmath363 has an eigenvalue which is not a root of unity , then , as shown in the proof of tits alternative @xcite ( or more simply using kronecker s lemma ) , there is a local field @xmath366 and a representation @xmath367 , such that @xmath368 has an eigenvalue with absolute value @xmath233 . this implies that the norm of @xmath369 in @xmath370 grows exponentially . if @xmath9 is a finite generating set , then the norms of the elements in @xmath371 are uniformly bounded . as the norm is a sub - multiplicative function , this implies that the length of @xmath364 in the word metric grows linearly with @xmath72 . recall that @xmath2 is locally arcwise connected , i.e. if @xmath17 is close enough to @xmath8 then they are connected by an arc of deformations @xmath203 . since the eigenvalues depend continuously on the matrix , and since the set of roots of unity is totally disconnected , we conclude that for any @xmath202 , all the eigenvalues of @xmath199 must be @xmath231 . hence @xmath199 , and in particular @xmath373 is unipotent . fix a generating set @xmath9 for @xmath361 and assume that @xmath7 as in corollary [ 4.7 ] . as @xmath374 is nilpotent , its zariski closure is nilpotent , and hence , by lie s theorem , @xmath375 is triangulable over @xmath270 . assume that @xmath376 has @xmath124 connected components . let @xmath377 . being a subgroup of @xmath244 , @xmath378 is triangulable over @xmath270 . since @xmath379 is made of unipotents , @xmath378 is generated by unipotents , and hence it is a unipotent group . it follows that @xmath380 is a unipotent algebraic group , and as such it is also connected . it might be possible to avoid the notion of @xmath57-elements and the use of theorem [ l - m - r ] , and to find a more elementary proof for proposition [ unipotent ] . this is not so hard in the special case where @xmath382 and @xmath383 is any lattice . moreover , for the special case of @xmath384 and @xmath385 theorem [ l - m - r ] also has an elementary proof ( see @xcite ) . according to @xcite , appendix 3 , since @xmath374 is unipotent , the set of common fixed points at @xmath228 @xmath387 is non - empty , and so is the set @xmath388 ( here @xmath389 denotes the tits distance on @xmath228 ) which clearly has diameter @xmath390 and thus has a unique chebyshev center @xmath391 . since @xmath392 normalizes @xmath361 , the point @xmath391 must be invariant under the action of @xmath386 . therefore @xmath386 is contained in the parabolic subgroup which corresponds to @xmath391 . let @xmath283 be a parabolic subgroup which contains @xmath393 , and let @xmath394 be a langland s decomposition of @xmath283 . note that the group @xmath395 is well defined , the group @xmath396 is determined up to conjugation by elements of @xmath395 , and the group @xmath397 is well defined . the group @xmath397 can be characterized as the set of all elements in @xmath283 which acts unimodularly on the lie algebra @xmath398 through the adjoint action , or as the set of all elements in @xmath283 which preserve the busemann functions and their level sets - the horospheres around the corresponding fixed points at @xmath228 . in fact we can choose @xmath283 so that for a finitely generated group @xmath294 , and a convergent sequence of homomorphisms @xmath295 , the property `` @xmath296 is contained in a connected unipotent group of dimension @xmath297 '' is preserved by taking a limit . the second statement follows from the fact that @xmath392 is a an arithmetic subgroup and hence a lattice in @xmath284 and in particular its zariski closure contain @xmath402 , where @xmath48 is the product of all non - compact factors of @xmath44 . hence , when @xmath17 is small enough , @xmath283 must contain a conjugate of @xmath402 , but this implies that it contains a conjugate of @xmath157 . since the canonical map @xmath403 is continuous and open , we have @xmath404 for some @xmath14 close to @xmath231 , and by replacing @xmath17 by its conjugation with @xmath321 , @xmath322 ( which is still a small deformation of @xmath8 ) , we may assume that @xmath405 . this is a subsequent of the fact that @xmath322 induces a small deformation of @xmath360 ( a cocompact lattice ) in @xmath82 together with either some classical theorems on lattices of nilpotent lie groups ( macev theory @xcite ) or , more simply , ehresmann - thurston s principle . as noted above @xmath410 is a lattice in @xmath411 . moreover , the projection of @xmath410 in @xmath44 is a lattice in @xmath44 , let s denote this lattice by @xmath412 . the morphism @xmath322 , composed with the projection @xmath413 , induces a small deformation @xmath414 of @xmath412 in @xmath44 . moreover , by theorem [ l - m - r ] and corollary [ u is unip ] , @xmath322 and hence @xmath415 takes unipotents to unipotents . hence , by an induction procedure on the real rank , starting with rank one , of which we already took care in the previous section , we conclude that : to conclude the proof , we now need , as in the rank one case , to introduce a notion of cusp , but this time it will depend on the group @xmath180 . by @xcite , theorem 4.2 , on the locally symmetric manifold @xmath416 , there exists a continuous and piecewise real analytic exhaustion function @xmath417 such that , for any @xmath418 , the sub - level set @xmath419 is a compact submanifold with corners of @xmath335 . moreover , the boundary of @xmath420 , which is a level set of @xmath421 , consists of projections of subsets of horospheres in @xmath42 . we can choose a sufficiently large @xmath422 in order to ensure that for any @xmath423 , @xmath424 is a collar neighborhood of the boundary of @xmath425 inside @xmath420 . in the following we let @xmath426 for some real @xmath423 . by the ehresmann - thurston principle , there is a @xmath43-structure @xmath427 on @xmath216 whose holonomy is @xmath17 and which induces a new @xmath43-structure on @xmath424 . the fundamental group of @xmath424 is a finite complex of groups @xmath428 where each simplex group is the intersection @xmath429 for some @xmath39-parabolic subgroup @xmath283 of @xmath0 and the inclusion order for simplices is the reversed inclusion order for the corresponding parabolic groups . in particular the vertex groups correspond to the @xmath3-conjugacy classes of maximal @xmath39-parabolics . the proof of proposition 3.3 in @xcite implies that any finite complex of groups as above leaves invariant a countable union of horospheres of @xmath42 ( level sets of busemann functions ) centered at the points of the ideal boundary of @xmath42 which corresponds to the vertices of the complex . in fact , leuzinger proves that the developing map of @xmath216 embeds @xmath430 in @xmath42 as the complement @xmath431 of countable union of open horoballs . these horoballs are disjoint if and only if @xmath180 is an arithmetic subgroup of a @xmath39-@xmath354 one group . the projection @xmath432 maps @xmath431 to @xmath420 which is a compact manifold with corners whose fundamental group is isomorphic to @xmath180 . the image of the developing map of @xmath433 with the @xmath43-structure induced from @xmath216 is @xmath434 . it is a @xmath435-invariant subset of @xmath42 . the image @xmath436 of the developing map of @xmath424 with the @xmath43-structure induced from @xmath427 is a subset of @xmath42 invariant under @xmath437 which converges toward @xmath438 on every compact subset as @xmath17 tends to @xmath8 . according to the above claims , when @xmath17 is sufficiently small , the group @xmath439 is still a complex of groups isomorphic to @xmath428 , and each simplex of the new complex is associated with a parabolic group @xmath440 of the same type as the parabolic @xmath157 associated to the corresponding simplex of @xmath428 . moreover @xmath441 is an isomorphism , its image @xmath442 is contained in @xmath443 and it is a lattice there . the group @xmath444 thus leaves invariant a countable family of horospheres as above . more precisely , we get that for @xmath17 sufficiently close to @xmath8 , the image @xmath436 of the developing map of @xmath445 with the @xmath43-structure induced from @xmath427 contains the subset of @xmath42 which is bounded between two families of horospheres as above . by adding the corresponding countably many horoballs , @xmath436 can be completed by a @xmath446-invariant subset of @xmath42 , whose boundary is one part of the boundary of @xmath436 with the reversed orientation and whose quotient @xmath226 by @xmath447 has a finite volume . as the restriction of the @xmath43-structures from @xmath427 and from @xmath226 coincide on @xmath424 , these two sets can be glued along @xmath433 in order to yield a complete @xmath43-manifold @xmath48 whose fundamental group is @xmath185 . note that @xmath48 has finite volume , being a union of a compact set and a finite volume set . the finiteness of the volume of each of the ( finitely many ) parts composing the `` cusp '' @xmath448 can be proved in the same way as one proves the finiteness of the volume of an ordinary siegel set . namely , by expressing the riemannian measure of @xmath42 in terms of the haar measures of @xmath449 and @xmath450 , and using fubiny theorem , integrating the function @xmath231 first along the horosphere corresponding to @xmath283 in @xmath451 ( here we get a finite number since @xmath452 is a lattice in @xmath453 ) , and then in the directions of the end - the simplex of the tits building which corresponds to @xmath283 . note that the riemannian measure of @xmath42 is obtained from the product measure of the corresponding horosphere and the haar measure of @xmath450 by multiplying by a factor which tends to zero exponentially fast as one moves toward the end . canary , r. d. ; epstein , d. b. a. ; green , p. notes on notes of thurston . analytical and geometric aspects of hyperbolic space ( coventry / durham , 1984 ) , 392 , london math . lecture note ser . , 111 , cambridge univ . press , cambridge , 1987 . margulis , g. a. non - uniform lattices in semisimple algebraic groups . lie groups and their representations ( proc . summer school on group representations of the bolyai jos math . budapest , 1971 ) , pp . halsted , new york , 1975 . nicolas bergeron : unit mixte de recherche 8628 du cnrs , laboratoire de mathmatiques , bt . 425 , universit paris - sud , 91405 orsay cedex , france + e - mail address : ` [email protected] ` + tsachik gelander : dept . of math , yale university , 10 hillhouse ave , new haven ct 06520 , usa + e - mail address : ` [email protected] `

using a sample of 3814 quasars from the early data release of the sloan digital sky survey , we confirm that high - ionization , broad , emission lines such as are significantly blueshifted with respect to low - ionization , broad , emission lines , such as , which are thought to be close to the systemic redshift . we examine the velocity shifts of the and emission lines with respect to [ ] and , respectively . emission line peaks have a range of shifts from a redshift of to blueshifts well in excess of as compared to . we confirm previous results that suggest an anti - correlation between the shift of the emission line peak and the rest equivalent width of the emission line . furthermore , by creating composite quasar spectra as a function of shift , we are able to study , in detail , the profiles of the line as a function of velocity shift . we find that the apparent shift of the emission line peak is not a shift so much as it is a lack of flux in the red wing for the composite with the largest apparent shift . this observation should strongly constrain models for the broad emission line region in quasars . the emission line blueshift and equivalent width of are also discussed in light of the well - known anti - correlation between the equivalent width of emission and continuum luminosity , otherwise known as the baldwin effect . we further discuss the emission line shift as a function of other quasar properties such as spectral index , radio and x - ray detection . we find a possible correlation between the emission line shifts and the radio properties of the quasars that is suggestive of orientation as the cause of the velocity shifts . finally , we explore whether the emission line blueshifts correlate with the presence of broad absorption line absorption troughs or with narrow , `` associated '' absorption and how these might be related to orientation .

astro-ph0204162

it has long been known that the redshifts derived from different quasar emission lines often do not agree with each other within typical measurement errors . ( 1982 ) performed the first detailed study of this phenomenon and concluded that high - ionization , broad , emission lines such as are shifted by a few hundred @xmath2 blueward of the redshifts determined from low - ionization , broad , emission lines such as . subsequent studies ( wilkes 1984 ; espey et al . 1989 ; corbin 1990 ; tytler & fan 1992 ; mcintosh et al . 1999 ; sulentic , marziani , & dultzin - hacyan 2000 ) have confirmed these shifts beyond any shadow of a doubt . using a sample of over 2200 active galactic nuclei ( agn ) , ( 2001 ) demonstrated that these effects are also present in the sample of quasars from the sloan digital sky survey ( sdss ; 2000 ) . these shifts affect many of the most important issues in quasar - related science ( both directly and indirectly ) . the most obvious area of research that is affected is in the modeling of the phenomena of agn , particularly the broad emission line region ( belr ; 1997 ; krolik 1999 ) . belr models based upon accretion disk winds ( murray & chiang 1997 ; proga , stone , & kallman 2000 ) will have different dynamics than cloud - based models ( blumenthal & mathews 1975 ; capriotti , foltz , & byard 1980 ) . these differences affect the line profiles and their locations in ways that should be testable . in addition to investigations of the belr , investigations regarding associated absorption ( foltz et al . 1986 ) , broad absorption lines ( weymann et al . 1991 ) , and the cosmic uv and x - ray background are affected by the apparent blueshift of the emission line . we will comment on each of these issues in turn . determining the nature of the so - called `` associated absorber '' ( foltz et al . 1986 ) population ( @xmath3 ) , requires understanding the velocity shift of the emission line in quasars . some associated absorbers are observed to have redshifts greater than the quasar redshift as determined from the emission line peak . however , as many authors have suggested , these apparent redshifts could simply be the result of the blueshift of the emission line relative to the systemic redshift of the quasar . given the average observed blueshift of the emission line , it is clear that the population of truly @xmath4 systems is smaller than it would appear , but it remains unclear if these systems can be completely explained as outflows that only appear to be infalling because of the observed shift of the emission line peak . in addition to the narrower associated absorbers , emission line shifts are important for broad absorption line ( bal ) quasars . previous investigations of the velocity shifts of quasar emission lines , such as that of ( 1990 ) , suggest a correlation between the velocity shift of the emission lines and the broad absorption line troughs observed in at least 10% of all quasars ( weymann 1997 ) . however , ( 1991 ) saw no such effect . whether or not the velocity shifts are an orientation effect has strong implications for bal quasars . if we can determine the nature of the emission line shifts , we can likely settle the issue of how bals are distributed relative to the plane of the accretion disk that is thought to feed agn . are bal flows found preferentially in the plane of the accretion disk , or are the flows more spherically symmetric in their distribution ? in addition to the more obvious problems associated with our inability to determine accurate redshifts for quasars , we must also consider the effect that the apparent inconsistency of emission line redshifts has on indirect quasar science . for example , the blueshift of the emission line causes an over - estimation of the ionizing flux in the inter - galactic medium , such as is described by ( 1999 ) . the issue of emission blueshift takes on even more importance if these shifts are correlated with the line of sight orientation of quasars , or with an orientation - type parameter such as the opening angle of a disk - wind . the most obvious ramification of such a possibility is the ability to determine orientation angles for radio - quiet quasars : until now , only a small percentage of quasars with distinct radio properties could be analyzed in terms of orientation . if a measure of orientation for radio - quiet quasars could be found and is robust ( whether it be related to the emission line shifts or not ) , it would likely revolutionize the study of quasars . the format of the paper is as follows . in 2 , we describe the data used in this analysis . throughout this paper we will follow ( 2002 ) and use a cosmology where @xmath5 . we also adopt a quasar - centered coordinate system in which outflows from the quasar along our line of sight ( which appear blueshifted ) have _ positive _ velocities . section 3 presents an analysis of the data in terms of composite quasar spectra , and the radio and x - ray properties of our sample . in 4 , we discuss the implications of our findings for modeling of the belr , the baldwin effect , the origin of intrinsic absorption , and orientation measures . we also give suggestions for future work . finally , 5 presents a summary of our results . the quasars were selected for spectroscopic followup from the sdss imaging survey which uses a wide - field multi - ccd camera ( gunn et al . 1998 ) ; quasar candidates were identified using a preliminary version of the sdss quasar target selection algorithm ( richards et al . objects were selected based on both their broad - band sdss colors ( fukugita et al . 1996 ; stoughton et al . 2002 ) and as optical matches to radio - detected quasars from the vla `` first '' survey ( becker , white , & helfand 1995 ) . all of the emission line properties are taken directly from the automated 1-d spectroscopic pipeline ( spectro ; frieman et al . , private communication ; 2002 , 4.10 ) outputs that are stored in the public database . we study a sample of 3814 bona - fide quasars ( @xmath6 , with at least one line broader than @xmath7 ) from the early data release ( edr ; 2002 ) quasar sample ( schneider et al . 2002 ) . though the sample is not technically homogeneous , it is adequate for the scientific goals of this work . the differences between the edr versions of quasar target selection ( stoughton et al . 2002 ) and the final version of quasar target selection ( richards et al . 2002 ) are largely for @xmath8 , whereas our sample is restricted to @xmath9 . in addition , we can not think of any reason why the edr quasar sample would be biased against ( or towards ) quasars with large ( or small ) emission line blueshifts . two sub - samples of quasars were created , one for the study of the velocity shifts between and , the other for the study of the velocity shifts between and [ ] . these samples contain 794 and 417 quasars , respectively . for the [ ] sample we take the redshift determined from [ ] to be the systemic ( center of mass ) redshift , since previous studies have shown that the redshifts determined from narrow forbidden lines agree well with redshifts determined from stellar absorption features ( vrtilek & carleton 1985 ) and 21 cm emission ( hutchings , gower , & price 1987 ) in the host galaxies of agn ; see the discussions in & fan ( 1992 ) and ( 1999 ) . to ensure that both the and [ ] emission lines are in each of the spectra considered , the redshifts were restricted to @xmath10 . for the sample , our spectra do not cover [ ] and we take the redshift determined from to be the systemic redshift , since low - ionization broad lines such as have been shown to have a smaller velocity offset from the forbidden narrow lines than do high - ionization broad lines ( e.g. , tytler & fan 1992 ) . in this sample , the redshifts were restricted to @xmath11 . these redshift restrictions ensure that both the and emission lines are observed in each of the spectra considered ( which span the range from @xmath12 to @xmath13 at a resolution of @xmath14 ) . the sample was further restricted by removing all broad absorption line quasars , since the bal troughs hinder accurate measurement of the line peaks . bal quasars were defined according to the computation of the traditional ( 1991 ) `` balnicity index '' using the composite quasar spectrum from ( 2001 ) to define the continuum level . this list of bal quasars is not yet final ; some weaker bals may move into or out of the sample . however , all strong bals ( balnicity @xmath15 ) should be accounted for . since the classification of weak bals is subjective anyway , we have proceeded with our analysis using our current list of bals . the `` systemic '' redshifts used in this analysis were determined using only the [ ] or emission lines , respectively . since our analysis is dependent upon how spectro measures the peaks of these lines , we describe some details of the process . spectro measures two sets of line parameters . the first are detected peaks in the spectrum which best match a list of common emission lines . once an object s redshift has been determined , either from the detected emission lines or from cross - correlation with template spectra , a second set of line measurements are made at the expected positions of a more comprehensive set of spectral features . in both cases a single gaussian is fit to the line . the first set are identified in the database as _ found _ and the second as _ measured_. the following concerns the _ measured _ line parameters , which are used in this paper . the code determines which lines are to be fit individually and which are to be fit as blends . the width , height and center of the line or blend of lines are then fit by a modified levenberg - marquardt algorithm . to enhance the robustness of the code , only a single shift from the expected positions is used for blended lines . the shift of the line center from the expected position is constrained to be @xmath16 by multiplying the @xmath17 values by a steep ramp function beyond this range . in some cases the expected positions are not the redshifted laboratory wavelengths , but the redshifted _ average _ line positions as determined by the composite sdss quasar spectrum ( vanden berk et al . 2001 ) ; the line wavelengths relevant here are @xmath18 and @xmath19 ( which are doublets , but quasar emission line widths are broad enough that the doublet is not resolved ) . these wavelengths are used to compute the redshifts as reported in the sdss database . in this analysis , we will determine the redshift for each individual line based on the spectro _ measured _ line peaks , but we will use the laboratory rest - frame wavelengths ( see 2001 ) as the expected line positions . the velocity distribution of emission line shifts with respect to is quite large as can be seen in figure [ fig : fig1 ] . we find the shifts by taking the _ measured _ line peaks for and from the sdss database and using the laboratory rest - frame wavelengths of the and doublets , ( @xmath20 and @xmath21 , respectively ) to determine the redshift of each line . the median of the distribution is found to be @xmath22 with a dispersion of @xmath23 . for individual quasars , the error in the blueshift with respect to can be @xmath24 ( @xmath25 ) . spot checking of good quality spectra to determine the blueshift by hand supports this conclusion . we do not believe that this error is skewed towards either positive or negative shifts , so the errors in the ensemble average should be much lower . in general , the errors will be biggest for those quasars with the largest blueshifts , since we find that those quasars have the weakest emission lines ; see [ sec : errors ] for a more detailed discussion regarding errors . in a similar manner to the shift , figure [ fig : fig2 ] shows the distribution of velocity differences between the and [ ] emission line peaks . again , we use the _ measured _ line peaks from the sdss database , and the laboratory wavelengths of doublet and the reddest of the [ ] triplet ( @xmath26 and @xmath27 , respectively ) in order to determine the redshift of each emission line and the blueshift between the emission lines . the peak of this distribution is near zero , but has an overall median shift of @xmath28 with a spread of @xmath29 . this shift is about half as large as that found by ( 2001 ) , who found a velocity offset of @xmath30 for ; quoted by ( 2001 ) becomes @xmath30 in the notation used herein . ] & fan ( 1992 ) found ambiguous results for the velocity shift of the emission line ( @xmath31 and @xmath32 with respect to [ ] for two different samples , respectively ) . in our sample , the emission line , while quite close to what is thought to be the systemic redshift in the ensemble average , can have significant deviations in individual quasars ( modulo measurement errors ) . other authors ( e.g. , mcintosh et al . 1999 ) have also studied the shift of the h@xmath33 emission line . we are not able to do so using the automated reductions of the sdss spectra . the spectro pipeline treats h@xmath33 and the nearby [ ] triplet as a blend with the same redshift ( fortunately , redshifts determined from h@xmath33 are typically similar to redshifts determined from [ ] ) . an analysis of the shift of the h@xmath33 emission line with respect to [ ] would thus require a substantial modification to the spectro pipeline and a reanalysis of all of the spectra : a task which is well beyond the scope of this paper . to assess the differences between the spectra of quasars with both large and small redshift differences , we created a set of four composite quasar spectra by combining the spectra of quasars with similar velocity shifts . we first sorted the list of quasars according to their redshift differences and then divided this list into four roughly equal lists of @xmath34 quasars each . composite quasar spectra were created by combining the spectra ( using the geometric mean , which is appropriate for power - law spectra ) in each of these four samples following the procedure of ( 2001 ) . table [ tab : tab1 ] gives the velocity offset properties of these samples ( denoted a , b , c , and d ) ; it includes the number of objects ( the number within the first survey area is given in parentheses ) , the average velocity offset of the emission line in the composite spectrum , along with the average , minimum and maximum velocity offsets of the emission line in the individual spectra . table [ tab : tab2 ] gives other properties of the samples , including the average redshift , the average absolute magnitude in the @xmath35-band , the average spectral index ( @xmath36 ) of the optical continuum , the number and fraction of radio and x - ray detected sources , along with the equivalent width and the full - width at half - maximum of the emission line ( in ) . we have binned the data into four different samples , in part , to minimize the effect of any errors in the emission line blueshift determination . errors can cause objects move from one bin to the next , but it is unlikely that many objects from sample a will move into sample d ( and vice versa ) as a result of errors . also , that the offset between the peaks of the emission lines for the composite spectra in [ sec : specanal ] is small suggests that the errors in the blueshift measurements are not a problem . finally , we have measured the blueshift of by hand ( using the mode of the upper 50% of the line ) in each of the composites . we find that the blueshifts are @xmath37 , and @xmath38 , for composites a , b , c , and d , respectively . these values are in excellent agreement with the average blueshifts of the individual quasars that contribute to each composite as can be seen by comparing columns 3 and 4 in table 1 , thus we are confident that measurement errors do not significantly effect our results . note that , as discussed above , the measured emission line centers have an upper limit of @xmath39 for the velocity difference of the line center from the accepted redshift of the quasar . some of the peaks are shifted by more than this @xmath1 , thus there is a `` pile - up '' at this velocity in the database . future versions of spectro will allow for shifts as large as @xmath40 . for those quasars affected by this limitation in spectro ( 28 in all ) , we have measured the velocity shifts by hand ; the emission line peak was determined using the mode of the upper 50% of the line as was the case in ( 2001 ) . the dashed histogram in figure [ fig : fig1 ] shows the hand calculated velocity shifts of these quasars . we find that the error on these shifts can be quite large and that the true upper velocity limit may be as high as or higher than @xmath41 . in figure [ fig : fig3 ] , we plot the spectra of the composites whose inputs had the smallest and largest emission line shifts , which we hereafter refer to as composites a and d , respectively . we also plot power - law continua with @xmath42 and a slightly bluer @xmath43 for composites a and d , respectively to guide the eye . the spectral indices were computed using a range of @xmath44 at both @xmath45 and @xmath46 . neither of these ranges are ideal continuum windows , see ( 2001 ) , but these are appropriate values to use in the available wavelength range . the composite spectra do not extend far enough into the red to use a better continuum window . for comparison , we have also created a composite quasar spectrum from all of the quasars in the edr quasar catalog ( schneider et al . 2002 ) , which differs from the sample used by ( vanden berk et al . the edr quasar composite has a spectral index of @xmath47 when measured in the manner described above . note , however , that bals , which may tend to be redder than average , are not excluded from the edr composite quasar spectrum . since these continua are clearly not optimal over the entire spectrum , we will use local continua for any further analysis . figure [ fig : fig4 ] shows the major emission line regions for all four of the ( normalized ) composite spectra ( a , b , c , and d ) with the red spectrum having the smallest shift ( composite a ) and the blue spectrum showing the largest shift ( composite d ) . here we have chosen to normalize the spectra in local continuum regions . the continuum regions [ ( @xmath48 ) , in ] are ( 1285:1355 ) , ( 1285:1355 ) , ( 1355:1465 ) , ( 1465:1695 ) , ( 1830:1975 ) , ( 2695:2955 ) for lyman-@xmath49 , , , , ] , and , respectively . of particular interest in figure [ fig : fig4 ] is the profile of the emission line as a function of velocity shift . by simple visual inspection , we confirm previous results that found an anti - correlation between the emission line velocity shift and equivalent width ( corbin 1990 ) . furthermore , after the spectra have been normalized by the continuum , it is clear that the shift of the emission line is not so much a shift of the centroid to the blue as it appears to be a suppression of flux in the red wing . the blue wing is quite similar in all four composite spectra , whereas the peak and the red wing are less pronounced as the velocity shift increases . we note that this conclusion depends significantly upon our continuum placement ; however , we feel that our procedure is appropriate ; see below for further discussion . this extinction in the red wing coupled with the similarity of the blue wings is more complex than is generally reported . however , such an effect is not entirely without precedent : see ( 2000 ) ; also both & koratkar ( 1995 ) and & boroson ( 1996 ) found enhanced baldwin effects ( 1977 , also see [ sec : baldwin ] ) in the red wing of . that the blueshift is apparently not a blueshift but rather a deficit of flux in the red wing _ significantly _ changes the interpretation of the emission line shifts . the shift is clearly not the result of an actual change in the ( apparent ) bulk velocity flow ( i.e. , it is not a blueward shift of the entire line profile ) ; instead , the shift may be caused entirely by obscuration or suppression of the lowest ( or most negative ) velocity gas ( in the quasar rest frame ) , since the blue wings of both the small and large blueshift composites are co - located . by dividing rather than subtracting the continuum , we implicitly assume that the flux in the emission lines is correlated with the continuum luminosity of the quasars , as expected since the belr is powered primarily by photoionization induced by the uv continuum . this correlation is not perfect ; higher - luminosity quasars have lower equivalent width emission lines ( the baldwin effect , 1977 ) . however , the continuum luminosity of quasars has a range of values at least an order of magnitude larger than the range of emission line equivalent widths ; for example , see , wampler , & gaskell ( 1989 ) , figure 5 and ( 1999 ) , figure 10.2 . if the continuum was completely decoupled from the lines , the range of values spanned by emission line eqws would be the larger of the two , since it would include intrinsic scatter plus uncorrelated variations produced by different continuum levels . as the opposite is true , dividing by the continuum is more appropriate than subtracting . however , it could be argued that for some cases , such as when trying to understand the line profiles , it would be more appropriate to subtract the continuum . thus , in figure [ fig : fig5 ] , we show the profiles of and after subtracting the continuum and scaling the emission lines to have the same peak strength . figure [ fig : fig5 ] demonstrates that the line profile is relatively symmetric in composite a , but is quite asymmetric in composite d , which has more blue flux than red flux as compared to composite a ( see also figure [ fig : fig6 ] ) . note , however , that the scaling used distorts the relative fluxes between the composites ; although composite d has more blue flux than red as compared to composite a , composite d has less overall blue flux ( see figure [ fig : fig4 ] ) . in any case , we emphasize that the interpretation of the profiles does depend significantly on how one handles the continuum . by visual inspection , we find that our composites are similar to those created by ( 1993 ) except that here we divide the sample by their redshift differences instead of by the fwhm of the emission line . in addition , we determine the redshift solely from the emission line . it is therefore interesting that we see many of the same properties as did ( 1993 ) , such as the anti - correlation between the rest equivalent width ( eqw ) and full - width at half maximum ( fwhm ) of the emission lines . although the rest equivalent widths of the emission lines are anti - correlated with the fwhm of the emission lines in our samples ( see table [ tab : tab2 ] ) , it is not clear that composite d , in fact , has the broadest lines . if the emission line profile is indeed absorbed in composite d as compared to composite a , then the fwhm of the emission line is largely unrelated to the velocity distribution of the emission line gas . we further compare to ( 1993 ) by plotting , in figure [ fig : fig6 ] , the profiles of each of our composites on top of reflections about the peaks of each of the profiles . in these plots we have normalized the continua using the same windows as for above ; we find no qualitative difference in the inverted profiles if instead we use ( 1480:1620 ) angstroms as the continuum window . the profiles are inverted around the observed peaks of the lines and not the expected line centers for . if we used the laboratory wavelength as the reflection point , the asymmetries in the more blueshifted composites would be even more dramatic . whereas ( 1993 ) find that their intermediate spectra are the most asymmetric , we find that the asymmetry in our spectra increases with increasing blueshift of the line peak . it is also possible to create difference spectra in the vicinity of , similar to those made by ( 1994 ) ; these are shown in figure [ fig : fig7 ] . these spectra have been normalized as above ; the difference between composite a and d is shown in black , a minus c in dark grey , and a minus b in light grey . note that the red flux missing in composites b , c , and d extends to at least @xmath50 nearly @xmath51 from the line center . these difference profiles are considerably more asymmetric than those of ( 1994 ) . this difference is probably the result of our use of the redshift to define the systemic redshift . if instead we aligned the peaks of the emission line in the composite spectra before we created the difference spectra , we would get more symmetric difference spectra . if the two component belr model discussed by ( 1993 ) is valid , then these difference spectra and the asymmetries of the line profiles place significant constraints upon the model ; given the shape of the difference spectra , we would prefer an obscured or suppressed disk - wind outflow as the preferred explanation for the belr ( see [ sec : discussion ] ) . emission lines other than are equally worthy of consideration . in particular , we note that , in figure [ fig : fig4 ] , the ] profile seems to be shifted significantly blueward in composite d. this observation is consistent with the findings of ( 1989 ) and ( 1990 ) . however , instead of interpreting this as a real shift of ] , we interpret this as an apparent shift due to an increase in the relative strength of ] @xmath52 ( see 2001 ) versus ] , or simply perhaps as a weakening of ] . alternatively , the changes could be due to ( uv34 ) ; see & wilkes ( 2001 ) , 5.2 . if the interpretation that the change in line profile results from a change in the relative strength of the lines is correct , the velocity difference for the ] peaks between composites a and d appears to be zero within the measurement errors ( however , the errors are quite large given the severity of the blending ) . in any case , it may still be possible to make use of this emission feature , since , for the purpose of measuring the shifts , it matters less why the line shifts than the fact that it shifts in the first place . the lyman-@xmath49 + blend in the upper left - hand panel of figure [ fig : fig4 ] is also of interest . the most obvious result is that the lyman-@xmath49 line is _ much stronger _ for composite a than for the composite d. however , this observation is mitigated by the fact that the line in composite b is also quite weak . this discrepancy may be the result of the fact that very few quasars ( @xmath53 per composite ) contribute to the composites in this region of the spectrum , thus the composite lyman-@xmath49 profiles ( and to a lesser extent , ) are on a much weaker footing than the other profiles discussed here . although more data is needed in the lyman-@xmath49 + region of the spectra , we note that a shift of the emission line is evident between composites a and d ( see figure [ fig : fig4 ] ) . in composite a , the peak is separated from the red wing of lyman-@xmath49 . however , in composite d , the emission line blends in with the red wing of lyman-@xmath49 consistent with a blueshift of . the blending is sufficiently strong that we do not attempt to measure the velocity shift of ; however , visual inspection of figure [ fig : fig4 ] indicates that the shift may be significantly larger than found by ( 2001 ) a finding which would bring more into line with the observed correlation between ionization potential and velocity shift ( vanden berk et al . 2001 ) . other lines , in addition to , show significant changes in strength in figure [ fig : fig4 ] . composite d has additional flux in the @xmath54 line in addition to the apparent increase in ] flux ( relative to ] ) , as discussed above . even more striking is the almost complete lack of @xmath55 in composite d , meaning that the emission line region is obscured or suppressed . alternatively , the flux in composite d may simply be broader , weaker and more blueshifted than in composite a. however , the `` more blueshifted '' part of this interpretation for is inconsistent with the results , since we have argued that the blueshift is not a bulk shift of the line profile to the blue , but rather a lack of red flux . since is a high - ionization line , its absence in the composite with the largest shift has interesting consequences . the blend of ] , 2 and just redward of this line also seems to be missing . finally , the extended emission ( i.e. , the @xmath56 bump ) appears to be similar in both composite a and d. perhaps as interesting as the differences in the composites in figure [ fig : fig4 ] is one notable lack of difference : . although is a high - ionization line and is similar in chemical structure to , we see little difference in the profiles between the composites . there is some evidence for a shift in the line peaks , but no evidence for dearth of flux in the red wing . we have no explanation for this fact , other than to note that this finding is consistent with previous results for the baldwin effect ( see [ sec : baldwin ] ) , and that the ionization potential for creation of is about equal to the ionization potential for destruction of . thus emission will tend to come from larger radii than emission , which may be a clue to the different behaviors of their profiles . since redshifts measured from are generally thought to be farther from systemic compared to those measured from [ ] , we have also created composite spectra covering these two emission peaks for the sake of comparison . figure [ fig : fig2 ] showed that the velocity shift of as compared to [ ] ranges from less than @xmath57 to @xmath0 . in figure [ fig : fig8 ] , we plot two composites of the spectra with the smallest and most negative shifts of with respect to [ ] , in black , and with the largest shifts , in grey . the inset of figure [ fig : fig8 ] shows the emission lines after the spectra have been normalized by a locally fit power - law continuum . as with the profiles , the profiles are in good agreement in the blue wing with an apparent lack of flux in the red wing of the composite with the largest shift . once again , the `` shift '' does not appear to be a shift of the centroid so much as a suppression of red flux in those emission lines that appear shifted . note that without follow - up observations , we have no way of knowing if the quasars with the largest blueshifts also have the largest blueshifts ( although this would be a reasonable assumption ) . a more thorough analysis of the emission line blueshifts would include gravitational redshifts ( e.g. , popovic et al . 1995 ) . however , we believe that they are probably not important , for a number of reasons . first , they are relatively small ; using values chosen to maximize the graviational redshift ( @xmath58 and a distance of 1 light - month between the black hole and the belr ) yields a gravitational reshift of only @xmath59 ( popovic et al . second , unless the blueshift is correlated with the mass of the central black hole , then there should not be a correlation between the emission line blueshifts and the strength of a gravitational redshift . finally , and most importantly , they are in the wrong direction : it is when has the largest _ blueshift _ that the redshift is least consistent with the redshift from narrow , forbidden emission features . we now turn to a comparison of the velocity shift composites as a function of a number of quasar properties . tables [ tab : tab1 ] and [ tab : tab2 ] give the properties of the four shifted composite spectra . each of the four composite spectra has a similar number of input spectra . we find that the four composites have relatively similar redshift and luminosity distributions , consistent with there being no strong correlation between these properties and the velocity shift of the emission line . we further tested this conclusion by creating luminosity and redshift composite spectra and found no significant emission line shifts between high- and low - luminosity quasars and between high- and low - redshift quasars . the redshift range covered by these composites is probably not large enough to draw any significant conclusions , however . on the other hand , the range in absolute magnitude , @xmath60 , is sufficient to test any significant trend with quasar luminosity . although the distributions appear similar , there is a small trend of increasing blueshift with luminosity ; see [ sec : baldwin ] for further discussion . we _ do _ find that there is a possible trend in the radio properties as a function of the velocity shift . greater velocity shifts are correlated with lower radio detection fraction as can be seen in table [ tab : tab2 ] ; this finding is consistent with the results of ( 1996 ) . although the absolute numbers are small , we find that composite a has a @xmath61 excess ( using the statistical definitions of 1986 ) of radio detections as compared to composite d. in addition , we find that there is a trend in radio flux density ; the radio - detected quasars are less radio bright with increasing blueshift ; the four sample d objects have an average flux density of only @xmath62 . to test that our results are not biased by the inclusion of more radio sources in sample a as compared to sample d , we have also re - created the composite spectra after removing all of the radio detected quasars . we find that the removal of these spectra produces no qualitative differences between the original and non - radio composite spectra . unfortunately , there are not enough radio - detected quasars in our sample ( only 41 of the 794 in the / sample , many of which are only faint radio sources ) to test if there is any correlation between velocity shift and quasar orientation as measured by radio spectral index or radio core dominance . nevertheless it is interesting to note that the sdss+first quasars in our sample are strongly biased against lobe - dominated quasars , since the sdss quasar survey ( richards et al . 2002 ) requires that first - selected quasars have a match to an sdss optical source within @xmath63 . this restriction will cause the sdss to fail to target ( as radio - selected quasars ) those objects which have strong radio lobes , but no core radio emission ( many of these objects will still be targeted as color - selected quasar candidates ) . although we would expect there to be unmatched double - lobe radio sources in our sample , a search of the first database out to @xmath64 and an examination of the images does not reveal many cases , and there is no obvious trend with blueshift . however , it may be that the high resolution of the first survey causes faint lobe emission to be missed . possible evidence in support of a lack of lobe - dominated sources is the fact that of the 18 radio - detected quasars in our sample that are brighter than @xmath65 at @xmath66 , 12 are clearly core dominated as determined by visual inspection of the first ( becker et al . 1995 ) images , and only one of the other six has clear radio lobes . thus , we might draw the conclusion that there is a dearth of radio - detected quasars in sample d because there is a selection effect that causes a correlation between large velocity shift and lobe - dominated quasars ( which are thought to be seen edge - on to the plane of the disk ) . see [ sec : orientation ] for further discussion . we find a similar correlation with the x - ray detections as determined from the edr quasar catalog ( schneider et al . 2002 ) ; however , it is attributable to the fact that quasars from composite a , for whatever reason , have more pointed rosat observations . there is no trend when considering only the rosat faint source catalog sources ( voges et al . . it would be interesting to examine the x - ray properties in more detail . specifically , if the velocity shift effect is caused by orientation , and if the correlation between radio core - dominance and soft x - ray energy index is real ( shastri et al . 1993 ) , then we might expect that the quasars in sample a will have flatter soft x - ray spectral indices than the quasars in sample d. the results presented herein will help to constrain models for the belr , in particular , it should be clear that there can not be any doubt as to the reality of the systematic blueshift of emission lines ; this shift _ must _ be accounted for in any agn models . this could be considered a strike against the model of ( 2000 ) , in which the blueshift of high - ionization emission lines relative to low - ionization emission lines has `` no obvious interpretation '' . however , it is equally important that these models incorporate the new results herein , primarily that the blueshift does not seem to be a net blueshift , but rather a blueshift resulting from a lack of flux in the red wing of the emission line . for cloud models , the similar blue wings of emission in the different composites would appear to favor an intrinsically isotropic velocity field in an isotropically - emitting belr , plus obscuration that preferentially removes flux from the red wing of however , such preferential obscuration may be difficult to produce . unfortunately , the parameter space available to belr models is huge , and it becomes difficult to distinguish between using observations to constrain a model and fine tuning a model to match the observations ex post facto . thus , a detailed discussion of how our results constrain belr models is beyond the scope of this paper . however , we briefly comment on the subject . first , we make two general assumptions : 1 ) that the belr region is stratified in ionization , specifically that the low - ionization region is located farther from the continuum source than the high - ionization region ( peterson 1993 ) , but see ( 1998 ) ; and 2 ) that an optically thick accretion disk is interior to both regions . we then hypothesize that the differences between composite a and d are an orientation effect with composite a representing a face - on configuration and composite d representing a more edge - on configuration . further assuming a cloud model for the belr with a roughly isotropic distribution of _ outflowing _ clouds , one possible ( but perhaps unlikely ) scenario could involve an equatorial screen that is optically thin in the face - on direction , but becomes optically thick to clouds on the side of the accretion disk facing away from the observer in the edge - on direction ( figure [ fig : fig9 ] ) . the high - ionization clouds are assumed to be outflowing faster than the low - ionization clouds , and the dust layer is assumed to be located mostly interior to the typical emission radius . such a `` model '' could produce emission profiles similar to those that we observe . however , if the clouds were infalling , it is difficult to imagine a way for obscuration to produce the observed results . we must also consider disk - like belr geometries , such as discussed by & kartje ( 1994 ) , ( 1997 ) , & chiang ( 1998 ) , and others . if we invoke a disk - wind type model instead of a spherically symmetric cloud model , then the situation may be quite different . in this case , the flow is not purely radial : the velocity differences can be due purely to orientation and a screen is not needed . however , since we observe that the emission line profiles are not simply shifted , but also they are weaker , a screen may still be necessary . for such a disk - wind model the most obvious source of a blueshift would be an outflow in the plane of the disk . in this case , a face - on orientation would yield no net shift , whereas an edge - on orientation could produce a net shift if the far side of the disk were obscured . however , we would still have to account for the fact that our observations suggest that there is not a net blueshift , just an apparent blueshift . in addition , although our hypothesis assumes that the proposed orientation effect is a result of the orientation of the disk itself , it could instead result from a change in the opening angle of the disk - wind , see [ sec : orientation ] for further discussion . we note that & chiang ( 1997 ) were able to use a varying outer radius for the emission line gas in a disk - wind model to explain the emission profiles in ( 1993 ) . & chiang ( 1997 ) also found that additionally varying the inclination angle did not produce a better fit . however , the ( 1993 ) composites all place the peak of the emission profile at zero velocity . since the emission line profiles of our four composites are so similar to theirs , it seems quite likely that their `` composite 4 '' should be blueshifted in a similar manner to our `` composite d '' ( and so on for the other composites ) . thus , a varying outer radius can not be the whole picture , and it would be interesting to re - address this issue . in any case , it should be obvious that these observations and further study of both cloud and disk - wind models , combined with the results of reverberation mapping data and analysis ( e.g. , korista et al . 1995 ; done & krolik 1996 ) will lead to a significantly better understanding of the broad line region in agn . the baldwin effect was originally defined as an anti - correlation between the equivalent width of emission and the continuum luminosity at @xmath67 in the spectra of quasars ( baldwin 1977 ) . while the original definition is still the dominant one , the baldwin effect has come to mean any anti - correlation between quasar luminosity or absolute magnitude and the strength of the broad emission lines . since our sample shows a very clear evolution of equivalent width from large equivalent width in sample a to a much smaller equivalent width in sample d , a discussion of how the baldwin effect might relate to our results is warranted . although the mean absolute magnitudes of our sample and a histogram of the individual absolute magnitudes in each sample ( not shown ) are quite similar , we were surprised to find that there is a significant difference between the samples in terms of their absolute magnitudes . the mean values of @xmath68 for samples a and d are @xmath69 and @xmath70 , respectively . this difference has a student s @xmath71 ( press et al . 1992 ) of @xmath72 with a probability of @xmath73 , indicating a significant difference between the samples . the differences between the adjacent samples ( e.g. , a and b ) are only marginally significant . to determine if our results are influenced by luminosity , we have created control samples by constraining the absolute magnitude distributions for sub - samples of each sample . we first bin the absolute magnitudes in bins of 0.5 mag , then count how many quasars are in each bin for each sample . we then created a sub - sample for each sample with only the minimum number of quasars in each bin . in this way , we create sub - samples with the same absolute magnitude distributions . a student s @xmath71 test confirms the similarity of the sub - samples in terms of their mean absolute @xmath35-band magnitudes . using these absolute magnitude normalized subsamples , we have recreated figure [ fig : fig4 ] and find no qualitative differences as compared to the full samples . we thus conclude that changes in @xmath68 are not driving the differences between the original samples . nevertheless , a better analysis in terms of the baldwin effect is still needed . our @xmath68 values sample both the continuum and emission line flux at @xmath74 , whereas the baldwin effect is usually in terms of the continuum flux at @xmath75 . more work is clearly needed to understand the exact relationship between our observations and the balwin effect . it is particularly interesting to note that the lines that show the most differences between samples a and d are also the ones that have the strongest reported baldwin effects , in particular , and ( osmer , porter , & green 1994 ) . we also see no strong differences in the profiles or blueshifts ; no baldwin effect is observed for this line ( osmer et al . these observations are unusual in the sense that both and are high ionization lines with very similar chemical structure . finally , if the baldwin effect and the blueshifts of quasar emission lines are related , one must take the blueshifts into account when determining the luminosity of the quasars . large blueshifts will cause the redshift and thus the luminosity of the quasar to be systematically underestimated . since we have shown that those quasars with large blueshifts also have weaker emission , large emission line blueshifts will cause the the baldwin effect to appear weaker than it should . since the troughs in bal quasars removes some of the emission flux , it is nearly impossible to ask whether individual bal quasars have small or large emission line blueshifts . however , we can study how the ensemble average of bals compares to the composites that we have created . first , we divide the bals into two types : low - ionization ( lobals ) and high - ionization ( hibals ) , where , in addition to absorption by , lobals show absorption in and other low - ionization lines . the hibal composite was created by combining all of the non - lobal bal - like objects that were originally rejected from our sample ; the redshifts are all taken from as they were for our sample . the lobal composite includes the lobals from the hibal sample and includes some lobals outside of the redshift range studied in the sample in order to improve the signal - to - noise of the composite spectrum . these bal composite spectra were created in exactly the same manner as composites a , b , c , and d. the composite bal spectra were then scaled as follows . a composite spectrum ( vanden berk et al . 2001 ) was fitted to each of them in the rest wavelength frame by minimizing a weighted @xmath17 merit function , normalized in the region @xmath76 . the spectral index and reddening , @xmath77 , were allowed to vary , using a reddening law with @xmath78 ( see , whiting , & webster 2000 ) , where @xmath79 is in @xmath80 and @xmath81 . the @xmath17 function was weighted at certain wavelengths to eliminate or reduce the effects of several emission and absorption features . the weight was 0 for the major emission lines from lyman-@xmath49 to , and also for the lyman-@xmath49 forest ; 0.5 for the region between and where bal troughs are most likely to be found ; and 1 for other wavelengths . the chi - square function was then minimized using a quasi - newton method , yielding best - fit values of the spectral index and reddening . see reichard et al . ( 2002 , in preparation ) for a more detailed discussion . in order to compare the bal composites to our a , b , c , and d composites , we also scale composites a , b , c , and d in the same manner as above . thus the composites from samples a , b , c , and d will have roughly the same ( overall ) continuum slope and shape as the bal composites , which allows comparison of their emission line features . from figure [ fig : fig10 ] , it is quite clear that the lobal composite is most similar to composite d : the red wing of matches quite well as does the ] blend . in addition , the lobal composite also shows a clear weakening of emission . the hibal composite is most similar to composite c ; the and ] emission line profiles are good matches . also note that , based upon the upper left - hand panel , it would appear that the bal outflows begin right at zero velocity with respect to the systemic redshift ( as defined by ) . since the viewing angles of bals are generally thought to be constrained to a relatively small angle from the accretion disk plane , that the lobal composite best matches composite d is consistent with the idea that composite d is composed of quasars that are preferentially observed in an edge - on configuration . the fact that the hibal composite is most similar to composite c is also consistent with the orientation hypothesis : hibals may differ from lobals in that the line of sight passes through somewhat less dense gas clouds that are more highly ionized . these correlations would seem to be inconsistent with , but do not necessarily rule out , the idea that bals may ( also ) be an evolutionary stage of all quasars ( becker et al . 2000 ) . not only does the blueshift of the emission line affect the study of bal absorption , it also affects our understanding of narrow , `` associated '' absorption ( foltz et al . these are absorption systems that are preferentially within @xmath82 of the quasar redshift ; they occur at a much higher rate ( per unit redshift ) than absorbers which are thought to be caused by intervening galaxies at redshifts much smaller than the quasars . since their discovery , two hypotheses have dominated the possible explanations , ( 1986 ) suggested that 1 ) they are due to outflows , similar to , but weaker than bals , or 2 ) they are due to a virialized cluster - like velocity dispersion . however , the fact that the emission line can be blueshifted changes the situation significantly ( mcintosh et al . 1999 ) . a full analysis of this issue is beyond the scope of this work ; however , we have performed a simple test that helps to shed light on this subject . one of us ( gtr ) examined the emission line regions of all the spectra that comprise composites a and d , and two samples of quasars ( one each from samples a and d ) with good signal - to - noise and strong associated absorption were constructed . these samples were further restricted to those objects showing the largest associated absorption redshifts ( i.e. , towards or in the red wing of ) . we present the emission line regions of some of the quasars from these biased sub - samples in figure [ fig : fig11 ] . the quasars on the left - hand - side are from sample a and have small emission line blueshifts with respect to , whereas the quasars on the right - hand - side are from sample d and have larger emission line blueshifts with respect to . what is interesting about figure [ fig : fig11 ] is that , in almost every case , the associated absorption is at or shortward of the systemic redshift ( dashed line , as measured from ) , and thus are consistent with outflows . large virial velocities are not necessary to explain the `` infalling '' absorbers . furthermore , in only one sample a case does the absorption occur entirely in the red wing of the emission profile . for this one exception , we suspect that a blueshift that has a larger error than normal is to blame ; when we remeasured the redshift by hand , we found that the redshift from the automated code was too small by @xmath83 . the expected location of using the revised redshift is indicated by the dotted line in the bottom panel on the left - hand side of figure [ fig : fig11 ] . however , given the expected errors in the individual measurements , it it unlikely that the errors will produce a symmetric distribution around the line center , but quite possible that the errors could explain any of the remaining apparent inflows . that is , if the velocity shifts of associated absorption lines were always referenced to the systemic redshift , we might find that associated absorbers are all outflows . since ( 1986 ) and others ( e.g. , richards et al . 2001 ) have found that the strong associated absorbers are predominantly found in steep - spectrum radio - loud quasars , it is interesting to think about them in terms of an orientation hypothesis . first , we note that the presence of associated absorption in our sample a quasars may be an argument _ against _ an orientation interpretation for the emission line blueshifts . steep - spectrum quasars are thought to have a more edge - on orientation , whereas flat - spectrum quasars are thought to be more face - on . thus the observation that associated absorbers are predominantly found in steep - spectrum quasars and that our hypothesized face - on quasars can have strong associated absorption is apparently inconsistent . however , it may be possible to reconcile this problem in the following manner . if we define `` associated absorption '' as any absorption that occurs within the profile of ( say within @xmath84 of the emission line peak regardless of whether the peak is blueshifted or not ) , _ and _ we assume that the associated absorbers are purely outflows , then we would expect to see a higher density of associated absorption per unit velocity in the objects in sample d ( assuming that they are seen edge - on ) . this is because there is a bigger range of velocities in which the absorbers can be found in sample d quasars , since the emission line is blueshifted . for example , in a quasar with no blueshift of the emission line , the observable range of associated absorption velocities is only @xmath40 , but in a quasar where the emission line is blueshifted by @xmath40 , the observable range of associated absorption is @xmath85 ; in both cases we _ assume _ that the full range over which associated absorption can be seen is @xmath85 , so a sample of quasars with blueshifted emission lines would be expected to have a higher density of associated absorption . if sample d objects tend to be steep - spectrum , lobe - dominated objects when they are radio sources , this could explain the excess of associated absorbers seen in steep - spectrum , lobe - dominated radio - loud quasars . if associated absorbers are really outflows , then they may be present at velocities much larger than @xmath40 from the emission redshift of the quasar , as was suggested by ( 2001 ) . thus , there may be a large population of narrow , intrinsic absorption systems in flat - spectrum face - on quasars , see ( 2001 ) and references therein . what causes the blueshifted emission lines in quasars and other similar phenomena ( e.g. , the baldwin effect ) ? orientation , specifically orientation of the plane of the accretion disk is certainly one possibility . we have presented some results that are consistent with this hypothesis . we have shown that bal quasars are most similar to the quasars from samples c and d. since most models of bals have the bal clouds confined to a region within a small angle of the disk ( or the disk - wind ) , this observation may be consistent with an orientation effect . the case for associated absorption , which is preferentially seen in steep - spectrum ( edge - on ) quasars , is less clear , but could also be consistent with an orientation effect . the radio properties of our sample may also support the hypothesis that orientation plays a key role in the blueshifting of the emission line . we find that there are significantly more radio - detected quasars in sample a than sample d and that the sample a objects are brighter at radio wavelengths . if the sdss+first matching algorithm is failing to find lobe dominated quasars , or if first is failing to detect some lobe dominated quasars , an orientation effect might explain the lack of radio - detected sources in sample d. furthermore that the sample a sources are bright , whereas the sample d sources are faint could be the result of an orientation - related beaming effect . an edge - on orientation for the largest shifts may also be consistent with the reduced emission and the lack of emission . these are both high - ionization lines that are likely to be formed closer to the source of ionizing radiation than low - ionization lines and thus would be easier to obscure than the low - ionization lines which are presumably formed further out . if the obscuration is restricted to the plane of the presumed accretion disk , then we might expect that some of the high - ionization lines would be completely or partially obscured in an edge - on orientation . furthermore , if the disk - wind model for the belr is true , then we might expect a broadening of the lines with increasing disk inclination angle as seen in h@xmath33 by & browne ( 1986 ) . , wilkes , & barthel ( 2000 ) find that systems which appear to be inclined according to their radio properties have broader emission line wings , which is consistent with the material being seen in a preferred direction . we do indeed see an increase in the fwhm of with an increase in our proposed viewing angle ( see table 2 ) . however , as we discussed earlier , it is not clear that this broadening of the line is real ; it may instead be due to partial obscuration of the line . a possible argument _ against _ the orientation hypothesis is the fact that & hunstead ( 1995 ) found significant differences in the strength of the @xmath56 bump between composite spectra that are core- and lobe - dominated ( core - dominated having stronger @xmath56 bump emission ) . in their sample , the @xmath56 bump is barely visible for steep - spectrum quasars and virtually disappears for _ compact _ steep - spectrum quasars . since core - dominance is thought to be correlated with orientation in radio - loud quasars , this change in @xmath56 bump strength may also be related to orientation . although we do see some differences in the strength of @xmath86 bump between composites a and d , they do not appear to be anywhere near as apparent as those found by & hunstead ( 1995 ) . if both our effects and the & hunstead ( 1995 ) effects are caused by orientation , we might expect to see a bigger difference in the fluxes between composites a and d. although the unified model ( e.g. , urry & padovani 1995 ) is well - established , more and more emphasis is being placed on properties other than orientation in terms of explaining differences between quasars . these include luminosity , the eddington ratio , accretion rate , black hole mass , black hole spin and age ( e.g. , laor 2000 ) . for example , & green ( 1992 ) argue that the first eigenvector from their principal component analysis ( `` eigenvector 1 '' which illuminates some correlations between intrinsic quasar properties ) is driven by something other than external orientation . even though the properties of our sample are very similar to the `` eigenvector 1 '' properties ( see & francis 1999 , table 2 ) , we prefer orientation as an explanation . however , this orientation effect need not be specific to the orientation of the accretion disk ( i.e. external orientation ) , but rather it could be related to the opening angle of a disk wind ( such as in 2000 , figure 7 ) , where some other quasar property is causing the opening angle of the disk wind to change , thus producing an orientation type effect ( i.e. internal orientation ) . a strong argument for this type of behavior ( or against orientation entirely depending on how you look at it ) is the intrinsic baldwin effect ( pogge & peterson 1992 ) . since a baldwin type effect is seen in individual quasars as their continuum luminosity varies ( but with a different slope than the `` external '' baldwin effect ) , the baldwin effect can not be entirely due to the orientation of the accretion disk ; the same is true for the emission line blueshifts if they are related to the baldwin effect . however , if the orientation effect is really the wind opening angle or the angle the wind makes with the disk axis and these angles can change on reasonably short timescales , then this might account for the major differences in quasar spectra . in any case , it is certainly possible that the size of the emission line blueshift may be a function of the orientation of the quasar . whether the orientation in question is external or internal remains to be seen . further study , particularly with regard to radio and x - ray properties , is needed to determine if this is really the case . some areas of future work are beyond the scope of this paper , but are clearly appropriate , including the following : \1 ) ideally , we would like to know the velocity offset between the and [ ] emission lines , which requires that we obtain ir spectra of the [ ] emission line region for those quasars where we already cover and . \2 ) similarly , we should attempt to extend our low redshift sample to cover the emission line region by obtaining uv spectra of these quasars . \3 ) uv spectra would also be appropriate for those objects where we can not see the lyman-@xmath49 emission line , since the profile of lyman-@xmath49 may be helpful in constraining belr models ( kallman & krolik 1986 ) . \4 ) if we could find a way to know how large the blueshift is likely to be without looking at the emission line , it would be possible to study how the blueshifts correlate with broad absorption line properties . \5 ) a detailed study of associated absorbers among the quasars studied herein is clearly in order . in many cases , this will mean getting higher signal - to - noise spectra . \6 ) since the blueshifts so strongly correlate with the equivalent width of the emission lines , a more detailed investigation of how this affects the baldwin effect is in order . \7 ) it is possible that emission from and could be influencing some of the changes that we attribute to other species . removing the iron emission complexes using the & wilkes ( 2001 ) iron template would address this issue . we have not chosen to attempt this procedure here , in part because the & wilkes ( 2001 ) iron template is not complete in the critical regions near and emission . \8 ) last , but perhaps most important , is that future agn `` models '' must not fail to take into account these emission line blueshifts . furthermore , they must be able to explain the emission line profiles . these should be key ingredients to any model . using two samples of quasars from the sdss early data release quasar catalog ( schneider et al . 2002 ) , we have studied the differences in redshift of quasars as measured by different emission lines . in one sample ( with 794 quasars in the redshift range , @xmath11 ) , we compared the redshifts as determined from the peak of the emission line to those as determined from the peak of the emission line . in the second sample ( with 417 quasars in the redshift range , @xmath87 ) , we compared the redshifts from to the redshifts from [ ] . our conclusions based on these two samples can be succinctly summarizes as follows : \2 ) this shift is apparently not a bulk shift , but rather is due to a lack of flux in the red wing of the emission line profile . this fact is a new result , but is not entirely without historical precedent . although this conclusion depends on the placement of the continuum , we feel that our handling of the continuum is appropriate and that this conclusion is robust . \4 ) the fwhm of the emission line increases with increasing blueshift of the peak of the emission line ; however , this effect may not be real . for example , if it is true that the red wing is simply attenuated for sample d quasars , then the fwhm will not have the same meaning that it does if there is no attenuation . \5 ) the blueshift of the emission line may correlate with the orientation of the quasar as measured by radio properties ( large blueshift having an equatorial viewing angle ) ; further work is clearly needed . \8 ) the spectra of bal quasars appear to be most similar to the spectra of quasars with large emission line blueshifts . if the emission line blueshift is correlated with orientation , this supports the conclusion that bals are correlated with orientation ( whether external or internal ) , and vice versa . \9 ) it is possible that redshifted `` associated '' absorption population may be explained by the blueshift of emission ( as has been suggested before ) . even if there are inflows , we find no need for extremely large velocities in the `` virialized '' cluster material explanation . we acknowledge helpful discussions with mike eracleous , julian krolik , arieh knigl , nahum arav , niel brandt , marianne vestergaard , and brad peterson . we also thank bev wills , michael brotherton , and sarah gallagher for their comments on the manuscript . we thank the anonymous referee for suggestions that helped improve the paper , particularly for the suggestion that we include more discussion of the baldwin effect . gtr , tar , and dps acknowledge support from nsf grant ast99 - 00703 . pbh acknowledges financial support from chilean grant fondecyt/1010981 and a fundacin andes grant . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss web site is http://www.sdss.org/. the participating institutions in the sdss are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , princeton university , the united states naval observatory , and the university of washington . crrrcccc a & 1.80 & @xmath69 & @xmath91 & 21/13.1 & 4/2.0 & 30.27 & 28.17 + b & 1.80 & @xmath92 & @xmath93 & 11/6.7 & 8/4.0 & 25.75 & 28.51 + c & 1.84 & @xmath94 & @xmath95 & 5/3.0 & 3/1.5 & 22.43 & 30.43 + d & 1.82 & @xmath70 & @xmath96 & 4/2.2 & 2/2.0 & 17.96 & 32.40 +

we determine the widths of three - body decays of sleptons , , in the presence of arbitrary slepton flavor violation and left - right mixing . these decays are important in scenarios in which the lightest supersymmetric particle is the gravitino , a generic possibility in models with gauge- and gravity - mediated supersymmetry breaking . three - body decays have been discussed previously assuming flavor conservation and left - right mixing in only the stau sector . flavor violation and general left - right mixing open up many new decay channels , which provide new avenues for precision mass measurements and may play an essential role in solving the standard model flavor problem . we present results for toy models with two - generation mixing , and discuss the implementation of these results in spice , a program that simplifies collider event simulations of flavor - violating supersymmetric models .

0904.1416

fermion masses are one of the least understood parts of the standard model ( sm ) . even the charged fermion masses span over five orders of magnitude from the top quark to the electron , begging for a theoretical explanation . the majority of these masses and mixing angles are precisely constrained by experiment . still , this wealth of data does not conclusively single out any theory of flavor . in the near future , this may change dramatically with the discovery of new particles at the weak scale . new particles may only deepen the mystery , as would be the case if a fourth generation were discovered . on the other hand , the masses and mixings of the new particles may be governed by the same principles that determine the sm fermion masses . in this case , rather than extending the fermion sector , the new particles will shed light on the existing fermion spectrum . weak - scale supersymmetry provides examples of both possibilities . in pure gauge - mediated models , for example , squark and slepton masses are set by flavor - blind contributions , with no connection to the sm fermion masses . however , in gravity - mediated models and hybrid models with both gauge- and gravity - mediated contributions , squark and slepton masses may receive contributions that are governed by flavor symmetries that also determine the sm fermion masses @xcite . the latter possibility leads to non - trivial flavor effects in high - energy experiments @xcite , opening the possibility for real progress on the sm flavor problem at the large hadron collider , as has been emphasized recently by many authors ( see , for example , refs . @xcite ) . here we study the implications of flavor violation for the three - body decays of charged sleptons . such processes are most relevant for colliders in models with a gravitino lightest supersymmetric particle ( lsp ) and a slepton next - to - lightest supersymmetric particle ( nlsp ) , a generic possibility in models with both gauge - mediated supersymmetry breaking @xcite and gravity - mediated supersymmetry breaking @xcite . we consider a situation where the lightest particles are a gravitino lsp , several light sleptons , and the lightest neutralino , with the masses of each light slepton less than the mass of the lightest neutralino . this results in the typical two - body decays of these light sleptons being highly suppressed or kinematically inaccessible , leaving the three - body decays as the dominant decay modes . in these cases , the three - body decays are also often the last visible step in cascade decays of squarks and gluinos , and so they impact nearly all supersymmetry searches and studies . three - body slepton decays have been studied previously in an important , flavor - conserving case , where the authors considered @xmath1 and @xmath2 , with the @xmath3 a mixture of left- and right - handed staus @xcite . these decays are characterised by two distinct channels : a `` charge - preserving '' channel @xmath4 with opposite - sign leptons and a `` charge - flipping '' channel @xmath5 with same - sign leptons . flavor - conserving three - body decays of squarks have also been considered @xcite , as have flavor - conserving three - body decays with sneutrinos as parent or daughter particles @xcite . our work generalizes the charged slepton analysis to the case of arbitrary lepton flavor violation ( lfv ) and arbitrary left - right mixing . in the presence of general lfv , any three - body decay of @xmath6 has up to 9 possible charge - preserving modes @xmath7 and 6 possible charge - flipping modes @xmath8 , where @xmath9 . in addition , lfv and left - right mixing bring additional complications that are absent in the flavor - conserving case , including new processes mediated by higgs and @xmath10 bosons , new final states with neutrinos and quarks , and new interference effects in charge - flipping processes . these complications are well worth confronting , however , as there is a wealth of information in these branching ratios , which may shed light on the sm flavor problem . these decays , if present , are also relevant more broadly , for example , for supersymmetric searches and precision mass measurements . for example , in the flavor - conserving examples studied previously @xcite , the final states necessarily contain @xmath11 leptons . since these decay with missing energy , they degrade searches based on energetic leptons , and they greatly reduce the prospects for precision mass measurements . with lfv , however , even if the lightest slepton is a stau , there may also be decay modes with two electrons , two muons , or an electron and a muon . even if these branching ratios are suppressed , they may be the more obvious signals at colliders , and they may also provide better opportunities for precision mass measurements . it is therefore of interest to know the size of these branching ratios , and what determines them . in the following three sections , we begin with a general discussion of three - body decays and move gradually to more specific scenarios and concrete calculations . in sec . [ sec : diagrams ] we present the new final states and new feynman diagrams that are relevant to three - body decays once general flavor and left - right mixing are introduced . in sec . [ sec : widths ] we discuss these results in more detail and explain the relative phenomenological importance of the various contributing diagrams in particular scenarios . in sec . [ sec : plots ] we then show concrete results in two toy models with two - generation slepton mixing to illustrate our results . finally , in sec . [ sec : conclusion ] , we present our conclusions and explain how our results have been incorporated into spice , a publicly available computation package for generating supersymmetric spectra and branching ratios in scenarios with arbitrary slepton mixing . we stress that , although we strive to give readers a intuitive feel for our results by considering concrete cases in the body of the paper , our analysis is valid for fully general lfv and left - right mixing . the complete , model - independent calculation is lengthy , but the full expressions for all three - body decay modes are given in a series of appendices . our conventions and notations are defined in appendix [ app : lagrangian ] . these are consistent with those of ref . @xcite , where full details may be found . as mentioned in the introduction , an important special case of three - body slepton decays has been discussed previously by ambrosanio , kribs , and martin @xcite . motivated by pure gauge - mediated models , these authors considered the flavor - conserving decays @xmath1 and @xmath12 . these decays are mediated solely by neutralinos . the charge - preserving modes @xmath7 are shown in fig . [ fig : preserving ] , and the charge - flipping modes @xmath13 are shown in fig . [ fig : flipping ] . the charge - flipping mode is made possible by the majorana nature of the neutralino , and detection of the charge - flipping mode would provide strong evidence that neutralinos are majorana fermions . mediated by neutralinos and @xmath14 mediated by charginos.,width=491 ] mediated by neutralinos.,width=491 ] in the presence of more general slepton mixing , the gauge eigenstates @xmath15 , @xmath16 , @xmath17 , @xmath18 , @xmath19 , and @xmath20 mix to form six mass eigenstates , @xmath21 , @xmath22 , with increasing mass , and the lepton - slepton - neutralino interactions are no longer flavor - diagonal . the neutralino diagrams of figs . [ fig : preserving ] and [ fig : flipping ] are then modified by the inclusion of @xmath23 mixing matrix factors at the interaction vertices . in addition , new diagrams contribute . if the initial and final state charged sleptons contain left - handed components , there is the chargino - mediated decay to neutrinos shown in fig . [ fig : preserving ] . this decay requires neither flavor violation nor left - right mixing . there are also charge - preserving decays mediated by neutral higgs bosons and the @xmath10 boson , as shown in fig . [ fig : preservinghz ] . the higgs bosons mediate decays to same flavor @xmath24 and @xmath25 pairs , and the @xmath10 diagram mediates decays to same flavor @xmath26 , @xmath24 , and @xmath27 pairs . bosons . the higgs scalar @xmath28 may be any of the neutral higgs bosons : @xmath29 , @xmath30 , @xmath31 , or ( in feynman gauge ) @xmath32.,width=491 ] both the @xmath28 and @xmath10 diagrams are present independent of lfv , but both require left - right mixing . in the case of the higgs - mediated diagrams , the higgs couplings to sleptons have the form @xmath33 , @xmath34 , and @xmath35 . the first is generated by left - right mixing terms in the lagrangian , and thus clearly requires left - right mixing to be non - zero . the other two come from @xmath36 terms , and in the absence of left - right mixing , they are flavor - diagonal in the separate @xmath37 and @xmath38 sectors , even with the presence of lfv in one or both . thus , in the absence of left - right mixing , the higgs coupling to sleptons is flavor - diagonal in the 6-dimensional space of left and right sleptons and does not contribute to three - body decays . similarly , the @xmath10 couples to the @xmath37 and @xmath38 sleptons separately , so in the absence of left - right mixing , its couplings are also flavor - diagonal . note , however , there are no photon - mediated diagrams at tree level , as the photon couples with equal strength to the left- and right - handed sleptons , and so its couplings are always flavor - diagonal . also note that in the case of non - zero left - right mixing but no flavor mixing , the @xmath28 and @xmath10 diagrams exist , but their contributions are only appreciable for certain mass orderings . namely , only when neither of the two mass eigenstates produced by left - right mixing posses kinematically allowed two - body decays will the three - body @xmath28 and @xmath10 contributions be relevant . the charge - flipping decay is unaffected by the higgs and @xmath10 boson diagrams , but in the presence of flavor violation , the fermions in the final state may be interchanged . this implies a new interference effect in the charge - flipping mode which is absent in the flavor - conserving case . in principle , there should also be a similar interference term in the neutrino decay modes if neutrinos possess majorana masses , but such a term is negligible in the limit of zero neutrino mass . in summary , for charge - preserving decays to charged leptons of different generations , the analysis of ref . @xcite may be straightforwardly modified to the general case . the only modification needed is the insertion of rotation matrices from flavor eigenstates to gauge interaction eigenstates . for charge - preserving decays to charged leptons of the same generation , however , the higgs and @xmath10 diagrams introduce new contributions to the decay width . there are also new charge - preserving decays to neutrinos , mediated by charginos and @xmath10 bosons , and to quarks , mediated by higgs and @xmath10 bosons . finally , in the case of the charge - flipping decays to like - sign leptons , the daughter leptons may be interchanged and thus the decay width must include an interference term between these two diagrams ; however , there are no higgs and @xmath10 contributions to this mode and so there are no further interference terms . given an understanding of the qualitatively new features introduced by flavor and left - right mixing described in sec . [ sec : diagrams ] , we can now calculate the resulting decay widths . the full results are lengthy and are given in the appendices . in this section , we summarize the results of our calculations and provide qualitative insight into the relative importance of each contributing mode . our slepton interaction lagrangian is @xmath39 + i \zeta^{(2)}_{ij } \left(\tilde{\ell}_i^ * \partial_{\mu } \tilde{\ell}_j - \tilde{\ell}_j \partial_{\mu } \tilde{\ell}_i^ * \right ) z^{\mu } \nonumber \\ & & + \sigma_{ij}^{(2 ) } \tilde{\ell}_i^ * \tilde{\ell}_j h^0 + \sigma_{ij}^{(3 ) } \tilde{\ell}_i^ * \tilde{\ell}_j h^0 + i \sigma_{ij}^{(4 ) } \tilde{\ell}_i^ * \tilde{\ell}_j a^0 + i \sigma_{ij}^{(5 ) } \tilde{\ell}_i^ * \tilde{\ell}_j g^0 \ , \label{lint}\end{aligned}\ ] ] where the coefficients @xmath40 , @xmath41 , @xmath42 , and @xmath43 contain gauge and yukawa couplings along with the mixing matrix elements for sleptons , higgs bosons , neutralinos , and charginos . these coefficients are defined in appendix [ app : lagrangian ] . here , the indices are summed over the six slepton , three lepton and neutrino , four neutralino , and two chargino mass eigenstates . we have kept only the terms relevant to the three - body decays in question . our lepton interaction lagrangian is @xmath44 the amplitudes for charge - preserving decays to charged leptons @xmath45 are @xmath46 where the indices @xmath47 , @xmath48 , @xmath49 , and @xmath50 correspond to the subscripts specifying the mass eigenstates of initial and final state particles in the decay mode @xmath51 , and @xmath52 is the kronecker delta function . here and throughout the rest of this section we suppress these indices on matrix elements and decay widths . the overall decay width is given by @xmath53 \right ) \\ & = & \gamma_{\tilde{\chi}^0 \tilde{\chi}^0 } + \delta_{km } \left(\gamma_{zz } + \gamma_{hh } + 2 \gamma_{z \tilde{\chi}^0 } + 2 \gamma_{h \tilde{\chi}^0 } + 2 \gamma_{zh } \right ) \ , \end{aligned}\ ] ] where @xmath54 is the integral over three - body phase space discussed in appendix [ app : phasespace ] which includes a sum over fermion spins , and we use an intuitively obvious notation for widths , so that , for example , @xmath55 is the partial width from @xmath56 . to develop a qualitative understanding of which matrix elements are typically dominant and sub - dominant , we first note that there is a suppression to the higgs and @xmath10 modes from left - right slepton mixing . in fact , the explicit higgs and @xmath10 couplings to sleptons depend on the left - right slepton mixing squared , @xmath57 in particular , for @xmath58 not significantly larger than the slepton mass , this is roughly the same order as yukawa suppression . we thus expect that , at the matrix element level , the higgs or @xmath10 radiated modes are generically suppressed by two powers of left - right mixing relative to the neutralino and chargino modes . second , to simplify phase space factors , we assume a typical amount of mass squared splitting @xmath59 ; we also neglect lepton masses when they are sub - dominant . under these assumptions , the difference in propagator structure between the neutralino mode and that of the higgs and @xmath10 modes becomes apparent . the neutralino propagator is inversely proportional to @xmath60 , while the higgs ( @xmath10 ) propagator is inversely proportional to @xmath61 , which simply reduces to @xmath62 if the slepton masses are not too far above @xmath63 . thus , in models where the lightest neutralino and light slepton masses are close , the lightest neutralino pole contribution will be enhanced over the higgs and @xmath10 mass - suppressed contributions . conversely , as the slepton and neutralino mass scale grows , the higgs and @xmath10 contributions will drop off more slowly than the neutralino contribution . the three - body decays with two neutrinos or two quarks have much the same form as above with minor changes ( refer to the appendices for details ) . these changes are as follows : for the neutrino mode , @xmath64 and @xmath65 , and the lepton masses are set to zero , hence simplifying the phase space calculation and removing the off - shell higgs contribution . for the quark modes , there is no off - shell fermion intermediary , and the higgs and @xmath10 modes are only modified with adjusted couplings and quark masses . the quark modes are , however , enhanced by the color factor and a sum over light flavors . because of these changes , however , all of these modes are suppressed relative to the di - lepton mode by left - right mixing . the quark modes only have contributions from the higgs and @xmath10 diagrams , and thus are suppressed by left - right mixing as noted above . in the majority of models with a light slepton nlsp , the @xmath66 gauge eigenstates are generically lighter than the @xmath67 gauge eigenstates ; for such a model the neutrino modes are suppressed since the neutrino only couples to @xmath67 gauge eigenstates while the light sleptons are primarily @xmath66 states . indeed , in such a model @xmath68 so the neutrino and quark decay widths are all suppressed by @xmath69 relative to the dilepton modes . finally , for the charge - flipping di - lepton decay , the matrix element is given by @xmath70 where the negative sign comes from fermi statistics . the decay width is then @xmath71 with @xmath72,\ ] ] and @xmath73 is a phase space factor which has a value of 1 for two outgoing leptons of different flavor and 1/2 for two outgoing leptons of the same flavor due to indistinguishable particle statistics . @xmath74 and @xmath75 both have the same basic form as the result from @xcite , where the only change is the insertion of flavor - mixing coefficients and , for @xmath75 , interchange of @xmath76 and @xmath77 between the two terms ; the new @xmath78 is presented in appendix [ app : gamma21 ] . we find that this charge - flipping decay width is of the same order of magnitude as the charge - preserving di - lepton mode , though the flavor structure is markedly different : this is discussed in the next section . to validate our results and investigate their phenomenological implications , we examine some simple cases of two - slepton mixing . in these examples we consider spectra with fairly degenerate sleptons and lightest neutralino . the slepton - neutralino degeneracy is motivated by simple gauge - mediated supersymmetry breaking scenarios with not too many messengers . larger splittings are , of course , possible if there are many messengers or in other frameworks , such as minimal supergravity . the sleptons are taken fairly degenerate so that large mixing angles are consistent with low energy constraints and so may be considered . we note , however , that these examples are merely illustrative , and our results are valid in any chosen framework with arbitrary mass splittings . first , we consider @xmath79 mixing parameterized by @xmath80 fig . [ fig : smuse ] shows the flavor - violating decay widths @xmath81 with electron and/or muon leptons as a function of the mixing angle @xmath82 . here both leptons are taken to be explicitly massless and left - right slepton mixing is set to zero , thereby cutting off the higgs and @xmath10 modes . @xmath83 and @xmath84 are assigned masses of 100 gev and 105 gev , respectively , and the lightest neutralino is given a mass of 110 gev ( heavier neutralino contributions are small in this case ) . the left plot in fig . [ fig : smuse ] , which shows the decay widths for the charge - preserving channel , demonstrates the typical structure of two - slepton mixing : all decay widths are at most @xmath85 periodic , the sleptons interchange roles ( @xmath86 and @xmath87 ) at @xmath88 , the @xmath89 and @xmath90 modes are equal at all mixing angles , and at @xmath91 all decay widths are equal . in contrast , the right plot in fig . [ fig : smuse ] , which shows the charge - flipping channel , demonstrates a different flavor structure . since the @xmath92 and @xmath93 modes in general contribute to the same decay width , all widths are @xmath94 periodic . also , the @xmath95 and @xmath96 modes are again equal at all mixing angles . in addition , note that the @xmath93 mode drops to zero at @xmath91 , because the charge - flipping mode has two diagrams at tree level which cancel at @xmath91 for the @xmath93 mode . then , as expected , the @xmath95 and @xmath97 modes have decay widths that are half the total width at @xmath91 . gev , @xmath98 gev , and @xmath99 gev.,title="fig:",width=302 ] gev , @xmath98 gev , and @xmath99 gev.,title="fig:",width=302 ] next we consider @xmath100 mixing , with massive leptons and a mixing angle @xmath101 given by @xmath102 figure [ fig : stause ] shows the widths for @xmath103 , where the leptons are electrons and/or taus . @xmath83 and @xmath104 have masses of 100 gev and 105 gev , and lightest neutralino mass is again 110 gev . left - right mixing is again neglected . the first notable feature in figure [ fig : stause ] is the separation between the @xmath105 and @xmath106 modes : the @xmath105 mode is almost unchanged numerically from the previous case , while the @xmath107 mode is suppressed by phase space constriction , as expected . likewise , the @xmath108 and @xmath109 modes are also suppressed relative to the massless case , but certainly less suppressed than the @xmath106 mode . note that the total decay width is no longer constant because of the nonzero tau mass . gev , @xmath110 gev , and @xmath99 gev.,title="fig:",width=302 ] gev , @xmath110 gev , and @xmath99 gev.,title="fig:",width=302 ] we have determined the decay widths of three - body slepton decays , including both `` charge - preserving '' processes @xmath111 and `` charge - flipping '' processes @xmath5 , in the presence of arbitrary lepton flavor violation ( lfv ) and arbitrary left - right mixing . such processes are particularly relevant in scenarios with a gravitino lsp and a slepton nlsp , where they are typically the dominant decay of some of the non - nlsp sleptons and are present as the last step in many susy cascade decays . arbitrary @xmath23 slepton mixing leads to many new diagrams and new decade modes , as well as new interference effects . our results are fully general , but we have illustrated them for two simple cases with 2-generation lfv . the branching ratios to new modes may be significant , even accounting for low - energy constraints on lfv . the new modes may substitute final state electrons and muons for taus , potentially providing more obvious signals and better prospects for precision mass measurements . in addition , the measurement of these branching ratios may help pin down the supersymmetric flavor parameters and shed light on the sm flavor problem . we note that our illustrative examples have assumed fairly small slepton mass splittings so that large mixing angles are consistent with low energy constraints . it is also possible , of course , that the splittings are larger and the mixing angles are smaller . the calculations presented here are valid also in those cases , and it would also be very interesting to determine whether such small mixings may be established through collider studies . the full calculation presented here is lengthy , but has been incorporated into spice : simulation package for including flavor in collider events , a publicly available code . spice takes as input a flavor - conserving model , such as minimal gauge - mediated supersymmetry breaking , and arbitrary lepton flavor - violating parameters , and generates the full supersymmetric spectrum , complete with flavor - violating branching ratios . the three - body decays discussed here are included in spice . details on obtaining and running spice are described in the spice user manual @xcite . we are grateful to y. nir for many helpful discussions and a careful reading of the manuscript . the work of jlf , ds , and fy was supported in part by nsf grant no . phy0239817 and no . phy0653656 and the alfred p. sloan foundation . ig thanks the uc irvine particle theory group for their hospitality while this work was in progress . this research was supported in part by the united states - israel binational science foundation ( bsf ) under grant no . 2006071 . the research of ig and ys is also supported by the israel science foundation ( isf ) under grant no . 1155/07 as given in eq . ( [ lint ] ) , the relevant flavor - general interactions for this study are given by the lagrangian @xmath39 + i \zeta^{(2)}_{ij } \left(\tilde{\ell}_i^ * \partial_{\mu } \tilde{\ell}_j - \tilde{\ell}_j \partial_{\mu } \tilde{\ell}_i^ * \right ) z^{\mu } \nonumber \\ & & + \sigma_{ij}^{(2 ) } \tilde{\ell}_i^ * \tilde{\ell}_j h^0 + \sigma_{ij}^{(3 ) } \tilde{\ell}_i^ * \tilde{\ell}_j h^0 + i \sigma_{ij}^{(4 ) } \tilde{\ell}_i^ * \tilde{\ell}_j a^0 + i \sigma_{ij}^{(5 ) } \tilde{\ell}_i^ * \tilde{\ell}_j g^0 \ .\end{aligned}\ ] ] the coefficients are @xmath112 \\ \nonumber \sigma_{ij}^{(2 ) } & = & - \left [ \left ( \frac{g m_w}{2 } ( 1 - \tan^2 \theta_w ) \sin ( \theta_h + \beta ) - \frac{g m_{\ell_c}^2 \sin \theta_h } { m_w \cos \beta } \right ) u_{c , i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell } ) } \right . \\ & & \quad + \left ( g m_w \tan^2 \theta_w \sin ( \theta_h + \beta ) - \frac{g m_{\ell_c}^2 \sin \theta_h } { m_w \cos \beta } \right ) u_{c+3,i}^{(\tilde{\ell } ) * } u_{c+3,j}^{(\tilde{\ell } ) } \nonumber \\ & & \quad \left . - \frac{g m_{\ell_c}}{2 m_w \cos \beta } \left(\mu \cos \theta_h + a_c^{\tilde{\ell } } \sin \theta_h \right ) \left ( u_{c , i}^{(\tilde{\ell } ) * } u_{c+3,j}^{(\tilde{\ell } ) } + u_{c+3,i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell})}\right ) \right ] \\ \nonumber \sigma_{ij}^{(3 ) } & = & \left [ \left ( \frac{g m_w}{2 } ( 1 - \tan^2 \theta_w ) \cos ( \theta_h + \beta ) - \frac{g m_{\ell_c}^2 \cos \theta_h } { m_w \cos \beta } \right ) u_{c , i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell } ) } \right . \\ & & \quad + \left ( g m_w \tan^2 \theta_w \cos ( \theta_h + \beta ) - \frac{g m_{\ell_c}^2 \cos \theta_h } { m_w \cos \beta } \right ) u_{c+3,i}^{(\tilde{\ell } ) * } u_{c+3,j}^{(\tilde{\ell } ) } \nonumber \\ & & \quad \left . + \frac{g m_{\ell_c}}{2 m_w \cos \beta } \left(\mu \sin \theta_h - a_c^{\tilde{\ell } } \cos \theta_h \right ) \left(u_{c , i}^{(\tilde{\ell } ) * } u_{c+3,j}^{(\tilde{\ell } ) } + u_{c+3,i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell})}\right ) \right ] \\ \sigma_{ij}^{(4 ) } & = & \frac{g m_{\ell_c } } { 2 m_w } ( \mu + a_c^{\tilde{\ell } } \tan \beta ) \left ( u_{c , i}^{(\tilde{\ell } ) * } u_{c + 3 , j}^{(\tilde{\ell } ) } - u_{c + 3 , i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell } ) } \right ) \\ \sigma_{ij}^{(5 ) } & = & \frac{g m_{\ell_c } } { 2 m_w } ( \mu \tan \beta + a_c^{\tilde{\ell } } ) \left ( u_{c , i}^{(\tilde{\ell } ) * } u_{c + 3 , j}^{(\tilde{\ell } ) } - u_{c + 3 , i}^{(\tilde{\ell } ) * } u_{c , j}^{(\tilde{\ell } ) } \right).\end{aligned}\ ] ] @xmath113 and @xmath114 are the slepton and sneutrino mixing matrices , respectively . these relate the slepton and sneutrino gauge eigenstates , @xmath115 to the mass eigenstates @xmath116 through the relations @xmath117 and @xmath118 . the mass eigenstates are defined in order of increasing mass . the neutralino gauge eigenstates , @xmath119 are related to the mass eigenstates @xmath120 by @xmath121 . similarly , the mixing of the negative charginos is @xmath122 finally , the neutral higgs doublet is given by @xmath123 where @xmath124 these coefficients are consistent with those of ref . @xcite , where a more complete discussion may be found . in the three - body decays @xmath125 , we label the initial and final state particle masses @xmath126 , @xmath127 , @xmath128 , and @xmath129 . we will maintain these indices throughout the appendix , with @xmath47 corresponding to the parent slepton , @xmath48 to the daughter slepton , @xmath49 to the daughter fermion with the same sign as the parent slepton , and @xmath50 to the other daughter fermion . the decay widths are the squared matrix elements integrated over 3-body phase space : @xmath130 where the dimensionless quantities @xmath131 have limits @xmath132\ , , \end{aligned}\ ] ] and we have defined @xmath133 throughout this work , we implicitly assume that the matrix element squared has been averaged and summed over initial and final state spins , respectively . the squared matrix element has the form @xmath134 where @xmath135 and @xmath136 are matrix elements corresponding to feynman diagrams with intermediate particle @xmath137 and @xmath138 . to evaluate the decay widths , we choose to integrate over @xmath139 analytically . the decay widths are then written in terms of dimensionless integrals over @xmath140 , which are performed numerically . in integrating over @xmath139 , it is convenient to note that all terms @xmath141 have a numerator that is either constant or linear in @xmath139 and a denominator that is proportional to the product of two propagators , each of which is either of the form @xmath142 or @xmath143 , where @xmath144 . we therefore define the following dimensionless functions of @xmath140 : @xmath145 where @xmath146 these functions will appear frequently in the decay widths to be discussed below . the first function @xmath147 contains an extra factor in the denominator to conform to the notation used in ref . the other functions are simply integrals over @xmath139 with all relevant combinations of propagators in the denominator and a numerator either constant or linear in @xmath139 . for the charge - preserving case , the matrix elements for all contributing modes are presented in eqs . ( [ eqn : preservingmestart])-([eqn : preservingmeend ] ) . the total decay width is @xmath148 \right ) \\ & = & \gamma_{\tilde{\chi}^0 \tilde{\chi}^0 } + \delta_{km } \left(\gamma_{zz } + \gamma_{hh } + 2 \gamma_{z \tilde{\chi}^0 } + 2 \gamma_{h \tilde{\chi}^0 } + 2 \gamma_{zh } \right ) \ , \end{aligned}\ ] ] where @xmath52 is the kronecker delta function . for @xmath55 the general case of independent @xmath49 and @xmath50 is taken ; for the remaining widths , @xmath149 is assumed to simplify the expressions . the neutralino width corresponds to the width given in ref . @xcite with the addition of flavor - violating vertices . the decay width is given by @xmath150 where @xmath151 labels the coefficients and integrals , and @xmath137 and @xmath138 label the neutralinos . the dimensionless integrals are defined by @xmath152 with coefficients @xmath153 \\ c_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(4 ) } & = & - 2 \text { re } \left [ \beta_{aik}^{1 * } \beta_{ajm}^1 \beta_{bik}^2 \beta_{bjm}^{1 * } + \beta_{aik}^{2 * } \beta_{ajm}^2 \beta_{bik}^1 \beta_{bjm}^{2 * } \right ] \\ c_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(5 ) } & = & - 4 \text { re } \left [ \beta_{aik}^{1 * } \beta_{ajm}^2 \beta_{bik}^2 \beta_{bjm}^{1 * } \right ] \\ c_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(6 ) } & = & - 4 \text { re } \left [ \beta_{aik}^{1 * } \beta_{ajm}^1 \beta_{bik}^2 \beta_{bjm}^{2 * } \right ] \ .\end{aligned}\ ] ] the @xmath10 boson contribution to the decay width is @xmath154 where the dimensionless integrals are @xmath155 i^{(1 ) } \left(x , r_z\right ) \\ i_{zz}^{(2 ) } & = & r_k^2 \int_{x_-}^{x_+ } dx \ , i^{(2 ) } \left(x , r_z\right ) + \left(x - 3 - r_j^2\right ) i^{(1)}\left(x , r_z\right ) \ , , \end{aligned}\ ] ] and the coefficients are @xmath156 the purely higgs - mediated contribution actually consists of several pieces with similar phase space structure . furthermore , it simplifies into the sum of contributions from the real and pseudoscalar higgs bosons , since the interference term between a real scalar and a pseudoscalar vanishes . the width is given by @xmath157 where the dimensionless integrals are @xmath158 where @xmath159 , and the coefficients are @xmath160 \\ c_{a^0 a^0 } & = & \left| \sigma_{ji}^{(4 ) } y_k^{(\ell ) } \sin \beta \right|^2 \\ c_{g^0 g^0 } & = & \left| \sigma_{ji}^{(5 ) } y_k^{(\ell ) } \cos \beta \right|^2 \\ c_{a^0 g^0 } & = & - \text{re}\left [ \sigma_{ji}^{(4 ) } \sigma_{ji}^{(5 ) * } y_k^{(\ell)2 } \sin \beta \cos \beta\right ] \ .\end{aligned}\ ] ] the width from the @xmath161 interference term is @xmath162 where the integrals are @xmath163\\ i_{z \tilde{\chi}_b^0}^{(2 ) } & = & r_k r_b \int_{x_-}^{x_+ } dx \ , \left [ g^{(2 ) } \left(x , r_z , r_b\right ) - \left(1 + x - r_j^2\right ) g^{(1)}\left(x , r_z , r_b\right ) \right ] \\ i_{z \tilde{\chi}_b^0}^{(3 ) } & = & r_k r_b \int_{x_-}^{x_+ } dx \ , \left [ g^{(2)}\left(x , r_z , r_b\right ) + \left(1 - x - r_j^2\right ) g^{(1 ) } \left(x , r_z , r_b\right ) \right ] \\ i_{z \tilde{\chi}_b^0}^{(4 ) } & = & r_k^2 \int_{x_-}^{x_+ } dx \ , \left [ \left(3 - 2x + r_j^2\right ) g^{(1 ) } \left(x , r_z , r_b\right ) \right ] \ , \end{aligned}\ ] ] and the coefficients are @xmath164 \\ c_{z \tilde{\chi}_b^0}^{(2 ) } & = & \text { re } \left[\left(2 \sin^2 \theta_w - 1\right ) \beta_{bik}^1 \beta_{bjm}^{2 * } + 2 \sin^2 \theta_w \beta_{bik}^2 \beta_{bjm}^{1*}\right ] \\ c_{z \tilde{\chi}_b^0}^{(3 ) } & = & \text { re } \left[2 \sin^2 \theta_w \beta_{bik}^1 \beta_{bjm}^{2 * } + \left(2 \sin^2 \theta_w - 1\right ) \beta_{bik}^2 \beta_{bjm}^{1*}\right ] \\ c_{z \tilde{\chi}_b^0}^{(4 ) } & = & \text { re } \left[2 \sin^2 \theta_w \beta_{bik}^1 \beta_{bjm}^{1 * } + \left(2 \sin^2 \theta_w - 1\right ) \beta_{bik}^2 \beta_{bjm}^{2*}\right ] \ .\end{aligned}\ ] ] the width from the @xmath165 interference is @xmath166 where the dimensionless integrals are @xmath167 \\ i_{h_a \tilde{\chi}_b^0}^{h(2 ) } & = & r_k \int_{x_-}^{x_+ } dx \ , \left(2 x - r^2 - 2 r_k^2\right ) g^{(1 ) } \left(x , r_{h_a},r_b\right ) \\ i_{h_a \tilde{\chi}_b^0}^{p(1 ) } & = & r_b \int_{x_-}^{x_+ } dx \ , \left [ g^{(2 ) } \left(x , r_{h_a},r_b\right ) + \left(x - 1 + r_j^2\right ) g^{(1 ) } \left(x , r_{h_a},r_b\right ) \right ] \\ i_{h_a \tilde{\chi}_b^0}^{p(2 ) } & = & r_k \int_{x_-}^{x_+ } dx \ , \left(1 - r_j^2\right ) g^{(1 ) } \left(x , r_{h_a},r_b\right)\ , , \end{aligned}\ ] ] and the coefficients are @xmath168 \\ c_{h^0 \tilde{\chi}_b^0}^{(2 ) } & = & \text { re}\left [ \sigma_{ji}^{(2 ) } y_k^{(\ell ) } \sin \theta_h \left(\beta_{bik}^1 \beta_{bjm}^{1 * } + \beta_{bik}^2 \beta_{bjm}^{2*}\right)\right ] \\ c_{h^0 \tilde{\chi}_b^0}^{(1 ) } & = & - \text{re}\left [ \sigma_{ji}^{(3 ) } y_k^{(\ell ) } \cos \theta_h \left(\beta_{bik}^1 \beta_{bjm}^{2 * } + \beta_{bik}^2 \beta_{bjm}^{1*}\right)\right ] \\ c_{h^0 \tilde{\chi}_b^0}^{(2 ) } & = & - \text{re}\left [ \sigma_{ji}^{(3 ) } y_k^{(\ell ) } \cos \theta_h \left(\beta_{bik}^1 \beta_{bjm}^{1 * } + \beta_{bik}^2 \beta_{bjm}^{2*}\right)\right ] \\ c_{a^0 \tilde{\chi}_b^0}^{(1 ) } & = & \text { re}\left [ \sigma_{ji}^{(4 ) } y_k^{(\ell ) } \sin \beta \left(\beta_{bik}^1 \beta_{bjm}^{2 * } - \beta_{bik}^2 \beta_{bjm}^{1*}\right)\right ] \\ c_{a^0 \tilde{\chi}_b^0}^{(2 ) } & = & \text { re}\left [ \sigma_{ji}^{(4 ) } y_k^{(\ell ) } \sin \beta \left(\beta_{bik}^1 \beta_{bjm}^{1 * } - \beta_{bik}^2 \beta_{bjm}^{2*}\right)\right ] \\ c_{g^0 \tilde{\chi}_b^0}^{(1 ) } & = & - \text{re}\left [ \sigma_{ji}^{(5 ) } y_k^{(\ell ) } \cos \beta \left(\beta_{bik}^1 \beta_{bjm}^{2 * } - \beta_{bik}^2 \beta_{bjm}^{1*}\right)\right ] \\ c_{g^0 \tilde{\chi}_b^0}^{(2 ) } & = & - \text{re}\left [ \sigma_{ji}^{(5 ) } y_k^{(\ell ) } \cos \beta \left(\beta_{bik}^1 \beta_{bjm}^{1 * } - \beta_{bik}^2 \beta_{bjm}^{2*}\right)\right]\ .\end{aligned}\ ] ] the @xmath169 interference decay width is @xmath170 where @xmath171 \\ h_a}^p & = & r_k \int_{x_-}^{x_+ } dx \ , \left(1 - r_j^2\right ) h^{(1 ) } \left(x , r_z , r_{h_a}\right ) \ , , \end{aligned}\ ] ] and @xmath172 \\ c_{z h^0 } & = & \text { re}\left[\sigma_{ji}^{(3 ) } y_k^{(\ell ) } \cos \theta_h \left(4 \sin^2 \theta_w - 1\right)\right ] \\ c_{z a^0 } & = & - \text{re}\left[\sigma_{ji}^{(4 ) } y_k^{(\ell ) } \sin \beta\right ] \\ c_{z g^0 } & = & \text { re}\left[\sigma_{ji}^{(5 ) } y_k^{(\ell ) } \cos \beta\right ] \ .\end{aligned}\ ] ] for the charge - flipping decay , the calculation is complicated by the fact that the same - sign daughter leptons create an interference term . it is convenient to break the matrix element into two parts , as shown in eqs . ( [ eqn : flippingmestart])-([eqn : flippingmeend ] ) . it is correspondingly convenient to separate the decay width into three terms @xmath173 where @xmath174 \ , \ ] ] and @xmath73 is 1 when the two outgoing leptons are of different generations and @xmath175 when they are of the same generation . from inspection of the matrix element , @xmath176 is identical to @xmath177 except for the coefficients in the @xmath178 vertex . then @xmath74 is identical to @xmath55 in the charge - preserving case with the substitutions @xmath179 and with the identical substitutions for the complex conjugates , where @xmath180 . from further inspection of the matrix elements , @xmath181 and @xmath182 differ only in the interchange of the two outgoing leptons . thus @xmath75 may be obtained from @xmath74 with the interchange @xmath183 . the @xmath78 width from the interference term is given by @xmath184 \nonumber \\ & = & \frac{m_i}{256 \pi^3 } \sum_{t=1}^8 \sum_{a , b=1}^4 \text { re } \left [ d_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(t ) } j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(t ) } \right ] \ , \end{aligned}\ ] ] where the integrals are @xmath185 f^{(1 ) } \left(x , r_a , r_b\right ) \right\ } \\ & & - \left(1 - x + r_k^2\right ) f^{(2 ) } \left(x , r_a , r_b\right ) \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(2 ) } & = & r_k r_m \int_{x_-}^{x_+ } dx \ , \left(1 + r_j^2 - r_k^2 - r_m^2 \right ) f^{(1 ) } \left(x , r_a , r_b\right ) \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(3 ) } & = & r_m r_b \int_{x_-}^{x_+ } dx \ , \left [ r^2 f^{(1 ) } \left(x , r_a , r_b\right ) - f^{(2 ) } \left(x , r_a , r_b\right ) \right ] \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(4 ) } & = & r_k r_b \int_{x_-}^{x_+ } dx \ , \left [ f^{(2 ) } \left(x , r_a , r_b\right ) - 2 r_m^2 f^{(1 ) } \left(x , r_a , r_b\right ) \right ] \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(5 ) } & = & r_m r_a \int_{x_-}^{x_+ } dx \ , \left(x - 2 r_k^2\right ) f^{(1 ) } \left(x , r_a , r_b\right ) \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(6 ) } & = & r_k r_a \int_{x_-}^{x_+ } dx \ , \left(r^2 - x\right ) f^{(1 ) } \left(x , r_a , r_b\right ) \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(7 ) } & = & r_a r_b \int_{x_-}^{x_+ } dx \ , \left [ \left(x - r^2\right ) f^{(1 ) } \left(x , r_a , r_b\right ) + f^{(2 ) } \left(x , r_a , r_b\right ) \right ] \\ j_{\tilde{\chi}_a^0 \tilde{\chi}_b^0}^{(8 ) } & = & 2 r_k r_m r_a r_b \int_{x_-}^{x_+ } dx \ , f^{(1 ) } \left(x , r_a , r_b\right ) \ , , \end{aligned}\ ] ] and the coefficients are @xmath186 the width is therefore @xmath190 \right ) \nonumber \\ & = & \gamma_{\tilde{\chi}^- \tilde{\chi}^- } + \delta_{km } \left ( \gamma_{zz } + 2 \gamma_{z \tilde{\chi}^- } \right ) \ , \end{aligned}\ ] ] where the partial widths are defined below . the chargino - mediated width is @xmath191 here the integral @xmath192 is the same as @xmath193 for the charge - preserving case , except the neutralino mass is replaced by the chargino mass and lepton masses @xmath128 and @xmath129 are set to zero . the @xmath10-mediated width is @xmath194 \nonumber \\ & = & \frac{m_i}{512 \pi^3 } \left| \frac{g \zeta_{ji}^{(2 ) } } { \cos \theta_w } \right|^2 i_{zz}^{(1 ) } \ .\end{aligned}\ ] ] here again the integral is the same as the @xmath195 integral for the charge - preserving case except the lepton masses @xmath128 and @xmath129 are set to zero . the @xmath10-chargino interference term is @xmath196 \nonumber \\ & & \int_{x_-}^{x_+ } dx \ , \left [ 2 \left(1 - x\right ) g^{(2 ) } \left(x , r_z , r_b\right ) + 2 \left(x - 1 + r_j^2\right ) g^{(1)}\left(x , r_z , r_b\right ) \right ] \nonumber \\ & = & - \frac{m_i}{512 \pi^3 } \text { re } \left [ \frac{g \zeta_{ji}^{(2 ) } } { \cos \theta_w } \gamma_{bik } \gamma_{bjm}^ * \right ] i_{z \tilde{\chi}_b^- } \ .\end{aligned}\ ] ] here the integral @xmath197 is the same as @xmath198 for the charge - preserving case , except the neutralino mass is replaced by the chargino mass and lepton masses @xmath128 and @xmath129 are set to zero . the decay modes with daughter quarks are much the same as those for daughter leptons except the neutralino intermediary contribution is removed and the different couplings are substituted . the matrix elements for the up - type quarks are the decay width to up - type quarks is @xmath202 \ , \end{aligned}\ ] ] with integrals from appendix [ app : dileptonp ] with quark masses substituted . the coefficients are @xmath203 \\ a_{a^0 a^0 } & = & \left| \sigma_{ji}^{(4 ) } y_k^{(u ) } \cos \beta \right|^2 \\ a_{g^0 g^0 } & = & \left| \sigma_{ji}^{(5 ) } y_k^{(u ) } \sin \beta \right|^2 \\ a_{a^0 g^0 } & = & \text { re } \left [ \sigma_{ji}^{(4)}\sigma_{ji}^{(5 ) * } y_k^{(u)2 } \sin \beta \cos \beta \right ] \\ a_{z h^0 } & = & - \text{re}\left [ \sigma_{ji}^{(2 ) } y_k^{(u ) } \cos \theta_h \left(\frac{8}{3 } \sin^2 \theta_w - 1\right)\right ] \\ a_{z h^0 } & = & - \text{re}\left [ \sigma_{ji}^{(3 ) } y_k^{(u ) } \sin \theta_h \left(\frac{8}{3 } \sin^2 \theta_w - 1\right)\right ] \\ a_{z a^0 } & = & - \text{re}\left [ \sigma_{ji}^{(4 ) } y_k^{(u ) } \cos \beta\right ] \\ a_{z g^0 } & = & - \text{re}\left [ \sigma_{ji}^{(5 ) } y_k^{(u ) } \sin \beta\right ] \ .\end{aligned}\ ] ] the decay width to down - type quarks is @xmath204 \ , \end{aligned}\ ] ] where again , the integrals are equivalent to those defined in appendix [ app : dileptonp ] with quark masses substituted . the coefficients are @xmath205 \\ b_{a^0 a^0 } & = & \left| \sigma_{ji}^{(4 ) } y_k^{(d ) } \sin \beta \right|^2 \\ b_{g^0 g^0 } & = & \left| \sigma_{ji}^{(5 ) } y_k^{(d ) } \cos \beta \right|^2 \\ b_{a^0 g^0 } & = & - \text{re } \left [ \sigma_{ji}^{(4)}\sigma_{ji}^{(5 ) * } y_k^{(d)2 } \sin \beta \cos \beta \right ] \\ b_{z h^0 } & = & - \text{re}\left [ \sigma_{ji}^{(2 ) } y_k^{(d ) } \sin \theta_h \left(\frac{4}{3 } \sin^2 \theta_w - 1\right)\right ] \\ b_{z h^0 } & = & \text { re}\left [ \sigma_{ji}^{(3 ) } y_k^{(d ) } \cos \theta_h \left(\frac{4}{3 } \sin^2 \theta_w - 1\right)\right ] \\ b_{z a^0 } & = & - \text{re}\left [ \sigma_{ji}^{(4 ) } y_k^{(d ) } \sin \beta\right ] \\ b_{z g^0 } & = & \text { re}\left [ \sigma_{ji}^{(5 ) } y_k^{(d ) } \cos \beta\right ] \ .\end{aligned}\ ] ] 99 c. d. froggatt and h. b. nielsen , nucl . b * 147 * , 277 ( 1979 ) . y. nir and n. seiberg , phys . b * 309 * , 337 ( 1993 ) [ arxiv : hep - ph/9304307 ] . y. grossman and y. nir , nucl . b * 448 * , 30 ( 1995 ) [ arxiv : hep - ph/9502418 ] . n. arkani - hamed , h. c. cheng , j. l. feng and l. j. hall , phys . lett . * 77 * , 1937 ( 1996 ) [ arxiv : hep - ph/9603431 ] . n. arkani - hamed , j. l. feng , l. j. hall and h. c. cheng , nucl . b * 505 * , 3 ( 1997 ) [ arxiv : hep - ph/9704205 ] . k. agashe and m. graesser , phys . d * 61 * , 075008 ( 2000 ) [ arxiv : hep - ph/9904422 ] . s. bar - shalom and a. rajaraman , phys . d * 77 * , 095011 ( 2008 ) [ arxiv:0711.3193 [ hep - ph ] ] . j. l. feng , c. g. lester , y. nir and y. shadmi , phys . d * 77 * , 076002 ( 2008 ) [ arxiv:0712.0674 [ hep - ph ] ] . g. d. kribs , e. poppitz and n. weiner , phys . d * 78 * , 055010 ( 2008 ) [ arxiv:0712.2039 [ hep - ph ] ] . y. nomura , m. papucci and d. stolarski , phys . d * 77 * , 075006 ( 2008 ) [ arxiv:0712.2074 [ hep - ph ] ] . y. nomura , m. papucci and d. stolarski , jhep * 0807 * , 055 ( 2008 ) [ arxiv:0802.2582 [ hep - ph ] ] . s. bar - shalom , a. rajaraman , d. whiteson and f. yu , phys . d * 78 * , 033003 ( 2008 ) [ arxiv:0803.3795 [ hep - ph ] ] . y. nomura and d. stolarski , phys . d * 78 * , 095011 ( 2008 ) [ arxiv:0808.1380 [ hep - ph ] ] . s. kaneko , j. sato , t. shimomura , o. vives and m. yamanaka , phys . d * 78 * , 116013 ( 2008 ) [ arxiv:0811.0703 [ hep - ph ] ] . g. hiller , y. hochberg and y. nir , jhep * 0903 * , 115 ( 2009 ) [ arxiv:0812.0511 [ hep - ph ] ] . j. hisano , m. m. nojiri and w. sreethawong , arxiv:0812.4496 [ hep - ph ] . s. dimopoulos , m. dine , s. raby and s. d. thomas , phys . rev . lett . * 76 * , 3494 ( 1996 ) [ arxiv : hep - ph/9601367 ] . j. l. feng and t. moroi , phys . d * 58 * , 035001 ( 1998 ) [ arxiv : hep - ph/9712499 ] . j. l. feng , a. rajaraman and f. takayama , phys . lett . * 91 * , 011302 ( 2003 ) [ arxiv : hep - ph/0302215 ] . j. l. feng , a. rajaraman and f. takayama , phys . d * 68 * , 063504 ( 2003 ) [ arxiv : hep - ph/0306024 ] . j. r. ellis , k. a. olive , y. santoso and v. c. spanos , phys . b * 588 * , 7 ( 2004 ) [ arxiv : hep - ph/0312262 ] . j. l. feng , s. f. su and f. takayama , phys . d * 70 * , 063514 ( 2004 ) [ arxiv : hep - ph/0404198 ] . j. l. feng , s. su and f. takayama , phys . d * 70 * , 075019 ( 2004 ) [ arxiv : hep - ph/0404231 ] . s. ambrosanio , g. d. kribs and s. p. martin , nucl . phys . b * 516 * , 55 ( 1998 ) [ arxiv : hep - ph/9710217 ] . a. djouadi and y. mambrini , phys . b * 493 * , 120 ( 2000 ) [ arxiv : hep - ph/0007174 ] . s. kraml and d. t. nhung , jhep * 0802 * , 061 ( 2008 ) [ arxiv:0712.1986 [ hep - ph ] ] . g. engelhard , j. l. feng , i. galon , d. sanford , and f. yu , [ arxiv:0904.1415 [ hep - ph ] ] .

consider the problem of estimating a multivariate normal mean with a known variance matrix , which is not necessarily proportional to the identity matrix . the coordinates are shrunk directly in proportion to their variances in efron and morris ( _ j . amer . statist . assoc . _ * 68 * ( 1973 ) 117130 ) empirical bayes approach , whereas inversely in proportion to their variances in berger s ( _ ann . statist . _ * 4 * ( 1976 ) 223226 ) minimax estimators . we propose a new minimax estimator , by approximately minimizing the bayes risk with a normal prior among a class of minimax estimators where the shrinkage direction is open to specification and the shrinkage magnitude is determined to achieve minimaxity . the proposed estimator has an interesting simple form such that one group of coordinates are shrunk in the direction of berger s estimator and the remaining coordinates are shrunk in the direction of the bayes rule . moreover , the proposed estimator is scale adaptive : it can achieve close to the minimum bayes risk simultaneously over a scale class of normal priors ( including the specified prior ) and achieve close to the minimax linear risk over a corresponding scale class of hyper - rectangles . for various scenarios in our numerical study , the proposed estimators with extreme priors yield more substantial risk reduction than existing minimax estimators .

1505.07607

a fundamental statistical problem is shrinkage estimation of a multivariate normal mean . see , for example , the february 2012 issue of _ statistical science _ for a broad range of theory , methods , and applications . let @xmath0 be multivariate normal with _ unknown _ mean vector @xmath1 and _ known _ variance matrix @xmath2 . consider the problem of estimating @xmath3 by an estimator @xmath4 under the loss @xmath5 , where @xmath6 is a _ known _ positive definite , symmetric matrix . the risk of @xmath7 is @xmath8 . the general problem can be transformed into a canonical form such that @xmath2 is diagonal and @xmath9 , the identity matrix ( e.g. , lehmann and casella @xcite , problem 5.5.11 ) . for simplicity , assume except in section [ sec3.2 ] that @xmath2 is @xmath10 and @xmath11 , where @xmath12 for a column vector @xmath13 . the letter @xmath14 is substituted for @xmath2 to emphasize that it is diagonal . for this problem , we aim to develop shrinkage estimators that are both minimax and capable of effective risk reduction over the usual estimator @xmath15 even in the heteroscedastic case ( i.e. , @xmath16 are not equal ) . an estimator of @xmath3 is minimax if and only if , _ regardless of _ @xmath17 , its risk is always no greater than @xmath18 , the risk of @xmath19 . for @xmath20 , minimax estimators different from and hence dominating @xmath21 are first discovered in the homoscedastic case where @xmath22 ( i.e. , @xmath23 ) . james and stein @xcite showed that @xmath24 is minimax provided @xmath25 . stein @xcite suggested the positive - part estimator @xmath26 , which dominates @xmath27 . throughout , @xmath28 . shrinkage estimation has since been developed into a general methodology with various approaches , including empirical bayes ( efron and morris @xcite ; morris @xcite ) and hierarchical bayes ( strawderman @xcite ; berger and robert @xcite ) . while these approaches are prescriptive for constructing shrinkage estimators , minimaxity is not automatically achieved but needs to be checked separately . for the heteroscedastic case , there remain challenging issues on how much observations with different variances should be shrunk relatively to each other ( e.g. , casella @xcite , morris @xcite ) . for the empirical bayes approach ( efron and morris @xcite ) , the coordinates of @xmath29 are shrunk directly in proportion to their variances . but the existing estimators are , in general , non - minimax ( i.e. , may have a greater risk than the usual estimator @xmath21 ) . on the other hand , berger @xcite proposed minimax estimators , including admissible minimax estimators , such that the coordinates of @xmath29 are shrunk inversely in proportion to their variances . but the risk reduction achieved over @xmath21 is insubstantial unless all the observations have similar variances . to address the foregoing issues , we develop novel minimax estimators for multivariate normal means under heteroscedasticity . there are two central ideas in our approach . the first is to develop a class of minimax estimators by generalizing a geometric argument essentially in stein @xcite ( see also brandwein and strawderman @xcite ) . for the homoscedastic case , the argument shows that @xmath27 can be derived as an approximation to the best linear estimator of the form @xmath30 , where @xmath31 is a scalar . in fact , the optimal choice of @xmath31 in minimizing the risk is @xmath32 . replacing @xmath33 by @xmath34 leads to @xmath35 with @xmath36 . this derivation is highly informative , even though it does not yield the optimal value @xmath37 . our class of minimax estimators are of the linear form @xmath38 , where @xmath39 is a nonnegative definite , diagonal matrix indicating the direction of shrinkage and @xmath31 is a scalar indicating the magnitude of shrinkage . the matrix @xmath39 is open to specification , depending on the variance matrix @xmath14 but _ not _ on the data @xmath29 . for a fixed @xmath39 , the scalar @xmath31 is determined to achieve minimaxity , depending on both @xmath14 and @xmath29 . @xcite minimax estimator corresponds to the special choice @xmath40 , thereby leading to the unusual pattern of shrinkage discussed above . the second idea of our approach is to choose @xmath39 by approximately minimizing the bayes risk with a normal prior in our class of minimax estimators . the bayes risk is used to measure average risk reduction for @xmath3 in an elliptical region as in berger @xcite . it turns out that the solution of @xmath39 obtained by our approximation strategy has an interesting simple form . in fact , the coordinates of @xmath29 are automatically segmented into two groups , based on their bayes `` importance '' ( berger @xcite ) , which is of the same order as the coordinate variances when the specified prior is homoscedastic . the coordinates of high bayes `` importance '' are shrunk inversely in proportion to their variances , whereas the remaining coordinates are shrunk in the direction of the bayes rule . this shrinkage pattern may appear paradoxical : it may be expected that the coordinates of high bayes `` importance '' are to be shrunk in the direction of the bayes rule . but that scheme is inherently aimed at reducing the bayes risk under the specified prior and , in general , fails to achieve minimaxity ( i.e. , it may lead to even a greater risk than the usual estimator @xmath21 ) . in addition to simplicity and minimaxity , we further show that the proposed estimator is scale adaptive in reducing the bayes risk : it achieves close to the minimum bayes risk , with the difference no greater than the sum of the 4 highest bayes `` importance '' of the coordinates of @xmath29 , simultaneously over a scale class of normal priors ( including the specified prior ) . to our knowledge , the proposed estimator seems to be the first one with such a property in the general heteroscedastic case . previously , in the homoscedastic case , @xmath41 is known to achieve the minimum bayes risk up to the sum of 2 ( equal - valued ) bayes `` importance '' of the coordinates over the scale class of homoscedastic normal priors ( efron and morris @xcite ) . the rest of this article is organized as follows . section [ sec2 ] gives a review of existing estimators . section [ sec3 ] develops the new approach and studies risk properties of the proposed estimator . section [ sec4 ] presents a simulation study . section [ sec5 ] provides concluding remarks . all proofs are collected in the . we describe a number of existing shrinkage estimators . see lehmann and casella @xcite for a textbook account and strawderman @xcite and morris and lysy @xcite for recent reviews . throughout , @xmath42 denotes the trace and @xmath43 denotes the largest eigenvalue . then @xmath44 and @xmath45 . for a bayes approach , assume the prior distribution : @xmath46 , where @xmath47 is the prior variance . the bayes rule is given componentwise by @xmath48 . then the greater @xmath49 is , the more @xmath50 is shrunk whether @xmath47 is fixed or estimated from the data . for the empirical bayes approach of efron and morris @xcite , @xmath47 is estimated by the maximum likelihood estimator @xmath51 such that @xmath52 morris @xcite suggested the modified estimator @xmath53 in our implementation , the right - hand side of ( [ eb - iter ] ) is computed to update @xmath51 from the initial guess , @xmath54 , for up to 100 iterations until the successive absolute difference in @xmath51 is @xmath55@xmath56 , or @xmath51 is set to @xmath57 so that @xmath58 otherwise . alternatively , xie _ et al . _ @xcite proposed empirical bayes - type estimators based on minimizing stein s @xcite unbiased risk estimate ( sure ) under heteroscedasticity . their basic estimator is defined componentwise by @xmath59 where @xmath60 is obtained by minimizing the sure of @xmath61 , that is , @xmath62 . in general , the two types of empirical bayes estimators , @xmath63 and @xmath64 , are non - minimax , as shown in section [ sec4 ] . for a direct extension of @xmath27 , consider the estimator @xmath65 and , more generally , @xmath66 , where @xmath67 is a scalar constant and @xmath68 a scalar function . see lehmann and casella @xcite , theorem 5.7 , although there are some typos . both @xmath69 and @xmath70 are spherically symmetric . the estimator @xmath69 is minimax provided @xmath71 and @xmath70 is minimax provided @xmath72 . no such @xmath73 exists unless @xmath74 , which restricts how much @xmath75 can differ from each other . for example , condition ( [ s - cond ] ) fails when @xmath76 and @xmath77 because @xmath78 and @xmath79 . berger @xcite proposed estimators of the form @xmath80 and @xmath81 , where @xmath67 is a scalar constant and @xmath68 a scalar function . then @xmath82 is minimax provided @xmath25 , and @xmath83 is minimax provided @xmath84 , regardless of differences between @xmath75 . however , a striking feature of @xmath82 and @xmath83 , compared with @xmath63 and @xmath64 , is that the smaller @xmath49 is , the more @xmath50 is shrunk . for example ( [ example ] ) , under @xmath82 , the coordinates @xmath85 are shrunk only slightly , whereas @xmath86 are shrunk as if they were shrunk as a 7-dimensional vector under @xmath27 . the associated risk reduction is insubstantial , because the risk of estimating @xmath87 is a small fraction of the overall risk of estimating @xmath3 . define the positive - part version of @xmath82 componentwise as @xmath88 the estimator @xmath89 dominates @xmath82 by baranchik @xcite , section 2.5 . berger @xcite , equation ( 5.32 ) , stated a different positive - part estimator , @xmath83 with @xmath90 , but the @xmath91th component may not be of the same sign as @xmath50 . given a prior @xmath92 , berger @xcite suggested an approximation of berger s @xcite robust generalized bayes estimator as @xmath93 x.\end{aligned}\ ] ] the estimator is expected to provide significant risk reduction over @xmath15 if the prior is correct and be robust to misspecification of the prior , but it is , in general , non - minimax . in the case of @xmath94 , @xmath95 becomes @xmath96 , in the form of spherically symmetric estimators @xmath97 , where @xmath68 is a scalar function ( bock @xcite , brown @xcite ) . the estimator @xmath98 is minimax provided @xmath99 and @xmath68 is nondecreasing . moreover , if @xmath100 , then @xmath98 is non - minimax unless @xmath101 . to overcome the non - minimaxity of @xmath95 , berger @xcite developed a minimax estimator @xmath102 by combining @xmath103 , @xmath95 , and a minimax estimator of bhattacharya @xcite . suppose that @xmath104 and the indices are sorted such that @xmath105 , where @xmath106 . define @xmath102 componentwise as @xmath107\frac{d_j}{d_j+\gamma_j } x_j,\end{aligned}\ ] ] where @xmath108 . in the case of @xmath94 , @xmath102 reduces to the original estimator of bhattacharya @xcite . the factor @xmath109 is replaced by @xmath110 in berger s @xcite original definition of @xmath102 , corresponding to replacing @xmath111 by @xmath112 in @xmath95 . in our simulations , the two versions of @xmath102 somehow yield rather different risk curves , and so do the corresponding versions of other estimators . but there has been limited theory supporting one version over the other . therefore , we focus on comparisons of only the corresponding versions of @xmath102 and other estimators . we develop a useful approach for shrinkage estimation under heteroscedasticity , by making explicit how different coordinates are shrunk differently . the approach not only sheds new light on existing results , but also lead to new minimax estimators . assume that @xmath113 ( diagonal ) and @xmath9 . consider estimators of the linear form @xmath114 where @xmath39 is a nonnegative definite , diagonal matrix indicating the _ direction _ of shrinkage and @xmath31 is a scalar indicating the _ magnitude _ of shrinkage . both @xmath39 and @xmath31 are to be determined . a sketch of our approach is as follows . a. for a fixed @xmath39 , the optimal choice of @xmath31 in minimizing the risk is @xmath115 b. for a fixed @xmath39 and a scalar constant @xmath116 , consider the estimator @xmath117 by theorem [ th1 ] , an upper bound on the risk function of @xmath118 is @xmath119,\end{aligned}\ ] ] where @xmath120 . requiring the second term to be no greater than 0 shows that if @xmath121 , then @xmath118 is minimax provided @xmath122 if @xmath121 , then the upper bound ( [ upper - bound ] ) has a minimum at @xmath123 . c. by taking @xmath123 in @xmath118 , consider the estimator @xmath124 subject to @xmath125 , so that @xmath126 is minimax by step ( ii ) . a positive - part estimator dominating @xmath126 is defined componentwise by @xmath127 where @xmath128 are the diagonal elements of @xmath39 . the upper bound ( [ upper - bound ] ) on the risk functions of @xmath129 and @xmath130 , subject to @xmath125 , gives @xmath131 we propose to choose @xmath39 based on some optimality criterion , such as minimizing the bayes risk with a normal prior centered at 0 ( berger @xcite ) . further discussions of steps ( i)(iii ) are provided in sections [ sec3.2][sec3.3 ] . we first develop steps ( i)(ii ) for the general problem where neither @xmath2 nor @xmath6 may be diagonal . the results can be as concisely stated as those just presented for the canonical problem where @xmath132 is diagonal and @xmath9 . such a unification adds to the attractiveness of the proposed approach . consider estimators of the form ( [ delta - form ] ) , where @xmath39 is not necessarily diagonal , but @xmath133 condition ( [ a - cond ] ) is invariant under a linear transformation . to see this , let @xmath134 be a nonsingular matrix and @xmath135 and @xmath136 . for the transformed problem of estimating @xmath137 based on @xmath138 with variance matrix @xmath139 , the transformed estimator from ( [ delta - form ] ) is @xmath140 . the application of condition ( [ a - cond ] ) to @xmath141 says that @xmath142 is nonnegative definite and therefore is equivalent to ( [ a - cond ] ) itself . for the canonical problem where @xmath113 ( diagonal ) , condition ( [ a - cond ] ) only requires that @xmath143 is nonnegative definite , allowing @xmath39 to be non - diagonal . on the other hand , it seems intuitively appropriate to restrict @xmath39 to be diagonal . then condition ( [ a - cond ] ) is equivalent to saying that @xmath39 is nonnegative definite ( and diagonal ) , which is the condition introduced on @xmath39 in the sketch in section [ sec3.1 ] . the risk of an estimator of the form ( [ delta - form ] ) is @xmath144 for a fixed @xmath39 , the optimal @xmath31 in minimizing the risk is @xmath145 replacing @xmath146 by @xmath147 and @xmath148 by a scalar constant @xmath149 leads to the estimator @xmath150 for a generalization , replacing @xmath67 by @xmath151 with a scalar function @xmath152 leads to the estimator @xmath153 we provide in theorem [ th1 ] an upper bound on the risk function of @xmath154 . [ th1 ] assume that @xmath68 almost differentiable ( stein @xcite ) . if ( [ a - cond ] ) holds and @xmath155 is nondecreasing , then for each @xmath3 , @xmath156,\end{aligned}\ ] ] where @xmath157 and @xmath158 . taking @xmath159 in ( [ r - upper - bound ] ) gives an upper bound on @xmath160 . requiring the second term in the risk upper bound ( [ r - upper - bound ] ) to be no greater than 0 leads to a sufficient condition for @xmath154 to be minimax . [ cor1 ] if ( [ a - cond ] ) holds and @xmath161 , then @xmath154 is minimax provided @xmath162 particularly , @xmath118 is minimax provided @xmath163 . for the canonical problem , inequality ( [ r - upper - bound ] ) and condition ( [ tan - cond2 ] ) for @xmath118 give respectively ( [ upper - bound ] ) and ( [ tan - cond ] ) . these results generalize the corresponding ones for @xmath69 and @xmath82 in section [ sec2 ] , by the specific choices @xmath164 or @xmath165 . the generalization also holds if @xmath67 is replaced by a scalar function @xmath166 . in fact , condition ( [ tan - cond2 ] ) reduces to baranchik s @xcite condition in the homoscedastic case . if @xmath167 , then the risk upper bound ( [ r - upper - bound ] ) has a minimum at @xmath168 . as a result , consider the estimator @xmath169 which is minimax provided @xmath167 . if @xmath170 ( berger @xcite ) , then @xmath171 and , by the proof of theorem [ th1 ] in the , the risk upper bound ( [ r - upper - bound ] ) becomes exact for @xmath118 . therefore , for @xmath170 , the estimator @xmath172 is uniformly best in the class @xmath118 , in agreement with the result that @xmath41 is uniformly best among @xmath27 in the homoscedastic case . the estimator @xmath126 has desirable properties of invariance . first , @xmath126 is easily shown to be invariant under a multiplicative transformation @xmath173 for a scalar @xmath174 . second , @xmath126 is invariant under a linear transformation of the inference problem . similarly as discussed below ( [ a - cond ] ) , let @xmath134 be a nonsingular matrix and @xmath175 , @xmath176 , and @xmath136 . for the transformed problem of estimating @xmath137 based on @xmath138 , the transformed estimator from @xmath126 is @xmath177 , whereas the application of @xmath126 is @xmath178 . the two estimators are identical because @xmath179 , @xmath180 , and hence @xmath181 . finally , we present a positive - part estimator dominating @xmath126 in the case where both @xmath182 and @xmath183 are symmetric , that is , @xmath184 similarly to ( [ a - cond ] ) , it is easy to see that this condition is invariant under a linear transformation . condition ( [ a - cond2 ] ) is trivially true if @xmath2 , @xmath6 , and @xmath39 are diagonal . in the , we show that ( [ a - cond2 ] ) holds if and only if there exists a nonsingular matrix @xmath134 such that @xmath185 , @xmath186 , and @xmath187 , where @xmath14 and @xmath188 are diagonal and the diagonal elements of @xmath14 or @xmath188 are , respectively , the eigenvalues of @xmath189 or @xmath39 . in the foregoing notation , @xmath190 and @xmath191 . for the problem of estimating @xmath192 based on @xmath138 , consider the estimator @xmath193 and the positive - part estimator @xmath194 with the @xmath91th component , @xmath195 where @xmath196 are the diagonal elements of @xmath188 . the estimator @xmath194 dominates @xmath197 by a simple extension of baranchik @xcite , section 2.5 . by a transformation back to the original problem , @xmath197 yields @xmath126 , whereas @xmath198 yields @xmath199 b x.\end{aligned}\ ] ] then @xmath200 dominates @xmath126 . therefore , ( [ r - upper - bound ] ) also gives an upper bound on the risk of @xmath201 , with @xmath202 , even though @xmath201 is not of the form @xmath154 . in practice , a matrix @xmath39 satisfying ( [ a - cond2 ] ) can be specified in two steps . first , find a nonsingular matrix @xmath134 such that @xmath185 and @xmath203 , where @xmath14 is diagonal . second , pick a diagonal matrix @xmath188 and define @xmath204 . the first step is always feasible by taking @xmath205 , where @xmath206 is a nonsingular matrix such that @xmath207 and @xmath208 is an orthogonal matrix @xmath208 such that @xmath209 is diagonal . given @xmath210 and @xmath14 , it can be shown that @xmath39 and @xmath200 depend on the choice of @xmath188 , but not on that of @xmath134 , provided that @xmath211 if @xmath212 for any @xmath213 . in the canonical case where @xmath113 and @xmath9 , this condition amounts to saying that any coordinates of @xmath29 with the same variances should be shrunk in the same way . different choices of @xmath39 lead to different estimators @xmath126 and @xmath200 . we study how to choose @xmath39 , depending on @xmath214 but _ not _ on @xmath29 , to approximately optimize risk reduction while preserving minimaxity for @xmath126 . the estimator @xmath201 provides even greater risk reduction than @xmath126 . we focus on the canonical problem where @xmath113 ( diagonal ) and @xmath9 . further , we restrict @xmath39 to be diagonal and nonnegative definite . as discussed in berger @xcite , any estimator can have significantly smaller risk than @xmath15 only for @xmath3 in a specific region . berger @xcite considered the situation where significant risk reduction is desired for an elliptical region @xmath215 with @xmath216 and @xmath217 the prior mean and prior variance matrix . see @xmath95 and @xmath102 reviewed in section [ sec2 ] . to measure average risk reduction for @xmath3 in region ( [ region ] ) , berger @xcite used the bayes risk with the normal prior @xmath218 . for simplicity , assume throughout that @xmath219 and @xmath220 is diagonal . we adopt berger s @xcite ideas of specifying an elliptical region and using the bayes risk to quantify average risk reduction in this region . we aim to find @xmath39 , subject to @xmath125 , minimizing the bayes risk of @xmath126 with the prior @xmath221 , @xmath222 , @xmath223 where @xmath224 denotes the expectation with respect to the prior @xmath221 . given @xmath39 , the risk @xmath225 can be numerically evaluated . a simple monte carlo method is to repeatedly draw @xmath92 and @xmath226 and then take the average of @xmath227 . but it seems difficult to literally implement the foregoing optimization . alternatively , we develop a simple method for choosing @xmath39 by two approximations . first , if @xmath125 , then taking the expectation of both sides of ( [ point - bound ] ) with respect to the prior @xmath221 gives an upper bound on the bayes risk of @xmath126 : @xmath228 where @xmath229 denotes the expectation with respect to the marginal distribution of @xmath29 in the bayes model , that is , @xmath230 . an approximation strategy for choosing @xmath39 is to minimize the upper bound ( [ bayes - bound ] ) on the bayes risk or to maximize the second term . the expectation @xmath231 can be evaluated as a 1-dimensional integral by results on inverse moments of quadratic forms in normal variables ( e.g. , jones @xcite ) . but the required optimization problem remains difficult . second , approximations can be made to the distribution of the quadratic form @xmath232 . suppose that @xmath232 is approximated with the same mean by @xmath233 , where @xmath234 is a chi - squared variable with @xmath235 degrees of freedom . then @xmath231 is approximated by @xmath236 . we show in the that this approximation gives a valid lower bound : @xmath237 a direct application of jensen s inequality shows that @xmath238 . but the lower bound ( [ bayes - bound2 ] ) is strictly tighter and becomes exact when @xmath239 . no simple bounds such as ( [ bayes - bound2 ] ) seem to hold if more complicated approximations ( e.g. , satterthwaite @xcite ) are used . combining ( [ bayes - bound ] ) and ( [ bayes - bound2 ] ) shows that if @xmath125 , then @xmath240 notice that @xmath126 is invariant under a multiplicative transformation @xmath241 for a scalar @xmath174 , and so is the upper bound ( [ bayes - bound3 ] ) . our strategy for choosing @xmath39 is to minimize the upper bound ( [ bayes - bound3 ] ) subject to @xmath125 or , equivalently , to solve the constrained optimization problem : @xmath242\\[-8pt ] & & \quad \mbox{subject to } \quad \sum_{j=1}^p ( d_j+ \gamma_j ) a_j^2 = \mbox{fixed}. \nonumber\end{aligned}\ ] ] the condition @xmath125 is dropped , because for @xmath20 , the achieved maximum is at least @xmath243 for some scalar @xmath244 . in spite of the approximations used in our approach , theorem [ th2 ] shows that not only the problem ( [ opt ] ) admits a non - iterative solution , but also the solution has a very interesting interpretation . for convenience , assume thereafter that the indices are sorted such that @xmath245 . [ th2 ] assume that @xmath20 , @xmath246 with @xmath247 and @xmath248 with @xmath249 ( @xmath250 ) . for problem ( [ opt ] ) , assume that @xmath251 with @xmath252 ( @xmath253 ) and @xmath254 , satisfied by @xmath255 . then the following results hold . a. there exists a _ unique _ solution , @xmath256 , to problem ( [ opt ] ) . b. let @xmath257 be the largest index such that @xmath258 . then @xmath259 , @xmath260 for @xmath261 , and @xmath262 where @xmath263 and @xmath264 the achieved maximum value , @xmath265 , is @xmath266 . c. the resulting estimator @xmath267 is minimax . we emphasize that , although @xmath39 can be considered a tuning parameter , the solution @xmath268 is _ data independent _ , so that @xmath269 is automatically minimax . if a data - dependent choice of @xmath39 were used , minimaxity would not necessarily hold . this result is achieved both because each estimator @xmath126 with @xmath125 is minimax and because a global criterion ( such as the bayes risk ) is used , instead of a pointwise criterion ( such as the frequentist risk at the unknown @xmath3 ) , to select @xmath39 . by these considerations , our approach differs from the usual exercise of selecting a tuning parameter in a data - dependent manner for a class of candidate estimators . there is a remarkable property of monotonicity for the sequence @xmath270 , which underlies the uniqueness of @xmath257 and @xmath268 . [ cor2 ] the sequence @xmath271 is nonincreasing : for @xmath272 , @xmath273 , where the equality holds if and only if @xmath274 the condition @xmath275 is equivalent to saying that the left side is greater than the right - hand side in the above expression for @xmath276 . therefore , @xmath257 is the smallest index @xmath272 with this property , and @xmath277 . the estimator @xmath267 is invariant under scale transformations of @xmath268 . therefore , the constant @xmath278 can be dropped from the expression of @xmath268 in theorem [ th1 ] . [ cor3 ] the solution @xmath279 can be rescaled such that @xmath280 then @xmath281 . moreover , it holds that @xmath282 the estimator @xmath267 can be expressed as @xmath283 the foregoing results lead to a simple algorithm for solving problem ( [ opt ] ) : a. sort the indices such that @xmath284 . b. take @xmath257 to be the smallest index @xmath285 ( corresponding to the largest @xmath286 ) such that @xmath272 and @xmath287 or take @xmath288 if there exists no such @xmath285 . c. compute @xmath289 by ( [ sol1])([sol2 ] ) . this algorithm is guaranteed to find the ( unique ) solution to problem ( [ opt ] ) by a fixed number of numerical operations . no iteration or convergence diagnosis is required . therefore , the algorithm is exact and non - iterative , in contrast with usual iterative algorithms for nonlinear , constrained optimization . the estimator @xmath267 has an interesting interpretation . by ( [ sol1])([sol2 ] ) , there is a dichotomous segmentation in the shrinkage direction of the coordinates of @xmath29 based on @xmath290 . this quantity @xmath291 is said to reflect the bayes `` importance '' of @xmath292 , that is , the amount of reduction in bayes risk obtainable in estimating @xmath292 in berger @xcite . the coordinates with high @xmath291 are shrunk inversely in proportion to their variances @xmath49 as in berger s @xcite estimator @xmath82 , whereas the coordinates with low @xmath291 are shrunk in the direction of the bayes rule . therefore , @xmath267 mimics the bayes rule to reduce the bayes risk , except that @xmath267 mimics @xmath82 for some coordinates of highest bayes `` importance '' in order to achieve minimaxity . in fact , by inequality ( [ sol - ineq ] ) , the relative shrinkage , @xmath293 , of each @xmath50 ( @xmath294 ) in @xmath267 versus the bayes rule is always no greater than that of @xmath295 ( @xmath296 ) . the expression ( [ delta - a ] ) suggests that there is a close relationship in beyond the shrinkage direction between @xmath297 and the bayes rule under the bayes model , @xmath298 . in this case , @xmath299 , and hence @xmath297 behaves similarly to @xmath300 . therefore , _ on average _ under the bayes model , the coordinates of @xmath29 are shrunk in @xmath269 the same as in the bayes rule , except that some coordinates of highest bayes `` importance '' are shrunk no greater than in the bayes rule . while this discussion seems heuristic , we provide in section [ sec3.4 ] a rigorous analysis of the bayes risk of @xmath267 , compared with that of the bayes rule . we now examine @xmath267 for two types of priors : @xmath301 and @xmath302 ( @xmath303 ) , referred to as the homoscedastic and heteroscedastic priors . for both types , @xmath304 are of the same order as the variances @xmath305 . recall that @xmath126 is invariant under a multiplicative transformation of @xmath39 . for both the homoscedastic prior with @xmath306 and the heteroscedastic prior _ regardless _ of @xmath307 , the solution @xmath308 can be rescaled such that @xmath309 denote by @xmath310 this rescaled matrix @xmath268 , corresponding to @xmath94 . then coordinates with high variances are shrunk inversely in proportion to their variances , whereas coordinates with low variances are shrunk symmetrically . for @xmath94 , the proposed method has a purely frequentist interpretation : it seeks to minimize the upper bound ( [ bayes - bound3 ] ) on the pointwise risk of @xmath126 at . for the homoscedastic prior with @xmath311 , the proposed method is then to minimize the upper bound ( [ bayes - bound3 ] ) on the bayes risk of @xmath126 with an extremely flat , homoscedastic prior . as @xmath311 , the solution @xmath268 can be rescaled such that @xmath312 denote by @xmath313 this rescaled matrix @xmath268 . then coordinates with low ( or high ) variances are shrunk directly ( or inversely ) in proportion to their variances . the direction @xmath313 can also be obtained by using a fixed prior in the form @xmath314 ( @xmath250 ) for arbitrary @xmath315 , where @xmath316 . finally , in the homoscedastic case ( @xmath23 ) , if the prior is also homoscedastic ( @xmath317 ) , then @xmath288 , @xmath318 , and @xmath269 reduces to the james stein estimator @xmath319 , _ regardless _ of @xmath320 and @xmath47 . the estimator @xmath267 is constructed by minimizing the upper bound ( [ bayes - bound3 ] ) on the bayes risk subject to minimaxity . in addition to simplicity , interpretability , and minimaxity demonstrated for @xmath267 , it remains important to further study risk properties of @xmath267 and show that @xmath269 can provide effective risk reduction over @xmath15 . write @xmath321 whenever needed to make explicit the dependency of @xmath268 on @xmath217 . first , we study how close the bayes risk of @xmath322 can be to that of the bayes rule , which is the smallest possible among _ all _ estimators including non - minimax ones , under the prior @xmath221 , @xmath323 . the bayes rule @xmath324 is given componentwise by @xmath325 , with the bayes risk @xmath326 where @xmath327 , indicating the bayes `` importance '' of @xmath292 ( berger @xcite ) . the upper bound ( [ bayes - bound3 ] ) on the bayes risk of @xmath328 gives @xmath329 because @xmath330 and hence @xmath331 by corollary [ cor3 ] . it appears that the difference between @xmath332 and @xmath333 tends to be large if @xmath257 is large . but @xmath334 can not differ too much from each other because by corollary [ cor1 ] , @xmath335 then the difference between @xmath336 and @xmath337 should be limited even if @xmath257 is large . a careful analysis using these ideas leads to the following result . [ th3 ] suppose that the prior is @xmath222 . if @xmath338 , then @xmath339 if @xmath340 , then @xmath341 throughout , an empty summation is 0 . there are interesting implications of theorem 3 . by ( [ loose - bound1 ] ) and ( [ loose - bound2 ] ) , @xmath342 then @xmath343 achieves almost the minimum bayes risk if @xmath344 . in terms of bayes risk reduction , the bound ( [ bayes - close ] ) shows that @xmath345 therefore , @xmath343 achieves bayes risk reduction within a negligible factor of that achieved by the bayes rule if @xmath346 . in the homoscedastic case where both @xmath22 and @xmath347 , @xmath267 reduces to @xmath319 , regardless of @xmath307 ( section [ sec3.3 ] ) . then the bounds ( [ tight - bound1 ] ) and ( [ tight - bound2 ] ) become exact and give efron and morris s @xcite result that @xmath348 or equivalently @xmath349 . it is interesting to compare the bayes risk bound of @xmath350 with that of the following simpler version of berger s @xcite estimator @xmath102 : @xmath351 by berger @xcite , @xmath352 is minimax and @xmath353 there seems to be no definite comparison between the bounds ( [ tight - bound1 ] ) and ( [ tight - bound2 ] ) on @xmath354 and the exact expression ( [ mb - bayes1 ] ) for @xmath355 , although the simple bounds ( [ loose - bound1 ] ) and ( [ loose - bound2 ] ) is slightly higher , by at most @xmath356 , than the bound ( [ mb - bayes2 ] ) . of course , each risk upper bound gives a conservative estimate of the actual performance , and comparison of two upper bounds should be interpreted with caution . in fact , the positive - part estimator @xmath357 yields lower risks than those of the non - simplified estimator @xmath102 in our simulation study ( section [ sec4 ] ) . the simplicity of @xmath267 and @xmath357 makes it easy to further study them in other ways than using the bayes ( or average ) risk . no similar result to the following theorem [ th4 ] has been established for @xmath102 or @xmath358 . corresponding to the prior @xmath359 , consider the worst - case ( or maximum ) risk @xmath360 over the hyper - rectangle @xmath361 ( e.g. , donoho _ et al . _ @xcite ) . applying jensen s inequality to ( [ point - bound ] ) shows that if @xmath362 , then @xmath363 which immediately leads to @xmath364 by the discussion after ( [ bayes - bound2 ] ) , a direct application of jensen s inequality to ( [ bayes - bound ] ) shows that the bayes risk @xmath225 is also no greater than the right - hand side of ( [ minimax - bound ] ) , whereas inequality ( [ bayes - bound2 ] ) leads to a strictly tighter bound ( [ bayes - bound3 ] ) . nevertheless , the upper bound ( [ minimax - bound ] ) on the worst - case risk of @xmath343 gives @xmath365 similarly as how ( [ bayes - bound3 ] ) leads to ( [ bayes - bound4 ] ) on the bayes risk of @xmath343 . therefore , the following result holds by the same proof of theorem 3 . [ th4 ] suppose that @xmath366 . if @xmath338 , then @xmath367 if @xmath340 , then @xmath368 there are similar implications of theorem [ th4 ] to those of theorem [ th3 ] . by donoho @xcite , the minimax linear risk over @xmath369 , @xmath370 , coincides with the minimum bayes risk @xmath337 , and is no greater than @xmath371 times the minimax risk over @xmath372 , @xmath373 . these results are originally obtained in the homoscedastic case ( @xmath374 ) , but they remain valid in the heteroscedastic case by the independence of the observations @xmath50 and the separate constraints on @xmath292 . therefore , a similar result to ( [ bayes - close ] ) holds : @xmath375 if @xmath344 , then @xmath269 achieves almost the minimax linear risk ( or the minimax risk up to a factor of @xmath371 ) over the hyper - rectangle @xmath369 , in addition to being globally minimax with @xmath3 unrestricted . the foregoing results might be considered non - adaptive in that @xmath328 is evaluated with respect to the prior @xmath376 or the parameter set @xmath372 with the same @xmath217 used to construct @xmath343 . but , by the invariance of @xmath126 under scale transformations of @xmath39 , @xmath328 is identical to the estimator , @xmath377 , that would be obtained if @xmath217 is replaced by @xmath378 for any scalar @xmath379 such that the diagonal matrix @xmath380 is nonnegative definite . by theorems 34 , this observation leads directly to the following adaptive result . in contrast , no adaptive result seems possible for @xmath102 . [ cor4 ] let @xmath378 and @xmath381 . then for each @xmath382 , @xmath383 & \le & r\bigl ( \delta^{\mathrm{bayes}}_{\gamma_\alpha } , \pi_{\gamma_\alpha}\bigr ) + \alpha^{-1 } \bigl(d_1^*+d_2^*+d_3^*+d_4^ * \bigr ) \\ & = & r^l(\mathcal h_{\gamma_\alpha})+ \alpha^{-1 } \bigl ( d_1^*+d_2^*+d_3^*+d_4^*\bigr),\end{aligned}\ ] ] where @xmath384 . for fixed @xmath217 , @xmath343 can achieve close to the minimum bayes risk or the minimax linear risk with respect to each prior in the class @xmath385 or each parameter set in the class @xmath386 under mild conditions . for illustration , consider the case of a heteroscedastic prior with @xmath387 . then @xmath388 can be reparameterized as @xmath389 . by corollary [ cor4 ] , for each @xmath307 , @xmath390 where @xmath391 and @xmath392 . therefore , if @xmath393 , then @xmath394 achieves the minimum bayes risk , within a negligible factor , under the prior @xmath395 for each @xmath396 . this can be seen as an extension of the result that in the homoscedastic case , @xmath41 asymptotically achieves the minimum bayes risk under the prior @xmath397 for each @xmath396 as @xmath398 . finally , we compare the estimator @xmath267 with a block shrinkage estimator , suggested by the differentiation in the shrinkage of low- and high - variance coordinates by @xmath267 . consider the estimator @xmath399 where @xmath400 is a cutoff index , and @xmath401 if @xmath402 is of dimension 1 or 2 . the index @xmath400 can be selected such that the coordinate variances are relatively homogeneous in each block . alternatively , a specific strategy for selecting @xmath400 is to minimize an upper bound on the bayes risk of @xmath403 , similarly as in the development of @xmath267 . applying ( [ bayes - bound3 ] ) with @xmath40 to @xmath404 in the two blocks shows that @xmath405 , where @xmath406 the first ( or second ) term in @xmath407 is set to 0 if @xmath408 ( or @xmath409 ) . then @xmath400 can be defined as the smallest index such that @xmath410 . but the upper bound ( [ bayes - bound4 ] ) on @xmath411 is likely to be smaller than the corresponding bound on @xmath412 , because @xmath413 for each @xmath414 by the cauchy schwarz inequality @xmath415 . therefore , @xmath267 tends to yield greater risk reduction than @xmath403 . this analysis also indicates that @xmath267 can be advantageous over @xmath403 extended to multiple blocks . the rationale of forming blocks in @xmath267 and @xmath416 differs from that in existing block shrinkage estimators ( e.g. , brown and zhao @xcite ) . as discussed in cai @xcite , block shrinkage has been developed mainly in the homoscedastic case as a technique for pooling information : the coordinate means are likely to be similar to each other within a block . nevertheless , it is possible to both deal with heterogeneity among coordinate variances and exploit homogeneity among coordinate means within individual blocks in our approach using a block - homoscedastic prior ( i.e. , the prior variances are equal within each block ) . this topic can be pursued in future work . we conduct a simulation study to compare the following 8 estimators , a. non - minimax estimators : @xmath63 by ( [ eb ] ) , @xmath64 by ( [ xkb ] ) , @xmath95 by ( [ rb ] ) with @xmath94 ; b. minimax estimators : @xmath417 by ( [ b+ ] ) , @xmath102 by ( [ mb ] ) with @xmath94 or @xmath418 for some large @xmath47 , @xmath419 by ( [ a+ ] ) with @xmath420 and @xmath313 . recall that @xmath421 corresponds to @xmath94 or @xmath422 and @xmath423 corresponds to @xmath424 with @xmath311 . in contrast , letting the diagonal elements of @xmath217 tend to @xmath57 in any direction in @xmath95 and @xmath102 leads to @xmath15 . setting @xmath217 to 0 or @xmath57 is used here to specify the relevant estimators , rather than to restrict the prior on @xmath3 . for completeness , we also study the following estimators : @xmath425 by ( [ b+ ] ) , @xmath95 with @xmath111 replaced by @xmath112 in ( [ rb ] ) , @xmath102 with @xmath109 replaced by @xmath110 in ( [ mb ] ) , and @xmath419 with @xmath426 replaced by @xmath427 in ( [ a+ ] ) , referred to as the alternative versions of @xmath428 , @xmath95 , @xmath102 , and @xmath429 respectively . the usual choices of the factors , @xmath111 , @xmath109 , and @xmath426 , are motivated to minimize the risks of the non - positive - part estimators , but may not be the most desirable for the positive - part estimators . as seen below , the alternative choices @xmath112 , @xmath110 , and @xmath430 can lead to risk curves for the positive - part estimators rather different from those based on the usual choices @xmath431 , @xmath109 , and @xmath426 . therefore , we compare the estimators @xmath432 , @xmath95 , @xmath102 , and @xmath419 and , separately , their alternative versions . each estimator @xmath7 is evaluated by the pointwise risk function @xmath433 as @xmath3 moves in a certain direction or the bayes risk function @xmath434 as @xmath435 varies in a set of priors on @xmath3 . consider the homoscedastic prior @xmath436 or the heteroscedastic prior @xmath437 for @xmath438 . as discussed in section [ sec3.3 ] , the bayes risk with the first or second prior is meant to measure average risk reduction over the region @xmath439 or @xmath440 . corresponding to the two priors , consider the direction along @xmath441 or @xmath442 , where @xmath197 gives the euclidean distance from 0 to the point indexed by @xmath197 . the two directions are referred to as the homoscedastic and heteroscedastic directions . we investigate several configurations for @xmath14 , including ( [ example ] ) and @xmath443 where @xmath444 is a chi - squared variable with @xmath285 degrees of freedom . in the last case , @xmath445 can be considered a typical sample from a scaled inverse chi - squared distribution , which is the conjugate distribution for normal variances . in the case ( [ example - group3 ] ) , the coordinates may be segmented intuitively into three groups with relatively homogeneous variances . in the case ( [ example - group22 ] ) , there is no clear intuition about how the coordinates should be segmented into groups . for fixed @xmath14 , the pointwise risk @xmath433 is computed by repeatedly drawing @xmath446 and then taking the average of @xmath447 . the bayes risk is computed by repeatedly drawing @xmath92 and @xmath448 and then taking the average of @xmath447 . each monte carlo sample size is set to @xmath449 . axis ( third row ) in the case ( [ example - group3 ] ) . left : non - minimax estimators @xmath63 ( @xmath450 ) , @xmath95 ( @xmath451 ) , @xmath64 ( @xmath452 ) . right : minimax estimators @xmath417 ( @xmath453 ) , @xmath102 with @xmath94 ( @xmath454 ) and @xmath455 ( @xmath456 ) , @xmath419 with @xmath420 ( @xmath457 ) and @xmath458 ( @xmath459 ) . ] axis ( third row ) in the case ( [ example - group3 ] ) , with the same legend as in figure [ fig1 ] . the alternative versions of @xmath428 , @xmath460 , @xmath102 , and @xmath419 are used . ] the relative performances of the estimators are found to be consistent across different configurations of @xmath14 studied . moreover , the bayes risk curves under the homoscedastic prior are similar to the pointwise risk curves along the homoscedastic direction . the bayes risk curves under the heteroscedastic prior are similar to the pointwise risk curves along the heteroscedastic direction . figure [ fig1 ] shows the pointwise risks of the estimators with the usual versions of @xmath428 , @xmath95 , @xmath102 , and @xmath419 and figure [ fig2 ] shows those of the estimators with the alternative versions of @xmath428 , @xmath95 , @xmath102 , and @xmath419 for the case ( [ example - group3 ] ) , with roughly three groups of coordinate variances , which might be considered unfavorable to our approach . for both @xmath310 and @xmath313 , the cutoff index @xmath257 is found to be 3 . see the supplementary material ( tan @xcite ) for the bayes risk curves of all these estimators for the case ( [ example - group3 ] ) and the results for other configurations of @xmath14 . a number of observations can be drawn from figures [ fig1][fig2 ] . first , @xmath461 , @xmath64 , and @xmath95 have among the lowest risk curves along the homoscedastic direction . but along the heteroscedastic direction , the risk curves of @xmath63 and @xmath64 rise quickly above the constant risk of @xmath29 as @xmath197 increases . moreover , all the risk curves of @xmath63 , @xmath64 , and @xmath95 along the @xmath462 axis exceed the constant risk of @xmath29 as @xmath463 increases . therefore , @xmath63 , @xmath64 , and @xmath460 fail to be minimax , as mentioned in section [ sec2 ] . second , @xmath417 or @xmath464 has among the highest risk curve , except where the risk curves of @xmath63 and @xmath64 exceed the constant risk of @xmath29 along the heteroscedastic direction . the poor performance is expected for @xmath465 or @xmath464 , because there are considerable differences between the coordinate variances in ( [ example - group3 ] ) . third , among the minimax estimators , @xmath419 with @xmath420 or @xmath313 has the lowest risk curve along various directions , whether the usual versions of @xmath428 , @xmath466 , and @xmath419 are compared ( figure [ fig1 ] ) or the alternative versions are compared ( figure [ fig2 ] ) . fourth , the risk curve of @xmath419 with @xmath420 is similar to that of @xmath419 with @xmath467 along the heteroscedastic direction . but the former is noticeably higher than the latter along the homoscedastic direction as @xmath197 increases , whereas is noticeably lower than the latter along the @xmath462 axis as @xmath463 increases . these results agree with the construction of @xmath310 using a heteroscedastic prior and @xmath313 using a flat , homoscedastic prior . their relative performances depend on the direction in which the risks are evaluated . fifth , @xmath102 with @xmath94 has risk curves below that of @xmath417 or @xmath464 , but either above or crossing those of @xmath419 with @xmath420 and @xmath458 . moreover , @xmath102 with @xmath455 has elevated , almost flat risk curves for @xmath197 from 0 to 16 . this seems to indicate an undesirable consequence of using a non - degenerate prior for @xmath102 in that the risk tends to increase for @xmath3 near 0 , and remains high for @xmath3 far away from 0 . the foregoing discussion involves the comparison of the risk curves as @xmath3 moves away from 0 between @xmath102 and @xmath468 specified with fixed priors . alternatively , we compare the pointwise risks at @xmath469 or @xmath442 and the bayes risks under the prior @xmath470 or @xmath471 between @xmath102 and @xmath357 specified with the prior @xmath470 for a range of @xmath197 . the homoscedastic prior used in the specification of @xmath102 and @xmath357 can be considered correctly specified or misspecified , when the bayes risks are evaluated under , respectively , the homoscedastic or heteroscedastic prior or when the pointwise risks are evaluated along the homoscedastic or heteroscedastic direction . for each situation , @xmath357 has lower pointwise or bayes risks than @xmath102 . see figure a2 in the supplementary material ( tan @xcite ) . the estimator @xmath267 and its positive - part version @xmath472 are not only minimax and but also have desirable properties including simplicity , interpretability , and effectiveness in risk reduction . in fact , @xmath267 is defined by taking @xmath473 in a class of minimax estimators @xmath126 . the simplicity of @xmath267 holds because @xmath126 is of the linear form @xmath474 , with @xmath39 and @xmath31 indicating the direction and magnitude of shrinkage . the interpretability of @xmath269 holds because the form of @xmath268 indicates that one group of coordinates are shrunk in the direction of berger s @xcite minimax estimator whereas the remaining coordinates are shrunk in the direction of the bayes rule . the effectiveness of @xmath267 in risk reduction is supported , in theory , by showing that @xmath269 can achieve close to the minimum bayes risk simultaneously over a scale class of normal priors ( corollary [ cor4 ] ) . for various scenarios in our numerical study , the estimators @xmath475 with extreme priors yield more substantial risk reduction than existing minimax estimators . it is interesting to discuss a special feature of @xmath154 and hence of @xmath118 and @xmath126 among linear , shrinkage estimators of the form @xmath476 where @xmath39 and @xmath134 are nonnegative definite matrices and @xmath477 is a scalar function . the estimator @xmath154 corresponds to the choice @xmath478 , which is motivated by the form of the optimal @xmath31 in minimizing the risk of @xmath479 for fixed @xmath39 . on the other hand , berger and srinivasan @xcite showed that under certain regularity conditions on @xmath477 , an estimator ( [ general ] ) can be generalized bayes or admissible only if @xmath480 . this condition is incompatible with @xmath481 , unless @xmath482 as in berger s @xcite estimator . therefore , @xmath126 including @xmath267 is , in general , not generalized bayes or admissible . this conclusion , however , does not apply directly to the positive - part estimator @xmath200 , which is no longer of the linear form @xmath474 . there are various topics that can be further studied . first , the prior on @xmath3 is fixed , independently of data in the current paper . a useful extension is to allow the prior to be estimated within a certain class , for example , homoscedastic priors @xmath483 , from the data , in the spirit of empirical bayes estimation ( e.g. , efron and morris @xcite ) . second , the bayes risk with a normal prior is used to measure average risk reduction in an elliptical region ( section [ sec3.3 ] ) . it is interesting to study how our approach can be extended when using a non - normal prior on @xmath3 , corresponding to a non - elliptical region in which risk reduction is desired . the following extends stein s @xcite lemma for computing the expectation of the inner product of @xmath484 and a vector of functions of @xmath29 . [ lem1 ] let @xmath485 be multivariate normal with mean @xmath3 and variance matrix @xmath2 . assume that @xmath486 is almost differentiable stein @xcite with @xmath487 for @xmath488 , where @xmath489 . then @xmath490,\ ] ] where @xmath491 is the matrix with @xmath492th element @xmath493 . a direct generalization of lemma 2 in stein @xcite to a normal random vector with non - identity variance matrix gives @xmath494 where @xmath495 is the row vector with @xmath91th element @xmath496 . taking the @xmath497th element of both sides of the equation gives @xmath498 where @xmath499 is the @xmath492th element of @xmath2 . summing both sides of the preceding equation over @xmath497 gives the desired result . proof of theorem [ th1 ] by direct calculation , the risk of @xmath154 is @xmath500 by lemma [ lem1 ] and the fact that @xmath501 , the third term after the minus sign in @xmath502 is @xmath503 by condition ( [ a - cond ] ) , @xmath504 is nonnegative definite . by section 21.14 and exercise 21.32 in harville @xcite , @xmath505 for @xmath506 . then the preceding expression is bounded from below by @xmath507 which leads immediately to the upper bound on @xmath508 . proof for condition ( [ a - cond2 ] ) we show that if condition ( [ a - cond2 ] ) holds , then there exists a nonsingular matrix @xmath134 with the claimed properties . the converse is trivially true . let @xmath509 be the unique symmetric , positive definite matrix such that @xmath510 . then @xmath511 is symmetric , that is , @xmath512 , because @xmath513 . moreover , @xmath514 and @xmath511 commute , that is , @xmath515 , because @xmath516 and @xmath511 is symmetric . therefore , @xmath517 and @xmath511 are simultaneously diagonalizable ( harville @xcite , section 21.13 ) . there exists an orthogonal matrix @xmath208 such that @xmath518 and @xmath519 for some diagonal matrices @xmath14 and @xmath188 . then @xmath520 satisfies the claimed properties . proof of inequality ( [ bayes - bound2 ] ) we show that if @xmath521 are independent standard normal variables , then @xmath522 . . then @xmath524 and @xmath525 are independent , @xmath526 , and @xmath527 . the claimed inequality follows because @xmath528 , @xmath529 , and @xmath530 @xmath531 by jensen s inequality . proofs of theorem [ th2 ] and corollary [ cor2 ] consider the transformation @xmath532 and @xmath533 , so that @xmath534 and @xmath535 . problem ( [ opt ] ) is then transformed to @xmath536 , subject to @xmath537 ( @xmath538 ) and @xmath539 , which is of the form of the special case of ( [ opt ] ) with @xmath540 ( @xmath250 ) . but it is easy to verify that if the claimed results hold for the transformed problem , then the results hold for original problem ( [ opt ] ) . therefore , assume in the rest of proof that @xmath540 ( @xmath250 ) . there exists at least a solution , @xmath268 , to problem ( [ opt ] ) by boundedness of the constraint set . let @xmath541 and @xmath542 . a key of the proof is to exploit the fact that , by the setup of problem ( [ opt ] ) , @xmath289 is automatically a solution to the problem @xmath543 the karush tucker condition for this problem gives @xmath544 where @xmath31 , @xmath545 ( @xmath546 ) , and @xmath547 satisfying @xmath548 ( @xmath250 ) are lagrange multipliers . first , we show that @xmath549 and hence @xmath550 for @xmath253 . if @xmath551 , then either @xmath549 for @xmath250 , or @xmath552 . the latter case is infeasible by the constraint @xmath553 . suppose @xmath554 . by ( [ a2 ] ) , @xmath549 for each @xmath555 . then @xmath556 for each @xmath557 because @xmath558 . second , we show that @xmath259 . if @xmath551 , then @xmath559 . . then @xmath560 by ( [ a2 ] ) . summing ( [ a3 ] ) over @xmath561 and ( [ a4 ] ) shows that @xmath562 . therefore , @xmath563 or equivalently @xmath259 . third , we show that @xmath564 and @xmath565 . for each @xmath566 and @xmath567 , @xmath568 by ( [ a2])([a3 ] ) and then @xmath569 because @xmath558 . the inequalities also hold for @xmath276 , by application of the argument to problem ( [ a1 ] ) with @xmath257 replaced by some @xmath566 . then @xmath570 because @xmath571 for each @xmath555 , @xmath572 , and @xmath257 is the largest element in @xmath573 . fourth , we show the expressions for @xmath574 and the achieved maximum value . by the definition of @xmath573 , @xmath575 for @xmath576 . by ( [ a2 ] ) , @xmath577 for @xmath578 . let @xmath579 and @xmath580 . then @xmath581 is a solution to the problem @xmath582 by the definition of @xmath573 , @xmath583 and hence @xmath584 lies off the boundary in the constraint set . then @xmath581 is a solution to the foregoing problem with the constraint @xmath585 removed . the problem is of the form of maximizing a linear function of @xmath586 subject to an elliptical constraint . straightforward calculation shows that @xmath587 and the achieved maximum value is @xmath588 , where @xmath589 @xmath590 . finally , we show that the sequence @xmath591 is nonincreasing : @xmath273 , where the equality holds if and only if @xmath592 . because @xmath593 or @xmath594 , this result implies that @xmath595 and hence @xmath268 is a unique solution to ( [ opt ] ) . let @xmath596 so that @xmath597 . by the identity @xmath598 and simple calculation , @xmath599 \frac { d_{k+1}^{-1}}{\sum_{j=1}^{k+1 } d_j^{-1 } } \nonumber \\[-6pt ] \\[-10pt ] & = & d_{k+1 } \frac { \ { r_k - ( k-2 ) \}^2}{r_k ( r_k+1)},\nonumber\end{aligned}\ ] ] where @xmath600 . therefore , @xmath601 . moreover , @xmath602 if and only if @xmath603 , that is , @xmath604 . proof of corollary [ cor3 ] it suffices to show ( [ sol - ineq ] ) . by corollary [ cor2 ] , @xmath605 and hence @xmath606 . then for @xmath294 . @xmath607 because @xmath608 for @xmath609 . proof of theorem [ th3 ] let @xmath610 so that @xmath611 , similarly as in the proof of theorem [ th2 ] . by equation ( [ a5 ] ) with @xmath612 and @xmath613 replaced by @xmath614 , @xmath615 by the relationship @xmath616 and simple calculation , @xmath617 if @xmath340 , combining the two preceding equation gives @xmath618 the first inequality follows because @xmath619 for @xmath620 and @xmath621 is increasing for @xmath622 with a maximum at @xmath623 . the second inequality follows because @xmath624 . therefore , if @xmath340 then @xmath625 if @xmath338 , then @xmath626 and hence @xmath627 this completes the proof . the author thanks bill strawderman and cunhui zhang for helpful discussions .

mean - field systems have been recently derived that adequately predict the behaviors of large networks of coupled integrate - and - fire neurons . the mean - field system for a network of neurons with spike frequency adaptation is typically a pair of differential equations for the mean adaptation and mean synaptic gating variable of the network . these differential equations are non - smooth , and in particular are piecewise smooth continuous ( pwsc ) . here , we analyze the smooth and non - smooth bifurcation structure of these equations and show that the system is organized around a pair of co - dimension two bifurcations that involve , respectively , the collision between a hopf equilibrium point and a switching manifold , and a saddle - node equilibrium point and a switching manifold . these two co - dimension 2 bifurcations can coalesce into a co - dimension 3 non - smooth bifurcation . as the mean - field system we study is a non - generic piecewise smooth continuous system , we discuss possible regularizations of this system and how the bifurcations which occur are related to non - smooth bifurcations displayed by generic pwsc systems .

1408.4767

recently , a class of two - dimensional integrate and fire type models have been developed which can be fit to properties of real neurons . this class of models includes the adapting , leaky integrate and fire neuron ( lif ) @xcite , the izhikevich model @xcite , the quartic integrate and fire model @xcite , and the adaptive exponential integrate and fire model ( adex ) @xcite . the models in this class are far simpler to fit and simulate than traditional conductance based models . nevertheless , these models still replicate the more complex behaviors of real neurons @xcite . these models have been fit to several different neuron types so that the behavior of large networks of these neuron models may be studied through numerical exploration of the parameter space . for example , this approach has been used to determine the role of various parameters in the generation of adaptation induced bursting in networks of ca3 pyramidal neurons @xcite while the numerical simulation of integrate and fire networks is far simpler and faster than that of conductance based models , numerical exploration of the parameter space is still a time - consuming process . furthermore , one can not easily perform direct bifurcation analysis on large networks . fortunately , a system of mean field equations has been derived for these large networks of two - dimensional integrate and fire neurons @xcite . this derivation assumes that the networks are all - to - all coupled and the neuronal parameters are homogeneous within the network . the resulting mean field system is a set of non - smooth differential equations governing the first moments of the adaptation variable and the synaptic coupling variable . the mean field system of equations is analytically derived from the original network , without any further fitting . thus , one can conduct bifurcation analysis ( either analytically or numerically ) on the mean field system with confidence that the results are representative of the behavior of the original network of model neurons , and possibly the original network of actual neurons . such a level of correspondence between the parameters for the individual neurons , and the resulting behavior of the full network is currently not possible with more sophisticated types of neuron models . however , analysis of the derived mean field system has an added level of difficulty as the system of equations is non - smooth . both classical bifurcation theory and the newer field of non - smooth bifurcation theory must be used to adequately understand the behavior of the mean field system , and thus the full network . here we explore , both analytically and numerically , many of the non - smooth bifurcations and phenomena that occur in the mean field system of equations from @xcite . the primary mean field system we use is that of the izhikevich model , with the neuronal models fit to hippocampal area ca3 pyramidal neuron data @xcite . we modify the parameters slightly as the neuronal model used in @xcite was an alteration of the default izhikevich model to better fit the action potential half - width observed in the data . we use this model primarily for two reasons : it is the most analytically tractable and the parameters have been fit to neuronal data . however , as we will see , many of the non - smooth bifurcations are present in the other models in the general class of two - dimensional adapting integrate and fire neurons . whenever possible we present our results in terms of this general class . we consider two - dimensional integrate and fire models of the form @xmath0 where @xmath1 represents the nondimensionalized membrane potential , and @xmath2 serves as an adaptation variable . time has also been non - dimensionalized . the dynamical equations ( [ de1])-([de2 ] ) are supplemented by the following discontinuities @xmath3 this particular notation was formally introduced in @xcite , along with a full bifurcation analysis of this general family of adapting integrate and fire neurons . members of this family include the izhikevich model , the adaptive exponential ( adex ) integrate and fire model , and the quartic integrate and fire model @xcite . these neurons can be coupled together via a synaptic gating variable , @xmath4 . the gating variable typically takes the form @xmath5 where @xmath6 is the time that the @xmath7th neuron fires its @xmath8th spike , @xmath7 is the index of the presynaptic neuron , and @xmath9 is the index of the postsynaptic neuron . the function @xmath10 varies depending on which synaptic pulse function is used . examples include the exponential synapse , the double exponential synapse , and the alpha synapse @xcite . for simplicity we restrict our attention to the exponential synapse , however the analysis can be extended to the other synaptic types without much difficulty . for the exponential synapse , @xmath10 is given by : @xmath11 given the form for @xmath10 , one can derive a differential equation for @xmath12 , the total synaptic input to the @xmath13 neuron @xcite . for example , for the exponential synapse the differential equation for @xmath14 is @xmath15 for all - to - all coupling , the function @xmath14 becomes identical for all the neurons , and can be replaced by a single variable @xmath4 , the global synaptic coupling function . in this case , the equations for the entire network are : @xmath16 the specific forms of @xmath17 we consider are : @xmath18 these forms can be arrived at through a suitable non - dimensionalization of the original equations for these models @xcite . note that the non - dimensionalization for the izhikevich model differs from the one used by @xcite and is from @xcite . these networks often display bursting , a oscillatory behaviour where the individual neurons alternate between firing and quiescence @xcite . the other common behaviour is tonic firing , where the neurons all fire at a constant rate . the transition between these two behaviours is a bifurcation of the full network . an example of this transition for a network of 1000 izhikevich neurons with all - to - all coupling is shown in figure [ fig1 ] . the parameters for this network can be found in table 1 . in figure [ fig1a ] , the neurons in the network fire spikes , and the mean - adaptation variable , @xmath19 , and the synaptic coupling variable , @xmath20 both converge to a stable steady state . in figure [ fig1b ] , the neurons fire synchronized bursts , and the pair of variables @xmath21 converge to a steady state limit cycle , representing the oscillation between firing and quiescence that the individual neurons undergo . this occurs as the current @xmath22 is decreased from @xmath23 in figure [ fig1a ] to @xmath24 in figure [ fig1b ] . as seen in figure [ fig1 ] , the network variables @xmath20 and @xmath19 can capture a great deal of information about the behavior of the entire network . further , their steady state behaviour undergoes a qualitative change as the parameter @xmath22 is decreased . thus , it would be advantageous to have a closed set of differential equations for these variables , as any qualitative change in the behavior of the full network should manifest itself as a bifurcation of the dynamical system for @xmath25 . in the following subsection , we derive such a system for these network variables . to derive the mean field system one begins by defining the population density function , @xmath26 , which is a probability density function for the location of of the variables @xmath27 in the phase space . that is , the probability of finding a neuron in the region @xmath28 of phase space is given by integrating @xmath29 over @xmath28 . starting from the full network model , one can derive ( in the limit that @xmath30 ) a partial differential equation that governs the evolution of the probability density function @xmath26 and predicts the large network dynamics of the original model @xcite . this partial differential equation , called the population density equation , takes the form @xmath31 where the term @xmath32 is referred to as the flux . the discontinuities and discrete jumps in the model neurons impose a boundary condition on the flux : @xmath33 this pde is coupled to an ode for @xmath20 , given by @xmath34 in order to reduce this system to a small , closed set of ordinary differential equations , one has to apply a series of approximations . the derivation is somewhat lengthy thus we refer the reader to @xcite for the exact details . after the approximations are made , the resulting mean field system is given by : @xmath35^{-1 } & \quad \text{if}\quad h(w , s)>0\\ 0 & \quad \text{if } \quad h(w , s)\leq 0 \end{cases } \label{mfc}\end{aligned}\ ] ] here @xmath20 and @xmath19 correspond to the mean network adaptation and global synaptic coupling variable . note that we have omitted the @xmath36 brackets denoting the average value of @xmath2 present in @xcite for simplicity and clarity . the function @xmath37 is the instantaneous network averaged firing rate , as a function of @xmath20 and @xmath19 . the function @xmath38 defines when the integral in makes sense . it defines _ switching manifold _ of the nonsmooth system . one can derive an expression for the switching manifold equation by determining when the denominator in @xmath39 first becomes zero somewhere in the @xmath40 phase space . to do this we find the minimum of @xmath41 , for @xmath42 $ ] , regarding @xmath20 and @xmath19 as fixed parameters . for the general class of models studied in @xcite , the function @xmath17 is assumed to be strictly convex , that is @xmath43 , and at least three times continuously differentiable . it follows that @xmath41 is also strictly convex as a function of @xmath1 , and hence its minimum on @xmath44 $ ] occurs at a critical point . the critical points ( as a function of @xmath1 ) are given by solving @xmath45 thus @xmath46 is the value of @xmath1 at which @xmath47 has a minimum . the leaky integrate - and - fire neuron is not strictly convex . in fact @xmath48 for this model , thus the minimum of @xmath47 occurs at the right endpoint of the interval . in either case , the minimum value defines the function @xmath38 @xmath49 and the switching manifold equation @xmath50 this latter expression is useful as we can think of @xmath51 as an @xmath52 and @xmath53 dependent rheobase current . anywhere in the phase space where @xmath54 the network is firing with mean firing rate given by @xmath55^{-1}. \label{fr}\ ] ] anywhere that @xmath56 the network is quiescent and the mean firing rate is 0 . there are a couple of important facts to note before we proceed further . first of all , @xmath57 , the rheobase current for the uncoupled , nonadapting neuron , which is governed by the equation @xmath58 based on the assumptions made on @xmath17 in @xcite , then this model neuron has a type - i firing profile . additionally , given that @xmath59 , we also have the following equation @xmath60 these two facts will prove important for our later analysis . to conclude we display some expressions for specific models . the rheobase currents are given by @xmath61 with corresponding minimum values of @xmath47 : @xmath62 for both the lif model and izhikevich model , the mean firing rate can be evaluated analytically : @xmath63 for the other models , the firing rate must be evaluated numerically . this can be done by integrating equation over @xmath44 $ ] treating @xmath64 , and @xmath20 as fixed parameters at each time step . this approach can be used to numerically analyze the bifurcation types of these equations using numerical bifurcation software , such as matcont , among others @xcite . however , the numerical integration method should be of high enough order accuracy for the numerical continuation results to be trusted . given the mean field system described above , one should consider whether numerical bifurcation results or analytical approaches should be taken . numerical bifurcation analysis can yield results which are accurate throughout the phase space , but require choosing a particular model and determining which parameters to fix and which to vary . analytical methods can yield model independent results and give insight into the role of various parameters in system behaviour , however , they are often restricted to particular regions of the parameter space and/or phase space , as we shall see . in order to proceed analytically , we need to sacrifice some of the complexity of the original mean field system . in particular , as the usual formulas of @xmath37 are difficult to deal with analytically , we approximate the firing rate for all the two dimensional integrate and fire models as follows : @xmath65 where the form of @xmath66 varies from model to model . there are two approaches to justify this particular approximation , one local and one global . the local method relies on the fact that all the neuron models fire with a type i profile , and as such , they have a characteristic firing rate near the switching manifold that is proportional to @xmath67 this basic result follows from the normal form for a saddle node bifurcation . this reduction is valid when @xmath68 is small . this occurs when @xmath69 and @xmath70 are small and @xmath22 is close to @xmath71 . we note that this approximation and very similar differential equations appear in @xcite . for example , equations ( 3.6)-(3.7 ) in @xcite are similar to ours however the interpretation for those particular equations was for the firing rate of an e / i coupled pair of neurons . additionally , the non - smooth nature of those equations is not explored to the best our knowledge in any source . for a more global approximation , one can fit a function @xmath72 to @xmath37 . for example , we have found that for the izhikevich network , the approximation @xmath73^{-1 } \approx \frac{1}{2}\sqrt{i - i^*(s , w ) } \label{mfs}\ ] ] is much better globally than the local approximation for type i firing neurons , which is given by : @xmath74 as shown in figure [ fig2 ] . as equations and only differ by a multiplicative constant , we take the firing rate to be in the form @xmath75 , which is valid for the izhikevich network globally , and for the other networks locally in the phase and parameter space . note that whichever constant is used , it can be merely absorbed into the @xmath76 and @xmath77 terms in the mean field dynamical equations . this particular approximation turns out to yield a system that is tractable to analysis of both the smooth and non - smooth bifurcations displayed by this system , and shows a considerable degree of accuracy with both the actual network , and the original mean - field system , as shown in figure [ fig1 ] . we reiterate that the non - smooth nature of these equations is often not considered . however , we will show that recent developments in non - smooth bifurcation theory allow us to analyze these equations more fully . with the above simplification , the approximate mean field system that we analyze is given by @xmath78 where the switching manifold varies depending on which neuron model the terms @xmath76 and @xmath77 are rescaled to absorb any constant term in ( [ fr1 ] ) or ( [ mfs ] ) . we will refer to equations ( [ s1])-([r1 ] ) from here on as the reduced mean - field system . we first classify the type of non - smoothness that this system exhibits . the system is smooth and has derivatives of all orders everywhere except on the switching manifold , i.e. , when @xmath79 . on the switching manifold , the system is continuous but not differentiable . thus , this is a piecewise - smooth continuous ( pwsc ) system . equivalently , it has a uniform order of discontinuity of 1 @xcite . in contrast , the order of discontinuity for the full ( non - simplified ) izhikevich mean field system is non - uniform . to see this , rewrite @xmath80 and consider the limit @xmath81 straightforward calculations show that this limit is @xmath82 , and hence the order of discontinuity of the model is @xmath83 , only if @xmath84 outside of this region the order of discontinuity is not @xmath83 . in general @xmath85 for physical reasons , so the order of discontinuity of the full izhikevich model is @xmath83 if @xmath70 is sufficiently small . thus the approximation simplifies the analysis as it has a uniform order of discontinuity , however , it will likely be a reasonable approximation of the izhikevich model only when this latter model has order of discontinuity @xmath83 , i.e. , if @xmath70 is sufficiently small . to summarize , for all the neuron models we consider the firing rate to be approximated by @xmath86 based on the type i profile for all the models , where @xmath87 varies depending on which model is used . these approximations are only local in nature , being valid only when @xmath88 , aside from the izhikevich model , as one can fit a more global approximation to the network averaged firing rate , as stated earlier . the equilibria of the mean field equations - depend on the sign of @xmath89 . if @xmath56 then the only equilibrium point is the trivial solution , @xmath90 , which is a stable node . this equilibrium corresponds to all the neurons being quiescent , @xmath91 , thus we will refer to it as the non - firing solution . it will only exist when the origin of the phase space lies in the region where @xmath92 , which corresponds to @xmath93 . alternatively , in the language of non - smooth dynamical systems theory , @xmath94 is virtual if @xmath95 and real if @xmath96 @xcite . if @xmath97 , nontrivial equilibria @xmath98 may exist . if they do then @xmath40 must satisfy @xmath99 where @xmath100 , and a similar equation holds for @xmath101 . equations ( [ work1 ] ) and ( [ work2 ] ) yield the following relationship @xmath102 thus the equilibria are given by @xmath103 where @xmath20 satisfies the nonlinear equation @xmath104 note that equation implies that @xmath105 . thus for an equilibrium to be a valid , it must satisfy @xmath106 ( which implies @xmath107 ) . the equilibrium condition for the quartic and adex models yield nonlinear equations without analytic closed form solutions . however , one can apply a power series ( assuming that @xmath70 is small ) to come up with an approximation to the steady solutions . note that @xmath108 is actually always a function of @xmath70 , as opposed to just @xmath20 , as it is given by solving the algebraic equation . thus , we can write down the following expansions for @xmath46 and @xmath109 @xmath110 where ( [ exp2 ] ) can be derived from using the relationship ( [ exp0 ] ) . since these expansions are valid if @xmath70 is sufficiently small , we will refer to them as the weak coupling expansions . if we use only the initial terms shown in equations ( [ exp1])-([exp2 ] ) , then the approximate solution obtained is exact in the case of the izhikevich model as all higher order terms vanish , and is the simplest analytical approximation to the other models . using these terms , we arrive at the following equation for the weak - coupling expansion to the equilibrium points : @xmath111 this equation can be solved , yielding two solution branches : @xmath112 we will denote the corresponding equilibria as @xmath113 . defining the new parameters @xmath114 the @xmath20 variable of the solution branches may be written @xmath115 note that @xmath116 . further , since @xmath117 is the minimum of @xmath17 and the reversal potential for an excitatory synapse is above the resting membrane potential , @xmath118 , we have @xmath119 . it follows that @xmath120 is a strictly positive function . in a similar way , one can solve for the equilibrium values of the full izhikevich model analytically . two equilibria @xmath121 of the same form are obtained with @xmath122 given by : @xmath123 introducing the parameters @xmath124 the steady states can again be written in the form . note , as a check of consistency , that @xmath125 , and @xmath126 for the izhikevich model . based on the form and the fact that @xmath116 , it is straightforward to show the signs of @xmath127 are as shown in figure [ fig3](a ) . since we require the equilibrium solutions to be positive , @xmath121 will have different regions of existence depending on the values of @xmath128 and @xmath129 . in particular , both equilibrium points exist when @xmath130 and @xmath131 in a wedge shaped region given by @xmath132 . only @xmath133 exists when @xmath134 . neither solution exists in other parts of the parameter space . the regions of existence of @xmath135 and the non - firing solution are show for the izhikevich model in figure [ fig3](b ) . away from the switching manifold , we can analyze the smooth bifurcations of the equilibria via linearization . the non - firing solution does not undergo any smooth bifurcations , as it lies in the region of phase space governed by the equations @xmath136 thus the non - firing solution is asymptotically stable when it exists and does not lie on the switching manifold , i.e. , for @xmath130 . to analyze the nontrivial equilibrium points , consider the jacobian of the mean field system in the region of phase space where @xmath54 @xmath137 we can simplify the jacobian by imposing the steady state condition @xmath138 . additionally , one can note that : @xmath139 applying these formulas yields the following simplified jacobian @xmath140 which has trace and determinant given by @xmath141 we can now discuss the stability of each equilibrium in its region of existence . first we apply the weak coupling expansion to the determinant : @xmath142 upon substitution of the equilibrium values of @xmath20 and using the definition of @xmath128 , we obtain @xmath143 since the sign of @xmath144 is strictly positive , and the equilibria are only defined when @xmath145 , we can immediately conclude that @xmath146 now , this implies that the equilibrium @xmath147 is always an unstable saddle . the equilibrium @xmath133 , however , can be a node or a focus and its stability is determined by the trace . we will discuss this further in section [ hopfsec ] . note that these results are only valid when @xmath148 is small for the qif and adex models , but are globally valid for the izhikevich model . we can use the equations for the trace and determinant to formulate necessary conditions for the equilibria to display certain smooth bifurcations . in particular , @xmath149 and @xmath150 are necessary conditions for an equilibrium to undergo a saddle - node bifurcation , while @xmath151 and @xmath152 are necessary conditions for a hopf bifurcation . having both @xmath149 and @xmath153 , is a necessary condition for a bogdanov - takens bifurcation . of course , to determine if these bifurcations actually occur requires checking additional genericity conditions . in the following section , we check these conditions where possible . as described above , necessary conditions for a saddle - node bifurcation are @xmath154 and @xmath150 . it is easy to see from that the first condition is satisfied for both @xmath135 when @xmath155 . it can be shown that the second condition is satisfied except at isolated points in the @xmath156 parameter space as determined in section [ hopfsec ] . in the following we will assume that we exclude these points . to pursue this further , we study the existence of the equilibria . from the previous subsection , we know that @xmath135 both exist if @xmath132 and neither exists if @xmath157 . when @xmath158 , the two equilibria collapse into a single equilibrium , with @xmath159 . we thus conclude that @xmath160 corresponds to a two - parameter curve of saddle - node bifurcation . rewriting this in terms of the original parameters yields the two - parameter bifurcation curve in terms of @xmath156 : @xmath161 thus , for fixed @xmath162 , @xmath163 is the value of the current that corresponds to a saddle - node bifurcation point . there are two things to note about @xmath164 . first , since @xmath165 is a strictly positive function , then @xmath166 with @xmath167 only if @xmath168 . also , this curve is only defined for @xmath169 . to see this , note that the saddle - node equilibrium , given by @xmath170 only exists if @xmath171 . since @xmath172 , @xmath173 only exists if @xmath169 . we shall see later that @xmath168 actually corresponds to a non - smooth co - dimension two bifurcation point . from the analysis of subsection [ stabsec ] , we know that only @xmath133 may undergo a hopf bifurcation and that det@xmath174 if @xmath175 . we thus conclude that the determinant condition for the hopf bifurcation is given by @xmath176 . to determine a necessary condition for the hopf bifurcation , we begin by applying the weak coupling expansion to the trace associated with @xmath177 : @xmath178 note that the first term is strictly negative and @xmath179 is a strictly positive function . setting the trace to zero and using equations - which define @xmath180 and @xmath177 yields @xmath181 solving for @xmath22 gives @xmath182 \stackrel{def}{= } i_{ah}(g ) \label{iahdef}\ ] ] we thus conclude that if @xmath183 and @xmath176 then the equilibrium @xmath177 has a pair of pure imaginary eigenvalues . recall that @xmath179 , @xmath120 , and @xmath165 , are positive functions . further , it is easy to check that @xmath184 . this leads to several observations . first , since the third equation in the sequence above can only be satisfied if @xmath185 , it follows that if @xmath186 then no hopf bifurcation occurs . second , from the first equation in the sequence above we must have @xmath187 in order for the equilibrium @xmath177 to exist at the hopf bifurcation . when @xmath188 @xmath189 and @xmath190 . we shall see later that the point @xmath191 , @xmath188 is a codimension-2 non - smooth bifurcation point . finally , if @xmath192 , then @xmath193 with @xmath194 only if @xmath195 . if @xmath196 , then it is possible for @xmath183 to intersect @xmath191 . we denote by @xmath197 the value of @xmath162 at the intersection point , if it exists . we can now determine the stability of the equilibrium @xmath133 by studying the trace equation . since the first term in this equation is strictly negative wherever it is defined ( when @xmath133 exists ) , the sign of the trace is determined by @xmath198 . since @xmath177 and @xmath179 are positive , it follows from the discussion above that in the case @xmath199 the trace negative , and hence @xmath200 is asymptotically stable , wherever @xmath133 exists . if @xmath201 then that the trace is negative ( and @xmath133 is asymptotically stable ) if @xmath202 or @xmath203 and @xmath204 . the trace is positive ( and @xmath133 is unstable ) if @xmath203 and @xmath205 . note that if @xmath22 is sufficiently close to @xmath206 then @xmath133 will have a pair of complex conjugate eigenvalues . in summary , for fixed @xmath162 with @xmath207 and @xmath208 , the equilibrium @xmath133 undergoes a hopf bifurcation at @xmath183 if @xmath176 . we do not attempt to determine the criticality of this bifurcation analytically , but will study it numerically in a later section . further , we can now state completely the conditions for the saddle - node bifurcation : for fixed @xmath162 with @xmath196 , the equilibria @xmath133 and @xmath147 undergo a saddle - node bifurcation when @xmath209 if @xmath210 . recall that necessary conditions for a bogdanov - takens bifurcation are @xmath211 and @xmath153 . thus , from the analysis of the last two subsections , bogdanov - takens bifurcations ( if they exist ) will occur at intersection point of the curves of saddle node and hopf bifurcations in the @xmath212 parameter space , i.e. , at values of @xmath162 such that @xmath213 , with @xmath214 . using the expressions for these curves gives @xmath215(g - g^ * ) \right]\\ 0 & = & a_2(g)\left(n(g)(g-\bar{g } ) - m(g)(g - g^*)\right)^2\\ & \updownarrow&\\ 0 & = & n(g)(g-\bar{g } ) - m(g)(g - g^*)\end{aligned}\ ] ] this latter equation may be simplified to a quadratic in @xmath162 : @xmath216 where @xmath217 analysis of the quadratic equation shows that if @xmath218 or @xmath219 then there are no roots with @xmath214 . however , if @xmath220 and @xmath221 is not too large then there can be up to two roots . thus there may be up to two bogdanov - takens points . the case @xmath222 is degenerate . there is a single root which is given by @xmath223 . this is not a standard / smooth bogdanov - takens point as the points @xmath224 and @xmath225 are not standard saddle - node and hopf points . we shall show later that this point arises at the collision of two codimension-2 non - smooth bifurcations , and thus corresponds to a codimension-3 non - smooth bifurcation . figure [ fig4 ] shows the smooth bifurcations for the mean field system corresponding to a network of izhikevich neurons with a the parameter values from @xcite . note that the hopf - bifurcation for both the full and reduced mean - field systems corresponds closely to the onset of bursting in the actual network , as noted in @xcite . for these parameter values , @xmath226 and no bogdanov - takens points are observed . figure [ fig5 ] shows the smooth bifurcations for the mean field system corresponding to networks of quartic integrate - and - fire neurons and adex neurons . in all figures the bifurcation curves derived from the weak coupling approximation of the model , i.e. equations and , are compared with curves for the full mean field model generated numerically in matcont @xcite . to study the non - smooth bifurcations for the mean field system , we will use the terminology and bifurcation classification for piecewise smooth continuous systems proposed in @xcite . we note that some care must be used when applying these ideas to our system . letting @xmath227^t$ ] and recalling the definition of the switching manifold , our system may be written in the general form used by @xcite : @xmath228 where @xmath229 however , @xmath230 is only defined for @xmath231 . in contrast , the work of @xcite assumes that both @xmath232 and @xmath230 are defined throughout the phase space . nevertheless , we able to classify a number of bifurcations in our system by analogy with the results in @xcite we will supplement our analysis with numerical studies of our example systems . in particular , we will perform a detailed study of the mean field system corresponding to a network of izhikevich neurons with parameters given in table [ table1 ] . recall that all the models we are considering have an equilibrium @xmath90 which exists ( and is a stable node ) if @xmath233 . when @xmath191 this equilibrium lies on the switching manifold @xmath234 . when @xmath134 , this equilibrium no longer exists as the origin is not an equilibrium of part of the mean field system corresponding to @xmath235 . in the terminology of non - smooth systems , the origin is a _ virtual equilibrium _ of the system for @xmath134 and undergoes a _ boundary equilibrium bifurcation _ ( beb ) when @xmath191 . the exact nature of this bifurcation depends on the value of @xmath162 , in particular , its relationship to @xmath236 , @xmath237 and @xmath197 . to determine the nature of the boundary equilibrium bifurcation , we begin by studying the nontrivial equilibria @xmath238 when @xmath191 . recalling the form for @xmath127 and noting that @xmath239 when @xmath191 , we find @xmath240 thus for @xmath241 , @xmath133 collides with @xmath94 at @xmath191 , and for @xmath196 , @xmath147 collides with @xmath94 . consider first the case @xmath242 ( which corresponds to @xmath218 ) . in this case there is no hopf bifurcation , so the results are straight forward . when @xmath241 @xmath133 is a sink which exists for @xmath134 . it collides with @xmath94 when @xmath191 and ceases to exist when @xmath130 . putting this together with the description of the existence and stability results for @xmath94 , we conclude that , for this range of @xmath162 values , the system undergoes a * persistence * beb at @xmath191 . this will be either a focus / node or node / node persistence beb depending on the classification of @xmath133 . when @xmath196 , recall that the equilibrium @xmath147 is a saddle when it exists ( for @xmath243 ) . since @xmath94 also exists for @xmath130 and is a stable node , we conclude that for @xmath244 there is a non - smooth saddle node beb at @xmath191 . now consider the case @xmath208 . for @xmath245 , analysis similar to that above shows the system undergoes a * persistence beb * at @xmath191 . figure [ beba ] shows this bifurcation for the mean field system corresponding to the izhikevich network with parameters as in table [ table1 ] . the situation for @xmath246 is similar , except that @xmath133 is now an unstable focus for @xmath134 . thus for this range of @xmath162 values , there is a focus / node persistence beb at @xmath191 . since @xmath133 is a source and @xmath94 is a sink , we may expect ( by analogy with the results in @xcite ) that a stable non - smooth limit cycle surrounding @xmath133 will be created as @xmath22 increases through @xmath71 . figure [ bebb ] confirms this for the mean field system corresponding to the izhikevich network . note that in this example , the amplitude of the limit cycle does not got to zero as @xmath22 approaches @xmath71 . ( see also in figure [ fig8c ] ) further , the period of the limit cycle diverges as @xmath247 . see figure [ fig8a ] . thus the limit cycle appears to be created in homoclinic - like bifurcation as @xmath22 increases through @xmath71 . we will thus refer to this as a * homoclinic persistence beb*. when @xmath196 , analysis similar to that above shows that there is a non - smooth saddle node beb at @xmath191 . based on the analysis of the equilibrium points , there is no reason to expect anything more to occur with this bifurcation . however , our numerical examples show two cases . figure [ bebd ] shows that a simple non - smooth saddle - node beb occurs for the mean field system corresponding to the izhikevich network with @xmath248 . figure [ bebc ] shows the bifurcation for the same system with @xmath249 . in this case there is a non - smooth limit cycle for @xmath134 that appears to be destroyed when @xmath191 . thus this bifurcation appears to be a non - smooth version of the saddle - node on an invariant circle ( snic ) bifurcation . we will refer to it as a * snic beb*. the transition between the two types of bebs that occur for @xmath196 will be discussed in a later section . based on our numerical results we hypothesize that a non - smooth limit cycle may be destroyed in a homoclinic - like bifurcation as @xmath22 decreases through @xmath71 . we support this hypothesis in two ways . first , consider the vector field in the neighbourhood of the origin . recall that the origin is always an attractor when it lies in the region where @xmath250 . in the region where @xmath251 , setting @xmath191 and retaining only the highest order terms in @xmath20 and @xmath2 gives : @xmath252 thus , for @xmath253 , and @xmath254 the vector field point away from the origin and the boundary equilibrium @xmath255 is a repeller in this region . since the boundary equilibrium point is as a repeller on one side of the switching manifold and and an attractor on the other , it is possible for a non - smooth homoclinic orbit to this equilibrium point to exist when @xmath191 . second , we show that under certain parameter conditions , if a non - smooth limit cycle surrounds the equilibrium @xmath200 , it must be destroyed when @xmath191 . to do this we show that trajectories that cross the switching manifold when @xmath71 lie within the basin of attraction of the origin . thus any non - smooth limit cycle must become homoclinic to the origin at @xmath191 . note that if @xmath256 , then @xmath257 and thus @xmath258 for some constant @xmath259 . assuming that the trajectory starts with @xmath260 on the switching manifold then @xmath261 where @xmath262 . now suppose this trajectory crosses the switching manifold again at @xmath98 . then @xmath263 clearly two solutions of this equation are @xmath260 and @xmath255 . dividing through by @xmath20 and simplifying one obtains @xmath264 where @xmath265 . if @xmath266 the left hand side of is monotonically increasing while the right hand side is a line with negative slope . hence @xmath260 is the unique intersection point . this means every trajectory that enters the region @xmath256 when @xmath191 is attracted to the origin . if @xmath267 , the left side of is now monotonically decreasing . unless the line is tangent the curve at @xmath268 there will always be another intersection point . rearranging the equation shows that this intersection point will occur for @xmath269 if @xmath270 is sufficiently small . for fixed @xmath162 , this means that any trajectory that starts on the switching manifold at @xmath271 with @xmath272 sufficiently small will be attracted to the origin . thus all non - smooth limit cycles that are close enough to the origin for @xmath134 will become homoclinic to the origin when @xmath191 . given how @xmath188 , and @xmath168 delimit the different types of beb bifurcations , it should be clear that these special points represent higher codimension bifurcations along the @xmath191 line . we shall explore these bifurcations further below , in addition to determining the geometrical meaning of these points . from the previous section , we concluded that the point @xmath274 is a special codimension - two bifurcation point where the boundary equilibrium bifurcation ( beb ) changes from a persistence beb to a non - smooth saddle - node . note that the smooth branch of saddle - node bifurcations found earlier actually emanates out from the codimension-2 point @xmath275 . we will show here that it does so in a highly non - generic way as the saddle - node equilibrium hits switching manifold tangentially at the beb , and is the only equilibrium point that interacts with the switching manifold in this way . we have seen that regardless of the parameter values , all the nontrivial equilibria lie on the curve @xmath276 . thus as any parameter is varied the nontrivial equilibrium will follow this curve , which has slope @xmath277 further , the only equilibrium that can be boundary equilibrium point is @xmath278 , the non - firing solution . now the switching manifold can be written as @xmath279 thus , at the slope of the switching manifold at the beb is @xmath280 equating and shows that the nontrivial equilibrium undergoing the beb will hit the switching manifold tangentially only if @xmath281 . from this it is straightforward to show that with @xmath168 fixed , the nontrivial equilibrium @xmath133 hits the switching manifold tangentially as @xmath282 @xmath283 . more interesting is to consider what happens when @xmath162 is varied . from our previous analysis we know that at the saddle - node bifurcation point , the saddle - node equilibrium , @xmath284 is defined by @xmath285 thus , as @xmath286 , @xmath287 . this implies that the saddle - node equilibrium hits the switching manifold tangentially at @xmath168 , @xmath191 . in summary the point @xmath288 is the collision between three branches of codimension-1 bifurcations : a pair of non - smooth boundary equilibrium bifurcations and a smooth branch of saddle - node bifurcations . the details of the beb involve depend on the relationship between @xmath236 and @xmath237 . if @xmath289 , bebs are simple : a simple node / focus or focus / focus persistence beb occurs for @xmath241 and a non - smooth saddle - node beb occurs for @xmath241 . the case @xmath208 is more complex due to the possible presence of limit cycles associated with the hopf bifurcation . in the case we studied numerically and described in section [ bebsec ] , for @xmath241 we observe a homoclinic persistence beb and for @xmath196 we observe a snic beb . while this bifurcation may be complicated , the bifurcation point can be determined analytically for all the models . of particular interest is the underlying physical interpretation . associated with this point is a region in the @xmath196 , @xmath130 quadrant of the parameter space with both stable firing and non - firing solutions , and hence bistability . this bifurcation point is shown in detail in figure [ fig10b ] . the andronov - hopf bifurcation described in section [ hopfsec ] leads to the creation of a limit cycle . as @xmath22 moves away from the bifurcation point , the amplitude of the limit cycle may increase enough that it hits the switching manifold tangentially , resulting in a grazing bifurcation . it is difficult to say much in general about the nature of this bifurcation , however , analysis similar to that in the last section shows that if @xmath130 then once a trajectory enters the region @xmath256 , it can not leave , but will be attracted to the origin . thus we expect that if a grazing bifurcation occurs for @xmath130 it will lead to the destruction of the limit cycle . to gain more insight , we performed a numerical study of the mean field system corresponding to the izhikevich network with parameter values as in table [ table1 ] . we first confirmed that the hopf bifurcation is subcritical , using matcont and by numerically simulating the time reversed system . we then showed that the unstable limit cycle generated by the hopf can undergo two different types of grazing bifurcations , depending on the value @xmath22 . for @xmath134 , the grazing bifurcation that occurs is a persistence type grazing , i.e. , the unstable limit cycle generated via the subcritical hopf bifurcation just becomes non - smooth after the grazing bifurcation . this is shown in figure [ fig7a ] . here , the limit cycle undergoes a grazing bifurcation at @xmath290 , and it persists past it . its amplitude rapidly increases past the grazing bifurcation , and it almost immediately undergoes a non - smooth saddle - node of limit cycles with a stable non - smooth limit cycle . for @xmath130 , the grazing bifurcation is a destruction type grazing as the limit cycle ceases to exist after the grazing for the reason discussed above . this is shown in figure [ fig7b ] . if the hopf were supercritical we would expect to see the same two types of grazing bifurcations . the only difference would be that the grazing bifurcation would occur for @xmath205 and we would not expect the saddle - node of limit cycles bifurcation to occur . the analysis of section [ bebsec ] showed that when @xmath208 the point @xmath291 is a codimension - two bifurcation point where the boundary equilibrium changes from a simple focus / node persistence beb to a homoclinic persistence beb . recall that the two parameter hopf bifurcation curve is given by @xmath183 as defined in section [ hopfsec ] . from the analysis in that section , the equilibrium point on the hopf curve is @xmath292 where @xmath293 . setting @xmath183 we see that as @xmath294 , @xmath295 and @xmath296 , that is the hopf equilibrium point undergoes a beb at @xmath291 . we thus refer to this point as a hopf boundary equilibrium bifurcation ( hopf beb ) . an alternative way to characterize the hopf beb is to fix @xmath191 and let @xmath297 . on @xmath191 , the mean field system for the izhikevich network may be approximated as follows : @xmath298 for @xmath98 in the vicinity of the origin . thus , we have @xmath299 for the trajectory of the homoclinic limit cycle . additionally , linearizing the switching manifold about the origin yields ; @xmath300 now , using these two equations we can solve for grazing bifurcations of the homoclinic limit cycle with the switching manifold at the origin . solving the grazing condition @xmath301 yields @xmath302 thus , the hopf beb bifurcation can be seen as a grazing bifurcation which destroys the non - smooth homoclinic limit cycle to the origin . our analysis so far shows three branches of bifurcation emanating from this codimension - two point : two non - smooth beb branches and a branch of hopf bifurcation . as shown in figure [ bebfig ] , for @xmath245 there is a simple persistence beb , while for @xmath303 there is a homoclinic persistence beb . we have numerically studied the bifurcations that occur in a neighbourhood of this point for the izhikevich model and find that that two more branches of bifurcation appear to emanate from this point we describe below . let @xmath162 be fixed with @xmath207 and consider the sequence of bifurcations involving limit cycles . at @xmath191 a stable non - smooth limit cycle is created in a homoclinic persistence beb , at @xmath304 an unstable smooth limit cycle is created in a subcritical hopf bifurcation . as @xmath22 increases the smooth limit cycle becomes non - smooth in a grazing bifurcation and then is destroyed along with the stable non - smooth limit cycle in a saddle - node of limit cycles . we wish to determine how the grazing and saddle - node of limit cycles bifurcations behave near @xmath188 . to do this we followed the stable non - smooth limit cycle along the hopf bifurcation curve . specifically , we numerically computed the amplitude and period of the limit cycle along the curve @xmath305 in the @xmath156 parameter space with @xmath306 . the results are shown in figure [ fig9 ] , specifically figure [ fig9a ] . the stable non - smooth limit cycle is computed using direct simulations of the ode system , where the system is initialized exterior to the limit cycle in the phase plane which ensures convergence . from this figure , we can see that the amplitude of the stable non - smooth limit cycles goes to 0 as @xmath306 . this implies that this limit cycle collapses to the origin @xmath255 . but as this bursting limit cycle is one part of the saddle - node of limit cycles bifurcation , then this bifurcation must also emerge from hopf beb . since the grazing bifurcation lies between the saddle - node of limit cycles and the hopf bifurcation , the persistence grazing bifurcation must also emerge from the point @xmath307 . the entire sequence of bifurcations near the hopf beb is shown in figures [ fig10a ] and [ fig10d ] . we briefly note that if @xmath308 , then we have @xmath309 which means that the hopf and saddle - node beb points coincide in a non - smooth codimension-3 bifurcation point . this bifurcation point may be thought of as a bogdanov - takens equilibrium point lying on a switching manifold . however , we note that there is no bogdanov - takens bifurcation ( or for that matter saddle - node or hopf bifurcations ) at this point in the classical sense , as the jacobian of the system diverges , and hence the conditions associated with these different smooth bifurcations can not be satisfied . this point appears to act as an organizing center for the bifurcation diagram , with all the non - smooth bifurcations emanating from it . due to the complexity of this point , we will leave its analysis for future work . however , it does illustrate how rich the non - smooth bifurcation sequence of this relatively simple pwsc system is . in addition to the two local non - smooth bifurcations that occur at @xmath188 , and @xmath168 , there appears to be a global codimension-2 bifurcation that occurs for these mean field systems . recall that there are two different types of grazing bifurcations , a destruction type ( which occurs for @xmath130 ) and a persistence type ( which occurs for @xmath134 . these are shown in figure [ fig7 ] . thus there is a co - dimension two point when the grazing bifurcation crosses @xmath191 . as for the other co - dimension two points , one may expect that would be a change in the beb bifurcations at this point . in the case we have studied numerically it appears that the beb changes from snic type before this transition to a regular non - smooth fold after . this is shown in figure [ fig10c ] and figure [ fig10d ] . note that this transition occurs for @xmath310 , i.e. , after the second intersection of the hopf curve with @xmath191 . it also appears that the saddle - node of non - smooth limit cycles bifurcation emanates from this point . note that this does not imply that there is a second impact with the hopf equilibrium and the switching manifold , as @xmath311 . this bifurcation results in the destruction of the homoclinic limit cycle that exists on @xmath191 , and it is very difficult to analyze , as it is a non - local co - dimension 2 non - smooth bifurcation . geometrically , however it occurs when the unstable smooth limit cycle ( generated via the hopf bifurcation ) grazes the switching manifold at @xmath191 . if the hopf bifurcation were supercritical instead of subcritical we would expect a similar codimension two point to occur ( if a grazing bifurcation occurred ) . however , it would occur for @xmath312 . again , due to the complexity of this particular bifurcation , further analysis is beyond the scope of this paper , and we leave it for future work . while the preceding analysis revealed a great deal of novelty and non - smooth bifurcations for the reduced mean - field system , in order for the non - smooth analysis to be useful , it has to be reflective of the phenomenon displayed by the actual network . here , we will demonstrate many of the non - smooth bifurcations predicted in the analysis are present in a full network of neurons . unfortunately however , one can not easily simulate the large network of neurons in such a way as to expose unstable equilibria and limit cycles . for example , the equilibrium point @xmath133 is a saddle in the mean - field , and short of somehow initializing the network of neurons on the stable manifold of the saddle , it can not be resolved via direct simulations . however , the unstable node @xmath133 can be resolved by modifying the network as follows . using the separation of time scales between the fast variable @xmath20 , and the slow variable @xmath2 , we replace the full network by the following : @xmath313 for @xmath314 . here the dynamics of @xmath20 are replaced entirely by its steady state , large network solution : @xmath315 , with @xmath2 replaced by the finite mean @xmath316 . we will refer to this network of neurons as the slow network . the resulting mean - field system for the slow network is simply a one - dimensional non - smooth ode , given by : @xmath317 the mean - field system for the slow network has the same steady states as the mean - field system for the full network , the two solutions @xmath318 , in addition to the non firing solution , @xmath319 . however , being a one dimensional system , no hopf bifurcations ( and thus oscillations ) are present in the mean - field system for the slow network . additionally , one can show that @xmath320 is always stable , and @xmath321 is always unstable as the eigenvalue for the steady state is given by @xmath322 the functions @xmath165 , and @xmath120 are identical to those for the mean - field system for the full network , thus we should expect that the slow network has a stable steady state , and undergoes a saddle node bifurcation as for the full network . as there can be no oscillations , we expect that @xmath320 exists and is stable for @xmath241 , and @xmath134 and @xmath196 , and @xmath323 . as the non - firing solution is also stable for @xmath130 , then we should expect bistability for @xmath243 . indeed , if we simulate the slow network with a slowly varying current that either decreases from current values greater then @xmath71 or increases from current values less then @xmath71 , we get bistability for @xmath196 and none for @xmath241 . this is shown in figure [ fig11 ] using the simulations of the slow network and the full network , we can piece together a pseudo - bifurcation diagram for the full network . this is shown in figure [ fig12 ] . the boundary equilibrium bifurcations that occur near the vicinity of @xmath236 are both present for the actual network . given the similarities between the bifurcation diagram for the actual network , and that predicted by the non - smooth mean - field equations , it would appear that in order to understand the bifurcations that occur for these networks , one has to consider non - smooth bifurcation theory . one might wonder as to whether or not the non - smooth nature of the mean - field system is a direct result of the non - smooth nature of the neurons , given the fact that they have discrete resets and jumps . however , this is not the case . in particular , the firing rate of all type i neurons in the vicinity of the saddle - node on an invariant circle bifurcation is always proportional to @xmath324 . assuming that the dynamics of the neurons voltage is much faster then the dynamics of the all the other intrinsic and synaptic currents , one could obtain mean - field equations very similar to that obtained explicitly here . for example , this is done in the finite network case in the work of @xcite . thus , one has to consider non - smooth bifurcations and bifurcation analysis when working with mean - field systems for type i neurons . the same is true of type 2 neurons , however the firing rate for these neurons changes discontinuously at @xmath71 , and thus it is likely that the mean - field systems for type 2 neurons would be completely non - smooth , as opposed to piecewise smooth continuous . through our analysis of the mean field systems for large networks of coupled neurons , a number of new non - smooth bifurcations have been discovered that have been previously been unknown in the literature . these include two co - dimension 1 branches of boundary equilibrium bifurcations that have homoclinic limit cycles at the bifurcation point , and can be thought of as generating / destroying non - smooth limit cycles . additionally , a pair of co - dimension 2 bifurcations have also been discovered that result from the collision of classical smooth branches of bifurcations with non - smooth bifurcations . these occur when either a hopf equilibrium point , or a saddle - node equilibrium point collide with a switching manifold . we have determined locally in a neighbourhood of these bifurcation points the resulting behavior of the system through analytical and numerical results . additionally , a global - codimension 2 bifurcation and the collision of a bogdanov - takens equilibrium point with a switching manifold was also discovered by analyzing these systems , however we leave their analysis for later work . given the analysis we have performed , then there are several predictions we can make , using the bifurcation curves for all the necessary branches . for example , it appears that the time scales @xmath221 and @xmath325 are crucial for determining the presence of bursting . if @xmath218 , then no bursting can occur , while if @xmath220 , there is a bell shaped region of bursting for @xmath134 , and @xmath207 . thus , if the adaptation time scale is smaller then the time scale of the synapses , adapting , recurrently coupled networks would not burst . this can be the case for example for weakly adapting neurons coupled together with nmda synapses , which have a long time scale . unfortunately however , while much of these bifurcations can be at least derived , one can not easily determine whether or not even the smooth bifurcations are generic in any sense . this is due to the fact that these particular systems can not be diagonalized very easily , due to the presence of the @xmath326 term , which has an unbounded derivative as @xmath327 . thus , center manifold theory can not be directly applied , and many of the genericity conditions can not be verified . in addition to these problems with regards to smooth bifurcations , one can not easily apply the existing non - smooth theory to these systems . the systems in the form @xmath328 are clearly piecewise smooth continuous , however unlike the vast majority of pwsc systems discussed in the literature , they fail to satisfy one critical constraint that these other systems have . in normal piecewise smooth continuous systems , given by @xmath329 where @xmath330 on @xmath331 , it is assumed that both @xmath332 and @xmath333 exists everywhere , and are smooth . in our system , @xmath326 only exists when @xmath334 and its first derivative only exists when @xmath254 . this renders much of the analysis on pwsc systems inapplicable . indeed , this system can not even be regularized in the normal way , via a teixeira type regularization scheme @xcite due to the fact that @xmath86 is undefined when @xmath335 . however , there are alternate ways to apply both center manifold theory , and simultaneously regularize this system . in particular , consider the three - dimensional system given by @xmath336 where @xmath337 is a small constant . in this singularly perturbed system , one can show that as @xmath338 , one recovers the piecewise smooth continuous system as when @xmath337 is small , we can regard @xmath20 and @xmath2 as fixed , and thus @xmath339 rapidly converges to the steady states @xmath82 , or @xmath86 , depending on the sign of @xmath340 . using this type of embedded system , which is entirely polynomial for the izhikevich mean field system , one has regularized the non - smooth system in a sense by embedding it as the fast system in a singular perturbation problem . thus , to actually determine the genericity properties of the bifurcations displayed above , in addition to how the non - smooth bifurcations discovered are related to the general smooth bifurcation theory , one can analyze either directly or numerically the embedded system for finite @xmath337 . preliminarily , we have found that the hopf beb bifurcation seems to be a bautin point under the regularization , and the saddle - node beb bifurcation point seems to be a bogdanov - takens point under the regularization . these are both co - dimension 2 smooth bifurcation , and they also explain the emergence of the non - smooth saddle - node of periodics in the hopf beb , which has a smooth saddle - node of periodics . additionally , the regularized bogdanov - takens has ( generically ) a branch of homoclinic bifurcations , which also exists in a non - smooth form for the saddle - node beb . however , as the embedded system , and its justification as a regularization are outside of the scope of this paper , we leave it for future work . in addition to the embedded regularization , this system is also unusual in the sense that there is a natural regularization for the mean - field system . suppose we consider the voltage equations to be perturbed by white noise : @xmath341 where @xmath342 and @xmath343 . in which case one can rigorously derive a mean - field system for this network of equations which is identical to the original mean - field system given in equations ( [ mfa])-([mfc ] ) only the firing rate is now given by : @xmath344^{-1}\end{aligned}\ ] ] where @xmath345 is an anti - derivative ( in @xmath1 ) of @xmath346 . as we shall do in forthcoming work @xcite one can rigorously show that this expression for @xmath347 is smooth with respect to @xmath20 and @xmath2 , and always defined , and converges to ( [ mfc ] ) as @xmath348 . thus , the mean - field system for a network with noise parameterized by @xmath349 , the variance in the noise , converges to the system ( [ mfa])-([mfc ] ) as the variance of the noise becomes negligible . but , since the mean - field system with noise is smooth , it can be thought of as a natural regularization for the non - smooth mean - field system . we remark that this is unusual in the field of non - smooth theory as generally a regularization is chosen or suggested , and is typically of the teixeira form @xcite . we leave the bifurcation analysis of the system with noise for future work . finally , one may ask if the non - smooth bifurcations we analyze here appear in other non - smooth systems or are generic in any way . to the best of our knowledge , the co - dimension 2 bifurcations are novel in the literature , according to a recent review @xcite . however , a hopf - bifurcation occuring on a discontinuity boundary ( a co - dimension 2 non - smooth bifurcation ) does occur in the example ( in section 6 ) in @xcite ( see figure 29 . ) . however , the system examined in @xcite is a fillipov system , and thus has a higher order of discontinuity . as to whether these bifurcations occur in a more generic system , we intend to explore this further with a more generic piecewise smooth continuous system that does not have undefined derivatives on the switching manifolds . in particular , we note that with a quadratic pwsc system , one can show that as for our saddle - node beb bifurcation , at the intersection between generic branches of persistence and non - smooth fold beb bifurcations , the equilibrium of @xmath333 must have a zero eigenvalue at this co - dimension 2 point . we intend to explore all the possible cases and their possible relationships to the co - dimension 2 non - smooth bifurcations we outline in this paper for future work . r. brette and w. gerstner . adaptive exponential integrate - and - fire model as an effective description of neuronal activity . , 94(5):36373642 , 2005 . a. buzzi , p.r . da silva , and m.a . teixeira a singular approach to discontinuous vector fields on the plane , 231:633655 , 2006 k.a . ferguson , c.y.l . huh , b. amilhon , s. williams , and f.k . skinner simple , biologically - constrained ca1 pyramidal cell models using an intact , whole hippocampus context , 2014 , 3:104 ( doi : 10.12688/f1000research.3894.1 ) w. nicola , and s.a . bifurcations of large networks of two - dimensional integrate and fire neurons , 35 : 87108 , 2013 . . nykamp . and d. tranchina . a population density approach that facilitates large - scale modeling of neural networks : analysis and an application to orientation tuning . , 8:1950 , 2000 . a. omurtag , b.w . knight , and l. sirovich . on the simulation of large populations of neurons . , 8:5163 , 2000 . ( in red ) using @xmath350 for @xmath8 predicted from topological normal form theory ( @xmath351 , green ) , which is only locally valid near the transition to firing from quiescence , or for a more global fit ( @xmath352 , blue ) . the specific @xmath8 used does not matter as it can be merely absorbed into the @xmath76 and @xmath77 parameters and the final bifurcation analysis is the same . ] .parameters for various network types and the mean - field systems . note that parameter above are dimensionless , where as in some of the cited sources they are in dimensional form only . [ cols="^,^,^,^",options="header " , ]

we report the detection of 10 new x - ray filaments using the data from the _ chandra _ x - ray satellite for the inner ( parsec ) around the galactic center ( gc ) . all these x - ray filaments are characterized by non - thermal energy spectra , and most of them have point - like features at their heads that point inward . fitted with the simple absorbed power - law model , the measured x - ray flux from an individual filament in the kev band is to ergs s and the absorption - corrected x - ray luminosity is ergs s at a presumed distance of 8 kpc to the gc . we speculate the origin(s ) of these filaments by morphologies and by comparing their x - ray images with the corresponding radio and infrared images . on the basis of combined information available , we suspect that these x - ray filaments might be pulsar wind nebulae ( pwne ) associated with pulsars of age yr . the fact that most of the filament tails point outward may further suggest a high velocity wind blowing away form the gc .

0710.2668

the galactic center ( gc ) is the only place where we can observe parsec details of various interaction in and around the galactic nucleus . advances in this research frontier rely primarily on observations at radio , infrared and x - ray wavelengths , because the optical band suffers seriously from a considerable extinction with an @xmath9 ( e.g. , becklin , matthews , neugebauer & willner 1978 ) . one of the most important discoveries in the gc region is perhaps the presence of many structured non - thermal radio filaments ( ntfs ) ( e.g. , yusef - zadeh , morris , & chance 1984 ; morris & serabyn 1996 ; larosa , kassim , lazio , & hyman 2000 ) . while these non - thermal radio filaments have been intensively studied , their origins and implications on the underlying physical processes around the gc region remain largely unclear . in the current analysis of x - ray filaments around the gc region , these x - ray emitting particles are usually expected to be fairly close to their acceleration zone and evolve very rapidly in time . thus , the x - ray study of the same region would be essential to probe the origin of these energetic particles . the _ chandra _ galactic center survey ( cgs ) , with its unprecedented high spatial resolution of @xmath10 and moderate spectroscopy capability , has already revealed remarkable x - ray structures ( including thousands of x - ray bright point sources and some filaments , as well as clumps of diffuse emission ) within the central @xmath11 pc of our galaxy ( e.g. , wang , lu , & lang 2002a ) . in this paper , we mainly concentrate on the nature of those thread - like linear structures or filaments as observed in x - ray bands . up to this point within @xmath12 ( @xmath13 pc at 8 kpc ) from where a @xmath14 black hole resides inside a compact region of less than @xmath15au ( e.g. , shen et al . 2005 ) , 5 x - ray filaments have been studied in details ( see also table 1 ) . for example , g359.95 - 0.04 , a comet - like filament at @xmath16 pc north of , is thought to be a ram - pressure confined pulsar wind nebula ( pwn ) ( wang et al . another prominent filament , g359.89 - 0.08 ( sgra - e ) , at @xmath17 pc southeast of , was first noticed by sakano et al . ( 2003 ) and an interpretation of a possible pwn origin was discussed in details by lu , wang & lang ( 2003 ) . an alternative picture of g359.89 - 0.08 as a source of synchrotron emission from relativistic particles accelerated by a shock wave of w28 snr was suggested recently by yusef - zadeh et al . ( 2005 ) ; in that same paper , they also detected a new x - ray filament g359.90 - 0.06 , which coincides spatially with a radio filament at @xmath18 pc southwest of . they explored the mechanism of an inverse compton scattering ( ics ) for the x - ray emission of g359.90 - 0.06 . another 2 filaments were found in more extensive regions . g0.13 - 0.11 in the radio arc region was first reported by wang , lu & lang ( 2003 ) and was suspected to be also a pwn . of particular interest is the x - ray filament g359.54 + 0.18 ; it associates with only one strand of the obviously bifurcated radio threads ( e.g. , yusef - zadeh et al . 2005 ; lu et al . 2003 ) . a common feature shared in the x - ray energy spectra of these filaments is that they all appear to be non - thermal . it should be noted however that any thermal component to these sources would likely be completely absorbed and unobservable , given the high foreground column density . in these earlier investigations , pulsar wind nebulae ( pwns ) and supernova remnants ( snrs ) seem to offer natural explanations for the appearance of such x - ray filaments . indeed , it is believed that a considerable number of supernovae should have happened in the gc region ( e.g. , figer et al . 1999 , 2004 ; wang et al . 2006 ) . one would naturally expect to find some of their end - products such as pulsars and snrs in the gc region . however , no radio pulsars have yet been found within @xmath19 of the gc ( wang et al . this might be caused by difficulties in radio observations ( cordes & lazio 1997 ; johnston et al . 2006 ) . seeking observational clues in x - ray bands might shed new light to the search for radio pulsars embedded in the gc region . another tempting idea is to use these x - ray filaments as potential tracers for the magnetic field and gas dynamics around the gc region , since the magnetic fields should have played a significant role in producing such non - thermal spectra and the thread - like shapes of these filaments and the gas motion is usually coupled with the magnetic fields . ( e.g. , chevalier 1992 ; boldyrev & yusef - zadeh 2006 ) . magnetic fields exist on all scales of the galaxy as well as in other spiral galaxies and generally trace out spiral patterns on large scales ( e.g. , beck & hoernes 1996 ; fan & lou 1996 ; zweibel & heiles 1997 ; wielebinski 2005 ; ferriere 2001 , 2007 ) , and great progress has been made in measuring them and inferring their influence by various means . for example , the observed high - energy cosmic ray anisotropy at a few times @xmath20 ( e.g. , amenomori et al . 2006 ) might be physically related to large - scale structures of galactic magnetic fields and inhomogeneous cosmic ray source distribution . using diffuse synchrotron radio emissions at 74 and 330 mhz frequencies produced by relativistic cosmic - ray electrons and the magnetic field around the galactic center , larosa et al . ( 2005 ) inferred a weak magnetic field of order of @xmath21 g on size scales @xmath22 based on a minimum - energy analysis . this is about 2 orders of magnitude lower than @xmath15 mg usually estimated for the gc region . very recently , cowin & morris ( 2007 ) argued that the assumption of larosa et al . ( 2005 ) that the magnetic field and cosmic rays are in a minimum - energy state across this region is unlikely to be valid . according to their model estimates , the mean magnetic field is at least 100 microgauss on a scale of several hundred parsecs and peaks at approximately 500 microgauss at the center of the diffuse nonthermal source ( dns ) . this is an important issue to be settled for the gc magnetic environment . most of the gc radio ntfs are found to be perpendicular to the galactic plane , implying a local poloidal magnetic fields of about milli - gauss strengths ( e.g. , yusef - zadeh & morris 1987 ) . however , recent observations revealed that the gc magnetic field may be more complex than a simple globally ordered dipolar field ( e.g. , larosa et al . 2004 ; nord et al . it might be possible for x - ray ntfs to also provide clues of the configuration of the local magnetic field as well as the interaction between it and the ambient gas flow . we report morphological and spectral properties of another 10 newly discovered x - ray filaments within a region of @xmath23 surrounding the ( roughly corresponding to a projected sky area of @xmath1 pc by @xmath1 pc at a presumed distance of 8 kpc to the gc ) . their plausible physical origins are discussed in section 4 . all error bars in the x - ray spectrum parameter measurements are at the 90% confidence level , and we express the fitted parameters in the format of @xmath24 , where @xmath25 is the best fit value while @xmath26 and @xmath27 are the lower and upper limits of the 90% confidence interval , respectively . for all the images in the paper , north is up and east is to the left . we shall adopt a distance of 8 kpc from the solar system to the gc throughout this paper . we note that during the review process of this manuscript , muno et al . ( 2007 ) submitted a paper also on x - ray filaments around the galactic center . this work takes advantage of a large number of observations by the advanced ccd imaging spectrometer ( acis - i ) onboard of _ chandra _ x - ray satellite aimed at the ( e.g. , baganoff et al . we utilize the archival data available by september 2005 , including the eleven observations taken before may 3 , 2002 , as listed by park et al . ( 2005 ) ( 1561b excluded ) , and the three more recent observations ( obs . i d 3549 , 4683 , and 4684 ) taken on june 19 , 2003 and july 5 and 6 , 2004 . the total resultant exposure time of the included 14 observations is @xmath28 760 ks . following the standard event reprocessing procedures from the package of chandra interactive analysis of observations ( ciao ) ( version 3.2.2 ) , we first processed individual observations , including the correction for charge transfer inefficiency , bad - pixel - removing , and light - curve cleaning etc . we then co - add the data from the 14 observations by the sky coordinates to generate the composite data set . the detailed data preparation procedures are described in wang et al . ( 2006 ) . when placed on axis at the focal plane of the grazing - incidence x - ray mirrors , the imaging resolution ( fwhm ) is determined primarily by the pixel size of the ccds of 0@xmath29492 . the ccds also measure the energies of incident photons with a resolution of @xmath30 ev ( depending on photon energy and distance from the read - out node ) ( e.g. , baganoff et al . 2003 ) . we used the x - ray data in @xmath31 kev energy band for the spectrum study . the ccd frames are read out every 3.2 s , which provides the nominal time resolution of the x - ray data ( e.g. , muno et al . 2006 ) . here , we focus on the inner @xmath28 @xmath32 region centered around the where the point spread function ( psf ) broadening is insignificant . the source region framed for extracting the spectrum of each source is chosen in such a way that it is small enough to minimize contaminations from the surroundings and yet large enough to contain the bulk of filament emission . for those x - ray filaments with a sufficient number of photons for statistics , we divide them into sub - regions to look for the spectral evolution along the filamentary structure . the background spectrum is extracted from one or several regions in the environs of the relevant filament under consideration . for each filament , we have used different background regions to test their effects on the final spectra . although the best fit parameters do vary slightly with different backgrounds , this variation is much smaller than the uncertainties of the current data statistics . most of the x - ray spectra are modelled by the absorbed power - law continuum ( @xmath33 for short hereafter ) in _ xspec_. the galactic coordinate of a filament is given by the position of its brightest part . incidentally , the size of the filament given in this paper should only be regarded as a rough estimate . 1 gives a panoramic view of the x - ray filaments in the gc region studied here . is buried deep in the lower right region where the emission is saturated in fig . 1 . we studied the images of each x - ray filament in details and the results are shown in the left columns of fig . 2 and fig . 3 . from the x - ray contours , we can see that there are signs for the existence of point - like sources in some filaments . the right columns of fig . 2 and fig . 3 show the @xmath31 kev counts image created from the composite event file of the 14 _ chandra _ observations as mentioned in the last section . the rectangular boxes drawn on the images represent the source regions ( including the sub - regions used for studying spectral evolution along a filament ) . 4 shows the final spectra of the ten filaments extracted from the `` entire '' source regions . all of them can be well fitted with the @xmath33 model , implying that the x - ray emissions we detected are dominated by the nonthermal component . it should be noted that if these sources emit any thermal components , they would likely be completely absorbed and unobservable . in fig . 5 , we give the joint - fit spectra extracted from the sub - regions of filaments f1 , f3 , f6 and f10 , whose photon counts are large enough to allow for a comparison of the sub - region spectra . the spectrum fitting results are summarized in table 1 . we have also sought for the counterparts of x - ray filaments in h , k infrared bands and in 20 cm radio band . the infrared images were produced by the 2 micro all sky survey project and were downloaded from the _ skyview _ website . the radio data were acquired by the very large array ( vla ) on july 23 , 2001 , with a beam size of 15@xmath34 ( lang et al . no obvious filamentary structures were found ( with an upper limit of 0.12 jansky / beam ) , except for filament f10 , which coincides with a statistically significant radio filament . in addition to the fitting parameters for filaments summarized in table 1 , we now describe properties of each x - ray filament below . * filament 1 ( f1 ) : g359.936 - 0.038 * filament f1 lies @xmath35 ( @xmath36 pc ) southwest of . the brightest part of f1 is near the celestial coordinates _ ( l , b)_=(@xmath37 , @xmath38 ) . the elongation occurs almost in the exact north - south direction with a point - like source residing at the head in the northeast part . it has a rough length of @xmath39 and a width of @xmath40 . its spectra can be well fitted by @xmath33 . we divided the filament into the head and tail regions , and fitted their respective spectra . although the the best - fit @xmath41 appears twice bigger than @xmath42 , the error bars are larger than the difference . therefore , the current data are not good enough to constrain the spectral evolution of f1 . * filament 2 ( f2 ) : g359.934 - 0.037 * filament f2 has a size of @xmath43 . the most concentrated part of f2 is near the celestial coordinates _ ( l , b ) _ = ( @xmath44 , @xmath45 ) , which is @xmath46 ( @xmath47 ) southwest of . the elongation is along the northeast - southwest direction with the head in the northeast end . similar to f1 , there is no obvious emission line and a simple power - law gives a fairly satisfactory fit . * filament 3 ( f3 ) : g359.965 - 0.053 * at @xmath48 ( @xmath283 pc ) northeast of lies filament f3 with a size of @xmath49 ; the elongation is from northeast ( ne ) to southwest ( sw ) . this filament was first mentioned by baganoff et al . ( 2003 ) as a curious linear feature " . they reported its existence in the @xmath50 kev and @xmath51 kev bands , but not in the @xmath52 kev band . our analysis also shows that the bulk of f3 emission comes from photons of energies below 6 kev . the x - ray morphology clearly shows that f3 contains two concentrations . in order to study their spectral evolution , we divide it into the `` head '' ( the ne concentration with stronger emission ) , `` middle '' ( between ne and sw ) , and `` tail '' ( the sw concentration with weaker emission ) regions . there appears a spectral steepening from the `` tail '' to `` head '' . however , since no obvious point - like source is detected in the `` head '' region , the `` head '' to the `` tail '' direction does not necessariely mean the electron flow direction , as will be further discussed in section 4.3 . the two - blob morphology is somewhat unexpected and one might wonder whether the three regions we choose are in fact coherent parts of the same source . actually , the spectral results may cast doubt on the coherence of the `` head '' and `` tail '' . first , the @xmath53 value in table 1 shows that the `` head '' region has a hydrogen column density of @xmath54 @xmath5 , while the `` tail '' region has a @xmath53 value of @xmath55 @xmath5 , which might suggests that the `` tail '' could be nearer to us than the head " is . second , the spectrum of the tail " shows slight signs of 6.7 kev emission line , which is not found in the head " and middle " regions . however , we find that these spectral `` differences '' are most likely not caused by the emitter proper . the higher @xmath53 to the head " region is quite probably due to a small molecular cloud in this direction ( e.g. , coil & ho 1999 , 2000 ) . 2 of coil & ho ( 2000 ) shows that the column density of the nh3 cloud in the `` head '' region is about 1/10 of that of the streamer at r.a.@xmath56 , dec@xmath5729 02 30 , while the equivalent @xmath53 of this streamer is @xmath58 @xmath5 ( coil & ho 1999 ) . so @xmath53 of the small nh3 cloud is @xmath59 @xmath5 , comparable to the @xmath53 difference between the `` head '' and `` tail '' spectra . the slight 6.7 kev emission line feature in the spectrum of the `` tail '' is likely due to an insufficient subtraction of the background emission . for comparison , we extract the spectrum of the dense diffuse emission near the tail " region and do find that it has a fairly strong 6.7 kev iron line ( see the upper panel of fig . 6 ) . therefore , f3 is most likely one source object in space . * filament 4 ( f4 ) : g359.964 - 0.056 * filament f4 is close to filament f3 in sky projection . the elongation occurs in the northeast - southwest direction . it has a size of @xmath49 . the spectrum also appears to be non - thermal . * filament 5 ( f5 ) : g359.959 - 0.028 * this filament lies @xmath60 ( @xmath61 ) northwest of , with a length of @xmath62 and a width of @xmath63 . the elongation is oriented along the northwest - southeast direction for fitting parameters ) . there exists a point source near the southeast end of f5 ( separation @xmath64 ) ; however , this is very likely a chance spatial coincidence and thus we do not include it in the source region . * filament 6 ( f6 ) : g359.969 - 0.038 * filament f6 lies @xmath65 ( @xmath66 ) northeast of . it has a width of @xmath40 and a length of @xmath67 . the elongation occurs in the northeast - southwest direction , and a point - like source resides at the southwest end . along the elongation , we divide f6 into the head ( sw ) and tail ( ne ) parts . the spectra of these two parts are the same within error bars . * filament 7 ( f7 ) : g359.920 - 0.029 * filament f7 also lies @xmath68 ( @xmath69 ) southwest of , with a size of @xmath70 . the elongation is along the northwest - southeast direction fairly close to the east - west direction . since there are no x - ray photons below 4 kev , the statistics are extremely poor for this filament and therefore , the energy spectrum can not be fitted to distinguish any models . nevertheless , we still fit it with an absorbed power - law model . * filament 8 ( f8 ) : g359.970 - 0.009 * filament f8 is located @xmath71 ( @xmath72 ) northwest of . the elongation is along the northwest - southeast direction . it has a width of @xmath73 and a length of @xmath74 . the energy spectrum of filament f8 also appears to be non - thermal . * filament 9 ( f9 ) : g359.974 - 0.00 * filament f9 is @xmath75 ( @xmath76 ) northwest of . its elongation is almost parallel to those of filaments f5 and f8 . it has a width of @xmath73 and a length of @xmath77 . the energy spectrum of filament f9 also appears to be non - thermal . * filament 10 ( f10 ) : g0.029 - 0.06 * this filament lies @xmath78 ( @xmath79 ) northeast of . with a width of @xmath80 and a length of @xmath81 , it is the longest x - ray filament so far identified within the central 15 pc of the galaxy center . the energy spectrum of filament f10 is rather flat and non - thermal , with the column density of a typical value for the absorption around the gc . to see the photon index evolution along the filament , we divide the filament into three parts : `` head '' ( i.e. where the assumed point - source resides ) , `` middle '' , and `` tail '' parts . we then conducted a joint fit for the energy spectra of these three regions . as listed in table 1 , there appears a spectral steepening from the `` head '' to `` tail '' . no ir counterparts are found for filament f10 . but quite interestingly , there is a 20 cm radio filament more or less coincident with f10 spatially ; implications of this spatial coincidence will be discussed in @xmath82 . while x - ray photon numbers may not be high enough to constrain the exact shapes of the energy spectra , all spectra appear featureless except for filament f3 showing weak iron line features ( see section 3.3 ) and can be well fitted with power - law models . fitting of some of these energy spectra with thermal emission models is acceptable statistically ; nevertheless , this always gives quite high temperatures , i.e. , @xmath83 kev . we therefore incline to the view that x - ray emissions from these filaments are non - thermal in nature . table 2 sums up the inferred parameters for these 15 non - thermal x - ray filaments in the inner @xmath84 around the gc . in addition to the 10 filaments studied in this paper , we also include the other 5 filaments , namely g359.89 - 0.08 , g359.90 - 0.06 , g359.95 - 0.04 , g359.983 - 0.046 , and g0.13 - 0.11 , reported and studied earlier in the literature . most of their hydrogen column density @xmath53 are of the order of @xmath85 , consistent with other @xmath53 estimates around the gc region . most of their photon indices @xmath86 fall within the range of @xmath87 . the chance of these x - ray filaments being background extragalactic sources is very small according to the spectral and morphological properties . for instance , it would be very difficult to explain the linear filamentary morphology using the hypothesis of extragalactic origins . therefore , these x - ray filaments are most likely unique objects around the gc region . as already discussed in section 1 , there were suggestions that these x - ray filaments may be ram - pressure confined pwne ( wang et al . 2003 , 2006 ) , or synchrotron emissions from mhd shocks associated snrs or emissions resulting from inverse compton scattering ( yusef - zadeh et al . 2005 ; figer et al . 1999 ) . the non - thermal x - ray emission mechanisms may be either synchrotron emission or inverse - compton scattering . magnetohydrodynamic ( mhd ) relativistic pulsar winds ( michel 1969 ; goldreich & julian 1970 ; kennel & coroniti 1984a , b ; lou 1996 , 1998 ) and mhd shock interactions of magnetized outflows ( e.g. , yu & lou 2005 ; yu et al . 2006 ; lou & wang 2006 , 2007 ) with the interstellar medium ( ism ) in snrs and pwne could provide high - energy electrons needed in these two radiation mechanisms ( e.g. , sakano et al . 1993 ; lu et al . 2003 ; wang et al . if these x - ray filaments are snrs , their elongations would probably represent mhd shock fronts and therefore , one would not expect to see a tendency of spectral softening along a filament . the two bright filaments f3 and f10 both show evidence for such a softening tendency . the fact that most of these filaments have point - like sources at the heads also againsts the snr origin . on the other hand , as nonthermal x - ray emission is only detected in several snrs within the entire galaxy , it would be highly unlikely that there are so many nonthermal snrs around the gc region . for this reason , we would argue that most of these x - ray filaments are not snrs . observed properties of these x - ray filaments may be more consistent with those of pwne . typical features of a pwn are : non - thermal x - ray spectrum , with photon index of @xmath88 and a x - ray luminosity @xmath89 range from @xmath90 to @xmath91 in @xmath92 kev band ( e.g. , gaensler & slane 2006 ; kaspi , roberts & harding , 2006 ) . the @xmath86 and @xmath89 of these 10 filaments are consistent with the values of a pwn . the existence of point - like x - ray sources as indicated by the image study also tends to favor a pwn scenario . we may estimate the ages of the putative pulsars with the x - ray luminosities of these x - ray filaments . li et al . ( 2007 ) studied statistically the nonthermal x - ray emission from young rotation powered pulsars and pwne . they noted that there exists a correlation between the pulsar age @xmath93 and the @xmath2 kev pwn luminosity @xmath94 , which can be expressed as @xmath95 . the x - ray luminosities of these 10 filaments are in the range of 0.2 - 2.2@xmath96 erg s@xmath6 . using this empirical formula , ages of these putative pulsars are possibly between @xmath97 to @xmath98 yr . however , given the dispersion about the above empirical relationship ( li et al . 2007 ) , the estimate may be uncertain probably by a factor of 10 . since the ages of pulsars with bright pwne are usually younger than a few tens of thousand years , one may doubt if a pulsar at the age of several @xmath99 years can produce a detectable x - ray nebula . however , the pwn of a relatively old pulsar can be enhanced in surface brightness and thus become detectable if the pulsar wind materials are confined to one direction . psr b0355 + 54 is @xmath100 yr old . it converts @xmath15% of its spin - down luminosity to the cometary - like x - ray nebula ( e.g. , tepedelenliolu & gelman 2007 ) . the old pulsar psr b1929 + 10 ( @xmath101 yr ) also converts @xmath102 of its spin - down luminosity @xmath103 erg s@xmath6 into the emission of the cometary nebula ( e.g. , becker et al . 2006 ) . the x - ray filaments identified in the gc region are similar to these two systems and thus probably powered by pulsars . now we discuss whether the number of x - ray filaments , if identified with pwne , would be consistent with the estimated star formation rate in the gc region . according to figer et al . ( 2004 ) , the star formation rate at the gc is about @xmath104 which is some 250 times higher that the mean star formation rate in the galaxy . in the field of view of our fig . 1 , we take a radius of about @xmath105 pc and estimate the star formation rate to be @xmath106 . if the mean mass of a star is 10@xmath107 , the frequency of supernova explosions would be @xmath108 per year , leading to about 40 pulsars in the field of fig . 1 younger than @xmath109 yr as estimated above . this number is roughly consistent with the 15 candidate pwne identified in the field . filament f10 bears certain unique features to be noted here . first , it has the longest linear structure with the entire image slightly bent towards the northeast , more or less like an arc . second , it is the farthest away from and thus has much less contamination from the strong diffuse x - ray emission of sgr a. third , there is an obvious 20 cm radio ntf coincident spatially with x - ray filament f10 . the spectral indices for different regions along filament f10 show evidence of spectral steepening from the head " to tail " ( see table 1 ) . when @xmath53 is fixed at the best fit value @xmath110 , the @xmath86 values for the head " , middle " , and tail " regions are 1.1(1.0 , 1.3 ) , 1.5(1.4 , 1.6 ) , and 1.8(1.7 , 1.9 ) , respectively . this might suggest an energetic particle flow direction from the southeast ( head ) to the northwest ( tail ) . a pulsar moving through the magnetized interstellar medium seems to give a plausible explanation of this scenario . indeed , the morphology of f10 does imply a point source in the head " region . the corresponding point spread function ( psf ) at g0.029 - 0.06 is an ellipse with a size of @xmath111 . for an updated x - ray versus spin - down luminosity correlation of rotation powered pulsars , a modified empirical relation is given by equation ( 3 ) of possenti et al . ( 2002 ) , namely , @xmath112 where @xmath113 is the x - ray luminosity in @xmath2kev energy band ; using this empirical relation , we would have a @xmath114 . since psrs j1747 - 2958 and b1929 + 10 convert about 2.5% and 2.1@xmath115 of their spin - down powers to their cometary x - ray nebulae ( e.g. , gaensler et al . 2004 ; becker et al . 2006 ) , the ratio @xmath89/@xmath116 of f10 ( @xmath117 ) indicates that the above estimate for @xmath116 is reasonable . the arc - like x - ray morphology of f10 and its coincidence with a radio ntf might be a good indicator of its interaction with the interstellar magnetic field environment of the gc region ( lang et al . 1999 ; wang et al . similar to the discussion about g0.13 - 0.11 by wang et al . ( 2002b ) , we may estimate the magnetic field strength @xmath118 in the current context . first , the lifetime @xmath93 of synchrotron x - ray emitting particles is given by @xmath119 , where @xmath120 is the x - ray photon energy in unit of kev ( a value of 4 kev is adopted here ) and @xmath121 is the magnetic field strength in the filament volume in units of mg . the simulations of bucciantini et al . ( 2005 ) show that the average flow speed in the tail is about 0.8 - 0.9 @xmath122 . for a sustained x - ray linear structure , we estimate by requiring @xmath123 . adopting a characteristic angular length @xmath124 of @xmath125 ( @xmath126 ) , we thus infer a magnetic field strength @xmath127 mg , similar to those in the bright radio ntfs ( e.g , yusef - zadeh & morris 1987 ; lang et al . 1999 ) . we try to outline a few plausible scenarios in the present context and discuss relevant aspects qualitatively . magnetized neutron stars move with peculiar speeds in the range of a few tens of kilometers per second ( a mean space velocities of @xmath128 for young pulsars ; hobbs et al . 2005 ; faucher - gigure & kaspi 2006 ) to well over one thousand kilometers per second ( @xmath129 ) and the surrounding ism is generally magnetized . generally speaking , a typical peculiar velocity of a neutron star is supersonic and super - alfvnic in a magnetized ism . neutron stars or pulsars have different ranges of surface magnetic field strengths : @xmath130 g for millisecond pulsars in binaries , @xmath131 g for a wide range of pulsars , and @xmath132 g inferred for several magnetars . several situations may happen . ( 1 ) if a pulsar does not involve an active pulsar wind , its peculiar motion through the surrounding magnetized ism would sustain an mhd bow shock by its magnetosphere as well as a magnetotail . the faster the pulsar moves , the more linear the system would appear . this is basically like a bullet moving through an air supersonically and generating a mach cone or wake . relativistic electrons can be produced at the mhd bow shock and synchrotron emissions can be generated and sustained at the same time . ( 2 ) in a binary system , the fast wind ( say , with a speed higher than 1000 km s@xmath6 ) from a companion star can blow towards a spinning magnetized pulsar in orbital motion . here , the situations of a companion fast wind blowing across a magnetized pulsar and a pulsar moving through the ism with a high speed are more or less equivalent . again , an mhd bow shock and a magnetotail can form in association with the pulsar system . the stronger the companion wind and the faster the pulsar moves , the more linear the pulsar system would appear . relativistic electron and/or positrons can be generated and sustained to power synchrotron emissions in the bow shock draped around the pulsar magnetosphere . for such a system , one might be able to detect the presence of the companion by various independent means . ( 3 ) for a pulsar emitting an active pulsar wind and with misaligned magnetic and spin axes , spiral forward and reverse shock pairs can be generated in the relativistic pulsar wind as a result of inhomogeneous wind and eventually the pulsar wind is stopped by the ism through a mhd termination shock ( e.g. , lou 1993 , 1996 , 1998 ) . ( 4 ) case ( 3 ) can also happen for a pulsar moving with a high peculiar velocity through a magnetized ism . ( 5 ) case ( 3 ) can also happen for a pulsar in binary orbital motion with the companion blows a powerful wind with a speed higher than 1000 km s@xmath6 . in both cases of ( 2 ) and ( 5 ) , the center of mass of the binary system may also move with a high speed through the ism . one can further speculate several possible combinations along this line of reasoning ( e.g. , chevalier 2000 ; toropina et al . 2001 ; romanova , chulsky & lovelace 2005 ) . there are two clumps ( referred to as the east and west clumps hereafter ) of diffuse x - ray emission surrounding filament f10 . although the west clump is also elongated , it is not called a filament because it contains many substructures . to see if these clumps are physically related to f10 , we extract their energy spectra separately ( see the middle and bottom panels of fig . the fitted parameters are listed in table 3 . the much higher absorbing column densities of the two clumps indicate strongly that they are located farther away from us than f10 is , while these two clumps themselves are almost at the same distance ( see table 3 ) . moreover , their x - ray emissions are very likely powered by the same mechanism , as hinted by the characters of their spectra , which can be fitted well with an absorbed power - law model plus a 6.4 kev emission line . the total emission comes mostly from the photon energy @xmath133 kev band , with a strong 6.4 kev neutral fe k line . in contrast , the fe line feature is not present in the spectrum of f10 . a possible explanation for this non - thermal , apparently broadened iron line emission at 6.4 kev is the collisions of low - energy cosmic - ray electrons with irons in molecular clouds ( e.g. , valinia et al . 2000 ) or by the radiative illumination from the gc massive black hole that was suggested to be very bright in the past ( e.g. , koyama et al . 1989 ) . in conclusion , f10 and the surrounding clumps do not seem to interact directly . the oritentation of the x - ray filaments provides an opportunity to probe the physics conditions of the gc region . as discussed in section 4.1 , the cometary shapes of the x - ray filaments imply that the pulsar wind particles are confined to one direction by the ambient materials . mechanism shaping the filamentary structure could be ordered magnetic fields ( e.g. , yusef - zadeh & morris 1987 ; lang , morris , & echevarria 1999 ) and/or high relative velocity between the pulsars and the surrounding gas ( e.g. , wang et al . 1993 ; shore & larosa 1999 ) . the magnetic field could be the product of the gas motion , and the magnetic field could also control the motion of the gas ( e.g. , heyvaerts et al . 1988 ; chevalier 1992 ) . radio ntfs suggests that the magnetic field is poloidal at large - scale ( e.g. , yusef - zadeh & morris 1987 ; lang , morris , & echevarria 1999 ) with some more complex smaller structures around the gc region ( e.g. , nord et al . 2004 ) . by looking at the x - ray images shown by figures 1 , 2 and 3 , there seems to be a tendency that the pwn tails point away from the gc , indicating that the pulsar wind particles are blown outward . this might imply the presence of a galactic wind of hot plasma blowing away from the center , given the high star formation rate ( and so plenty of hot gas ) in this region . pulsars may have typical peculiar velocities of @xmath134 ( e.g. , hobbs et al . 2005 ; faucher - gigure & kaspi 2006 ) and we would expect them to move in random directions . the tendency for the structures of ten pwns to orient away from the center seems to suggest that the galactic wind has a speed comparable to or greater than @xmath135 . in the above scenario , the particle flow direction of the x - ray filament f3 should be from the southwest ( closer to sgr a * ) to the northeast . this suggests that the pulsar , the origin site of the particles , is actually in the `` tail '' region defined in fig 2 . then the evident spectral steepening from the southwest to the northeast ( see section 3 ) can be naturally expained . therefore , the spectral evolution along f3 also supports the existence of a radial high velocity wind in the gc region . while the x - ray filaments may not completely overlap with their radio ntfs , their overall orientations are similar . this is supported by the four x - ray ntfs ( including filament f10 in our analysis ) that have radio counterparts : g359.54 + 0.18 overlays exactly on the northern part of the two radio filaments ( e.g. , yusef et al . 2005 ; lu et al . 2003 ) ; g359.89 - 0.08 and its radio counterpart sgra - e overlaps partly and extends in the same direction , with a centroid offset of @xmath136 ( e.g. , yusef et al . 2005 ; lu et al . 2003 ) ; g359.90 - 0.06 ( sgra - f ) ( e.g. , yusef et al . 2005 ) and g0.029 - 0.06 ( f10 ) ( see our subsection 3.10 ) also show similar spatial property . generally speaking , the x - ray ntfs tend to be shorter than the radio ntfs . this centroid offset and smaller extent of x - ray filaments could be both explained by the much shorter synchrotron cooling lifetime in x - ray than in radio ( e.g. , ginzburg & syrovatskii 1965 ) . to summarize , the most important properties of the gc x - ray filaments in this _ chandra _ data analysis are their non - thermal spectra , and of course , the apparent thread - like linear morphology with point - like sources residing at the head " . the scenario of pulsars ploughing through the interstellar medium and high energy particles from the pwn interacting with the surrounding magnetic field seems to be a plausible interpretation . however , this relationship with pulsars can only be confirmed by the direct detection of periodic flux variations from the putative point - sources . future faster timing , higher sensitivity and spatial resolution of x - ray instruments , together with radio and near - ir follow - up observations , should help to reveal the true identity of these x - ray filaments . meanwhile , these filaments could also provide some clues to the searching for gc pulsars , as well as the study of small - scale magnetic field structures and gas dynamics around the gc . by our suggestion , one might get the impression that there are many pulsar bow shocks within @xmath1pc of the galactic center . actually in this scenario , there should be even more due to the unavoidable projection effect ; some x - ray bright sources may be heading directly towards or away from us with their ` tails ' behind or in front superposed with their ` heads ' . we now only catch those suspects readily identified by their linear or arc - like morphologies . by all astronomical standards , our galactic center is not very active at the present epoch even though there is a black hole with a mass of @xmath14 lurking at the center near sgr a@xmath137 ( e.g. , shen et al . 2005 ) . nevertheless , the neighborhood of our galactic center might have been active in the past with starbursts and magnetized galactic winds and so forth . while interacting with winds or flows of the ism , the compact remnants of these massive stars ( produced during the starburst phase ) with a distribution of peculiar speeds might then have a higher concentration around the galactic center . a typical peculiar speed of a few hundred kilometers per second would be supersonic and super - alfvnic in a magnetized ism . with the projection effect taken into account , the physical linear extension of such a pwn should be jointly determined by the local ism flow direction and the direction of the remnant peculiar velocity . the authors thank the referee and editor e. d. feigelson for constructive comments leading to a significant improvement of the manuscript . we thank q. d. wang for very helpful discussion , thank c.c . lang for sending us the 20 cm radio image before publication . this research has been supported in part by the special funds for major state basic science research projects of china , by the nsfc grant 10573017 at the institute of high energy physics , by nsfc grants 10373009 and 10533020 at the tsinghua university , by the tsinghua center for astrophysics , and by the srfdp 20050003088 and the yangtze endowment from the ministry of education at tsinghua university . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , funded by the national aeronautics and space administration and the national science foundation . baganoff , f. k. , maeda , y. , morris , m. , bautz , m. w. , brandt , w. n. , cui , w. , doty , j. p. , feigelson , e. d. , garmire , g. p. , pravdo , s. h. , ricker , g. r. , & townsley , l. k. 2003 , apj , 591 , 891 beck r. , hoernes p. , 1996 , nature , 379 , 47 becker , w. , kramer , m. , jessner , a. , taam , r. e. , jia , j. j. , cheng , k. s. , mignani , r. , pellizzoni , a. , de luca , a. , slowikowska , a. , & caraveo , p. a. 2006 , apj , 645 , 1421 becklin , e. e. , matthews , k. , neugebauer , g. , & willner , s. p. 1978 , apj , 219 , 121 bian , f .- y . & lou , y .- q . 2005 , mnras , 363 , 1315 boldyrev , s. , & yusef - zadeh , f. 2006 , apj , 637 , l101 bucciantini , n. 2002a , a&a , 387 , 1066 bucciantini , n. 2002b , a&a , 393 , 629 bucciantini , n. , amato , e. , & del zanna , l. 2005 , a&a , 434 , 189 cheng , k. s. , taam , r. e. , & wang , w. 2004 , apj , 617 , 480 cheng , k. s. , taam , r. e. , & wang , w. 2006 , apj , 641 , 427 chevalier , r.a . 1992 , apj , 397 , l39 chevalier , r. a. 2000 , apj , 539 , l45 cordes , j. m. , & chernoff , d. f. 1998 , apj , 505 , 315 cordes , j. m. , & lazio , t. j. w. 1997 , apj , 475 , 557 cowin , b. j. , & morris , m. 2006 , baas , vol . 38 , p.1147 faucher - gigure , c .- a . , & kaspi , v. m. 2006 , apj , 643 , 332 ferriere , k. m. 2001 , reviews of modern physics , 73 , 1031 ferriere , k. m. 2007 , eas publications series , 23 , 3 figer , d. f. , kim , s. s. , morris , m. , serabyn , e. , rich , r. m. , & mclean , i. s. 1999 , apj , 525 , 750 figer , d.f . , majarro , f. , gilmore , d. , morris , m. , kim , s. s. , serabyn , e. , mclean , i. s. , gilbert , a. m. , graham , j. r. , larkin , j. e. , levenson , n. a. , & teplitz , h. i. 2002 , apj , 581 , 258 figer , d.f . , rich , r. m. , kim , s.s . , morris , m. , & serabyn , e. 2004 , apj , 601 , 319 gaensler , b. m. , van der swaluw , e. , camilo , f. , kaspi , v. m. , baganoff , f. k. , yusef - zadeh , f. , & manchester , r. n. 2004 , , 616 , 383 gaensler , b. m. , slane , p. o. , 2006 , ara&a , 44 , 17 ginzburg , v. l. , & syrovatskii , s. i. 1965 , ara&a , 3 , 297 goldreich , p. , & julian , w. h. 1970 , 160 , 971 gotthelf , e. v. , & olbert , c. m. 2002 , in asp conf . ser.271 , neutron stars in supernova remnants , ed . p. o. slane & b. m. gaensler ( san francisco : asp ) , 171 gotthelf , e. v. 2003 , apj , 591 , 361 gotthelf , e. v. 2004 , astro - ph/0610376 han , j. l. 2007 , eprint / arxiv:0705.4175 heyvaerts , j. , norman , c. , & pudritz , r.e . 1988 , apj , 330 , 718 ho , p. t. p. , jackson , j. m. , barrett , a. h. , & armstrong , j. t. 1985 , apj , 288 , 575 hobbs , g. , lorimer , d. r. , lyne , a. g. , & kramer , m. , 2005 , mnras , 360 , 974 johnston , s. , kramer , m. , lorimer , d. r. , lyne a. g. , mclaughlin , m. , klein , b. , manchester , r. n. 2006 , mnras , 373 , l6 kaspi , v. m. , roberts , m. s. e. , harding , a. k. 2006 , in : compact stellar x - ray sources . edited by walter lewin & michiel van der klis . cambridge astrophysics series , no . cambridge , uk , p. 279 - 339 kawai , n. , tamura , k. , & shibata , s. 1998 , in neutron stars and pulsars thirty years after the discovery , ed . n. shibazaki , n. kawai , s. shibata & t. kafune ( tokyo : universal academy ) , 449 kennel , c. f. , coroniti , f. v. 1984a , apj , 283 , 694 kennel , c. f. , coroniti , f. v. 1984b , apj , 283 , 710 koyama , k. , awaki , h. , kunieda , h. , takano , s. , tawara , y. , yamauchi , s. , hatsukade , i. , & nagase , f. 1989 , nature , 339 , 603 lang , k. r. 1999 , astrophysical formulae , volume ii , ( springer ) , 87 lang , c. c. , morris , m. , & echevarria , l. 1999 , apj , 526 , 727 lang , c. c. et al . , 2007 , in preparation larosa , t. n. , brogan , c. l. , shore , s. n. , lazio , t. j. , kassim , n. e. , nord , m. e. 2005 , apj , 626 , 23 larosa , t. n. , kassim , n. e. , lazio , t. j. w. , & hyman , s. d. 2000 , aj , 119 , 207 li , x. h. , & lu , f. j. 2007 , , in preparation lou , y .- q . 1987 , apj , 322 , 862 lou , y .- q . 1993 , apj , 414 , 656 lou , y .- q . 1996 , mnras , 279 , 129 lou , y .- q . 1998 , mnras , 294 , 443 lou , y .- q . 2001 , apj , 563 , l147 lou , y .- q . , & bai x. n. 2006 , mnras , 372 , 81 lou , y .- q . , & wang , w. g. 2007 , mnras , 378 . l54 ( astro - ph/0704.0223 ) lu , f. j. , wang , q. d. , aschenbach , b. , durouchoux , p. , & song , l. m. 2002 , apj , 568 , l49 lu , f. j. , wang , q. d. , & lang , c.c . 2003 , aj , 126 , 319 manchester , r. n. 2003 , www.atnf.csiro.au/research/pulsar/catalogue/ michel , f. c. 1969 , apj , 158 , 727 morris , m. , & serabyn , e. 1996 , , 34 , 645 muno , m. p. , bauer , f. e. , bandyopadhyay , & wang , q. d. 2006 , , submitted , astro - ph/0601627 muno , m.p . , baganoff , f.k . , brandt , w.n . , morris , m.r . , & starck , j.l . 2007 apj , submitted ( astro - ph/0707.1907 ) nord , m. e. , lazio , t. j. w. , kassim , n. e. , hyman , s. d. , larosa , t. n. , brogan , c. l. , & duric , n. 2004 , aj , 128 , 1646 park , s. , et al . 2005 , apj , 631 , 964 possenti , a. , cerutti , r. , colpi , m. , & mereghetti , s. 2002 , a&a , 387 , 993 romanova , m. m. , chulsky , g. a. , lovelace , r. v. e. 2005 , apj , 630 , 1020 sakano , m. , warwick , r. s. , decourchelle , a. , & predehl , p. 2003 , mnras , 340 , 747 seward , f. d. , & wang , z. r. 1988 , apj , 332 , 199 shen , z .- q . , lo , k. y. , liang , m .- c . , ho , paul t. p. , & zhao , j .- h . 2005 , nature , 438 , 62 shore , s.n . , & larosa , t.n . 1999 , apj , 521 , 587 spitkovsky , a. 2006 , apj , 648 , 51 tepedelenliolu , e. & gelman , h. 2007 , apj , 658 , 1183 toropina , o. d. , romanova , m. m. , toropin , yu . m. , lovelace , r. v. e. 2001 , apj , 561 , 964 valinia , a. , tatischeff , v. , arnaud , k. , ebisawa , k. , & ramaty , r. 2000 , apj , 543 , 733 wang , q.d . , li , z.y . , & begelman , m.c . 1993 , nature , 364 , 127 wang , q. d. , gotthelf , e. v. , & lang , c. c. 2002a , , 415 , 148 wang , q. d. , lu , f. j. , & lang , c. c. 2002b , apj , 581 , 1148 wang , q. d. , lu , f. j. , & gotthelf , e. v. 2006 , , 367 , 937 wang , w. g. , & lou , y .- q . 2007 , apss , in press wang , x. f. , yang , y. b. , zhang , t. m. , ma , j. , zhou , x. , li , w. d. , lou , y .- q . , & li , z. w. , 2005 , apj , 626 , l89 wielebinski , r. , 2005 , cosmic magnetic fields . richard wielebinski and rainer beck . lecture notes in physics volume 664 , p89 willingale , r. , aschenbach , b. , griffiths , r. g. , sembay , s. , warwick , r. s. , becker , w. , abbey , a. f. , & bonnet - bidaud , j .- 2001 , a&a , 365 , l212 wu , y. , & lou , y .- q . 2006 , mnras , 372 , 992 yu , c. , & lou , y .- q . 2005 , mnras , 364 , 1168 yu , c. , lou , y .- q . , bian , f .- y . , & wu , y. 2006 , mnras , 370 , 121 yusef - zadeh , f. , morris , m. , & chance , d. 1984 , , 310 , 557 yusef - zadeh , f. , wardle , m. , muno , m. , law , c. , & pound , m. 2005 , advances in space research , 35 , 107 zweibel , e. g. , & heiles , c. 1997 , nature , 385 , 131 _ left : _ the 0.5 - 8.0 kev x - ray images of f1 to f6 with contours overlaid . these images are smoothed adaptively with a gaussian to achieve a signal - to - noise ratio of 10 . for the same data , these contours have the same stepsize of 4 counts arcsec@xmath138 but different starting values , i.e. , 20 counts arcsec@xmath138 for f1 to f4 and 16 counts arcsec@xmath138 for f5 and f6 . right : _ the count maps of the same regions as in the left column . the color bars are in logarithmic scale . for f1 , f2 , f5 and f6 , the lower and upper limits are 4 counts arcsec@xmath138 and 120 counts arcsec@xmath138 , while for f3 and f4 the two limits are 12 counts arcsec@xmath138 and 120 counts arcsec@xmath138 , respectively . rays are displayed needs to be given in the caption . + whole & 1.0(0.2 , 1.5 ) & 0.9(-0.3 , 1.6 ) & 8.0 ( 7.3 , 10.2 ) & 0.92 ( 0.84 , 1.15 ) & 29.8/30 + joint - fit & set @xmath53 the same + head(n ) & 0.8(0.3 , 1.7 ) & 0.8(-0.2 , 2.0 ) & & & 26.3/28 + tail(s ) & 0.8(0.3 , 1.7 ) & 1.6(0.1 , 3.3 ) & & & 26.3/28 + + whole & 0.9(0.2 , 2.1 ) & 0.8(0.3 , 2.0 ) & 4.7 ( 4.2 , 5.3)&0.53 ( 0.44 , 0.97 ) & 13.2/20 + joint - fit & set @xmath53 the same + head(ne ) & 0.9(0 , 2.7 ) & 0.8(-0.8 , 2.9 ) & & & 17.1/14 + tail(sw ) & 0.9(0 , 2.7 ) & 1.2(-0.6 , 3.4 ) & & & 17.1/14 + + whole & 1.4(1.1 , 1.7 ) & 2.3(1.8 , 2.8 ) & 11.9 ( 11.4 , 12.5 ) & 2.2 ( 1.9 , 2.78 ) & 72.2/71 + head & 1.9(1.3 , 2.6 ) & 3.1(2.2 , 4.3 ) & & & 31/35 + middle & 1.8(0.6 , 4.2 ) & 2.2(0.6 , 5.3 ) & & & 3.6/10 + tail & 0.7(0.3 , 1.1 ) & 1.4(0.9 , 1.8 ) & & & 23.5/24 + joint - fit & set @xmath53 the same + head & 1.2(0.9 , 1.5 ) & 2.0(1.5 , 2.7 ) & & & + middle & 1.2(0.9 , 1.5 ) & 1.5(0.8 , 2.2 ) & & & + tail & 1.2(0.9 , 1.5 ) & 2.4(1.7 , 2.8 ) & & & + + whole & 0.6(0.1 , 1.2 ) & 1.9(1.2 , 3.0 ) & 4.0 ( 3.4 , 5.2 ) & 0.47 ( 0.41 , 0.75 ) & 13.8/19 + + whole & 0.8(0.4 , 1.5 ) & 1.4(0.9 , 2.5 ) & 4.3 ( 3.6 , 4.8 ) & 0.52 ( 0.43 , 0.79 ) & 19.5/24 + head(se ) & 0.5(0 . , 1.4 ) & 0.9(-0.4 , 2.6 ) & & & 9.9/10 + tail(nw ) & 0.8(0 . , 5.9 ) & 0.8(-1.2 , 6.8 ) & & & 2.7/2 + joint fit : & set @xmath53 the same + head(se ) & 0.6(0.2 , 1.5 ) & 1.1(-0.3 , 2.1 ) & & & + tail(nw ) & 0.6(0.2 , 1.5 ) & 0.9(0.2 , 1.5 ) & & & + + whole & 0.9(0.4 , 1.4 ) & 1.4(0.9 , 2.3 ) & 6.6 ( 5.7 , 7.3 ) & 0.84(0.69 , 1.07 ) & 35/40 + head(sw ) & 0.7(0 . , 2.5 ) & 1.1(-0.2 , 3.8 ) & & & 18/13 + tail(ne ) & 0.7(0.1 , 1.5 ) & 1.3(0.6 , 2.3 ) & & & 25.4/23 + joint fit : & set @xmath53 the same + head(sw ) & 0.8(0.3 , 1.5 ) & 1.2(0.3 , 2.3 ) & & + tail(ne ) & 0.8(0.3 , 1.5 ) & 1.6(0.7 , 2.0 ) & & + + whole & 0.(0 . , 3.6 ) & 0.9(-0.2 , 4.4 ) & 2.8 ( 1.6 , 3.0 ) & 0.21 ( 0.19 , 1.96 ) & 5.1/5.0 + + whole & 1.5(0.6 , 3.0 ) & 2.3(0.8 , 4.6 ) & 3.0 ( 2.3 , 3.5 ) & 0.6 ( 0.4 , 1.7 ) & 12.4/12 + + whole & 0.7(0 . , 1.5 ) & 0.6(-0.9 , 1.9 ) & 4.7 ( 4.1 , 5.1 ) & 0.41 ( 0.40 , 0.46 ) & 5.8/13 + + + whole & 0.5(0.4 , 0.7 ) & 1.2(0.9 , 1.5 ) & 10.4 ( 9.9 , 11.1 ) & 1.06 ( 1.02 , 1.16 ) & 48.9/55 + head & 1.0(0.6 , 1.6)&1.8(1.3 , 2.8 ) & & & 27/21 + middle & 0.4(0 . , 1.1 ) & 1.0(0 . , 2.0 ) & & & 8.7/13 + tail & 0.3(0 . , 0.9 ) & 1.0(0.5 , 1.5 ) & & & 22/18 + + head & 0.7(0.4 , 1.0)&1.1(0.6 , 1.8 ) & & & + middle & 0.7(0.4 , 1.0)&1.5(0.8 , 2.2 ) & & & + tail & 0.7(0.4 , 1.0)&1.8(1.4 , 2.2 ) & & & + + head & 0.7 & 1.1(1.0 , 1.3 ) & & & + middle & 0.7 & 1.5(1.4 , 1.6 ) & & & + tail & 0.7 & 1.8(1.7 , 1.9 ) & & & + llccccccl 1&@xmath139&g359.95 - 0.04&2@xmath1408&@xmath141 & 1.9 ( 1.8 , 2.1)&1.4 ( 1.3 , 1.5)&10 . & no + 2&f1&g359.936 - 0.038&5@xmath14017&@xmath142 & 0.9 ( -0.3 , 1.6)&1.0 ( 0.2 , 1.5)&@xmath143&no + 3&f2&g359.934 - 0.037&3@xmath1409&48@xmath29 & 0.8 ( 0.3 , 2.0)&0.9 ( 0.2 , 2.1 ) & @xmath144&no + 4&f3&g359.965 - 0.053&4@xmath14012&@xmath48 & 2.3 ( 1.8 , 2.8 ) & 1.4 ( 1.1 , 1.7)&@xmath145&no + 5&f4&g359.964 - 0.056&4@xmath14012&@xmath48 & 1.9 ( 1.2 , 3.0)&0.6 ( 0.1 , 1.2)&@xmath144&no + 6&f5&g359.959 - 0.028&4@xmath14010&@xmath146 & 1.4 ( 0.9 , 2.5)&0.8 ( 0.4 , 1.5)&@xmath144&no + 7&f6&g359.969 - 0.038&5@xmath14016&@xmath147 & 1.4 ( 0.9 , 2.3 ) & 0.9 ( 0.4 , 1.4)&@xmath148&no + 8&f7&g359.920 - 0.029&4@xmath14015&@xmath48 & 0.9 ( -0.2 , 4.4 ) & 0.0 ( 0.0 , 3.6)&@xmath149&no + 9&@xmath150cannonball&g359.983 - 0.046&&@xmath151&1.6 ( 1.2 , 2.2)&1.6 ( 0 , 4.8)&3.1&no + 10&f8&g359.970 - 0.009&4@xmath14010&@xmath152 & 2.3 ( 0.8 , 4.6 ) & 1.5 ( 0.6 , 3.0)&@xmath153&no + 11&f9&g359.974 - 0.00&4@xmath1407&@xmath154 & 0.6 ( -0.9 , 1.9)&0.7 ( 0.0 , 1.5)&0.4&no + 12&@xmath155sgra - e&g359.89 - 0.08&8@xmath14030 & @xmath156&2.0 ( 1.5 , 2.5)&3.7 ( 2.5 , 5.1)&@xmath157&yes ( radio ) + 13&f10&g0.029 - 0.06&6@xmath14047&@xmath158 & 1.2 ( 0.9 , 1.5 ) & 0.5 ( 0.4 , 0.7)&1.1&yes ( radio ) + 14&@xmath159sgra - f&g359.90 - 0.06&3@xmath1406&@xmath160 & ( 0.8 , 8.2)&(1.7 , 7.9)&@xmath161&yes ( radio ) + 15&@xmath162&g0.13 - 0.11&2@xmath14040&@xmath163 & 1.8 ( 1.4 , 2.5)&0.6 ( 0.4 , 0.9 ) & 3.2&yes ( radio ) + lccccccc e. c. & 1.4 ( 1.2 , 1.6 ) & 1.0 ( 0.5 , 1.2)&6.1 ( 5.9 , 6.6 ) & 8.1 ( 7.5 , 8.4 ) & 6.42 ( 6.41 , 6.43 ) & 0.75 ( 0.68 , 0.83 ) & 112/135 + w. c. & 2.3 ( 1.6 , 2.9 ) & 1.0 ( 0.4 , 1.6 ) & 5.4 ( 5.0 , 5.7 ) & 8.7 ( 7.8 , 10.9 ) & 6.39 ( 6.38 , 6.40 ) & 1.07 ( 0.94 , 1.23 ) & 126/114 +

radial velocity ( rv ) monitoring of solar - type visual binaries has been conducted at the ctio / smarts 1.5-m telescope to study short - period systems . data reduction is described , mean and individual rvs of 163 observed objects are given . new spectroscopic binaries are discovered or suspected in 17 objects , for some of them orbital periods could be determined . subsystems are efficiently detected even in a single observation by double lines and/or by the rv difference between the components of visual binaries . the potential of this detection technique is quantified by simulation and used for statistical assessment of 96 wide binaries within 67pc . it is found that 43 binaries contain at least one subsystem and the occurrence of subsystems is equally probable in either primary or secondary components . the frequency of subsystems and their periods match the simple prescription proposed by the author ( 2014 , aj , 147 , 87 ) . the remaining 53 simple wide binaries with a median projected separation of 1300au have the distribution of the rv difference between their components that is not compatible with the thermal eccentricity distribution but rather matches the uniform eccentricity distribution .

1511.01335

binary and multiple systems are ubiquitous products of star formation @xcite . the gas condensing into stars contains excessive angular momentum that is transported outwards during collapse and accretion and stored partially in the orbital momentum of binaries . while the basic physics is known , no current theory is able to predict the multiplicity fraction and the distributions of separations , mass ratios , or eccentricities , although large - scale numerical simulations show promising agreement with observations @xcite . formation of binary stars is related to the stellar mass function and sets the scene for subsequent evolution ( mass transfer , blue stragglers , x - ray binaries , and type i supernovae ) . on the other hand , the gas left around newly born stars as a material for planet formation is inherited from the epoch of mass accretion and is related to multiplicity . the common origin of stars , planets , and binaries should therefore be studied jointly . solid observational data on binary statistics are difficult to get because of various selection effects , and this is exacerbated for stellar hierarchies containing three or more stars . yet , hierarchical multiple systems give additional insights into star formation from their period ratios , relative orbit orientation , and structure . this work complements the study of hierarchical multiplicity of solar - type stars in the 67-pc volume ( * ? ? ? * ; * ? ? ? * hereafter fg67a and fg67b ) . the large size of the fg-67 sample defined in fg67a ( 4867 stars ) and modest and well - controlled observational selection allowed reliable estimates of the fraction of hierarchical systems , 13% , and of their statistics . one unexpected result was the large fraction of subsystems in the secondary components of wide binaries and the emerging correlation between subsystems , so that 4% of targets turned out to be 2 + 2 quadruples ( two close binaries on a wide orbit around each other ) . however , this result relied on extrapolation from the small number of well - studied secondary components . while the main targets of the fg-67 sample were the subject of spectroscopic and imaging observational campaigns , the secondaries were generally neglected . recent imaging studies of secondary components @xcite have largely remedied the situation , but _ close _ secondary subsystems accessible only to spectroscopy still remain to be discovered . exploring these close subsystems is the main motivation of the present work . i report here rv monitoring of solar - type stars in the fg-67 sample that are either known to be spectroscopic binaries but lack the orbit or belong to wide visual binaries . most targets have southern declinations . the main goal is characterization of subsystems in the secondary components and determination of their periods , complementing previous efforts in this respect . extensive previous rv data were furnished by the geneva - copenhagen survey , gcs @xcite , which covered mostly the main targets . as a continuation of this campaign , @xcite monitored the rvs of wide pairs , discovering several subsystems . @xcite reported rv monitoring of both wide and close ( unresolved on the slit ) visual binaries , mostly on the northern sky . a similar , but much smaller survey of southern multiples is published by @xcite . @xcite measured precise rvs of wide binaries in search of stellar or planetary subsystems . these survey - type works are complemented by studies of individual objects published by various authors ( e.g. * ? ? ? although this paper is devoted to wide binaries , rvs of all observed stars are given . presentation of newly determined orbits is deferred to a future paper , as their lengthy discussion would distract from the main topic , while binaries needing more data are still being observed . the instruments , data reduction , and raw results are covered in section [ sec : obs ] . then in section [ sec : simul ] i evaluate through simulation the proposed method of detecting subsystems from a single observation . section [ sec : widebin ] reports on subsystems in a volume - limited sample of wide binaries . in section [ sec : ecc ] the rv difference in wide binaries without subsystems is analyzed to constrain their eccentricities . the work is summarized in section [ sec : sum ] . the observations reported in this paper were obtained with the 1.5-m telescope located at the cerro tololo interamerican observatory in chile and operated by the smarts consortium . the observing time was allocated through noao ( programs 10b-0022 , 14b-0009 , and 15a-0055 ) . the observations were made by the telescope operators in service mode . the goal of the 2010 campaign was to determine unknown spectroscopic orbits of close solar - type binaries and to observe components of wide binaries that lacked rv coverage . at least three observations of each target were planned in the 20 allocated nights . however , less time was actually scheduled , so most targets were observed just once , and only a few were pointed several times . this data set is therefore too small for addressing the goals of the original proposal . nevertheless , one orbit was determined and several subsystems were discovered . in 2010 , the spectra were taken with the fiber echelle ( fech ) instrument . it was the de - commissioned echelle spectrometer from the blanco 4-m telescope moved to the coud room of the 1.5-m telescope and fed by the fiber . in this program the image of the fiber projected into the spectrograph was not masked by the slit and its width defined the spectral resolution of @xmath1 . the detector was a sit ccd of 2014@xmath22048 pixels format with the arcon controller . the calibration spectrum of the thorium - argon lamp fed through the same fiber was taken before or after each target . in 2014 and 2015 , observations were made with the chiron spectrograph @xcite that replaced fech in 2011 . stars brighter than @xmath3 mag were observed in the slicer mode with a spectral resolution of @xmath4 , while for fainter targets the fiber mode with @xmath5 was used . when both components of a wide binary were observed , the same mode was chosen . the thorium - argon calibrations were recorded for each target . the observing campaigns of 2014 and 2015 targeted secondary components of wide binaries in a `` snapshot survey '' designed to detect subsystems from just one observation by rv difference between the components . i selected binaries from the fg-67 sample south of @xmath6 declination with separations no less than 8 to ensure that the components are well resolved on the entrance fiber aperture . only the secondaries with @xmath7 magnitudes between 8 and 11 were chosen to shorten the exposure time , maximizing the sample coverage . secondaries with prior rv coverage were removed from the program ; however , they are included in the statistical analysis in section [ sec : widebin ] . primary components were observed as well if they lacked prior rv data comparable in quality to the gcs ( two or three observations with a time base of at least a year ) . also , monitoring of several known solar - type spectroscopic binaries was continued for computing their orbits . the reduced and wavelength - calibrated spectra delivered by the chiron pipeline were retrieved from the smarts center at the yale university . the availability of this service has greatly enhanced this program by allowing rapid analysis of the rvs and flexible scheduling of new observations when needed . the fech data were reduced by the author using a simplified and adapted version of the chiron pipeline software @xcite . i extracted echelle orders 84 to 123 covering the wavelength range from 4500 to 6850 . at shorter wavelengths , there is little signal in the spectra of red stars and quartz lamp , while longer wavelengths are affected by telluric absorptions and are not used in the cross - correlation . a set of 10 quartz spectra was co - added and used to extract the stellar and arc spectra in a boxcar window of @xmath88 pixels . the position of the spectra on the ccd , defined by the fiber , was stable to within 0.5 pixels . cosmic rays were removed during the order extraction , but the current algorithm did not detect all these events and some spectra do contain a few residual spikes . while the extracted chiron spectra are corrected for small - scale variations of the detector sensitivity using the quartz lamp calibration , this correction was not applied to the fech data because the quartz `` flats '' for fech deviate from unity by only a few percent , less than the typical noise . the fech wavelength calibration typically used 1400 thorium - argon lines with a scatter of 0.1 pixels or 0.36 km s@xmath9 between the measured line positions and their 6-th order polynomial approximation . each science spectrum was paired with the wavelength calibration closest in time ( typically within 3 minutes ) . the spectra were cross - correlated with the digital binary mask based on the solar spectrum stored in the noao archive ( see * ? ? ? * for more details ) . figure [ fig : example ] shows data on a faint target . the extracted fech spectrum has @xmath10250 counts ( the gain is 1.4 el / adu ) . however , the cross - correlation function ( ccf ) is very smooth . its gaussian fit leads to @xmath11 km s@xmath9 . the amplitude of the gaussian is @xmath12 and its dispersion is @xmath13 km s@xmath9 . the rms residual to the fit is 0.7% . the rvs are derived without applying any zero - point correction from rv standards , they are tied directly to the solar spectrum . the ccf @xmath14 is computed over the rv range of @xmath15 km s@xmath9 . a gaussian is fitted to the portion of the ccf within @xmath16 of the minimum . when two or three gaussians are fitted , the data around all the minima are used jointly . after the first iteration , the centers and dispersions are determined , and in the second iteration the fitting area is adjusted accordingly . the ccf model with @xmath17 gaussians is @xmath18 \label{eq : fit}\ ] ] and contains @xmath19 free parameters @xmath20 . it is possible to fix some parameters , but this option , handy for de - blending overlapping ccfs , was not used in the data reported below . formal errors of the fitted parameters do not reflect their real precision , being dominated by systematic , rather than random , deviations of the ccf from its model . i do not provide these errors in the data table . the real rv precision estimated from residuals to orbital fits is about 0.1 km s@xmath9 for both chiron and fech . for example , 4 observations of hip 105585c taken with chiron over a time span of 333days have rms scatter of 40 m s@xmath9 . the bright star hip 5896 with a very wide ccf ( @xmath21 km s@xmath9 ) observed 11 times with fech shows the rms rv scatter of 0.28 km s@xmath9 . for 21 apparently single stars the rvs were measured both here and in the gcs . the average difference chiron@xmath22gcs is 0.47 or 0.36 km s@xmath9 ( with or without rejecting one outlier , respectively ) . the rms scatter of the difference is 0.58 km s@xmath9 . this minor systematics is ignored in the following , being most likely attributable to the coravel instruments used by the gcs . more accurate rvs @xcite match the chiron data better . l rrl | l rrl | l rrl 5896 & 7.79 & 10 & 1f & 37645b & @xmath2216.39 & 1 & 1c & 98274 & @xmath2219.31 & 1 & 1f + 6772a & 49.87 & 1 & 1c & 38908b & 16.98 & 1 & 1f & 99572 & 27.27 & 1 & 1f + 6772b & 50.43 & 1 & 1c & 40452b & 21.19 & 1 & 1f & 99651a & 24.27 & 2 & 1f + 7601 & @xmath2210.61 & 12 & 3co & 41353a & 7.31 & 2 & 1f & 100895 & @xmath2214.97 & 1 & 1f + 8486b & @xmath226.65 & 1 & 1f & 41353b & 28.51 & 3 & 2f * & 100896 & @xmath225.81 & 11 & 2f + 9497 & 17.81 & 1 & 1f & 44579b & 0.60 & 1 & 1f & 101443 & 23.03 & 1 & 1f + 9642 & 44.90 & 10 & 2co & 45734b & @xmath227.78 & 3 & 2f & 101551b & 21.38 & 2 & 1f + 9902a & 11.48 & 2 & 1f & 45734a & 0.14 & 2 & 2f & 101551a & 21.41 & 2 & 1c + 9902b & 11.03 & 2 & 1f & 45838a & 54.95 & 1 & 1f & 102418c & 11.34 & 1 & 1c + 9911b & @xmath2238.69 & 2 & 1f & 46236b & 30.95 & 2 & 1f & 102655b & @xmath222.34 & 1 & 1c + 9911a & @xmath2239.13 & 1 & 1f & 47862b & @xmath2214.46 & 1 & 1f & 102655a & @xmath222.64 & 1 & 1c + 10579b & 44.28 & 1 & 1f & 49030b & 15.83 & 1 & 1f & 102945 & @xmath2222.12 & 1 & 1f + 10579a & 38.94 & 1 & 1f & 59272b & 2.74 & 1 & 1f & 103578 & @xmath2231.01 & 1 & 1f + 10754 & 44.82 & 2 & 1c & 59690b & 25.10 & 2 & 1f & 104687b & @xmath2221.17 & 1 & 1f + 11024a & 45.19 & 1 & 1f & 60353b & 4.35 & 1 & 1f & 104687a & @xmath2220.47 & 1 & 1f + 11024b & 45.19 & 2 & 1f & 60749d & @xmath221.77 & 1 & 1f & 105569 & @xmath221.97 & 1 & 2f + 11324a & 34.35 & 1 & 1f & 61595b & @xmath2215.33 & 1 & 1f & 105569c & 3.79 & 1 & 1f + 11324b & 32.90 & 1 & 1f & 64478a & @xmath2240.23 & 10 & 3fo & 105585c & 3.63 & 4 & 1c + 11417 & 62.92 & 1 & 2f & 64478b & 20.03 & 22 & 2f*o & 105585a & @xmath221.80 & 4 & 2c + 11537 & 30.23 & 4 & 2f * & 64498b & @xmath2215.73 & 1 & 1f & 105879d & 35.90 & 2 & 1f + 11783b & @xmath2228.26 & 1 & 1c & 64498a & @xmath2212.04 & 1 & 1f & 105879a & 40.17 & 3 & 2f*o + 11783a & @xmath2224.58 & 1 & 1c & 66121b & @xmath2228.41 & 1 & 1f & 105947 & 22.13 & 1 & 2f * + 11909 & 6.14 & 1 & 1f & 66676a & 1.40 & 2 & 1f & 106438a & @xmath2229.24 & 1 & 1f + 12326a & 17.00 & 1 & 1f * & 66676b & 2.37 & 1 & 1f & 106438b & @xmath2228.99 & 1 & 1c + 12326b & 19.21 & 1 & 1f & 67246b & @xmath2230.37 & 1 & 1f & 106632a & 16.73 & 1 & 1c + 12361 & 19.36 & 1 & 1f & 67408b & 2.89 & 1 & 1f & 106632b & 16.31 & 1 & 1c + 12764a & 14.75 & 2 & 1f & 69220b & 49.95 & 1 & 1f & 107299 & 37.31 & 1 & 1f + 12780b & @xmath223.44 & 9 & 1f*o & 71682c & 13.22 & 2 & 1f & 107300 & 37.31 & 1 & 1f + 12780a & @xmath226.64 & 7 & 2fo & 72235a & 9.13 & 1 & 1f & 109035c & @xmath2214.62 & 1 & 1f + 12884 & 43.78 & 1 & 1f & 72235b & 9.08 & 1 & 1f & 109035b & 4.74 & 1 & 1f + 13498 & 13.88 & 8 & 3co & 74975b & 55.08 & 1 & 1f & 109035a & @xmath228.67 & 1 & 1f + 13725 & 10.28 & 1 & 1f & 76435c & 7.34 & 2 & 1f * & 109951 & @xmath2224.04 & 1 & 2f * + 14194a & 35.32 & 1 & 1c & 76435a & 5.47 & 3 & 1f & 110091 & 16.05 & 1 & 1f + 14194b & 31.86 & 3 & 2c * & 76888a & 8.06 & 1 & 1f & 110447a & 2.47 & 1 & 1c + 14307b & 19.89 & 3 & 2c * & 76888b & 7.90 & 1 & 1f & 110447b & 1.37 & 1 & 1c + 14519b & 17.70 & 2 & 1c & 79730b & @xmath2240.08 & 1 & 1c & 110712a & 14.57 & 1 & 1f + 14519a & 15.87 & 2 & 1c * & 83701b & @xmath221.09 & 1 & 1c & 110712b & 15.28 & 1 & 1f + 16860b & @xmath2224.93 & 1 & 1c & 85342b & @xmath2215.85 & 1 & 1c & 111903 & 41.68 & 1 & 1f * + 22611b & 46.03 & 1 & 1c & 87813b & @xmath225.94 & 3 & 1c & 112201b & @xmath2212.87 & 1 & 1c + 22611a & 45.72 & 1 & 1c & 89805b & 1.61 & 1 & 1f & 112325 & @xmath2228.07 & 1 & 1f + 22611c & 46.39 & 1 & 1c & 89805a & 0.87 & 1 & 1f & 113386a & 0.46 & 1 & 1f + 22683b & 4.71 & 1 & 1c & 89808 & 8.14 & 1 & 1f & 113386b & 0.85 & 2 & 1f + 22693b & @xmath220.65 & 1 & 1c & 91837c & @xmath223.65 & 1 & 1f & 113579 & 7.03 & 1 & 1f + 23926b & 45.09 & 1 & 1f & 91837a & @xmath223.58 & 1 & 1f & 113597 & 20.55 & 5 & 3f * + 24711b & @xmath2210.79 & 1 & 1c & 92140 & 4.60 & 9 & 1f & 114167b & 4.81 & 1 & 1f + 25082b & 23.23 & 1 & 1f & 94310 & @xmath2215.12 & 1 & 1f & 114167a & 4.25 & 1 & 1f + 27922b & 44.03 & 1 & 1f & 94389 & 23.49 & 1 & 1f & 115087 & 15.21 & 10 & 1fo + 28790b & 34.92 & 10 & 2c*o & 95106a & 9.35 & 2 & 1f & 116063b & 4.30 & 1 & 1f + 28790a & 17.20 & 8 & 1co & 95106b & 9.29 & 5 & 1f * & 116063a & 4.03 & 1 & 1f + 31711c & 31.58 & 1 & 1c & 95116b & @xmath2241.12 & 3 & 1c & 117081a & 21.79 & 1 & 1f + 35261c & 23.83 & 3 & 2c & 95116a & @xmath2241.93 & 1 & 1c & 117081b & 21.24 & 1 & 1f + 36165b & 64.86 & 1 & 1c & 95847 & @xmath2223.95 & 1 & 1f & 117391b & 17.74 & 1 & 1c + 36165a & 66.23 & 1 & 1c & 97508c & 8.04 & 2 & 1f & 117666 & 9.82 & 1 & 2f * + 36640b & 56.25 & 1 & 1c & 97508a & 8.32 & 1 & 1f & & & & + 37645a & @xmath2219.52 & 1 & 1c & 97548 & @xmath228.46 & 1 & 1f & & & & table [ tab : report ] lists all 163 observed targets . column ( 1 ) identifies the target by the _ hipparcos _ number of the main component and , where necessary , the component letter . as most ( but not all ) targets belong to the fg-67 sample , additional information such as equatorial coordinates , proper motions , and magnitudes in various pass - bands can be retrieved from the fg67a paper . the following columns list the average rv , the number of observations , and the tag that combines the maximum number of fitted gaussians ( e.g. 2 for a double - lined binary ) with the instrument code , c for chiron and f for fech . asterisk after the tag signals a spectroscopic system discovered in this survey , the letter o means that spectroscopic orbit has been computed . table [ tab : rv ] , published in full electronically , gives individual observations . its column ( 1 ) contains the target name ( as in table 1 ) . columns ( 2 ) to ( 5 ) contain the heliocentric julian date and the parameters @xmath23 from the gaussian fits ( equation [ eq : fit ] ) . if more than one component was fitted , the parameters of additional components are listed separately , with the same date . the last column ( 6 ) contains a tag that combines the number @xmath24 of the fitted gaussian with the instrument code . the table contains 343 individual observations , 233 made with fech and 110 with chiron . l c ccc l 5896 & 55490.6235 & 7.784 & 0.040 & 33.504 & 1f + 5896 & 55510.5567 & 7.880 & 0.040 & 33.680 & 1f + 6772a & 56863.8045 & 49.866 & 0.201 & 5.638 & 1c + 6772b & 56863.7976 & 50.427 & 0.280 & 6.495 & 1c + 8486b & 55436.9363 & @xmath226.649 & 0.299 & 4.973 & 1f + 9497 & 55436.9452 & 17.811 & 0.223 & 5.318 & 1f + 9902a & 55449.8915 & 11.836 & 0.072 & 18.887 & 1f l l 6896 & f6iv star belongs to the nearby ( 20pc ) visual quadruple system @xmath25 tuc . the rv is 1.1 km s@xmath9 and variable ( gcs ) , + & 7.8 km s@xmath9 and constant here . _ hipparcos _ acceleration . orbital period of several years ? + 7601 & sb3 , two orbits and speckle resolution , periods 19.4 and 614 days . + 9497 & close visual triple system without prior rv data . the ccf has only one narrow dip . + 9642 & sb2 discovered in gcs , orbit @xmath26d . + 10579 & algol - type eclipsing binary ds cet , but the ccfs of a and b are narrow . + 10754 & the component a = hip 10621 at 836is optical , different rv . + 11417 & sb2 discovered by d. latham ( 2012 , private communication ) ; here double - lined with equal components . + 11537 & * new sb2 , _ hippparcos _ acceleration . possibly resolvable by speckle interferometry . + 11783 & a was resolved at soar at 02 , which explains its rv difference with b. + 12326 & * a has asymmetric ccf , rv differs from b and c. possibly resolvable by speckle . + 12780 & * b is new sb1 , orbit @xmath27d . a is sb2 and visual binary . + 12884 & eclipsing binary cn hyi in a 78-year visual binary . the rv here refers to the visual secondary with narrow ccf . + & two wide , low - contrast dips of the eclipsing pair are also present in the ccf , but they were not measured . + 13498 & sb3 discovered by @xcite , orbit with @xmath28d , correction of the visual orbit . + 14194 & * b is new sb2 with a period on the order of a year . + 14307 & * b has asymmetric ccf . it was resolved at soar at 019 , estimated period 25yr . + 14519 & * rv difference of 1.8 km s@xmath9 between a and b indicates a subsystem in one of the components . + 22683 & the b - component hip 22668 at 796has different rv , it is either optical or contains a subsystem . + 26611 & three components at 100and 52 in non - hierarchical configuration have matching rvs . + 28790 & * orbits for a ( @xmath29d ) and b ( new sb2 , @xmath30d ) . + 35261 & c is an sb2 discovered by @xcite . + 37645 & a is a known sb resolved at soar at 020 , hence the rv difference with b. + 41353 & * b = hd 71842 is a new sb2 , at 297from a , optical . for this reason orbit determination is not attempted . + 45734 & young quadruple : a is a close pair , while b has double lines @xcite . + & the ccfs of b are of strange rectangular shape , possibly triple - lined , while the ccf of a is asymmetric . + 64478 & * b is a new sb2 , orbit @xmath31d . updated orbit of a , @xmath32d + 66676 & b is a triple system resolved at soar , but its ccf is narrow , while the rvs of a and b match . + 72235 & b was resolved at soar at 04 , but its ccf is narrow , while the rv matches that of a. + 76435 & * rv(c ) differs from rv(a ) . c was resolved at soar at 56mas , estimated period 4yr . + 85342 & b = hip 85326 is a spectroscopic and acceleration binary resolved at soar . + 87813 & b is a spectroscopic binary discovered by @xcite . + 92140 & sb1 with slow rv variation ( fech , also n. gorynya , private commuication , 2013 ) . + 95106 & * b = hip 95110 is a new sb1 with period of a few months . a was resolved at soar at 027 . + 95116 & hip 96979 has matching rv , similar proper motions and parallaxes . too distant to be a bound binary , however . + 95708 & ab is a 65-year visual binary , its ccf is asymmetric . the rv(c ) is similar , physical triple . + 98274 & close visual binary , its ccf is asymmetric . this is the first rv measurement . + 100896 & the rvs match the 160-day sb2 orbit by @xcite , b = hip 100895 is physical . + 102945 & fast axial rotation ( @xmath33 km s@xmath9 ) ; it is a 08 visual binary with evolved component . + 105585 & ab is a visual binary i 337 , seen as sb2 here . all data can be fitted by a 207-year orbit with @xmath34 . + 105879 & * a is sb2 , preliminary orbit with @xmath35yr , also _ hipparcos _ acceleration . + 105947 & * sb2 and 20.7-year visual binary with a well - defined orbit . km s@xmath9 is 2.2 times larger + & than predicted , caused by a subsystem ? + 109951 & * double ccf , suspected subsystem in the secondary component . + & @xmath37 km s@xmath9 is 3 times larger than predicted by the visual orbit . + 110091 & close visual binary chr 107 with estimated period of @xmath106years . the ccf is slightly asymmetric . + 110792 & b = hip 110719 at 206 was resolved at soar at 019 , explaining the rv difference . + 111903 & * rv(a ) differs by 3.4 km s@xmath9 from the rv reported in gcs , hence new sb ? + 113597 & * new sb3 . it is a 18 visual binary , while hip 113579 at 581also belongs to the system . + 115087 & sb1 discovered by gcs , orbit @xmath38d . + 117666 & * the 30-year visual orbit of a 2700 predicts @xmath39 of 2.7 km s@xmath9 , here @xmath40 km s@xmath9 , hence a subsystem . short comments on new spectroscopic systems or other objects for which the new data are interesting in some or other way are assembled in table [ tab : disc ] . asterisks mark 17 systems or subsystems discovered in this survey . orbital periods are given here for reference only , pending publication of the full analysis . some stars with asymmetric ccfs , presumably binaries with components of comparable luminosity and small rv difference , were resolved recently by speckle interferometry at the soar telescope @xcite . subsystems with large mass ratios are readily detected by the appearance of double or blended lines in a single spectrum . when the visual orbit is known , the rv difference for the moment of observation can be computed using estimated masses . if the observed rv difference is discrepant even after re - adjusting the visual orbit , the presence of a subsystem in one of the components is strongly suspected . this is the case of hip 105947 , 109951 , and 117666 . the alternative idea of detecting subsystems in wide binaries from the rv difference between the components ( as in hip 12780 , 14519 , 76435 ) is explored here by means of simulation . we need to decide whether the wide binary contains a subsystem ( hypothesis h1 ) or is just a binary ( hypothesis h0 ) , based on a single measurement of the rv difference between its components @xmath41 . the quantity @xmath42 is measured with some error . moreover , the wide binary is itself in orbital motion , so the rvs of its components are not exactly equal even without subsystems . first , i establish the distribution of the orbital rvs of spectroscopic binaries . the rv semi - amplitude @xmath43 depends on the orbital inclination @xmath44 and eccentricity @xmath45 as @xmath46 here @xmath47 ( the rv semi - amplitude in a circular orbit at @xmath48 , in kms@xmath9 ) is related to the orbital period @xmath49 ( in years ) and the component masses @xmath50 and @xmath51 ( in solar - mass units ) . the coefficient 29.8 km s@xmath9 is the average orbital velocity of the earth . a large number ( @xmath52 ) of spectroscopic binaries was simulated . their rv @xmath53 was calculated at random orbital phase . the influence of masses and periods is eliminated by studying the distribution of dimensionless quantity @xmath54 . only the random effects of inclination , eccentricity , and phase remain . i assume isotropic orbit orientation which corresponds to the uniform distribution of @xmath55 . the distribution of @xmath56 depends on the distribution of orbital eccentricity . four cases are considered : ( 1 ) @xmath57 ( circular orbits ) , ( 2 ) sine distribution @xmath58 , ( 3 ) uniform distribution @xmath59 , and ( 4 ) linear distribution @xmath0 , often called `` thermal '' . in the additional case 2a , the eccentricity of case 2 is multiplied by 0.7 to get the average @xmath60 typical of spectroscopic binaries . figure [ fig : simul ] shows the results . in the case ( 1 ) , the strict inequality @xmath61 holds . combination of random phase and random inclination results in the uniform distribution of @xmath56 in the interval @xmath62 $ ] , so its cumulative distribution is linear , @xmath63 . larger rv variations are produced in eccentric orbits . their probability is small , increasing with the increasing fraction of eccentric orbits . on the other hand , the median of the rv distribution depends on the eccentricity much less . a spectroscopic binary can be detected by a large rv variation . the probability of such event @xmath64 is defined by the cumulative distribution @xmath65 . i found a good analytical approximation for the case 2 , @xmath66 } \label{eq : fx}\ ] ] ( crosses in figure [ fig : simul ] ) . if the eccentricity is distributed uniformly ( case 3 ) , the formula ( [ eq : fx ] ) is still good at small @xmath56 . the observed rv difference @xmath39 includes a contribution from the orbital motion in wide binary . a subsystem is detectable from a single observation when @xmath39 significantly exceeds the orbital velocity of the outer binary . the latter is estimated statistically using simulations . the relevant normalization factor is now @xmath67 instead of @xmath47 . i denote the `` long '' period of the wide binary by @xmath68 , its mass sum by @xmath69 , and the corresponding amplitude of the rv difference by @xmath70 . for the long period of @xmath71years and a mass sum of 1 @xmath72 , the amplitude is 0.1 of the earth speed or 3 km s@xmath9 . considering that the errors of rvs measured with chiron are on the order of 0.1 km s@xmath9 , it is clear that the rv difference is dominated by the motion in wide binaries , while the measurement errors are negligible in comparison . in the case of wide binaries , we do not know their true orbital periods @xmath68 and estimate them by assuming that projected separation equals semi - major axis @xmath73 . considering that @xmath74 , the estimated amplitude becomes @xmath75 . this has been accounted for in the simulations . normalized rv amplitudes using projected separations ( @xmath76 instead of @xmath70 ) are slightly smaller because , statistically , @xmath76 is larger than @xmath70 . cumulative distributions for different cases are plotted in figure [ fig : sim5 ] . the probability that the measured rv difference @xmath42 is compatible with the hypothesis h0 ( no subsystems ) is estimated through cumulative distribution @xmath77 as @xmath78 let @xmath79 be a small number characterizing the false alarm probability ( detection of a non - existent subsystem ) . the hypothesis h0 is rejected if @xmath80 , or @xmath81 . adopting @xmath82 , the detection limit for subsystems translates to @xmath83 , where @xmath84 for circular outer orbits , @xmath85 for the sine eccentricity distribution , and @xmath86 for the uniform distribution of eccentricity in the outer orbits . now the tools are in place to estimate the probability of detecting a subsystem in a wide binary from a single measurement of @xmath87 . for each combination of the mass ratio @xmath88 and period @xmath89 in the subsystem , the amplitude @xmath47 is computed by ( [ eq : k1 ] ) . the detection threshold is @xmath90 or @xmath91 . the probability of such event is estimated by the approximate formula ( [ eq : fx ] ) . a single measurement of @xmath42 provides thus statistical constraints on the presence of subsystems in both components of a wide binary . this reasoning does not apply to wide binaries with known subsystems , e.g. containing close inner visual binaries , where @xmath42 is likely produced by motion in those subsystems , rather than in the wide binary itself . period ratio vs. mass ratio in detectable subsystems is plotted in figure [ fig : bindet ] with the following assumptions : @xmath92 , sine eccentricity distribution in inner subsystems , and @xmath93 . note the `` fuzzy '' character of this detection technique : the chance to miss a subsystem is substantial for all parameters . this is the penalty for using just a single rv measurement . the detection curves are described by @xmath94 where @xmath95 is determined by inverting ( [ eq : fx ] ) , namely 0.15 , 0.39 , 0.63 , 0.95 , and 1.60 for probabilities of 0.1 , 0.3 , 0.5 , 0.7 , and 0.9 respectively . subsystems with @xmath96 can be detected by double lines independently of the outer period . several detections in this work belong to this category . however , doubling of the ccf is observable when the rv difference in the inner subsystem exceeds a few km s@xmath9 , in which case the single - measurement strategy also works . for simplicity , i do not account for this additional detection channel in the statistical analysis of the next section . the list of wide binaries in the fg-67 sample was created using the following criteria : separation larger than @xmath97 , @xmath7 magnitudes of the secondary from 8 to 11 , declination south of @xmath6 . the original list of 112 binaries was cleaned from optical systems and one white dwarf companion , leaving 98 wide binaries . table [ tab : widelist ] concentrates relevant information on the selected binaries . its column ( 1 ) is the _ hipparcos _ number of the primary component and designations of the wide - binary components . columns ( 2 ) to ( 5 ) give the parallax ( mostly from * ? ? ? * ) , the visual magnitudes of both components , and the separation , respectively . column ( 6 ) contains the characteristic amplitude of the orbital motion in the wide binary @xmath76 computed from the projected separation and masses of the components given in fg67a . then in columns ( 8) and ( 9 ) the average rvs of the primary and secondary components are given , while column ( 9 ) gives reference codes for these rvs , explained in the notes to the table . when only one code is given , it covers both components . in two cases i found only low - accuracy rvs in simbad . if there are several rv sources , i gave preference to this work and/or selected the same source for both components where available . the average rvs of spectroscopic subsystems are replaced by s2 or s1. the presence of subsystems in the primary and secondary components of each binary is indicated in columns ( 10 ) and ( 11 ) , respectively , using the codes of fg67a . briefly , c means subsystems wider than @xmath98 , v and v stand for close resolved subsystems , s2 and s2 mean double - lined spectroscopic binaries , s and s1 mean single - lined binaries , a stands for acceleration binaries . a plus sign indicates sub - subsystems ( spectroscopic pair in the close visual binary hip 64478a ) . the last column ( 10 ) gives _ hipparcos _ or other identifications of secondary components , where available . not all subsystems are absolutely certain , e.g. the spectroscopic subsystems in both components of hip 36165 announced in the gcs . i do not include the potential subsystem in hip 14519 discovered by the @xmath39 method . the apparently non - hierarchical system hip 22611 is removed from the following analysis , leaving 96 wide pairs . subsystems are found in 43 pairs , 25 of which have subsystems in the primary component , 28 in the secondary , and 10 in both ( @xmath99 ) . obviously , the frequency of subsystems in the primary and secondary components is statistically the same . this sample gives a weak evidence of the correlated occurrence of subsystems in both components , noted previously in @xcite and fg67b . if the occurrence of subsystems were statistically independent , the expected number of wide binaries with subsystems in both components would be @xmath100 , slightly less than the actual number 10 . figure [ fig : sec ] compares periods and mass ratios of subsystems in the secondary components with their detection probability . the probability is calculated by the prescription of fg67a . the additional factor resulting from @xmath101 measurements is incorporated for the 53 binaries without subsystems . the four subsystens with unknown periods and mass ratios are depicted by crosses located arbitrarily . the addition of spectroscopic coverage has dramatically improved the detection at @xmath102 days in comparison with imaging - only surveys such as @xcite . such plot for the primary subsystems is similar , with slightly deeper detection contours . it is instructive to compare the frequency and periods of subsystems with the statistical model developed in fg67b . the basic assumption is that subsystems are chosen randomly from the log - normal distribution of periods , keeping only dynamically stable multiples . i adopted the following parameters of the underlying period distribution : @xmath103 , @xmath104 , the mass - ratio exponent @xmath105 , and the total binary frequency @xmath106 ( see fg67b for description of those parameters ) . two - dimensional distribution in the @xmath107 space was multiplied by the detection probability and the dynamical truncation , providing the expected period distribution in the subsystems and their total number . it is compared to the actual cumulative distribution of secondary periods in figure [ fig : cum ] . the four yet unknown periods were given arbitrary values close to 100days . the model predicts 26.6 sevcondary subsystems for @xmath105 and 23.1 for @xmath108 , while the actual number is 28 . this calculation was repeated for the primary subsystems , giving their predicted numbers of 32.4 and 29.1 for @xmath105 and @xmath108 , respectively ; the actual number is 25 . the numbers match within statistical errors , as do the period distributions ( with the caveat related to unknown periods ) . therefore , the data collected here support the statistical description of multiplicity proposed in fg67b . in this section , i study the distribution of the observed rv difference in 53 wide binaries without known subsystems . the cumulative distribution of the normalized quantity @xmath109 is plotted in figure [ fig : dv ] and compared to the simulated distributions presented in figure [ fig : sim5 ] . a good agreement with the case-3 simulation ( uniformly distributed eccentricity ) and a poor match with the `` thermal '' eccentricity distribution are obvious . the observed distribution is widened by the rv errors ( both random and systematic ) and by yet undetected subsystems . so , the real distribution should be even narrower than the observed one . this means that the eccentricities of wide binaries are likely even smaller than in the case 3 . @xcite suggested that very wide binaries can form through ejection of one star from compact triple systems . outer orbits of such ejected stars mostly have large eccentricities with typical values of @xmath100.7 ( see their figure 2 ) . the wide binaries studied here have separations on the order of @xmath71au , as in the above paper . however , eccentricities of those binaries without subsystems appear to be moderate , as follows from the rv differences between their components . moreover , the frequency of subsystems in the primary and secondary components is similar , which speaks against the ejection scenario . interestingly , @xcite found that eccentricities of binaries in the 25-pc volume are distributed approximately uniformly independently of their period . however , they based this conclusion on known spectroscopic and visual orbits which do not include binaries as wide as studied here . results of the rv survey of nearby stars are reported . i focused on wide visual binaries and aimed at detecting subsystems in their components by a presence of double lines or by a substantial rv difference between the components . this latter `` snapshot '' technique is studied by simulation , showing that the major factor limiting the subsystem detection is the orbital motion in the wide binary itself . the survey readily detects subsystems with short periods . non - detections are interpreted statistically , constraining the parameters of subsystems and their overall frequency . as most of the primary components of wide binaries in this sample already have rv coverage in the literature , i observed fainter and less - well studied secondaries . five new subsystems in wide binaries were discovered in this project . the frequency of subsystems in the primary and secondary components turns out to be statistically the same . the data match the crude model proposed in fg67b , where subsystems are chosen randomly from the general period distribution and are constrained only by dynamical stability . an interesting additional result concerns wide binaries without subsystems . the distribution of the observed rv difference between their components places some constraints on the distribution of the orbital eccentricities , which are not measured directly owing to very long periods . the results of this work rule out the `` thermal '' distribution @xmath110 and match instead the uniform distribution . this suggests that dynamical interactions in @xmath111-body stellar systems were not dominant in shaping the orbits of wide binaries . i thank the operators of the 1.5-m telescope for executing observations of this program and the smarts team at yale for scheduling and pipeline processing . this work used the simbad service operated by centre des donnes stellaires ( strasbourg , france ) , bibliographic references from the astrophysics data system maintained by sao / nasa , and the washington double star catalog maintained at usno . nordstrm , b. , mayor , m. , andersen , j. et al . 2004 , a&a , 418 , 989 ( gcs ) pourbaix , d. , tokovinin , a. a. , batten , a. h. et al . 2004 a&a , 424 , 727 ( sb9 ) raghavan , d. , mcalister , h. a. , henry , t. j. et al . 2010 , apjs , 190 , 1 reipurth , bo , & mikkola , s. 2012 , nature , 492 , 221 l r rr rr rr l l l l 2292 a , b & 17.6 & 7.9 & 9.4 & 839.3 & 0.2 & 9.6 & 9.6 & gcs , l02 & & v & b=2350 + 2888 a , b & 22.9 & 6.8 & 8.5 & 329.8 & 0.5 & s2 & s1 & sb9,t06 & s2 & s1 & b=2848 + 3203 a , b & 37.7 & 7.0 & 9.9 & 8.5 & 2.6 & 12.0 & n / a & n04,cfa & & & + 3795 ab , c & 15.6 & 7.7 & 10.8 & 152.3 & 0.6 & 21.40 & n / a & gcs & v & v & + 6772 a , b & 16.6 & 8.2 & 11.1 & 389.1 & 0.3 & 49.9 & 50.4 & chi & & & b=6804 + 9902 a , b & 22.6 & 7.6 & 10.5 & 52.2 & 0.9 & 11.5 & 11.0 & chi & & & + 9911 a , b & 27.1 & 7.0 & 10.5 & 79.2 & 0.8 & 39.1 & 38.7 & chi & & & + 11024 a , b & 24.4 & 8.0 & 10.4 & 10.3 & 1.9 & 45.2 & 45.2 & chi & & & + 11137 a , b & 17.0 & 8.9 & 9.4 & 34.7 & 0.9 & 26.3 & 26.6 & l02 & & & b=11134 + 11783 a , b & 37.5 & 4.8 & 8.8 & 345.3 & 0.5 & 24.6 & 28.3 & chi & s , a , v & & b=11759 + 12780 a , b & 24.2 & 7.0 & 8.5 & 12.5 & 2.4 & s2 & s1 & chi & v , s2 & s1 & + 14194 a , b & 18.8 & 7.6 & 10.0 & 8.7 & 2.2 & 35.3 & s2 & chi & & s2 & + 14307 a , b & 18.7 & 7.6 & 8.6 & 38.3 & 1.1 & 20.4 & s2 & gcs , chi & & v , s2 & b=14313 + 14519 a , b & 18.8 & 9.1 & 10.6 & 60.3 & 0.7 & 15.9 & 17.7 & chi & & s1 ? & + 15527 a , b & 28.1 & 7.4 & 8.5 & 253.0 & 0.4 & 39.5 & 39.6 & gcs & & & b=15526 + 16860 a , b & 21.1 & 9.0 & 10.9 & 31.5 & 1.0 & 26.1 & 24.9 & gcs , chi & & & b=16858 + 18888 a , b & 15.9 & 8.2 & 8.6 & 595.2 & 0.3 & 33.7 & 33.7 & gcs & & v & b=18958 + 21923 a , b & 23.0 & 7.2 & 10.2 & 141.6 & 0.6 & 14.5 : & s2 & sim , h12 & & s2 & b=21946 + 22611 a , c & 16.9 & 6.7 & 8.7 & 51.9 & 0.9 & 45.7 & 46.4 & chi & & & + 22611 a , b & 16.9 & 6.8 & 9.0 & 99.6 & 0.7 & 45.7 & 46.0 & chi & & & b=22604 + 22826 a , b & 19.9 & 7.2 & 8.7 & 415.9 & 0.3 & 12.1 & 12.0 & gcs & & & b = hd 31222 + 23693 a , b & 85.9 & 4.8 & 9.0 & 321.7 & 0.6 & 0.8 & 0.7 & gcs , chi & & & b=23708 + 23926 a , b & 18.7 & 6.8 & 10.3 & 10.1 & 1.9 & 43.8 & 45.1 & gcs , chi & & & b=23923 + 24711 a , b & 15.3 & 8.5 & 10.6 & 13.4 & 1.5 & 11.3 & 10.8 & gcs , chi & v & & b=24712 + 25082 a , b & 17.6 & 7.1 & 10.1 & 9.5 & 1.9 & 25.1 & 23.2 & gcs , chi & & & + 27922 a , b & 42.4 & 7.6 & 10.6 & 10.6 & 2.3 & 42.1 & 44.0 & gcs , chi & & & + 31711 ab , c & 47.0 & 6.3 & 9.8 & 808.6 & 0.4 & 28.1 & 31.6 & gcs , chi & v , s , a & & c=31878 + 33705 a , b & 26.8 & 6.7 & 8.6 & 323.9 & 0.4 & 16.3 & 16.4 & gcs & & & b=33691 + 34065 ab , c & 60.5 & 5.6 & 8.8 & 184.9 & 0.9 & 88.1 & s1 & gcs , s11 & c & a , s1,v & c=34052 + 34065 a , b & 60.5 & 5.6 & 7.0 & 21.0 & 2.7 & 88.1 & 88.0 & gcs & & & b=34069 + 36165 a , b & 31.2 & 7.1 & 8.1 & 17.7 & 1.8 & 66.7 & 66.3 & gcs & s & s & b=36160 + 36485 ab , d & 22.0 & 7.4 & 8.0 & 112.0 & 0.8 & 1.0 & s1 & gcs , h12 & c & s1,a , v & d=36497 + 36640 a , b & 35.8 & 6.0 & 8.7 & 23.4 & 1.7 & 54.8 & 56.3 & gcs , chi & & & b=36642 + 37645 a , b & 19.6 & 7.1 & 10.4 & 9.6 & 2.3 & s1 & 16.4 & cfa , chi & s1,v & & + 38908 a , bc & 61.7 & 5.6 & 9.9 & 60.6 & 1.3 & 14.7 & 17.0 & gcs , chi & & v & + 40452 a , b & 25.9 & 7.7 & 9.8 & 32.0 & 1.1 & 20.5 & 21.2 & gcs , chi & & & + 42488 a , b & 27.5 & 7.3 & 8.6 & 25.9 & 1.4 & 20.3 & 20.2 & gcs , c11 & s & & b=42491 + 44579 a , b & 15.5 & 8.7 & 9.3 & 140.5 & 0.4 & 2.2 & 0.6 & gcs , chi & & & b = hd 77759 + 45734 a , b & 17.6 & 8.4 & 9.7 & 9.0 & 2.5 & s2 & s2 & chi & v , s & s2 & + 45838 c , ab & 16.5 & 7.5 & 9.2 & 135.0 & 0.6 & 52.6 & 55.0 & gcs , chi & & v & ab = hd 60815 + 46236 a , b & 21.4 & 7.0 & 10.2 & 19.3 & 1.4 & 30.2 & 31.0 & gcs , chi & & & + 47058 a , b & 21.3 & 7.9 & 10.5 & 8.1 & 2.1 & 60.0 & s1 & gcs , cfa & & s1 & + 47839 a , b & 19.0 & 8.1 & 8.2 & 18.8 & 1.8 & 2.2 & 2.4 & gcs & & & b=47836 + 47862 a , b & 15.8 & 7.2 & 10.8 & 9.6 & 1.8 & 16.5 & 14.5 & gcs , chi & & & + 48785 a , b & 23.9 & 7.7 & 8.4 & 30.6 & 1.2 & 30.4 & 31.0 & gcs & & & b=48786 + 49030 a , b & 16.7 & 7.7 & 10.5 & 24.2 & 1.1 & 14.7 & 15.8 & gcs , chi & & & + 49520 a , b & 16.9 & 8.8 & 9.0 & 9.5 & 1.9 & 0.4 & 0.3 & d06 & & v & + 56242 a , b & 42.9 & 6.4 & 9.2 & 15.5 & 2.1 & 4.9 & 4.9 & gcs & & & + 57148 a , bc & 16.0 & 8.4 & 9.8 & 24.9 & 1.2 & 4.3 & 4.3 & lco & & v & bc=57146 + 58067 a , b & 24.7 & 8.3 & 8.5 & 73.4 & 0.8 & 5.9 & 5.6 & gcs & & & b=58703 + 58813 a , b & 18.3 & 8.1 & 8.8 & 23.2 & 1.2 & 6.0 & 5.5 & gcs & & & b=58815 + 59272 a , b & 45.0 & 7.0 & 9.7 & 9.4 & 2.7 & 1.6 & 2.7 & gcs , chi & & & + 59690 a , b & 18.1 & 7.6 & 9.7 & 23.2 & 1.2 & 24.2 & 25.1 & gcs , chi & & & b=59687 + 60353 a , b & 33.4 & 6.6 & 10.5 & 20.7 & 1.7 & 4.3 & 4.4 & gcs , chi & & & b=60352 + 60749 ab , d & 15.9 & 8.5 & 10.2 & 24.7 & 1.3 & 3.9 & 1.8 & gcs , chi & v & & d=60750 + 61595 a , b & 17.8 & 8.3 & 10.9 & 8.4 & 1.9 & 13.7 & 15.3 & gcs , chi & & & + 64478 a , b & 23.7 & 6.3 & 9.4 & 25.1 & 1.9 & s2 & s2 & s90,chi & v+s2 & s2 & + 64498 a , b & 18.0 & 7.6 & 9.9 & 9.3 & 2.2 & 12.0 & 15.7 & chi & v & & + 66121 a , b & 28.4 & 6.5 & 9.3 & 22.1 & 1.5 & 28.9 & 28.4 & gcs , chi & & & b=66125 + 66676 a , bc & 16.9 & 8.4 & 9.2 & 77.5 & 0.7 & 1.3 & 2.4 & chi & & v & bc = hd 118735 + 67246 a , b & 31.6 & 6.4 & 10.2 & 488.5 & 0.3 & 30.9 & 30.4 & gcs , chi & & & b=67291 + 67408 a , b & 33.9 & 6.7 & 10.2 & 11.6 & 2.2 & 3.2 & 2.9 & gcs , chi & & & + 69220 a , b & 20.8 & 8.4 & 9.7 & 57.5 & 0.8 & 48.9 & 50.0 & gcs , chi & & & b=69224 + 71682 a , cd & 23.4 & 7.1 & 10.0 & 123.0 & 0.7 & 11.1 & 13.2 & gcs , chi & a & v & cd=71686 + 72235 a , b & 24.0 & 8.6 & 10.8 & 9.1 & 2.1 & 9.1 & 9.1 & chi & & v & + 74930 a , b & 20.0 & 7.2 & 8.1 & 13.3 & 1.8 & 35.9 & 36.7 & gcs & & & b=74931 + 74975 a , b & 39.4 & 5.1 & 10.1 & 11.4 & 2.6 & 54.2 & 55.1 & gcs , chi & & & + 75790 ab , c & 17.1 & 7.0 & 10.1 & 9.6 & 2.3 & 12.2 & 10.9 & lco & v & & + 76435 a , c & 20.6 & 9.1 & 10.6 & 13.5 & 1.8 & 5.4 & 7.3 & chi & & v & + 76888 a , b & 15.0 & 9.3 & 9.4 & 23.2 & 1.1 & 8.1 & 7.9 & chi & & & b=76891 + 78738 a , b & 39.6 & 7.5 & 8.1 & 11.8 & 2.3 & 31.8 & 32.3 & gcs & & & b=78739 + 79730 a , b & 21.1 & 7.2 & 8.7 & 19.8 & 1.4 & 40.6 & 40.1 & gcs , chi & & & + 83701 a , b & 17.1 & 7.9 & 10.0 & 97.2 & 0.6 & 1.3 & 1.1 & gcs , chi & & & + 85342 a , b & 21.1 & 7.0 & 8.9 & 127.2 & 0.6 & 15.9 & 16.6 & gcs & & a , s , v & b=85326 + 90355 a , b & 27.3 & 7.9 & 8.4 & 608.4 & 0.3 & s1 & 18.0 & h12 & a , s1 & & b=90365 + 91837 ab , c & 16.0 & 7.7 & 10.2 & 32.6 & 1.2 & 4.2 & 4.1 & lco & v & & + 93772 a , b & 15.2 & 7.0 & 8.9 & 9.4 & 1.9 & 18.2 & 18.7 & gcs , ts02 & & & + 95106 a , b & 21.2 & 8.2 & 10.3 & 13.7 & 1.9 & 13.7 & s1 & gcs , fech & v , s & s & b=95110 + 95116 a , b & 15.9 & 7.1 & 9.3 & 8.8 & 2.0 & 41.9 & 41.1 & chi & & & + 96979 a , b & 15.1 & 6.9 & 8.7 & 27.3 & 1.1 & 41.9 & 41.6 & gcs & & & b=96976 + 97508 ab , c & 17.8 & 7.4 & 10.4 & 13.9 & 1.9 & 8.3 & 8.0 & chi & v & & + 99729 a , b & 16.4 & 7.8 & 8.1 & 43.3 & 0.9 & 0.9 & 0.6 & n02 & & & b=99727 + 101551 a , b & 16.7 & 7.9 & 9.6 & 10.4 & 1.7 & 21.4 & 21.4 & chi & & & b=101549 + 102418 a , cd & 15.3 & 8.8 & 10.5 & 539.2 & 0.3 & 10.7 : & 11.3 & sim , chi & & v & + 102655 a , b & 18.6 & 8.5 & 9.7 & 391.3 & 0.3 & 2.6 & 2.3 & chi & & & b = hd 198016 + 103311 ab , c & 21.9 & 7.4 & 10.7 & 324.9 & 0.4 & 4.5 & 6.9 & sacy & v & & + 105585 ab , c & 18.0 & 8.9 & 9.0 & 18.1 & 1.6 & s2 & 3.6 & chi & v , s2 & & c=105569 + 105879 a , d & 15.8 & 7.3 & 10.0 & 44.0 & 0.9 & s2 & 35.9 & chi & s2,a & & + 106438 a , b & 20.4 & 7.8 & 9.8 & 40.3 & 0.9 & 29.2 & 29.0 & chi & & & + 106632 a , b & 16.5 & 8.8 & 10.0 & 11.4 & 1.6 & 16.7 & 16.3 & chi & & & b=106633 + 110447 a , b & 15.0 & 9.1 & 10.2 & 12.6 & 1.4 & 1.4 & 2.5 & chi & & & b=110444 + 110712 a , b & 43.4 & 6.1 & 8.9 & 20.6 & 2.1 & 14.6 & 15.3 & fech & & v & b=110719 + 112201 a , b & 18.5 & 8.3 & 9.8 & 179.0 & 0.4 & 13.3 & 12.9 & gcs , chi & & & + 113386 a , b & 16.3 & 7.6 & 9.2 & 9.1 & 1.9 & 0.5 & 0.9 & chi & & & + 113579 a , b & 32.5 & 7.6 & 9.7 & 581.2 & 0.3 & 7.0 & s3 & fech & & v & b=113597 + 114378 a , b & 40.3 & 6.5 & 10.3 & 31.7 & 1.4 & s1 & 1.4 & griff01 & s1 & & + 114702 a , b & 25.6 & 8.1 & 8.6 & 24.9 & 1.6 & 30.0 & s1 & gcs , t06 & & v+s1 & b=114703 + 116063 a , b & 32.6 & 7.2 & 9.1 & 36.4 & 1.2 & 4.0 & 4.3 & fech & & & + 117391 a , b & 17.5 & 7.6 & 10.1 & 109.3 & 0.5 & 17.2 & 17.7 & gcs , chi & & &

we introduce a strongly correlated mean - field theory of the ionic hubbard model on the triangular lattice with alternating stripes of site energy using barnes - coleman slave bosons . we study the paramagnetic phases of this theory at three quarters filling , where it is a model of , , and . this theory has two bands of fermionic quasi - particles : one of which is filled or nearly filled and hence weakly correlated ; the other is half - filled or nearly half - filled and hence strongly correlated . further results depend strongly on the sign of the hopping integral , . the light band is always filled for , but only becomes filled for for , where is the difference in the site energies of the two sublattices . a metal charge transfer insulator transition occurs at for and for . in the charge transfer insulator complete charge disproportionation occurs : one sublattice is filled and the other in half filled . we compare our results with exact diagonalisation calculations and experiments on , and discuss the relevance of our results to and . we propose a resolution of seemingly contradictory experimental results on . many experiments suggest that there is a charge gap , yet quantum oscillations are observed suggesting the existence of quasiparticle states at arbitrarily low excitation energies . we argue that the heavy band is gapped while the light band , which contains less than 1 charge carrier per 100 unit cells , remains ungapped .

0904.0505

experiments on , where @xmath7 is na , rb , or k , show a wide range of strongly correlated phases.@xcite , in particular , has attracted much attention following the discovery of superconductivity when water is intercalated into the system.@xcite the phase diagram , with doping @xmath8 , shows many interesting phases,@xcite including a ` curie - weiss metal ' , a type antiferromagnetism ( ferromagnetic layers stacked antiferromagnetically ) and an ` insulating ' phase seen only at @xmath9 . the latter phase is particularly puzzling as many probes [ including resistivity,@xcite optical conductivity,@xcite and angle resolved photoemission spectroscopy ( arpes)@xcite ] suggest that it is insulating , however , shubnikov - de haas oscillations are also observed,@xcite suggesting the state is metallic . the aims of this paper are to present a simple variational theory of , where @xmath7=na , rb , or k , and to attempt to reconcile these seemingly contradictory experiments . in the co atoms form a triangular lattice and the simplest model of the band structure , a single band triangular lattice with nearest neighbour hopping only , gives good agreement with arpes experiments on .@xcite the doping of the system is controlled by the concentration of @xmath7 ions , @xmath8 , with the single band being half filled at @xmath10 and filled at @xmath11 . however , the @xmath7 ions order so as to minimise the mutual coulomb repulsion.@xcite therefore , the coulomb potential due to the @xmath7 ions is different at different co atoms and , because the @xmath7 ions are ordered , this gives rise to an ordered arrangement of potentials at the vertices of the triangular lattice . thus an effective hamiltonian for is the ionic hubbard model.@xcite in addition to the interest in the ionic hubbard model is of significant interest in its own right . the ionic hubbard model on half filled bipartite lattices has attracted interest because it undergoes a transition from a mott insulator to a band insulator.@xcite furthermore , away from half filling and on frustrated lattices the ionic hubbard model shows a subtle interplay between charge and spin ordering and metallic and insulating phases.@xcite the hamiltonian of the ionic hubbard model is @xmath12 where @xmath0 is the hopping amplitude between nearest neighbour sites only , @xmath13 is the effective on - site coulomb repulsion between two electrons , @xmath14 is the site energy , @xmath15 annihilates ( creates ) an electron on site @xmath16 with spin @xmath17 and @xmath18 . previous studies of this model on frustrated lattices included both analytical and numerical studies of zigzag ladders,@xcite dynamical mean field theory ( dmft ) studies on infinite dimensional fcc@xcite and two - dimensional triangular lattices@xcite and exact diagonalisation on small triangular lattices.@xcite a more extensive discussion of previous work on the ionic hubbard model is given in ref . however , here we aim to provide a simple variational description that captures as much of the strongly correlated physics of this model as possible . the wide range of numerical techniques , described above , that have been applied to the ionic hubbard model have not , previously , been complemented by a commensurate effort to develop simple variational approaches . the history of the theory of strongly correlated electrons shows that progress has often been made when accurate numerical techniques are combined with such variational calculations . therefore , our theory , which can be straightforwardly generalised to other potential arrangements , lattices , fillings , etc . , is complementary to the previous numerical work . in many different na ordering patterns are seen at different values of @xmath8,@xcite and each of these correspond to a different @xmath19.@xcite in principle , each of these different @xmath19 correspond to a different hamiltonian . therefore , for simplicity and definiteness , we specialise to the hamiltonian relevant to @xmath9 . this model has two sublattices with different site energies , @xmath20 for a - sites and @xmath21 for b - sites,@xcite and with the sublattices arranged in stripes as shown in fig . [ fig : unit - cells]b . thus the hamiltonian is @xmath22 where @xmath23 annihilates ( creates ) an electron with spin @xmath17 in an orbital centred on site @xmath16 belonging to sublattice @xmath24 and @xmath25 indicates that the sum is over nearest neighbours on the appropriate sublattices only . hamiltonian ( [ ham ] ) is a natural instance of hamiltonian ( [ ham1 ] ) for initial study on both theoretical and experimental grounds . experimentally , the behaviour of , particularly , is very different from that of with @xmath8 either a little larger or smaller than @xmath26 . further , many of the phenomena observed at @xmath9 seem to have a natural explanation in terms of the ionic hubbard model.@xcite we have previously studied the quarter - filled ionic hubbard model with a stripe potential by exactly diagonalising small clusters.@xcite we found a complicated interplay between the charge and spin degrees of freedom and between insulating and metallic states . we also found that these calculations provide a possible framework for understanding a wide variety of experiments on , , and . however , the most interesting regime ( i.e. , @xmath27 , see section [ sect : natheory ] ) , in terms of its relevance to experiments , is also the most challenging to investigate theoretically . this provides additional motivation to investigate the simpler mean - field theory presented below . comparison between our mean - field theory and these exact diagonalisation calculations will be made , where possible , below . the rest of this paper is organised as follows . in the remainder of this introduction we briefly review the most pertinent experiments on and previous theories aimed at explaining these measurements . in section [ sect : formal ] we give the formal details of our theory and derive a set of self - consistency equations for the ground state . in section [ sect : numeric ] we report the results of numerical solutions of these equations and compare our results with experiments on and previous theoretical studies of the triangular lattice ionic hubbard model . in section [ sect : conc ] we draw our conclusions . in appendix [ sect : non - int ] we describe the band structure of the model ( [ ham ] ) for non - interacting electrons . there has been far less experimental work on or than . therefore , in this section , we focus mainly on experiments on . the brief review below brings out two particular puzzles that theory needs to explain about the ground state . the first is how it is some properties are consistent with a metallic ground state and that others are more consistent with an insulating ground state . the second puzzles concerns how it is that the ground state has a large magnetic moment but a small amount of charge ordering . measurements of the relaxation rates for @xmath28co nmr found two distinct co sites , consistent with a charge ordered state.@xcite however , the authors noted that the degree of charge disproportionation is rather small . more recent @xmath28co and @xmath29na nmr measurements find no signature of differences in the charge state of the cobalt ions at the two distinct sites,@xcite suggesting that all the co atoms are in the charge state co@xmath30 , where @xmath31 gives an upper bound on the extent of charge ordering.@xcite this is in contrast , to the case of @xmath32 or 0.75 for which nmr clearly detecting a charge disproportionation.@xcite high resolution neutron crystallography detects small differences in co - o bond lengths , and an analysis based on bond valence sums , is consistent with a charge order of @xmath33.@xcite very recent nmr measurements were able to detect a very small charge ordering along the co(1 ) chains , below 51 k,@xcite but the authors do not discuss the magnitude of the difference in charge between two sublattices . as the temperature decreases the intralayer resistivity increases monotonically . no feature is seen at the magnetic ordering temperature ( 88 k , see below ) . above 51 k , it is weakly temperature dependent with values of a few m@xmath34 cm,@xcite characteristic of a `` bad '' or incoherent metal which does not have well - defined quasi - particle excitations.@xcite below 51 k , the resistivity increases significantly , consistent with an activated form with energy gap of about 10 mev.@xcite although , it is usually stated that the transition at 51 k is a `` metal - insulator '' transition , we stress that it is really above 51 k is a bad metal and below 51 k the experimental evidence is not all consistent with the claim that is an insulator ( see section [ sect : metal ] ) . applying hydrostatic pressure causes the resistivity to decrease and the temperature where the resistivity becomes activated to increase . above about 13 gpa the resistivity has a metallic temperature dependence.@xcite in contrast , for @xmath8=0.75 is metallic at ambient pressure but becomes insulating above about 23 gpa . resistivity measurements found that the charge gap apparent from the resistivity was suppressed by magnetic fields parallel to the layers , larger than about 35 t , with hysteresis between 15 and 40 t , at low temperatures.@xcite a field perpendicular to the layers does not suppress the `` insulating '' state.@xcite the thermopower and hall coefficient have small and weakly temperature dependent positive values above 100 k.@xcite however , they change sign near 88 k , obtain large negative values below 50 k , and then decrease towards zero at low temperatures . the latter is distinctly different from the behaviour of conventional semiconductors and insulators , for which the thermopower and hall constant both diverge as the carrier density vanishes with decreasing temperature . quasi - particle features are only observed in angle resolved photoemission spectroscopy ( arpes ) below about 150 k and reveal a fermi surface consistent with a hole - like band for a tight - binding model on the triangular lattice with an effective hopping integral of @xmath35 mev,@xcite about six times smaller than that predicted by lda calculations . a gap begins to open up over the fermi surface below about 60 k , and has a magnitude of about 6 - 11 mev at 10 k.@xcite measurements of the frequency dependence of the conductivity show no evidence of a drude peak , consistent with the bad metal behaviour , seen in the static transport quantities.@xcite below about 100 k there is a loss of spectral weight at frequencies below about 100 @xmath36 , consistent with the opening of a charge gap . this leads to a peak in @xmath37 at @xmath38 @xmath36.@xcite in comparison for na@xmath39coo@xmath40 , with @xmath8 away from 0.5 , a drude peak is present at low temperatures and the total spectral weight of the optical conductivity scales with @xmath41.@xcite shubnikov - de haas oscillations are observed in the field range of 15 - 30 t.@xcite a fast fourier transform of the oscillatory part of the conductivity has peaks at frequencies of 150 t and 40 t. these frequencies correspond to pockets with cross - sectional areas of 0.25 % and 0.06 % of the area of the undistorted hexagonal brillouin zone ( cf . figure 1d ) . the effective mass associated with these orbits are ( in units of the free electron mass ) @xmath42 and @xmath43 , respectively . these measurements are extremely surprising given the range of other evidence suggesting that has an insulating ground state . muon spin rotation ( @xmath44sr ) experiments first saw evidence for magnetic order below about 86 k , and slight effects at 50 k and 30k.@xcite below 88 k , there is a splitting of nmr lines , consistent with the development of commensurate antiferromagnetic order.@xcite the change in the resistivity from a bad metal to an activated behaviour at 51 k has little effect on the magnetic state . elastic neutron scattering also detects long - range antiferromagnetic order below 88 k , with a magnetic moment of @xmath45 per magnetic cobalt ion.@xcite the authors interpret this order in terms of alternating rows of magnetic co@xmath46 and non - magnetic co@xmath47 ions . ( i.e. , complete charge order ) . however , we stress that the observation of such magnetic order does not require charge ordering to be present . simple classical arguments suggest the magnetic moment should be less than @xmath48 where @xmath49 is the average number of electrons on the @xmath50 sublattice . hence , the charge disproportionation observed in the crystallography experiments discussed above , would imply a moment of about @xmath51 , about one quarter of the observed value . both the bulk magnetic susceptibility and the knight shift are weakly temperature dependent , with a peak around 300 k , and a magnitude of about 5 @xmath52 emu / mol.@xcite this suggests an antiferromagnetic exchange between magnetic ions of order a few hundred kelvin . at other values of @xmath8 in , the ordering of sodium ions has been found to be important for the existence of various ordered phases.@xcite no insulating state is seen in the corresponding misfit cobaltates,@xcite which supports the claim that na - ordering is necessary for the insulating state . furthermore , cooling a material at different rates found that the presence of sodium ordering can drive an additional magnetic phase transition at @xmath53 and 0.85.@xcite choy _ et al_. @xcite considered an extended hubbard model that included the coulomb interaction between electrons on neighbouring sites , @xmath54 , but neglected the na - ordering and the ` ionic ' term in eq . ( [ ham ] ) . they argued that @xmath54 stabilises a charge ordered state and considered an effective low - energy hamiltonian for this charge ordered state . by fine tuning the parameters in their effective hamiltonian choy _ et al_. were able to reproduce the observed temperature dependence of the resistivity and the hall coefficient . lee _ et al_. @xcite have studied via lda+u calculations that allow for the effects of na - ordering . they find a first order metal - charge ordered insulator transition as @xmath13 is increased . the charge disproportionation and the opening of the gap both occur at a single phase transition in this weak coupling theory . we have reported exact diagonalisation calculations for hamiltonian ( [ ham ] ) of finite lattices.@xcite these studies suggested that , for parameters relevant to , the system is a covalent insulator , with little charge disproportionation and a small gap @xmath55 , rather than a charge ordered insulator , with strong charge disproportionation and a gap @xmath56 . this prediction is consistent with the experimental measurements of the charge disproportionation.@xcite further , these calculations predict a gap of the same order of magnitude as in seen in the arpes@xcite and the resistivity,@xcite predict the large moment seen in neutron scattering,@xcite and reproduce the main features seen in the optical conductivity.@xcite zhou and wang@xcite also studied a model that incorporates the effects of na - ordering within the gutzwiller approximation and found it could explain many of the experimental observations . there has been significantly less work on and . lee and pickett@xcite have reported lda calculations for and compared these with equivalent calculations for . they found that the @xmath57 band is rather narrower in than and that the fermi surface is more complicated in as there are several small pockets along the @xmath58-@xmath59 line . they speculated that these pockets , some of which are quite well nested near the @xmath58 point , may lead to enhanced magnetic tendencies in . clearly an important question is : what parameters in hamiltonian ( [ ham ] ) correspond to the various possible choices of @xmath7 in ? various atomistic calculations that address this question have been presented for @xmath7=na , but for other @xmath7 there are not yet suitable estimates . calculations@xcite on small clusters of give @xmath60 ev , @xmath61 ev , and @xmath62 ev . for bulk the lda yields @xmath63 ev @xcite and @xmath64 ev @xcite , and electrostatic ewald calculations give @xcite @xmath65 ev . furthermore , coo@xmath40 , which is described by hamiltonian ( [ ham ] ) at half filling with @xmath66 , is observed to be a strongly correlated metal rather than a mott insulator.@xcite this suggests that @xmath67 , where @xmath68 is the critical value for formation of a mott insulator on the triangular lattice.@xcite hence , realistic parameters for may be in the range @xmath69 , and @xmath70 . the negative sign of @xmath0 is natural if the hopping between co sites is dominated by the contribution from hopping via an intermediate o site . to leading order this gives @xmath71 , where @xmath72 is the direct hopping between a co atom and a neighbouring o atom , @xmath73 is the energy of an electron in a @xmath74 orbital centred on a co atom , and @xmath75 is the energy of an electron in a @xmath76 orbital centred on an o atom ( note that one expects that @xmath77 ) . we now investigate paramagnetic phases of the @xmath78 limit of the ionic hubbard model by deriving a mean field theory using barnes - coleman slave bosons.@xcite the @xmath78 limit great simplifies the analysis and is not unreasonable in spite of our estimate , above , that @xmath69 . our previous exact diagonalization calculations@xcite show that , provided @xmath79 the physics of this model is insensitive to particular value of @xmath80 . specifically , our results for @xmath81 show only minor quantitative differences from those for @xmath82 . we begin by making the particle - hole transformation @xmath83 . we then introduce the slave bosons . following , for example , ref . , we write @xmath84 where @xmath85 is the completely antisymmetric tensor , @xmath86 annihilates ( creates ) a fermion with spin @xmath17 at site @xmath16 on sublattice @xmath24 , @xmath87 annihilates ( creates ) a boson , corresponding to a site that is doubly occupied by holes , at site @xmath16 on sublattice @xmath24 , and @xmath88 annihilates ( creates ) a boson , corresponding to a site containing no holes , at site @xmath16 on sublattice @xmath24 . if the number of holes is less than or equal to the number of lattice sites and @xmath89 there is zero weight for configurations containing any sites doubly occupied ( with holes ) in the the ground state or any excited states at finite energy above the ground state . therefore , @xmath90 can be ` deleted ' giving us @xmath91 it is clear from eq . ( [ eqn : hole ] ) that the @xmath92 s are hole - like operators . this transformation is exact , in the limit @xmath78 , provided we impose the constraint @xmath93 on the system . hence , we find that the hamiltonian ( [ ham ] ) may be written as @xmath94 we make a mean - field approximation by replacing the bosons by the expectation value @xmath95 , which gives , @xmath96 we now assume that the bosonic mean field is homogeneous ( @xmath97 ) and introduce the lagrange multipliers , @xmath98 to enforce , on average , the local constraints [ via the term @xmath99 . thus one finds that @xmath100 upon performing a fourier transform one finds that there are two bands , which we denote the bonding ( - ) and antibonding ( + ) bands , @xmath101 \right\ } \\&\equiv & \sum_{{\bf k}\sigma}\left ( \begin{array}{c } \hat{\psi}_{{\bf k}+\sigma } \\ \hat{\psi}_{{\bf k}-\sigma } \end{array } \right)^{\dagger } \left ( \begin{array}{cc } e_{{\bf k}+\sigma } & 0 \\ 0 & e_{{\bf k}-\sigma } \end{array } \right)\left ( \begin{array}{c } \hat{\psi}_{{\bf k}+\sigma } \\ \hat{\psi}_{{\bf k}-\sigma } \end{array } \right ) \notag\\ & & + n\left\{\frac\delta2\left(q_{aa } -q_{bb } \right ) -\sum_{\nu}\left[\mu(1+q_{\nu\nu } ) + \lambda_{\nu}(1-q_{\nu\nu } ) \right ] \right\}\end{aligned}\ ] ] where @xmath102 and @xmath103 are defined in the reduced ( @xmath104 ) brillouin zone of the model ( fig . [ fig : unit - cells]e ) , @xmath105 is the number of unit cells , i.e. , half the number of lattice sites , and the dispersion relations , @xmath106 , of the ` quasiholes ' , which are destroyed ( created ) by @xmath107 , are given by @xmath108^{2}+16t^{2}|q_{ab}|^{2}\cos^{2}\frac{k_{x}}2\cos^{2}\frac{k_{y}}2}\end{aligned}\ ] ] where @xmath109 , @xmath110 , @xmath111 , and @xmath112 . hence we find that @xmath113^{2}+16t^{2}|q_{ab}|^{2}\cos^{2}\frac{k_{x}}2\cos^{2}\frac{k_{y}}2 } \right\}\hat{\psi}_{{\bf k}\alpha\sigma}^{\dagger}\hat{\psi}_{{\bf k}\alpha\sigma } \notag\\ & & + 2n\left[\left(\frac\delta2+\lambda_-\right)q_{- } - \mu(1+q_{+})-\lambda_+(1-q_{+ } ) \right],\end{aligned}\ ] ] where @xmath114 . it is interesting to compare this with the solution of the model with @xmath115 [ see appendix [ sect : non - int ] and particularly eq . ( [ eqn : dispersion ] ) ] . this enables a straightforward identification of the physical meaning of the mean fields @xmath116 and @xmath117 and the lagrange multipliers @xmath118 , see table [ table : meanings ] . [ cols="<,^,^",options="header " , ] [ table : meanings ] it follows from the the approximation that the bosonic mean - field is homogeneous and the constraint of one particle per site that , at three - quarters filling , we need only consider a two site model ( i.e. , one unit cell ) , which must contain exactly one boson . thus , the bosonic wavefunction may be written as @xmath119 where @xmath120 is a real number between 0 and 1 . thus @xmath121 therefore , the total energy of the model is @xmath122^{2}+16t^{2}(q_{+}^{2}-q_{-}^{2})\cos^{2}\frac{k_{x}}2\cos^{2}\frac{k_{y}}2 } \right\}n_{{\bf k}\alpha\sigma } \notag \\ & & + 2n\left[\left(\frac\delta2+\lambda_-\right)q_{- } - \mu(1+q_{+})-\lambda_+(1-q_{+ } ) \right ] , \label{eqn : free - energy}\end{aligned}\ ] ] where @xmath123 . the phase @xmath124 does not change the energy , thus , the solution has a u(1 ) degeneracy above that found in the slave boson mean field theory of the hubbard model@xcite due to the two sublattice structure of the problem . it is straightforward to show that @xmath125 follows from the requirement that @xmath126 , where the total number of electrons is @xmath127 for @xmath128coo@xmath129 and we have specialised to @xmath130 , which is the relevant filling . this result fits with our intuition that , on average , there should be one boson ( i.e. , one doubly occupied site ) per unit cell . similarly , @xmath131 follows from the requirement that @xmath132 is an extremum with respect to @xmath133 . again , this fits with our intuition that there is , on average , one quasihole fermion ( singly occupied site ) per unit cell . three more complicated intertwined self - consistency conditions can also be derived : [ eqn : selfconst ] @xmath1342t\cos k_x - 16t^2q_{-}\cos^{2}\frac{k_x}{2}\cos^{2}\frac{k_y}{2}}{\sqrt{\left[2tq_-\cos k_x + \lambda_- \right]^2 + 16t^2(q_{+}^{2}-q_{-}^{2})\cos^{2}\frac{k_x}{2}\cos^{2}\frac{k_y}{2 } } } \right\}n_{{\bf k}\alpha\sigma } \label{eqn : lambda_--self - const}\end{aligned}\ ] ] follows from the requirement that @xmath132 is a minimum with respect to @xmath135 ; @xmath136 ^ 2 + 16t^2(q_{+}^{2}-q_{-}^{2})\cos^{2}\frac{k_x}{2}\cos^{2}\frac{k_y}{2 } } } \right\}n_{{\bf k}\alpha\sigma } \label{eqn : q_--self - const}\end{aligned}\ ] ] follows from the requirement that @xmath132 is an extremum with respect to @xmath137 ; and @xmath138 ^ 2 + 16t^2(q_{+}^{2}-q_{-}^{2})\cos^{2}\frac{k_x}{2}\cos^{2}\frac{k_y}{2 } } } \right\}n_{{\bf k}\alpha\sigma } + \lambda_+ \notag\\\end{aligned}\ ] ] follows from the requirement that @xmath132 is a minimum with respect to @xmath139 . in order to compare our results with arpes experiments on na@xmath140coo@xmath40 and k@xmath140coo@xmath40 we calculate the spectral function , @xmath141 , where @xmath142 is the electronic propagator . in real space and time the one electron propagator is given by @xmath143 fourier transforming and taking the trace , one finds that @xmath144\delta(e_{{\bf k}\alpha\sigma}-\omega),\label{eqn : arpes}\end{aligned}\ ] ] where @xmath145 is the fermi function , @xmath146 is the quasiparticle weight , and @xmath147 therefore , we find that , in contrast to mean field slave boson theories of other models , the quasiparticle weight is momentum dependent in the theory of the ionic hubbard model . however , this does not mean that all non - local correlations induced by the on - site coulomb repulsion are included in the present approach . with any minimisation over a finite parameter space it is important to check explicitly whether any of the end points are the global minima.@xcite therefore we must consider the special case @xmath148 separately from the general set of self consistency conditions [ eqs . ( [ eqn : selfconst ] ) ] . let us consider the case of large ( how large will be determined later ) positive @xmath4 such that @xmath149 , recall that @xmath150 . ( [ eqn : early ] ) then yields @xmath151 \notag \\ & & -n\left(\frac\delta2 + 3\mu+\lambda_a\right).\end{aligned}\ ] ] therefore @xmath152 , as is required to ensure three electrons per unit cell . further , one finds that @xmath153 for all @xmath154 and @xmath17 from the requirement that @xmath132 is an extremum with respect to @xmath155 , which we expect from @xmath156 and the constraint of one particle per cite . finally , @xmath157 due to the requirement that @xmath132 is an extremum with respect to @xmath158 , which is simply the expected constraint that the a sublattice is half filled with quasiholes . note , however , that because of the infinite @xmath13 or , equivalently , the constraint of one particle per site , the half filled a sublattice is insulating . this is also reflected in the fact that @xmath159 , i.e. there is zero quasiparticle weight for fermions on the a sublattice . so the state @xmath160 , @xmath156 ( or , equivalently , @xmath150 , @xmath149 ) describes a charge transfer insulator , which has the characteristics of both mott ( a sublattice ) and band ( b sublattice ) insulators . thus , we find that , @xmath161 which is clearly just the classical energy of paramagnetic insulating state . as discussed in section [ sect : natheory ] , choy _ et al_.@xcite have studied an empirical model hamiltonian for holes and doublons motivated by the presumed low energy processes about an antiferromagnetically ordered charge transfer insulating state . this empirical model gives good agreement with the measured temperature dependence of both the resistivity and the hall coefficient . it is therefore interesting to compare our eq . ( [ eqn : effectiveh ] ) with the hamiltonian studied by choy _ et al_. [ their eq . ( 2 ) ] . upon making the following notational changes to and substitutions into eq . ( [ eqn : effectiveh ] ) : @xmath162 ; @xmath163 ; @xmath164 ; @xmath165 ; @xmath166 ; @xmath167 ; @xmath168 ; @xmath169 and neglecting constant terms one finds that the two hamiltonians are identical . notice , in particular , that the renormalised ionic potential , @xmath137 , in our theory plays the role of the inter - site interactions , @xmath170 , in that of choy _ et al _ .. also their non - magnetic regime ( @xmath171 ) corresponds to @xmath172 in our slave boson theory . the ground state studied by choy _ et al_. corresponds to the charge transfer insulator we only find for @xmath173 . thus , at least in this regime , the two theories appear to be equivalent . in order to get agreement with the experimentally measured temperature dependence of the resistivity and the hall coefficient choy _ et al_. require a specific temperature dependence of ( in our notation ) @xmath174 , @xmath175 , and @xmath137 . although we will not present finite temperature calculations below , such a temperature dependence is a quite reasonable expectation for our mean field theory . indeed slave boson mean - field theories of other strongly correlated models have boson mean - fields , similar to @xmath176 , that decrease monotonically with increasing temperature and vanish at some coherence temperature , @xmath177 , the fermi temperature of the non - interacting system.@xcite we have solved the self consistent equations ( [ eqn : selfconst ] ) on an @xmath178 reciprocal space mesh for @xmath179 and 1000 . for the @xmath180 solutions we also tightened the convergence criteria by a factor of 50 . we find that the solutions are well converged as the difference between the @xmath179 and @xmath180 self consistent solutions is no larger than a few parts in a thousand for any of the parameters . by way of an example , in fig . [ fig : q_- ] we report both sets of results , it can be seen the difference in the results is significantly smaller than the width of the lines . therefore , we report only the @xmath180 results in the subsequent figures . in fig . [ fig : q_- ] we plot the real space charge disproportionation , @xmath181 , where @xmath182 is the average number of _ electrons _ per sublattice site , against . it can be seen that the results are strongly dependent on the sign of @xmath0 . the system is metallic for @xmath183 ; @xmath184 corresponds to exactly one fermion on each site of the a - sublattice and exactly one boson on each site of the b - sublattice . the metal - insulator transition ( mit ) occurs at @xmath6 for @xmath3 and @xmath5 for @xmath1 . we also plot the reciprocal space charge disproportionation ( band filling ) in fig . [ fig : q_- ] . for @xmath1 the bonding band is filled for all non - zero @xmath185 ( for @xmath66 the unit cell is halved so there is only a single band ) . for @xmath3 the bonding band is only partially filled for @xmath186 . as we will discuss in detail below , the quasiparticles in the bonding ( - ) band are significantly lighter than those in the , nearly half - filled , antibonding ( + ) band . it is interesting to observe therefore that for @xmath3 , @xmath187 and @xmath1 , @xmath188 the antibonding band is half - filled and , even though @xmath89 , the system remains metallic . at first sight this is rather surprising as the system is two - dimensional and even quasi - one - dimensional . however , because the bands arise from the hybridisation of the a and the b sublattices the usual real space arguments that explain the mott insulating state do not apply here . there is no interaction between the quasiparticles other than the real space constraint of one fermion ( or boson ) per site . therefore , unless there is complete charge disproportionation the system remains metallic . it is interesting to contrast the results of the slave boson calculations with our previous exact diagonalisation results for finite clusters with @xmath82.@xcite at least at low , the slave boson theory gives excellent agreement with the exact diagonalisation for some physical quantities , e.g. , the level of charge disproportionation ( cf . [ fig : q_- ] ) . but , there is an important qualitative difference between the slave boson calculations and the exact diagonalisation results : the nature of the insulating state . in the slave boson calculations an insulating state is only realised when there is complete charge disproportionation leaving one sublattice completely filled and the other half - filled . the filled sublattice acts as a band insulator and , because @xmath89 , the half filled sublattice becomes a mott insulator . this charge transfer insulator is very different from the covalent insulating state predicted from the exact diagonalisation calculations.@xcite these calculations predict that the insulating state occurs at quite small values of in spite of there being only rather weak charge disproportionation . this insulating state depends crucially on the hybridisation between the two sublattices and is analogous to a covalent insulator.@xcite our mean field theory neglects non - local correlations that are included in the exact diagonalisation calculations and cause the covalent insulating state . preliminary results suggest that dmft , which also neglects non - local correlations but is more sophisticated than our mean field theory , can not describe the covalent insulator phase either . in figs . [ fig : f - t<0 ] and [ fig : f - t>0 ] we plot the ground state energy , @xmath132 , as a function of @xmath135 for various values of and the self consistently calculated curve of @xmath132 against @xmath135 with as a parametric parameter . it can be seen from these plots that the self consistent solutions are the true minima . further , the absence of double well structures in the plots of @xmath132 against @xmath135 strongly suggest that the mit is a second order phase transition in the slave boson theory . we now turn to analyse the properties of the fermionic quasiholes . as these are non - interacting particles one can straightforwardly calculate the properties of the quasiholes . it is interesting to compare these results with the band structure of the non - interacting ( @xmath115 ) ionic hubbard model , which we discuss in appendix [ sect : non - int ] . before discussing the band structure in detail we should point out that the bands are , to some extent , an artefact of the mean field approximation . in the exact solution correlations that are not captured by the mean field approximation may destroy the description of the system in terms of extended bands ( cf . ref . ) nevertheless , if the mean field theory does provide a reasonable description of the ionic hubbard model then one expects that some of the features of the band structure described below do survive in the exact solution . further , the bands are well defined in the mean - field slave - boson theory . therefore , it is legitimate to calculate their properties in this context and this is extremely useful for building intuition about this problem . figs . [ fig : bs - t<0 ] and [ fig : bs - t>0 ] show how the dispersion relations of the fermionic quasiholes vary with for each sign of @xmath0 ( also see the animations online@xcite ) . these should be compared with the equivalent plots ( figs . [ bs - lt ] and [ bs - gt ] ) for the non - interacting system . it can be seen that at @xmath66 , for both signs of @xmath0 , the quasiparticle bands are significantly narrower than the equivalent non - interacting electron bands . as increases the antibonding band narrows and eventually becomes flat at the mit . while the bonding band broadens somewhat as increases these conclusions are consistent with changes observed in the density of states , which we plot in figs . [ fig : dos - t<0 ] and [ fig : dos - t>0 ] ( also see the animations online@xcite ) , calculated by treating the fermions as non - interacting particles . these figures should be contrasted with the equivalent plots of the non - interacting system ( figs . [ dos - lt ] and [ dos - gt ] ) . for both signs of @xmath0 as is increased the antibonding band narrows into the quasiparticle peak , familiar from the mott - hubbard transition.@xcite note that , as the fermions are hole - like , the states with @xmath189 are filled and the states with @xmath190 are unoccupied . + + + + we plot the the fermi surfaces of the quasiholes in figs . [ fig : fs - t<0 ] and [ fig : fs - t>0 ] . both signs of @xmath0 exhibit strong nesting , suggesting that there may be magnetic instabilities , which we do not consider here . for @xmath3 and @xmath186 there are small two - dimensional fermi pockets of light quasiholes arising from the bonding band , as well as the quasi - one - dimensional sheets arising form the antibonding band . for @xmath191 the small fermi pocket occupies @xmath192 of the brillouin zone . as increases the fermi pocket shrinks and it vanishes at @xmath193 . in the materials the unit cells are twice the size of that of our model . this results in a zone folding of the brillouin zone in the @xmath8-direction , cf . [ fig : unit - cells ] . this will result in the formation of additional small pockets around the @xmath58 and @xmath194 points ( of the brillouin zone , fig . [ fig : unit - cells]f ) , which are formed from fermions in the antibonding band . if one simply folds our results into the brillouin zone the pockets from the bonding band occupy 1.4% of the brillouin zone for @xmath191 ( as the brillouin zone has halved in size , while the pocket area remains unchanged ) and the pockets around the @xmath58 and @xmath194 points arising from the antibonding band occupy 8.6% and 9.9% of the brillouin zone respectively . the combination of small and large @xmath80 significantly changes the dispersion relations of the quasiparticles from those of the non - interacting electrons , compare section [ sect : bs - quasiholes ] with appendix [ sect : non - int ] and particularly figs . [ fig : bs - t<0 ] and [ fig : bs - t>0 ] with figs . [ bs - lt ] and [ bs - gt ] . first , the antibonding band has a much weaker dispersion than for @xmath115 than for @xmath89 . this is seen in arpes experiments on and , where the measured fermi velocity is several times smaller than the value calculated from the lda with no sodium ion ordering ( cf . 2i of ref . ) . as @xmath4 increases weight gets shifted from the antibonding to the bonding band . we plot two different reciprocal space cuts of the calculated spectral function for representative values of and @xmath3 in figs . [ fig : arpes - gm ] and [ fig : arpes - gk ] ( also see the animations online@xcite ) . as shown in fig . [ fig : unit - cells ] , the unit cells of @xmath195coo@xmath40 are twice as large as the unit cell of hamiltonian ( [ ham ] ) . in order to facilitate comparison with experiment , we have plotted these results in the brillouin zone of na@xmath140coo@xmath40 by ` folding over ' the results for hamiltonian ( [ ham ] ) into the brillouin zone of the actual materials . at @xmath196 eq . ( [ eqn : arpes ] ) gives a plane of @xmath197-functions ( in the three dimensional @xmath154 , @xmath198 space ) . it is well known that @xmath199}{2\sqrt{\pi b}}.\label{eqn : broadening}\ ] ] therefore we have broadened the results by replacing the @xmath197-function with a gaussian with @xmath200 . we report our results with the fermi factor appropriate for @xmath201 k for @xmath202 ev . @xmath203 @xmath204 @xmath205 + + @xmath206 @xmath207 + @xmath203 @xmath204 @xmath205 + + @xmath206 @xmath207 + it can be seen from figs . [ fig : arpes - gm ] and [ fig : arpes - gk ] that as is increased the antibonding band remains at the fermi energy , but loses spectral weight , consistent with the calculated density of states for fermions ( figs . [ fig : dos - t<0 ] and [ fig : dos - t>0 ] ) . meanwhile , the bonding band is shifted down in energy . in the @xmath208-m cut ( fig . [ fig : arpes - gm ] ) we see that the antibonding band narrows and loses intensity as is increased . in the @xmath208-k cut ( fig . [ fig : arpes - gk ] ) we see that the bonding band moves from the fermi level at @xmath66 to lower energies such that it is completely filled at larger values of . at first sight fig . [ fig : arpes - gk ] might appear to suggest that the bonding band also narrows as increases , however it is clear from fig . [ fig : arpes - gm ] that this is not the case . rather , as increases the system becomes more one dimensional ( for @xmath209 , the system consists of uncoupled one - dimensional chains ) , increasing the range of @xmath210 in the cuts shown in fig . [ fig : arpes - gm ] and decreasing the range of @xmath210 in the cuts shown in fig . [ fig : arpes - gk ] ( cf . figs . [ fig : bs - t<0 ] and [ fig : fs - t<0 ] ) . as we are considering a quasi - two - dimensional metal the cyclotron effective mass , @xmath211 , associated with shubnikov - de haas oscillations of the part of the fermi surface arising from the @xmath212 band is@xcite @xmath213 where @xmath214 is the renormalised density of states per spin at the fermi energy , @xmath215 , arising from the @xmath212 band . we plot the variation of the cyclotron mass with for parameters relevant to , i.e. , @xmath216 ev , a basal lattice constant , @xmath217 , and a unit cell basal area of @xmath218 , in fig . [ fig : cyclo ] . the cyclotron mass for the bonding band associated with the small pockets is about one quarter of that for heavy electrons associated with the antibonding band . in order to observe the states arising from the antibonding band , i.e. , the quasi-1d sheets on the fermi surface , in quantum oscillation experiments a field that is sufficiently large to cause magnetic breakdown between the two bands must be applied . the field required for magnetic breakdown and the large effective mass associated the antibonding band make it more difficult to observe the breakdown orbit than than the small pockets in quantum oscillation experiments . for realistic parameters for , @xmath219 and @xmath3 ( cf . section [ sect : natheory ] ) , the slave boson theory presented above suggests that the system is metallic . however , an important question , which we can not conclusively answer because of the uncertainty in the precise value of relevant to , is : is large enough in that the bonding band is completely filled ? for @xmath186 there are two features in the fermi surface ( cf . [ fig : fs - t<0 ] ) : small hole - like pockets of light fermions arising from the bonding ( nearly filled ) band , and quasi - one - dimensional sheets of heavier electron - like quasiparticles . this seems to be quite consistent with quantum oscillations,@xcite which are observed to have a frequency correspond to orbits that encompass less than 1% of the first brillouin zone , in apparent violation of luttinger s theorem . a natural explanation of this experiment is that balicas _ et al_. @xcite are only able to observe the , small , light , quasiparticles in the bonding band and do not see oscillations arising from the antibonding band , either because the quasiparticles in the antibonding band are two heavy , or because the field they applied is insufficient to cause magnetic breakdown of the ` gap ' between the quasi - one - dimensional sheets and the two - dimensional pockets . however , for @xmath220 the small fermi pockets are absent as the bonding band is filled , and our results would be inconsistent with the quantum oscillations observed by balicas _ et al_. the experimental situation , particularly in na@xmath140coo@xmath40 , is somewhat confusing . several experiments suggest an insulating state : resistivity versus temperature has a negative gradient at low temperatures,@xcite and angle resolved photoemission spectroscopy ( arpes)@xcite and optical conductivity@xcite both observe a gap , all of which suggests that below 51 k na@xmath140coo@xmath40 is a insulator . however , quantum oscillations are observed,@xcite which suggest that there are quasiparticles and that the system is metallic with a small fermi surface ( occupying less that 1% of the cross - sectional area of the first brillouin zone ) and an effective mass similar to the free electron mass . below we propose a picture , inspired by both the mean field slave boson theory ( above ) and our previous exact diagonalisation calculations for the same ionic hubbard model,@xcite that offers a possible reconciliation of these seemly contradictory experiments . however , it is important to stress that this is a tentative proposal and does not correspond to the solution found in either theory . nevertheless , we will argue , below , that this picture is plausible given the limitations of the mean field slave boson theory and the finite size effects in the exact diagonalization calculations . further calculations will be required to discover whether this proposal is indeed supported by the theory of the ionic hubbard model . the fundamental definition of a metal is a substance with charge carrying excitations at arbitrarily low energies above the ground state . in most situations it follows that the resistivity monotonically increases with temperature as increasing the temperature decreases the quasiparticle lifetime . however , this is not necessarily the case . for example , in many strongly correlated systems , such as the organic superconductors,@xcite the charge transport is incoherent at high temperatures - this leads to a resistivity with a broad maximum slightly above the temperature where quasiparticle coherence is lost.@xcite above , we have shown that the slave boson theory of hamiltonian ( [ ham ] ) produces two bands . we will assume some features of these bands survive even in the exact solution even if the simple single particle states , of the slave boson theory , do not . the antibonding band is narrow and it is likely that coherent quasiparticles do not emerge from these states until quite low temperatures . within slave boson theories the coherence temperature can be estimated from the bose condensation temperature , @xmath221 of the slave bosons,@xcite which , in 2d , is proportional to the density of bosons , @xmath222 . for @xmath191 there are 0.993 quasiholes per unit cell in the antibonding band , corresponding to @xmath223 , and the density of bosons decreases as is increased , going to zero at = 1.5 . estimating the bose condensation temperature ( cf . refs . and ) by either @xmath224 , where @xmath225 is the basal plane lattice constant ( 2.82 for ) and @xmath226 is the mass of the boson ( which is the simplest approximation given that the boson corresponds to two electrons ) , or @xmath227 yields @xmath228 k for @xmath223 . however it is well known that these simple formulae dramatically overestimate the coherence temperature as they neglect gauge fluctuations.@xcite thus the coherence temperature ( @xmath229 k ) is expected to be extremely low in the experimentally relevant parameter regime . this low coherence temperature is consistent with the observed incoherent transport evident from the resistivity above @xmath23051 k. as the temperature is lowered below 51 k the resistivity rises rapidly , suggesting that a gap opens . the slave boson calculations do not predict a true gap unless @xmath173 . experiments observe magnetic ordering,@xcite which we have not considered here , below 88 k , but , as the gap does not appear to open until 51 k , one can not conclude that the magnetic order causes the gap to open . however , previous exact diagonalisation calculations@xcite suggest that in the experimentally relevant parameter regime , @xmath27 , non - local correlations , not captured by our mean field treatment , drive the formation of a covalent insulator . we propose that a gap opens on the antibonding band at 51 k due to non - local correlations , analogous to the covalent insulator . but , we also propose that the bonding band remains ungapped . this proposal is rather similar to the theory of the orbital - selective mott transition . the question of whether , in multi - band systems , one band can becoming insulating , due to electronic correlations , while others remain metallic has been extensively discussed recently,@xcite motivated , in part , by the volume collapse in ce ( ref . ) , the heavy fermion behaviour of ca@xmath231sr@xmath232ruo@xmath233 ( ref . ) and the nodal anti - nodal differentiation in the cuprates.@xcite although the debate is ongoing , the current theoretical consensus is that orbital - selective mott transitions can occur for some parameter regimes of some models.@xcite therefore , testing our proposed picture will require calculations beyond the mean - field slave - boson theory presented above . given the important differences between the current mean field calculations and our previous exact diagonalisation calculations , particularly the critical value of for the metal - insulator transition , it is important to ask , whether this explanation is likely to hold in the exact solution and , more importantly , for the real material . firstly , we note that the real space charge disproportionation calculated from our mean field theory is in very good agreement with exact diagonalisation calculations in the experimentally relevant regime . this gives one hope that the filling of the bands calculated from the mean field theory may also be reasonable ( although non - local coulomb correlations could cause additional hybridisation effects , cf . ref . ) . secondly , we note that the finite size clusters that were , necessarily , studied in the exact diagonalisation calculations mean that it is extremely difficult to accurately study the bonding band . taking a , typical , 18 site ( 9 unit cell ) cluster , a density of 0.007 holes per unit cell means that one would expect less than 1/16@xmath234 of a hole in the bonding band in the exact diagonalisation calculation . however , exact diagonalisation calculations do accurately reproduce the band filling of the analytical exact solution for @xmath115.@xcite therefore , it is not clear , at present , whether the exact diagonalisation calculations can resolve the tiny hole pockets suggested by our calculations and the quantum oscillation experiments . as the exact diagonalisation calculations do not rule out the possibility that small weakly correlated fermi pockets containing a tiny number of charge carries remain even in the low temperature ` insulating ' state , let us briefly discuss how this would produce a consistent explanation of the experiments on . it is important to note that the two bands act as parallel channels for charge transport . above 51 k charge transport in the antibonding band will be incoherent ( assuming the coherence temperature @xmath235 k , as we have argued above ) and the resistivity in this channel can be estimated to be @xmath236 , where @xmath237 is the interlayer spacing , i.e. , at or above the mott - ioffe - regel limit . below 51 k a gap opens on the antibonding band , but as the gap is small , @xmath238 ( ref . ) the resistivity in the antibonding channel is not increased by orders of magnitude from the resistivity value in the incoherent metallic state . however , because of the extremely low charge carrier concentration in the bonding band ( metallic channel ) the contribution of the bonding band to the conductivity will be even smaller than that of the antibonding band ( incoherent / insulating channel ) . therefore the metallic channel will be unable to ` short circuit ' the incoherent / insulating channel . this contention is supported by the fact that even above 51 k the resistivity is larger than the mott - ioffe - regel limit . thus the observation that the resistivity decreases with increasing temperature is not necessarily inconsistent with the observation of quasiparticles in quantum oscillation experiments . foo _ et al_. @xcite found that the conductivity , @xmath17 , of na@xmath239coo@xmath40 is very similar to that of na@xmath240coo@xmath40 and that both of these are an order of magnitude larger than the conductivity of . as @xmath241 , one expects that the contribution of the electrons in the bonding band to the conductivity of will be two orders of magnitude smaller than the total conductivity of na@xmath239coo@xmath40 or na@xmath240coo@xmath40 . therefore foo _ et al_. s results are consistent with the , counterintuitive , picture that the insulating band short - circuits the metallic band ! arpes , which measures @xmath242 , has been reported in the directions @xmath208-m and @xmath208-k , i.e. , in the directions of the cuts shown in figs . [ fig : arpes - gm ] and [ fig : arpes - gk ] respectively . however , our slave boson calculations do not capture the gap on the antibonding band required in our proposal . therefore the gap observed in the arpes is likely to correspond to the ` covalent insulator ' gap on the antibonding band . optical conductivity does not observe a drude peak , however the height of the drude peak is proportional to the charge carrier density , which would make the tiny fermi pockets suggested by our slave boson calculations difficult to observe . as one expects that the fermi pockets do not contribute significantly to either the resistivity or the hall coefficient the correspondence of our mean - field slave boson theory to the empirical low energy hamiltonian studied by choy _ et al_. @xcite ( discussed in section [ sect : choy ] ) become important . the non - local correlations on the antibonding band open up a gap on the quasi - one - dimensional sheets of the fermi surface . this provides the gap assumed by choy _ et al_. and suggests that their results will carry across to our model . this suggests that the measured temperature dependence of both the resistivity and the hall coefficient in have a natural explanation in terms of the ionic hubbard model . however , clearly further calculations are require to test the validity of this _ ad hoc _ proposal and its relevance to . we have presented a mean - field slave boson theory of the ionic hubbard model on the triangular lattice with alternating stripes of site energy . this model has two bands : one of which is weakly correlated and nearly filled or filled , the other is nearly half - filled or half - filled and hence strongly correlated . the results depend strongly on the sign of @xmath0 . the light band is always filled for @xmath1 , but only becomes filled at @xmath193 for @xmath3 . a metal - charge transfer insulator transition occurs at larger ( 5.0 for @xmath1 and 8.0 for @xmath3 ) , when complete charge disproportionation occurs and one sublattice is filled and the other in half filled . we have also proposed a speculative picture of . in particular , we have argued that the observed quantum oscillations arise from quasiparticles in the bonding band , but these are not seen in the zero field resistivity and several other experiments because the low density of charge carriers in the bonding band mean that the metallic bonding band is not able to ` short circuit ' the incoherent / insulating antibonding band . calculations beyond the mean field slave boson calculations , which we have presented above , will be required to test this picture . another important avenue for future work will be to allow for magnetic ordering in the theory as this may improve the degree of agreement with experiments and allow one to compare with other experiments such as nmr@xcite and neutron scattering.@xcite we thank h. alloul , j. bobroff , m .- h . julien , and r. r. p. singh for helpful discussions . numerical calculations were performed on the australian partnership for advanced computing national facility under the a grant from the merit allocation scheme . this work was supported by the australian research council ( arc ) under the discovery program ( project numbers dp0557532 and dp0878523 ) , mec ( project ctq2005 - 09385- c03 - 03 ) and micinn ( project ctq2008 - 06720-c02 - 02 ) . b. j. p. was supported by the arc under the queen elizabeth ii scheme ( dp0878523 ) . j. m. acknowledges financial support from the ramn y cajal program . r.h.m . was supported by the arc under the apf scheme ( dp0877875 ) . for @xmath115 there are two bands , bonding ( - ) and antibonding ( + ) , and whose dispersion relations are @xmath243 where @xmath102 and @xmath103 are defined in the reduced ( @xmath244 ) brillouin zone of hamiltonian ( [ ham ] ) , shown in fig . [ fig : unit - cells]e . these dispersion relations are plotted in figs . [ bs - lt ] and [ bs - gt ] for @xmath3 and @xmath1 respectively . to calculate where the bonding band becomes filled we need to consider the maximum of @xmath245 . this is complicated by the fact the chemical potential , @xmath44 , is a function of @xmath4 . for @xmath1 the top of the bonding band is at @xmath246 . thus we find that the bonding band become full at @xmath247 . however , for @xmath248 or @xmath249 and @xmath1 one finds that @xmath250 and this condition is satisfied . therefore the bonding band is filled , and hence the antibonding band is half filled , for an infinitesimal @xmath4 . note that this is not the case for @xmath66 as we require at least an infinitesimal @xmath4 to double the size of the unit cell and cause the splitting into two bands , cf . figs . [ fig : unit - cells]a and [ fig : unit - cells]b . for @xmath3 the top of the bonding band is @xmath251 and we find the bonding band is filled for @xmath252 . solving this condition numerically we find that @xmath253 . in figs . [ fig : ni - fs - t<0 ] and [ fig : ni - fs - t>0 ] we plot the fermi surfaces for non - interacting electrons for each sign of @xmath0 at a range of . the first point to note is that at low the two fermi surfaces are very different . as @xmath254 both fermi surfaces become quasi - one - dimensional ; but , even in this limit the two fermi surfaces are different in important ways as for @xmath3 the fermi surface is electron like , while of @xmath1 the fermi surface is hole like . only when we reach the limit , and both fermi surfaces are straight lines , are the two fermi surfaces the same . we plot the dos of the non - interacting ( @xmath115 ) system in figs . [ dos - lt ] and [ dos - gt ] for @xmath3 and @xmath1 respectively . notice that the plots are very different for the two different signs of @xmath0 . in particular at @xmath66 the fermi energy is at the van hove singularity for @xmath1 , but is not for @xmath3 . this suggests that the fermi liquid will be less stable for @xmath1 than for @xmath3 . indeed we find that , in the slave boson theory , the mit occurs at @xmath5 for @xmath1 and @xmath6 for @xmath3 , consistent with this expectation ( cf . [ fig : q_- ] ) . however , as is increased the van hove singularity is moved away from the fermi energy even for @xmath1 , so this , weak coupling , description is clearly not the whole story of the strong coupling theory . m. z. hasan , y .- d . chuang , d. qian , y. w. li , y. kong , a. p. kuprin , a. v. fedorov , r. kimmerling , e. rotenberg , k. rossnagel , z. hussain , h. koh , n. s. rogado , m. l. foo , and r. j. cava , phys . . lett . * 92 * , 246402 ( 2004 ) ; h .- b . yang , s .- c . wang , a. k. p. sekharan , h. matsui , s. souma , t. sato , t. takahashi , t. takeuchi , j. c. campuzano , r. jin , b. c. sales , d. mandrus , z. wang , and h. ding , _ ibid_. * 92 * , 246403 ( 2004 ) . m. roger , d. j. p. morris , d. a. tennant , m. j. gutmann , j. p. goff , j. u. hoffmann , r. feyerherm , e. dudzik , d. prabhakaran , a. t. boothroyd , n. shannon , b. lake , and p. p. deen , nature * 445 * , 631 ( 2007 ) . a. p. kampf , m. sekania , g. i. japaridze , and ph . brune , j. phys . : condens . matter * 15 * 5895 ( 2003 ) ; also see references therein and citations thereof for further details of the extensive literature on the ionic hubbard model in one - dimension . the dependence on the direction of the field may be explained by ( i ) the larger g - factor for fields parallel to the layers , and/or ( ii ) that the transitions at 51 k and 88 k have a larger effect on the bulk magnetic susceptibility parallel to the layers , @xmath255 than on @xmath256 ( see fig . 1(c ) in ref . and fig . 8 in ref . ) . animated versions of figs . [ fig : bs - t<0 ] , [ fig : bs - t>0 ] , [ fig : dos - t<0 ] , [ fig : dos - t>0 ] , [ fig : arpes - gm ] , and [ fig : arpes - gk ] can be viewed , respectively , at http://www.youtube.com/watch?v=8l8yvfhhs74 , http://www.youtube.com/watch?v=oqbfseimi3s , http://www.youtube.com/watch?v=q5bmss3vnaq , http://www.youtube.com/watch?v=fz9xkblzbke , http://www.youtube.com/watch?v=ejnveweo0oa , and http://www.youtube.com/watch?v=p2ytkyuzfca . the animations show the same data as is reported in the associated figure ; each frame of the movie represents a different value of , specifically each subsequent frame represents an increment of 0.1 in . a. liebsch , europhys . * 63 * , 97 ( 2003 ) ; phys . . lett . * 91 * , 226401 ( 2003 ) ; phys . b * 70 * , 165103 ( 2004 ) ; phys . rev . lett . * 95 * , 116402 ( 2005 ) ; a. koga , n. kawakami , t. m. rice , and m. sigrist , phys . lett . * 92 * , 216402 ( 2004 ) ; r. arita and k. held , phys . b * 72 * , 201102(r ) ( 2005 ) ; m. ferrero et al . b * 72 * , 205126 ( 2005 ) ; l. demedici , a. georges , and s. biermann , phys . rev . b * 72 * , 205124 ( 2005 ) ; m. ferrero , f. becca , m. fabrizio , and m. capone , phys . rev . b * 72 * , 205126 ( 2005 ) ; a. regg , m. inderland , s. pilgram , and m. sigrist , eur . phys . j. b. * 48 * , 55 ( 2005 ) ; k. inaba and a. koga , phys . rev . b * 73 * , 155106 ( 2006 ) ; p. g. j. van dongen , c. knecht , and n. blmer , phys . status solidi ( b ) * 243 * , 116 ( 2006 ) ; t. a. costi and a. liebsch , phys . lett . * 99 * , 236404 ( 2007 ) ; n. blmer et al . , j. magn . magn . mater . * 310 * , 922 ( 2007 ) ; k. bouadim , g. g. batrouni , and r. t. scalettar , , 226402 ( 2009 ) .

we use tools from geometric statistics to analyze the usual estimation procedure of a template shape . this applies to shapes from landmarks , curves , surfaces , images etc . we demonstrate the asymptotic bias of the template shape estimation using the stratified geometry of the shape space . we give a taylor expansion of the bias with respect to a parameter describing the measurement error on the data . we propose two bootstrap procedures that quantify the bias and correct it , if needed . they are applicable for any type of shape data . we give a rule of thumb to provide intuition on whether the bias has to be corrected . this exhibits the parameters that control the bias magnitude . we illustrate our results on simulated and real shape data . shape , template , quotient space , manifold

1610.01502

the shape of a set of points , the shape of a signal , the shape of a surface , or the shapes in an image can be defined as follows : the remainder after we have filtered out the position and the orientation of the object @xcite . statistics on shapes appear in many fields . paleontologists combine shape analysis of monkey skulls with ecological and biogeographic data to understand how the _ skull shapes _ have changed in space and time during evolution @xcite . molecular biologists study how _ shapes of proteins _ are related to their function . statistics on misfolding of proteins is used to understand diseases , like parkinson s disease @xcite . orthopaedic surgeons analyze _ bones shapes _ for surgical pre - planning @xcite . in signal processing , the _ shape of neural spike trains _ correlates with arm movement @xcite . in computer vision , classifying _ shapes of handwritten digits _ enables automatic reading of texts @xcite . in medical imaging and more precisely in neuroimaging , studying _ brain shapes _ as they appear in the mris facilitates discoveries on diseases , like alzheimer @xcite . what do these applications have in common ? position and orientation of the skulls , proteins , bones , neural spike trains , handwritten digits or brains do not matter for the study s goal : only _ shapes _ matter . mathematically , the study analyses the statistical distributions of _ the equivalence classes of the data _ under translations and rotations . they project the data in a quotient space , called the _ shape space_. the simplest - and most widely used - method for summarizing shapes is the computation of the mean shape . almost all neuroimaging studies start with the computation of the mean brain shape @xcite for example . one refers to the mean shape with different terms depending on the field : mean configuration , mean pattern , template , atlas , etc . the mean shape is an average of _ equivalence classes of the data _ : one computes the mean after projection of the data in the shape space . one may wonder if the projection biases the statistical procedure . this is a legitimate question as any bias introduced with this step would make the conclusions of the study less accurate . if the mean brain shape is biased , then neuroimaging s inferences on brain diseases will be too . this paper shows that a bias is indeed introduced for the mean shape estimation under certain conditions . we review works on the shape space s geometry as a quotient space , and existing results on the mean shape s bias . * shapes of landmarks : kendall analyses * the theory for shapes _ of landmarks _ is introduced by kendall in the 1980 s @xcite . he considers shapes of @xmath1 labeled landmarks in @xmath2 . the size - and - shape space , written @xmath3 , takes also into account the overall size of the landmarks set . the shape space , written @xmath4 , quotients by the size as well . both @xmath3 and @xmath4 have a riemannian geometry , whose metrics are given in @xcite . these studies model the probability distribution of the data directly in the shape space @xmath4 . they do not consider that the data are observed in the space of landmarks @xmath5 and projected in the shape space @xmath4 . the question of the bias is not raised . * shapes of landmarks : procrustean analyses * procrustean analysis is related to kendall shape spaces but it also considers shapes of landmarks @xcite . kendall analyses project the data in the shape space by explicitly computing their coordinates in @xmath4 . in contrast , procrustean analyses keep the coordinates in @xmath5 : they project the data in the shape space by `` aligning '' or `` registering '' them . orthogonal procrustes analysis `` aligns '' the sets of landmarks by rotating each set to minimize the euclidean distance to the other sets . procrustean analysis considers the fact that the data are observed in the space @xmath5 but does not consider the geometry of the shape space . the bias on the mean shape is shown in @xcite with a reducto ad absurdum proof . but there is no geometric intuition given about how to control or correct the phenomenon . * shapes of curves * the curve data are projected in their shape space by an alignment step @xcite , in the spirit of a procrustean analysis . the bias of the mean shape is discussed in the literature . unbiasedness was shown for shapes of signals in @xcite but under the simplifying assumption of no measurement error on the data . some authors provide examples of bias when there is measurement error @xcite . their experiments show that the mean signal shape may converge to pure noise when the measurement error on simulated signals increases . the bias is proven in @xcite for curves estimated from a finite number of points in the presence of error . but again , no geometric intuition nor correction strategy is given . we are missing a global geometric understanding of the bias . which variables control its magnitude ? is it restricted to the mean shape or does it appear for other statistical analyses ? how important is it in practice : do we even need to correct it ? if so , how can we correct it ? our paper is addressing these questions . we use a geometric framework that unifies the cases of landmarks , curves , images etc . [ [ contributions ] ] contributions + + + + + + + + + + + + + we make three contributions . first , we show that statistics on shapes are biased when the data are measured with error . we explicitly compute the bias in the case of the mean shape . second , we offer an interpretation of the bias through the geometry of the shape space . in applications , this aids in deciding when the bias can be neglected in contrast with situations when it must be corrected . third , we leverage our understanding to suggest several correction approaches . [ [ outline ] ] outline + + + + + + + the paper has four sections . section 1 introduces the geometric framework of shape spaces . section 2 presents our first two contributions : the proof and geometric interpretation of the bias . section 3 describes our third contribution : the procedures to correct the bias . section 4 validates and illustrates our results on synthetic and real data . we introduce two simple examples of shape spaces . we will refer to them constantly to provide intuition . first , we consider two landmarks in the plane @xmath6 ( figure [ fig : simple ] ( a ) ) . the landmarks are parameterized each with 2 coordinates . for simplicity we consider that one landmark is fixed at the origin on @xmath6 . thus the system is now parameterized by the 2 coordinates of the second landmark only , e.g. in polar coordinates @xmath7 . we are interested in the shape of the 2 landmarks , i.e. in their distance which is simply @xmath8 . second , we consider two landmarks on the sphere @xmath9 ( figure [ fig : simple ] ( b ) ) . one of the landmark is fixed at the origin of @xmath9 . the system is now parameterized by the 2 coordinates of the second landmark only , i.e. @xmath10 . the shape of the two landmarks is the angle between them and is simply @xmath11 . the data are objects @xmath12 that are either sets of landmarks , curves , images , etc . we consider that each object @xmath13 is a point in a riemannian manifold @xmath14 . we restrict in this paper to finite dimensional manifolds in order to avoid complications . we have @xmath15 in the plane example : a flat manifold of dimension 2 . we have @xmath16 in the sphere example : a manifold of constant positive curvature and of dimension 2 . by definition , the objects shapes are their equivalence classes @xmath17\}_{i=1}^n$ ] under the action of some finite dimensional lie group @xmath18 : @xmath18 is a group of continuous transformations that models what does not change the shape . the action of @xmath18 on @xmath14 will be written with `` @xmath19 '' . in our examples , the rotations are the transformations that leave the shape of the systems invariant . let us take @xmath20 a rotation . the action of @xmath20 on the landmark @xmath21 is illustrated by a blue arrow in figures [ fig : leadingcases1 ] ( a ) for the plane and ( d ) for the sphere . we observe that the action does not change the shape of the systems : the distance between the two landmarks is preserved in ( a ) , the angle between the two landmarks is preserved in ( d ) . the equivalence class of @xmath13 is also called its orbit and written @xmath22 . the equivalence class / orbit of @xmath21 is illustrated with the blue dotted circle in figure [ fig : leadingcases1 ] ( a ) for the plane example and in figure [ fig : leadingcases1 ] ( b ) for the sphere example . the orbit of @xmath21 in @xmath14 is the submanifold of all objects in @xmath14 that have the same shape as @xmath21 . the curvature of the orbit as a submanifold of @xmath14 is the key point of the results in section [ sec : quant ] . the _ shape space _ is by definition the space of orbits . this is a quotient space denoted @xmath23 . one orbit in @xmath14 , i.e. one circle in figure [ fig : leadingcases1 ] ( b ) or ( e ) , corresponds to a point in @xmath24 . the shape space is @xmath25 in the plane example . this is the space of all possible distances between the two landmarks , see figure [ fig : leadingcases1 ] ( c ) . the shape space is @xmath26 $ ] in the sphere example . this is the space of all possible angles between the two landmarks , see figure [ fig : leadingcases1 ] ( f ) . we consider that the action of @xmath18 on @xmath14 is _ isometric with respect to the riemannian metric of @xmath14_. this implies that the distance @xmath27 between two objects in @xmath14 does not change if we transform both objects in the same manner . in the plane example , rotating the landmark @xmath28 and another landmark @xmath29 with the same angle does not change the distance between them . the distance in @xmath14 induces a quasi - distance @xmath30 in @xmath24 : @xmath31 . the distance between the shapes of @xmath28 and @xmath29 is computed by first registering / aligning @xmath28 onto @xmath29 by the mean of @xmath20 , and then using the distance in the ambient space @xmath14 . in the plane example , the distance between two shapes is the difference in distances between the landmarks . one can compute it by first aligning the landmarks , say on the first axis of @xmath6 . then , one uses the distance in @xmath6 . both object space @xmath14 and shape space @xmath24 are stratified because of the notion of isotropy group . the _ isotropy group of @xmath13 _ is the subgroup of transformations of @xmath18 that leave @xmath13 invariant . for the plane example , every @xmath32 has isotropy group the identity and @xmath33 has isotropy group the whole group of 2d rotations . objects on the same orbit , i.e. objects that have the same shape , have conjugate isotropy groups . the _ orbit type _ of an orbit is the corresponding conjugate class . _ principal shapes _ are shapes with smallest isotropy group conjugation class . in the plane example , @xmath34 is the set of objects with principal shapes . it corresponds to @xmath35 in the shape space and is colored in blue on figure [ fig : leadingcases1 ] ( c ) . _ singular shapes _ are shapes with larger isotropy group conjugation class . in the plane example , @xmath33 is the only object with singular shape . it corresponds to @xmath36 in @xmath37 and is colored in red in figure [ fig : leadingcases1 ] ( c ) . the ( connected components of ) the _ orbit types _ form a stratification of @xmath14 , called the _ orbit - type stratification of @xmath14_. the principal type is predominant in the following sense : the set of principal strata , which we denote @xmath38 , is open and dense in @xmath14 . this means that there are objects with non - degenerated shapes almost everywhere . the stratification of @xmath14 into orbit types strata gives a stratification of the shape space @xmath39 . also , non - degenerated shapes are dense in the shape space . we have focused on an intuitive introduction of the concepts . we refer to @xcite for mathematical details . from now on , the mathematical setting is the following : we assume a proper , effective and isometric action of a finite dimensional lie group @xmath18 on a finite dimensional riemannian manifold @xmath14 . we recall that the data are the @xmath12 that are sets of landmarks , curves , images , etc . we interpret the data @xmath13 s as random realizations of the generative model : @xmath40 where the observed object @xmath41 is a shape @xmath42 with a given position or parameterization @xmath43 and observed with noise @xmath44 . here exp(p , u ) denotes the riemannian exponential of @xmath45 at point @xmath46 . the @xmath47 are themselves i.i.d . realizations of random variables . drawing them lead to the following three step interpretation of the generative model [ eq : genmodel ] . [ [ step-1-generate - the - shape - y_i - in - mg ] ] step 1 : generate the shape @xmath48 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we assume that there is an probability density of shapes in @xmath23 , with respect to the measure on @xmath24 induced by the riemannian measure of @xmath14 . the @xmath49 s are i.i.d . samples drawn from this distribution . for example , it can be a gaussian as illustrated in figure [ fig : genmodel1 ] on the shape spaces for the plane and sphere examples . this is the variability that is meaningful for the statistical study , whether we are analyzing shapes of skulls , proteins , bones , neural spike trains , handwritten digits or brains . we assume in this paper that the distribution is simply a dirac at @xmath50 which we call the _ template shape_. this is the most common assumption in these generative models @xcite . [ [ step-2-generate - its - positionparameterization - g_i - in - g ] ] step 2 : generate its position / parameterization @xmath43 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we can not observe shapes in @xmath23 . we rather observe objects in @xmath14 , that are shapes posed or parameterized in a certain way . we assume that there is a probability distribution on the positions or parameterizations of @xmath18 , or equivalently a probability distribution on principal orbits with respect to their intrinsic measure . we assume that the distribution does not depend on the shape @xmath49 that has been drawn . the @xmath51 s are i.i.d . from this distribution . for example , it can be a gaussian as illustrated in figure [ fig : genmodel2 ] on the shape spaces for the plane and sphere examples . [ [ step-3-generate - the - noise - epsilon_i - in - t_g_i - cdot - y_im ] ] step 3 : generate the noise @xmath44 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the observed @xmath13 s are results of noisy measurements . we assume that there is a probability distribution function on @xmath52 representing the noise . we further assume that this is a gaussian centered at @xmath53 , the origin of the tangent space @xmath52 , and with standard deviation @xmath0 , see figures [ fig : genmodel3 ] . the parameter @xmath0 will be extremely important in the developments of section , as we will compute taylor expansions around @xmath54 . other generative models may be considered in the literature . we find in @xcite the model : @xmath55 and in @xcite the model : @xmath56 . our goal is to unveil the variability _ of shapes in @xmath23 _ while we in fact observe the noisy _ objects @xmath13 s in @xmath14_. first , we focus on the case where the variability in the shape space is assumed to be a dirac at @xmath50 ( step 1 of generative model ) . our goal is thus to estimate the template shape @xmath50 . one may consider the maximum likelihood estimate of @xmath50 : @xmath57 we have hidden variables , the @xmath20 s . the expectation - maximization ( em ) algorithm would be the natural implementation for computing the ml estimator . but the em algorithm is computationally expensive , above all for tridimensional images . thus , one usually relies on another procedure that is an approximation of the em . @xcite . [ [ estimating - the - template - shape - with - the - frchet - mean - in - the - shape - space ] ] estimating the template shape with the frchet mean in the shape space + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one initializes the estimate with @xmath58 . then , one iterates the following two steps until convergence : @xmath59 ( 1 ) is an estimation of the hidden observations @xmath51 and an approximation of the e - step of the em algorithm . ( 2 ) is the m - step of the em algorithm : the maximization of the surrogate in the m - step amounts to the maximization of the variance of the projected data . this is exactly the minimization of the squared distances to the data of ( 2 ) . the procedure converges because it decreases at each step a cost bounded below by zero . the estimator computed with this procedure is : @xmath60 the term @xmath61 in equation [ eq : frechet ] is the distance in the shape space between the shapes of @xmath50 and @xmath13 . thus , we recognize in equation [ eq : frechet ] the frchet mean on the shape space . the frchet mean is a definition of mean on manifolds @xcite : it is the point that minimizes the squared distances to the data in the shape space . all in all , one projects the probability distribution function of the @xmath13 s from @xmath14 to @xmath39 and computes its `` expectation '' , in a sense made precise later . we illustrate the procedure with the examples of the plane and the sphere . we take @xmath28 , @xmath29 , @xmath62 three objects in @xmath6 in figure [ fig : estimator ] ( a ) and on @xmath9 in figure [ fig : estimator ] ( b ) . step ( 1 ) is the registration / alignment step . one filters out the position / parameterization component , i.e. the coordinate on the orbit . one projects the objects @xmath28 , @xmath29 , @xmath62 in the shape space @xmath24 using the blue arrows . step ( 2 ) is the computation of the frchet mean of the registered data . we implemented the generative model and the estimation procedure on the plane and the sphere in shiny applications available online : https://nmiolane.shinyapps.io/shinyplane and https://nmiolane.shinyapps.io/shinysphere . we invite the reader to look at the web pages and play with the different parameters of the generative model . figure [ fig : shinyplane1 ] shows screen shots of the applications . [ fig : shinyplane1 ] our main result is to show that this procedure gives an inconsistent estimate of the template shape @xmath50 of the generative model . the estimator @xmath63 converges when the number of data goes to infinity . however it has an asymptotic bias with respect to the parameter @xmath50 it is designed to estimate : @xmath64 $ ] . this is a vector at the tangent space of @xmath39 at the real parameter @xmath50 . the vector represents how much one has to shoot from @xmath50 to get the estimated parameter @xmath63 . it simply writes @xmath65 $ ] for linear spaces . we could also consider the variance of the estimator . the variance is defined as @xmath66)^2]$ ] . in the limit of an infinite sample , we have : @xmath67 . this is why we focus on the asymptotic bias . we first compute the asymptotic bias for the examples of the plane and the sphere to give the intuition . the probability distribution function of the @xmath13 s comes from the generative model . this is a probability distribution on @xmath6 for the plane example , parameterized in polar coordinates @xmath7 like figure [ fig : simple ] . so we can compute the projected distribution function on the shapes , which are the radii @xmath8 here . this is done simply by integrating out the distribution on @xmath11 , the position on the circles . this gives a probability distribution on @xmath37 for the plane example . we write it @xmath68 . we remark that @xmath69 does not depend on the probability distribution function on the @xmath70 s of step 2 of the generative model . we can also compute @xmath71 in the sphere example : we integrate over @xmath72 the probability distribution function on @xmath73 . figure [ fig : pdfs ] ( a ) shows @xmath69 for the plane example , for a template @xmath74 . we plot it for two different noise levels @xmath75 and @xmath76 . note that here @xmath69 is the rice distribution . figure [ fig : pdfs ] ( b ) shows @xmath69 for the sphere example , for a template @xmath77 . we plot it for different noise levels and @xmath75 and @xmath76 . in both cases , the x - axis represents the shape space which is @xmath37 for the plane example and @xmath78 $ ] for the sphere example . the green vertical bar represents the template shape , which is 1 in both cases . the red vertical bar is the expectation of @xmath69 in each case . it is @xmath63 , the estimate of @xmath50 we see on these plots that @xmath69 is not centered at the template shape : the green and red bars do not coincide . @xmath69 is skewed away from 0 in the plane example and away from @xmath36 and @xmath79 in the sphere example . the skew increases with the noise level @xmath0 . the difference between the green and red bars is precisely the bias of @xmath63 with respect to @xmath50 . figure [ fig : bias ] shows the bias of @xmath63 with respect to @xmath50 , as a function of @xmath0 , for the plane ( left ) and the sphere ( right ) . increasing the noise level @xmath0 takes the estimate @xmath63 away from @xmath50 . the estimate is repulsed from @xmath36 in the plane example : it goes to @xmath80 when @xmath81 . it is repulsed from @xmath36 and @xmath79 in the sphere example : it goes to @xmath82 when @xmath83 . one can show numerically that the bias varies as @xmath84 around @xmath85 in both cases . this is also observed on the shiny applications @xcite at https://nmiolane.shinyapps.io/shinyplane and https://nmiolane.shinyapps.io/shinysphere . these examples already show the origin of the asymptotic bias of @xmath63 . _ the bias comes from the curvature of the template s orbit . _ figure [ fig : curvature ] shows the template s orbit in blue , in ( a ) for the plane and ( b ) for the sphere . in both cases the black circle represents the level set @xmath0 of the gaussian noise . the probability of generating an observation @xmath13 outside of the template s shape orbit is bigger than the probability of generating it inside : the grey area in the black circle is bigger than the white area in the white circle . there will be more registered data that are greater than the template . their expected will therefore be greater than the template and thus biased . we prove this in the general case in the next section . we show the asymptotic bias of @xmath63 in the general case and prove that it comes from the external curvature of the template s orbit . we show it for @xmath50 a principal shape and for a gaussian noise of variance @xmath84 , truncated at @xmath86 . our results will need the following definitions of curvature . the _ second fundamental form @xmath87 _ of a submanifold @xmath88 of @xmath14 is defined on @xmath89 by @xmath90 , where @xmath91 denotes the orthogonal projection of covariant derivative @xmath92 onto the normal bundle . mean curvature vector @xmath93 of @xmath88 _ is defined as : @xmath94 . intuitively , @xmath87 and @xmath93 are measures of extrinsic curvature of @xmath88 in @xmath14 . for example an hypersphere of radius @xmath95 in @xmath2 has mean curvature vector @xmath96 . [ th : pdf ] the probability distribution function on the shapes induced by the generative model is : @xmath97 here @xmath98 is the riemannian logarithm @xmath99 $ ] of the shape @xmath100 $ ] at the template shape @xmath50 , @xmath87 is the second fundamental form of the orbit of @xmath98 , @xmath95 is the ricci curvature , @xmath101 are constants independent of @xmath98 and @xmath0 . the proof is given in appendix [ app : proofpdf ] . the exponential in the expression of @xmath69 belongs to a gaussian distribution centered at @xmath102 . this is a riemannian gaussian centered at the template shape @xmath50 , because @xmath98 are coordinates at the tangent space of @xmath39 at @xmath50 . however the whole distribution @xmath69 differs from the gaussian because of the @xmath98-dependent term in the right parenthesis . this induces a skew of the distribution away from the singular shapes , as observed for the examples in figure [ fig : pdfs ] . [ th : bias ] the asymptotic bias of the template s shape estimation writes : @xmath103 here @xmath93 is the mean curvature vector of the template shape s orbit and @xmath84 the variance of the noise on the objects . the proof is given in appendix [ app : proofbias ] . this generalizes the quadratic behavior observed in the examples on figure [ fig : bias ] . the asymptotic bias has a geometric origin : it comes from the external curvature of the template s orbits , see figure [ fig : curvature ] . we can vary two parameters in equation [ eq : bias ] : @xmath50 and @xmath0 . the external curvature of orbits generally increases when @xmath50 is closer to a singularity of the shape space ( see section 1 ) @xcite . the singular shape of the two landmarks in @xmath6 arises when their distance is 0 . in this case , the mean curvature vector is @xmath104 : it is inversely proportional to @xmath105 , the radius of the orbit . @xmath105 is also the distance of @xmath50 to the singularity @xmath36 . [ [ beyond - y - being - a - principal - shape ] ] beyond @xmath50 being a principal shape + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + our results are valid when the template @xmath50 is a principal shape . this is a reasonable assumption as the set of principal shapes is dense in the shape space . what happens when @xmath50 approaches a singularity , i.e. when @xmath50 changes stratum in the stratified space @xmath24 ? taking the limit @xmath106 in the coefficients of the taylor expansion is not a legal operation . therefore , we can not conclude on the taylor expansion of the bias for @xmath106 . indeed , the taylor expansion may even change order for @xmath106 . we take @xmath107 with the action of @xmath108 and the template @xmath109 : @xmath110 the bias is linear in @xmath0 in this case . [ [ beyond - sigma-1 ] ] beyond @xmath111 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the assumption @xmath111 is reasonable as we hope that the noise on the data is not too large . nevertheless it would be very interesting to study the asymptotic bias for any @xmath0 , including large noises ( @xmath112 ) . the distribution over the @xmath13 s in @xmath14 will be spread on the whole manifold @xmath14 . we can not rely on local computations on @xmath14 ( at the scale of @xmath0 ) anymore . we have to make global assumptions on the manifold @xmath14 . the plane example is the canonical example of a flat manifold . the sphere example is the canonical example of manifold with constant ( positive ) curvature . the bias as a function of @xmath0 is plotted in figure [ fig : bias ] . it leads us to the conjecture that the estimate converges towards a barycenter of shape space s singularities when the noise level increases . singularities have a repulsive action on the estimation of each template s shape . such repulsive force acts on each estimators . as a result , the estimators of the mean shape finds an equilibrium position : the barycenter . [ [ beyond - one - dirac - in - q - several - templates ] ] beyond one dirac in @xmath24 : several templates + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we have considered so far that there is a unique template shape @xmath50 : the generative model has a dirac distribution at @xmath50 in the shape space . what happens for other distributions ? we assume that there are @xmath113 template shapes @xmath114 . observations are generated in @xmath14 from each template shape @xmath115 with the generative model of section 2 . our goal is to unveil the structure of the shape distribution , i.e. the @xmath113 template shapes here , given the observations in @xmath14 . the distributions on shapes projected on the shape space is a mixture of probability density functions of the form of equation [ eq : f ] . its modes are related to the template shapes . the k - means algorithm is a very popular method for data clustering . we study what happens if one uses k - means algorithms on shapes generated with the generative model above . the goal is to cluster the shape data in @xmath113 distinct and significant groups . one performs a coordinate descent algorithm on the following function : @xmath116 in other words , one minimizes @xmath117 by successively minimizing on the assignment labels @xmath118 s and the cluster s centers @xmath119 s . given the @xmath118 , minimizing @xmath117 with respect to the @xmath119 s is exactly the simultaneous computation of @xmath113 frchet means in the shape space . one looks for meaningful well separated clusters ( high inter - clusters dissimilarity ) whose members are close to each other ( high intra - cluster similarity ) . in other words , the quality of the clustering is evaluated by the following criterion : @xmath120 which is the dissimilarity between clusters quotiented by the diameter of the clusters . in the absence of singularity in the shape space , the projected distribution looks like figure [ fig : kmeans ] ( a ) and @xmath121 . the criterion is worse in the presence of singularities . figure [ fig : kmeans ] illustrates this behavior for the plane example . we consider any two clusters @xmath122 and call @xmath123 , @xmath124 the estimated centroids . the criterion @xmath125 writes : @xmath126 even in the best case with correct assignments to the clusters @xmath127 and @xmath128 , the k - means algorithm looses an order of validation when computed on shapes . [ [ beyond - the - finite - dimensional - case ] ] beyond the finite dimensional case + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + our results are valid when @xmath14 is a finite dimensional manifold and @xmath18 a finite dimensional lie group . some interesting examples belong to the framework of infinite dimensional manifold with infinite dimensional lie groups . this is the case for the lddmm framework on images @xcite . it would be important to extend these results to the infinite dimensional case . we take @xmath107 with the action of @xmath108 . we have a analytic expression of @xmath69 in this case @xcite . figure [ fig : finitedims ] shows the influence of the dimension @xmath129 for the probability distribution functions on the shape space and for the bias . the bias increases with @xmath129 . this leads to think that it appears in infinite dimensions as well . we propose two procedures to correct the asymptotic bias on the template s estimate . they rely on the bootstrap principle @xcite , more precisely a parametric bootstrap . as such , they are directly applicable to any type of data . we assume that we know the variance @xmath130 from the experimental setting . the first procedure is called iterative bootstrap . algorithm [ alg : iterative ] details it . figure [ fig : iterativebootstrap ] illustrates it on the plane example . algorithm [ alg : iterative ] starts with the usual template s estimate @xmath131 , see figure [ fig : iterativebootstrap ] ( a ) . at each iteration , we correct @xmath63 with a better approximation of the bias . first , we generate bootstrap data by using @xmath63 as the template shape of the generative model . we perform the template s estimation procedure with the frchet mean in the shape space . this gives an estimate @xmath132 of @xmath133 . the bias of @xmath134 with respect to @xmath133 is @xmath135 . it gives an approximation of the bias @xmath136 , see figure [ fig : iterativebootstrap ] ( b ) . we correct @xmath63 by this approximation of the bias . this gives a new estimate @xmath137 , see figure [ fig : iterativebootstrap ] ( c ) . we recall that the bias @xmath136 depends on @xmath50 , see theorem [ th : bias ] . @xmath137 is closer to the template @xmath50 than @xmath133 . thus , the next iteration gives a better approximation @xmath138 of @xmath136 . we correct the initial @xmath63 with this better approximation of the bias , etc . the procedure is written formally for a general manifold @xmath14 in algorithm [ alg : iterative ] . [ alg : iterative ] * input : * objects @xmath12 , noise variance @xmath84 + * initialization : * + @xmath139\}_{i=1}^n)$ ] + @xmath140 + * repeat : * + generate bootstrap sample @xmath141 from @xmath142 + @xmath143_i\}_{i=1}^n)$ ] + @xmath144 + @xmath145 + @xmath146 + * until convergence : * @xmath147 + * output : @xmath148 * in algorithm [ alg : iterative ] , @xmath149 denotes the parallel transport from @xmath150 to @xmath151 . for linear spaces , @xmath152 , @xmath153 , @xmath154 . algorithm [ alg : iterative ] is a fixed - point iteration @xmath155 where : @xmath156 in a linear setting we have simply @xmath157 . one can show that @xmath158 is a contraction and that @xmath50 , the template shape , is the unique fixed point of @xmath158 ( using the local bijectivity of the riemannian exponential and the injectivity of the estimation procedure ) . thus the procedure converges to @xmath50 in the case of an infinite number of observations @xmath159 . figure [ fig : iterativebootstrap_fixedpoint ] illustrates the convergence for the plane example , with a gaussian noise of standard deviation @xmath160 . the template shape @xmath161 was initially estimated at @xmath162 . algorithm [ alg : iterative ] corrects the bias . figures [ fig : iterationsplane ] and [ fig : iterationssphere ] show the iterations of iterative bootstrap for the plane and the sphere example . the second procedure is called the nested bootstrap . algorithm [ alg : nested ] details it . figure [ fig : nestedbootstrap ] illustrates it on the plane example . algorithm [ alg : nested ] starts like algorithm [ alg : iterative ] with @xmath163 , see figure [ fig : nestedbootstrap ] ( a ) . it also performs a parametric bootstrap with @xmath164 as the template , computes the bootstrap replication @xmath165 and the approximation @xmath166 of @xmath167 , see figure [ fig : iterativebootstrap ] ( b ) . now algorithm [ alg : nested ] differs from algorithm [ alg : iterative ] . we want to know how biased is @xmath166 as an estimate of @xmath167 ? this is a valid question as the bias depends on the template @xmath50 , see theorem [ th : bias ] . we want to estimate this dependence . we perform a bootstrap , nested in the first one , with @xmath168 as the template . we compute the estimate @xmath169 and the approximation @xmath170 of @xmath166 , see figure [ fig : iterativebootstrap ] ( c ) . we observe how far @xmath170 is from @xmath166 . this gives the blue arrow , which is the bias of @xmath170 as an estimate of @xmath166 , see figure [ fig : iterativebootstrap ] ( d ) . the blue arrow is an approximation of how far @xmath166 is from @xmath167 . we correct our estimation of the bias ( in red ) by the blue arrow . we correct @xmath171 by the bias - corrected estimate of its bias , see figure [ fig : iterativebootstrap ] ( e ) . [ alg : nested ] * input : * objects @xmath12 , noise variance @xmath84 + * initialization : * + @xmath139\}_{i=1}^n)$ ] + * bootstrap : * + generate bootstrap sample @xmath172 from @xmath173 + @xmath174_i\}_{i=1}^n)$ ] + @xmath175 + * nested bootstrap : * + for each @xmath127 : * generate bootstrap sample @xmath176 from @xmath177 * @xmath178_i\}_{k=1}^n)$ ] @xmath179 + @xmath180 + * output : @xmath137 * one may use the iterative bootstrap or the nested bootstrap depending on the experimental setting . we illustrate them both on the plane example in figure [ fig : comparison ] . the advantages of the iterative bootstrap are the following . it corrects perfectly the bias of @xmath63 in the case of a very large number of observations @xmath181 . it can be used to experimentally compute the mean curvature vector @xmath93 of each orbit of a group action . one probes the orbit s curvature by `` feeling it '' with a riemannian gaussian on @xmath14 and projecting on the shape space . its drawbacks are the following . it works only with very large @xmath181 . it is not robust as it uses the generative model several times . if the generative model is far from being true , then the iterative bootstrap fails . the advantages of the nested bootstrap are the following . it is a standard statistical procedure that is more robust with respect to variations of the generative model . even if generative model is different from the one that we assume , the nested bootstrap performs well . moreover , it does not need as many data as the iterative bootstrap . its drawback is that it does not correct perfectly the bias , especially when the noise is important . these simulations give a rule of thumb for when the bias needs to be corrected . this is when the noise @xmath0 is comparable to the distance of the template @xmath50 to the singularity . we perform a simulation using the iterative bootstrap on triangles . we randomly generate @xmath182 triangles in @xmath6 . the mean triangle is chosen by taking two coordinates randomly from a uniform distribution on @xmath183 $ ] . then we add bivariate gaussian noise on each landmark . these experiments are illustrated in figure [ fig : iterationstriangles ] . the number of iterations required for the convergence of algorithm 1 with respect to the noise level are shown in figure [ fig : iterationstriangles ] . we observe the convergence in the three experiments for less than 10 iterations . now we go to real triangle data . we have 24 images of rhesus monkeys eyes , acquired with a heidelberg retina tomograph @xcite . for each monkey , an experimental glaucoma was introduced in one eye , while the second eye was kept as control . one seeks a significant difference between the glaucoma and the control eyes . on each image , three anatomical landmarks were recorded : @xmath184 for the superior aspect of the retina , @xmath185 for the nose side of the retina , and @xmath186 for the side of the retina closest to the temporal bone of the skull . the data are matrices @xmath12 where the landmark coordinates form the rows . for the onh example , @xmath14 is the space of @xmath187 landmarks in 3d , @xmath188 and the rotations act isometrically on each object @xmath13 . * analysis * this simple example illustrates the estimation of the template shape . we use the following procedure to compute the mean shape for each group . we initialize @xmath63 with @xmath28 and repeat the following two steps until convergence : @xmath189 figure [ fig : onh ] shows the mean shapes @xmath190 of the control group ( left ) and @xmath191 of the glaucoma group ( right ) in orange , while the initial data are in grey . the difference between the two groups is quantified by the distance between their means : @xmath192 m . we want to determine if this analysis presents an bias that significantly changes the estimated shape difference between the groups . we use the nested bootstrap to compute an approximation of the asymptotic bias on each mean shape , for a range of noise s standard deviation in @xmath193 . the asymptotic bias on the template shape of the glaucoma group is @xmath194 @xmath195 and of the control group is @xmath196 . the corrected template shape differences are @xmath197 . in particular , for @xmath198 , we observe that the bias in the template shape are respectively @xmath199 for the healthy group and @xmath200 for the glaucoma group . this follows the rule - of - thumb : the bias is more important for the healthy group , for which the overall size is smaller than the glaucoma group , for a same noise level . the bias of the template shape estimate accounts for less than @xmath201 in this case , which is less than @xmath202% of the shapes sizes . this computation guarantees that this study has not been significantly affected by the bias . we estimate the impact of the bias on statistics on protein shapes . a standard hypothesis in biology is that structure ( i.e. shape ) and function of proteins are related . fundamental research questions about protein shapes include structure prediction - given the protein amino - acid sequence , one tries to predict its structure - and design - given the shape , one tries to predict the sequence needed . one relies on experimentally determined 3d structures gathered in the protein data base ( pdb ) @xcite . they contain errors on the protein s atoms coordinates . average errors range from 0.01 @xmath203 to 1.76 @xmath203 , which is of the magnitude of the length of some covalent bonds . these values are averaged over the whole protein and in general , the main - chain atoms are better defined than the side - chain atoms or the atoms at the periphery . this is illustrated on figure where we have plot the b - factor ( related to coordinates errors @xcite ) as a colored map on the atoms for proteins of pdb - codes 1h7w and 4hbb . [ [ proteins - radius - of - gyration ] ] protein s radius of gyration + + + + + + + + + + + + + + + + + + + + + + + + + + + + a biased estimate of a protein shape has consequences for studies on proteins folding . stability and folding speed of a protein depend on both the estimated shape of the denatured state ( unfolded state ) and of the native state ( folded state ) . one may study if compact initial states yield to faster folding . the protein compactness is represented by the protein s radius of gyration , defined as : @xmath204 , where @xmath185 is the number of non - hydrogen atoms , @xmath205 , @xmath206 are resp . the coordinates of atoms and centers and @xmath207 their mass . note that we assume ( as it is usually the case ) that all masses @xmath207 for non - hydrogen atoms are equal and that hydrogen atoms have mass @xmath36 . error on atoms coordinates give a bias on the estimate of the radius of gyration : @xmath208 the radius of protein hjsj ( 85 residues ) is known to be around 10 @xmath203 . the error on @xmath209 is of @xmath210% with an average error of positions on the atoms of @xmath211 . it is 8.6% for an error of @xmath212 . the error will be greater if one consider binding sites at the periphery of the proteins rather than the whole protein . indeed sites size is smaller and they have less atoms . one could think about doing clustering on radii of gyration using the k - means algorithm on shapes . the index @xmath125 of section [ sec : correction ] is : @xmath213 clustering on radii of gyration may lead to a misleading indicator . @xmath125 indicates that the clustering performs better that it actually does . [ [ false - positive - probability - in - proteins - motif - detection ] ] false positive probability in protein s motif detection + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the relation between a protein s shape and function is linked to its motifs , which define the supersecondary structure . motifs have biological properties : for example the helix turn helix motif @xciteis responsible for the binding of dna within several prokaryotic proteins . automatic motif detection is another challenge in the study of protein shapes . we investigate the impact of bias on the false positive probability estimation in motif detection . let us consider a set @xmath214 of proteins each with @xmath215 atoms . one is interested in the motifs of @xmath1 atoms that can be detected in the protein s set , where @xmath216 . we define @xmath0 that represents an allowed error zone . the number of detected motifs increases if : ( i ) one decreases @xmath1 , or ( ii ) one increases @xmath0 , or ( iii ) increases @xmath181 . thus how many detected motifs actually come from chance , with respect to the parameters @xmath1 , @xmath0 , @xmath181 ? the false positives probability indicates when one detects truth and when one detects noise . the usual estimate of the false positive probability is @xmath217 . here @xmath218 is the volume of the error zone allowed . @xmath219 is the total volume of the protein @xcite , thus the a ball of radius the radius of gyration . thus @xmath219 may be biased and overestimated . the probability of false positive is underestimated . we consider the example of @xcite . one tries to find motifs between the tryptophan repressor of escherichia coli ( pdb code 2wrp ) and the cro protein of phage 434 ( pdb code 2cro ) . these two proteins are known to share the helix - turn - helix motif . the radius of gyration of 2wrp is @xmath220 , the total volume is : @xmath221 . we assume an error zone that takes the form of a diagonal covariance matrix with standard deviations @xmath222 . we get the error zone volume : @xmath223 and the estimation of the false positive probability : @xmath224 . we find that @xmath225 is underestimated by @xmath226 using the expression of the radius of gyration s bias . we apply the rule of thumb of section [ sec : correction ] to determine when the bias needs a correction in the computation of a brain template from medical images . here @xmath14 and @xmath18 will be infinite dimensional . nevertheless we apply our results to get intuition for this application . in neuroimaging , a template is an image representing a reference anatomy . computing the template is often the first step in medical image processing . then , the subjects anatomical shapes may be characterized by their spatial deformations _ from the template_. these deformations may serve for ( i ) a statistical analysis of the subject shapes , or ( ii ) for automated segmentation by mapping the template s segmented regions into the subject spaces . in both cases , if the template is not centered among the population , i.e. if it is biased , then the analyzes and conclusions could be biased . we are interested in highlighting the variables that control the template s bias . the framework of large deformation diffeomorphic metric mapping ( lddmm ) @xcite embeds the template estimation in our geometric setting . the lie group of diffeomorphisms acts on the space of images as follows : @xmath227 the isotropy group of @xmath228 writes : @xmath229 . its lie algebra @xmath230 consists of the infinitesimal transformations whose vector fields are parallel to the level sets of @xmath228 : @xmath231 . the orbit of @xmath228 is : @xmath232 . the `` shape space '' is by definition the space of orbits . two images that are diffeomorphic deformations of one another are in the same orbit . they correspond to the same point in the shape space . topology of an image is defined as the image s properties that are invariant by diffeomorphisms . consequently , the shape space is the space of the images topology , represented by the topology of their level sets . we get a stratification of the shape space when we gather the orbits by orbit type . a stratum is more singular than another , if it has higher orbit type , i.e. larger isotropy group . the manifold @xmath14 has an infinite stratification . one changes stratum every time there is a change in the topology of an image s level sets . singular strata are toward simpler topology . `` principal '' strata are toward a more complicated topology . indeed , the simpler the topology of the level sets is , the higher is the `` symmetry '' of the image . thus the larger is its isotropy group . note that strata with smaller isotropy group ( more detailed topology ) do not represent `` singularities '' from the point of view of a given image and do not influence the bias . in fact , such strata are at distance 0 : an infinitesimal local change in intensity can create a maximum or minimum , thus complexifying the topology . using the rule - of - thumb of section [ sec : correction ] , the template s bias depends on its distance @xmath105 to the next singularity , at the scale of @xmath0 the intersubjects variability . the template is biased in the regions where the difference in intensity between maxima and minima is of the same amplitude as the variability . the template may converge to pure noise in these regions . we introduced tools of statistics on manifolds to study the properties of template s shape estimation in medical imaging and computer vision . we have shown its asymptotic bias by considering the shape space s geometry . the bias comes from the external curvature of the template s orbit at the scale of the noise on the data . this provides a geometric interpretation for the bias observed in @xcite . we investigated the case of several templates and the performance k - mean algorithms on shapes : clusters are less well separated because of each centroid s bias . the variables controlling the bias are : ( i ) the distance in shape space from the template to a singular shape and ( ii ) the noise s scale . this gives a rule - of - thumb for determining when the bias is important and needs correction . we proposed two procedures for correcting the bias : an iterative bootstrap and a nested bootstrap . these procedures can be applied to any type of shape data : landmarks , curves , images , etc . they also provide a way to compute the external curvature of an orbit . our results are exemplified on simulated and real data . many studies use the template s shape estimation algorithm in molecular biology , medical imaging or computer vision . their estimations are necessarily biased . but these studies often belong to a regime where the bias is not important ( less than @xmath233 ) . for example , the bias is important in landmark shapes analyses when the landmarks noise is comparable to the template shape s size . studies are rarely in this regime . we have considered shapes belonging to infinite dimensional shape spaces . our results do not apply to the infinite dimensional case . we have used them to gain intuition about it . the bias might be more important in infinite dimensions and need the correction we have suggested . biblabel[1]#1 . 10 , _ the riemannian geometry of orbit spaces . the metric , geodesics , and integrable systems _ , publ debrecen , 62 ( 2003 ) . , _ towards a coherent statistical framework for dense deformable template estimation _ , journal of the royal statistical society . , 69 ( 2007 ) , 329 . , _ estimating the template in the total space with the frchet mean on quotient spaces may have a bias . _ , proceedings of the fifth international workshop on mathematical foundations of computational anatomy ( mfca15 ) , 2015 , pp . 131142 . , _ convergent stochastic expectation maximization algorithm with efficient sampling in high dimension . application to deformable template model estimation _ , computational statistics & data analysis , 91 ( 2015 ) , pp . 4 19 . , _ the protein data bank _ , nucleic acids res , 28 ( 2000 ) , pp . 235242 . , _ on the consistency of frchet means in deformable models for curve and image analysis _ , electronic journal of statistics , ( 2011 ) , pp . 10541089 . , _ a deconvolution approach to estimation of a common shape in a shifted curves model _ , ann , 38 ( 2010 ) , pp . 24222464 . , _ the helix - turn - helix dna binding motif . _ , journal of biological chemistry , 264 ( 1989 ) , pp . 19036 . , _ analyse biomtrique de lanneau pelvien en 3 dimensions propos de 100 scanners _ , revue de chirurgie orthopdique et traumatologique , 100 ( 2014 ) , pp . s241 . , _ statistical shape analysis _ , john wiley & sons , new york , 1998 . , _ bootstrap methods : another look at the jackknife _ , the annals of statistics , 7 ( 1979 ) , pp . 126 . , _ morphometrics for nonmorphometricians _ , springer , 2012 . , _ brain templates and atlases _ , neuroimage , 62(2 ) ( 2012 ) , pp . 911922 . , _ intrinsic shape analysis : geodesic principal component analysis for riemannian manifolds modulo lie group actions . _ , statistica sinica , 20 ( 2010 ) , pp . 1100 . , _ riemannian structures on shape spaces : a framework for statistical inferences _ , in statistics and analysis of shapes , 2006 , pp . 313333 . , _ the diffusion of shape _ , advances in applied probability , 9 ( 1977 ) , pp . 428430 . , _ shape manifolds , procrustean metrics , and complex projective spaces _ , bulletin of the london mathematical society , 16 ( 1984 ) , pp . 81121 . , _ signal estimation under random time - warpings and nonlinear signal alignment _ , in advances in neural information processing systems 24 , 2011 , pp . 675683 . , _ the riemannian structure of euclidean shape spaces : a novel environment for statistics _ , the annals of statistics , 21 ( 1993 ) , pp . 12251271 . , _ euclidean distance matrix analysis ( edma ) : estimation of mean form and mean form difference _ , mathematical geology , 25 ( 1993 ) , pp . 573602 . . , _ mapping the effects of a@xmath234 levels on the longitudinal changes in healthy aging : hierarchical modeling based on stationary velocity fields _ , in proceedings of medical image computing and computer assisted intervention ( miccai ) , vol . 6892 of lncs , springer , 2011 , pp . 663670 . , _ curvature explosion in quotients and applications _ , j. differential geom . , 85 ( 2010 ) , pp . 117140 . , _ biased estimators on quotient spaces _ , proceedings of the 2nd international of geometric science of information ( gsi2015 ) , ( 2015 ) . , _ nonparametric statistics on manifolds and their applications to object data analysis _ , taylor & francis group , 2016 . , _ toward a generic framework for recognition based on uncertain geometric features _ , videre : journal of computer vision research , 1 ( 1998 ) , pp . 5887 . , _ intrinsic statistics on riemannian manifolds : basic tools for geometric measurements _ , journal of mathematical imaging and vision , 25 ( 2006 ) , pp . 127154 , _ a geometric algorithm to find small but highly similar 3d substructures in proteins . _ , bioinformatics , 14 ( 1998 ) , pp . 516522 . , _ riemannian geometry _ , encyclopaedia of mathem . sciences , springer , 2001 . , _ easy web applications in r. _ , 2013 . url : http://www.rstudio.com / shiny/. , _ error estimates of protein structure coordinates and deviations from standard geometry by full - matrix refinement of * @xmath235*b- and * @xmath236*b2-crystallin _ , acta crystallographica section d , 54 ( 1998 ) , pp . 243252 . , _ shapes and diffeomorphisms _ , applied mathematical sciences , springer london , limited , 2012 . here @xmath21 is a point in @xmath14 . we consider that @xmath21 belongs to a principal orbit @xmath237 . this will have no impact on the integration because the set of principal orbits is dense in @xmath14 . we write @xmath238 the projection of @xmath21 in the shape space . we write the template shape @xmath50 . we write its estimate @xmath63 . we take a normal coordinate system centered at the template @xmath50 . we have the decomposition @xmath239 , where @xmath88 is the orbit of @xmath50 . we note that @xmath240 , so that @xmath241 are _ not _ the coordinates of @xmath238 in the normal coordinate system at @xmath50 , see figure [ fig : notations ] . we denote @xmath129 the dimension of @xmath14 , @xmath46 the dimension of the principal orbits and @xmath242 the dimension of the quotient space . we write coordinates in @xmath24 with indices @xmath243 $ ] and coordinates in an orbit with indices @xmath244 $ ] . the generative model implies the following riemannian normal distribution on the objects : @xmath245 the distance @xmath246 expressed in the normal coordinate system at @xmath50 is simply @xmath247 . the riemannian measure @xmath248 at @xmath21 in the normal coordinate system at @xmath50 has the taylor expansion : @xmath249 . we recognize the riemannian curvature tensor @xmath95 . we truncate the riemannian gaussian : @xmath250 where @xmath184 is the normalization coefficient coming of the univariate truncated gaussian at @xmath251 . @xmath252 , @xmath253 and @xmath254 refer to geodesic balls of radius @xmath255 in their respective spaces . we denote @xmath256 and remark that @xmath257 where @xmath242 is the dimension of the quotient space and @xmath46 the dimension of the principal orbits . we represent @xmath88 in @xmath260 as the graph of a smooth function from @xmath261 to @xmath262 , around @xmath263 by : @xmath264 the local graph @xmath18 is : @xmath265 . its 0-th and 1-th order derivatives are zero because the graph goes through @xmath238 and is tangent at @xmath266 . its second order derivative is @xmath267 and represents the best quadratic approximation of the graph . we need the coordinate of @xmath238 in the ncs at @xmath50 . we parallel transport @xmath270 from @xmath268 to @xmath50 : @xmath271 . the @xmath36-th order term of the parallel transport at @xmath268 in a ncs at @xmath50 is 0 : @xmath272 this concludes . we take an ncs at @xmath50 . we denote @xmath98 the coordinate of @xmath238 in the shape space . we compute the induced probability distribution @xmath69 on shapes by integrating the distribution on the orbit of @xmath21 out of @xmath273 : @xmath274 the point @xmath21 has coordinates @xmath275 where @xmath276 is the integration coordinate . @xmath277 the lie group action is isometric so the riemannian metric splits onto @xmath280 at any point . moreover , the taylor expansion of the metric around @xmath21 still respects the splitting at the third order . the measure @xmath281 for @xmath276 close enough to @xmath21 is the restriction of this taylor expansion to the orbit . we have : @xmath282 so that : @xmath283 where @xmath95 is the ricci curvature . for the last term ( 3 ) , we assume again that the graph of the orbit is smooth enough so that it admits a taylor serie at @xmath263 , ie : @xmath286 . by plugging this , we get the moments with the same manipulation as in the proof of theorem 1 . as the moments of order @xmath187 ( and of odd order ) are zero : @xmath287 . we want to compute the systematic bias . to this aim , we compute the mean of the distribution of shapes in @xmath24 , in a coordinate system centered at the real @xmath50 . it writes : @xmath289 the coordinates of @xmath21 in the euclidean coordinate system at @xmath50 are simply @xmath290 . thus @xmath291 . from the lemma above : @xmath292 we denote ( 1 ) the first term of the sum and ( 2 ) the second term of the sum . we compute them independently . @xmath296 where the last line comes from the definitions of the moments of the truncated gaussian . we perform a taylor expansion under the integration , in @xmath0 , using the chain rule : @xmath303 again , the terms in @xmath304 integrate to 0 . we have : @xmath305 the term @xmath306 is precisely the normalization constant of the truncated gaussian @xmath307 , so that : @xmath308 the coefficient @xmath309 comes from the fact that we have truncated the gaussian . its expression is independent of @xmath0 because we have truncated at a multiple of @xmath0 . now we show that the second term ( 2 ) is of order @xmath310 . we assume that the local graph of the orbit is smooth enough so that it has a multivariate taylor serie : @xmath286 . we have : @xmath312 here the @xmath313 depend on @xmath241 . we first perform a majoration on each on them by a @xmath314 ( integration on a compact ball ) . @xmath315 we compute : @xmath316 so that : @xmath317 where we recognize the moment of the truncated gaussian , which are non zero only for even power . this gives : @xmath318 .

in this article we investigate averaging properties of fully nonlinear pdes in bounded domains with oscillatory neumann boundary data . the oscillation is periodic and is present both in the operator and in the neumann data . our main result states that , when the domain does not have flat boundary parts and when the homogenized operator is rotation invariant , the solutions uniformly converge to the homogenized solution solving a neumann boundary problem . furthermore we show that the homogenized neumann data is continuous with respect to the normal direction of the boundary . our result is the nonlinear version of the classical result in for divergence - form operators with co - normal boundary data . the main ingredients in our analysis are the estimate on the oscillation on the solutions in half - spaces ( theorem [ thm : planar ] ) , and the estimate on the mode of convergence of the solutions as the normal of the half - space varies over _ irrational _ directions ( theorem [ continuity ] ) .

1302.5386

let us consider a bounded domain @xmath0 in @xmath1 containing the closed unit ball @xmath2 ( see figure 1 ) . let @xmath3 $ ] be a hlder continuous function which is periodic with respect to the orthonomal basis @xmath4 of @xmath1 . more precisely , @xmath5 satisfies @xmath6 with @xmath7 and @xmath5 as given above , we are interested in the limiting behavior of the following problem : @xmath8 here @xmath9 is the outward normal vector at @xmath10 and @xmath11 is a uniformly elliptic , fully nonlinear operator ( see section 2 for the precise definitions and conditions on @xmath11 ) . the set @xmath12 is introduced to avoid discussions of compatibility conditions on the neumann boundary data . the operators under discussion include , for example , the divergence or non - divergence form operator @xmath13 where @xmath14 are hlder continuous and @xmath15 for positive constants @xmath16 and @xmath17 . let us define @xmath18 due to @xcite , there exists a uniformly elliptic operator @xmath19 such that any solution of @xmath20 in @xmath0 with _ fixed _ boundary data on @xmath21 converges uniformly to the solution of @xmath22 in @xmath0 ( see theorem [ original_convergence ] ) . therefore the question under investigation is whether the oscillatory boundary data changes the averaging behavior of @xmath23 . as we will discuss below , several difficulties arise in answering this question due to the nonlinear ( or non - divergence form ) nature of the problem as well as the geometry of the domain . let us state our main result . first we introduce the following cell problem " for given @xmath24 and @xmath25 . @xmath26 [ main ] let @xmath27 and , @xmath28 and @xmath29 be as given above . * for given normal direction @xmath30 and any @xmath25 there exists a unique homogenized neumann data @xmath31 for solutions of @xmath32 . * suppose that @xmath33 only depends on the eigenvalues of @xmath34 . then there exists a continuous function @xmath35 , given as the continuous extension of @xmath31 over @xmath36 , such that the following holds : + suppose @xmath0 is a bounded domain in @xmath1 such that @xmath21 is @xmath37 and does not contain any flat part . let @xmath23 solve @xmath28 . then @xmath23 converges locally uniformly to @xmath38 , which is the unique solution of the homogenized problem @xmath39 * the assumption on @xmath40 in theorem [ main ] ( b ) seems to be necessary condition to achieve the continuity of the homogenized slope @xmath41 , at least from numerical experiments ( see @xcite ) . note that this assumption is equivalent to the rotation and reflection invariance of @xmath40 . for more discussion on the role of the assumption in the analysis , see the discussion in section 4 . + * we point out that the restriction on the geometry of @xmath21 is necessary : in fact , if @xmath21 is locally @xmath42 with @xmath43 and @xmath44 , then the ( normalized ) distribution of boundary data @xmath45 on @xmath46 changes a lot as @xmath47 . consequently in this case there is no unique limit of @xmath23 . @xmath48_discussions on previous results _ our problem is classical for the case of uniformly elliptic , divergence - form equations @xmath49 with the _ co - normal _ boundary data @xmath50 for - , a corresponding result to theorem [ main ] was proved in the classical paper of bensoussan , lions and papanicolau @xcite with explicit integral formula for the limiting operator as well as the limiting boundary data . recently , a corresponding result was shown for systems of divergence type operators with oscillatory dirichlet boundary data in convex domains , by gerard - varet and masmoudi @xcite ( also see @xcite ) . for nonlinear or non - divergence type operators , or even for linear operators with oscillatory neumann boundary data that is not co - normal , most available results concern half - space type domains whose boundary goes through the origin . in @xcite , tanaka considered some model problems in half - space whose boundary is parallel to the axes of the periodicity by purely probabilistic methods . in @xcite arisawa studied special cases of problems in oscillatory domains near half spaces going through the origin , using viscosity solutions as well as stochastic control theory . generalizing the results of @xcite , barles , da lio and souganidis @xcite studied the problem for operators with oscillating coefficients , in half - space type domains whose boundary is parallel to the axes of periodicity , with a series of assumptions which guarantee the existence of approximate corrector . in @xcite the continuity property of the averaged neumann boundary data in half - spaces , with respect to the normal direction , was studied in the case of homogeneous operator @xmath11 . recently , barles and mironescu @xcite showed a corresponding result to @xcite for oscillatory dirichlet boundary data . @xmath48 _ main ingredients and challenges _ the main steps in extending aforementioned results from half - space type domains to general domains are the following . the plan is to use these solutions in strip domains to approximate those in general domains . for the stability of such approximation one requires the distribution of @xmath5 to be invariant on @xmath51 regardless of the choice of @xmath16 , which is the case if @xmath52 is not a multiple of vectors in @xmath53 : we call such vectors _ irrational_. for irrational @xmath52 we show that there is a unique linear profile the solutions converges to , by proving an oscillation lemma as well as using a quantitative version of weyl s equi - distribution theorem . secondly , to establish sufficient stability to address the general domains , some estimate on the convergence mode of @xmath54- solutions in the strip domain , in terms of the variation of @xmath52 , is necessary . establishing this estimate is our second , and most challenging , main step ( theorem [ continuity ] ) . for problems with divergence - form structure , such estimates were obtained in @xcite and @xcite ( also see @xcite ) by means of integral formulas . in our setting we must proceed by maximum principle - type arguments , which requires careful perturbation of the boundary data as well as construction of delicate barriers which describes the averaging behavior of solutions up to the neumann boundary . the proof of theorem 4.1 is based on the observation that , in meso - scopic " scale , the distribution of @xmath5 on two hyperplanes with close irrational normals are similar ( see section 4.1 . for a heuristic discussion of this fact ) . we adopt a multi - scale homogenization argument to address separately the effects on the solution caused by ( a ) microscopic oscillation of @xmath5 near the neumann boundary ( b ) the difference in normal directions and ( c ) the oscillation present in the operator @xmath11 . we mention that a significant difficulty arises due to the presence of the @xmath55-dependence in @xmath11 in addition to the oscillations in @xmath5 . to get around this difficulty , we use localization arguments as well as the existing homogenization results ( e.g. @xcite and @xcite ) to show homogenization occurs away from the neumann boundary . such strategy works since near the neumann boundary the oscillation of the first derivative ( @xmath5 ) dominates the behavior of solutions . @xmath48 _ outline of the paper _ in section 2 we introduce some notations as well as preliminary results which will be used in the rest of the paper . in section 3 we first study the averaging properties of the operator in the strip ( half - space type ) domains , to show that if the hypersurface is normal to an irrational direction then there is a unique homogenized neumann data in the limit @xmath47 . besides the complications that the inhomogeities in the operator @xmath11 cause , the proof of averaging phenomena in this setting is due to the weyl s distribution theorem , whose quantitative version that we need is borrowed from @xcite and is stated in theorem [ lemma - m ] . in section 4 we prove the main estimate on the mode of convergence of @xmath54-solutions in the strip domain as the normal @xmath52 varies around a reference direction . after a heuristic description of the proof , we prove a series of lemmas which describe the behavior of solutions in different regions of the strip domain , divided in terms of the distance to the neumann boundary . based on the lemmas then we are able to construct suitable barriers to obtain the desired estimate ( theorem 4.1 ) . lastly in section 5 we prove theorem [ main ] using the aforementioned estimate , the non - flat geometry of @xmath21 , and the stability of viscosity solutions . we first introduce the following categories of normal directions , following @xcite . * @xmath24 is a rational direction if @xmath56 . + * @xmath24 is an irrational direction if @xmath52 is not a rational direction . hyperplanes with irrational directions represent surfaces on which the boundary data does not change too much when one translates it in the normal direction . the following lemma quantifies this fact . the proof is a straightforward application of weyl s equi - distribution theorem . [ lemma 2.7 , @xcite ] [ lemma - m ] for @xmath57 and @xmath58 , let us define @xmath59 . then the following is true for @xmath60 : * suppose that @xmath52 is a rational direction . then there exists a constant @xmath61 depending only on @xmath52 such that , for any @xmath62 , there is @xmath63 satisfying @xmath64 * suppose that @xmath52 is an irrational direction . then there exists a mode of continuity @xmath65 and a dimensional constant @xmath66 such that the following is true : for any @xmath62 , there exists @xmath67 such that @xmath68 and @xmath69 * if @xmath52 is an irrational direction , then for any @xmath70 and @xmath71 , there exists @xmath63 such that @xmath72 the power @xmath73 appearing in ( ii ) of above lemma is arbitrarily chosen , and can be replaced with any @xmath74 . the mode of continuity @xmath75 , of course , depends on the choice of @xmath76 . next we introduce the properties of the operator @xmath11 . as before , let @xmath77 denote the normed space of symmetric @xmath78 matrices . in this paper we assume that the function @xmath79 satisfies the following conditions : * @xmath80 for any @xmath34 , @xmath81 and @xmath82 . in particular it follows that @xmath83 . * @xmath11 is locally lipschitz continuous in @xmath84 , and there exists a constant @xmath85 such that for all @xmath86 and @xmath87 @xmath88 * there exists constants @xmath89 such that @xmath90 * for any @xmath34 and @xmath81 , @xmath91 a typical example of @xmath11 satisfying ( f1)-(f4 ) is the non - divergence type operator @xmath92 where @xmath93 is periodic , lipschitz continuous and @xmath94 satisfies @xmath95 . another example is the bellman - issacs operators arising from stochastic optimal control and differential games @xmath96 where @xmath97 is a two - parameter family of uniformly elliptic operators of the form . we refer to @xcite for further examples of @xmath11 . let @xmath0 and @xmath12 be as given before , and let @xmath98 be a continuous function . to incorporate both @xmath28 and the homogenized problem @xmath99 , let us introduce a definition of viscosity solutions for the following problem : @xmath100 where @xmath9 is the outward normal at the boundary point @xmath10 . see @xcite for a game - interpretation of @xmath101 . the following definition is equivalent to the ones given in @xcite : * an upper semi - continuous function @xmath102 is a viscosity subsolution of @xmath101 if @xmath38 can not cross from below any @xmath103 which satisfies @xmath104 * a lower semi - continuous function @xmath102 is a viscosity supersolution of @xmath101 if @xmath38 can not cross from above any @xmath105 which satisfies @xmath106 * @xmath38 is a viscosity solution of @xmath101 if @xmath38 is both a viscosity sub- and supersolution of @xmath101 . the existence and uniqueness of viscosity solutions in bounded domains are consequences of the following comparison principle . [ thm : comp ] [ section v , @xcite ] let @xmath107 be a domain in @xmath1 whose boundary consists of @xmath51 and @xmath108 , and let @xmath11 satisfy @xmath109 . suppose @xmath110 are bounded and continuous . if @xmath38 and @xmath111 satisfy * @xmath112 in @xmath0 ; * @xmath113 on @xmath108 ; * @xmath114 on @xmath51 . where @xmath115 is continuous , and ( a ) and ( c ) should be interpreted in the viscosity sense . then @xmath113 in @xmath0 . next we state regularity results for solutions of @xmath116 which will be used later let us recall the pucci extremal operators : for a symmetric @xmath78 matrix @xmath117 , we define @xmath118 and @xmath119 where @xmath120 s are the eigenvalues of @xmath117 . the operator @xmath121 then satisfies the assumptions ( f1)-(f4 ) . moreover note that , due to , we have @xmath122 in particular , the difference @xmath123 of two viscosity solutions of @xmath124 satisfies both @xmath125 and @xmath126 in the viscosity sense . [ lemma - reg ] let @xmath38 be a viscosity solution of @xmath127 in a domain @xmath0 . then for any @xmath74 and a compact subset @xmath128 of @xmath0 , we have @xmath129 with @xmath130 and @xmath85 depending only on @xmath131 . [ lemma - interior ] let @xmath11 satisfy @xmath109 , and @xmath38 be a continuous function in @xmath132 solving @xmath133 then there exists @xmath74 and @xmath85 , depending only on @xmath134 and @xmath135 , such that @xmath136 [ thm : reg2 ] let @xmath137 for given bounded function @xmath138 , the following is true for @xmath38 satisfying @xmath139 * @xmath140 for @xmath141 given in theorem [ lemma - interior ] with the estimate @xmath142 * if , in addition , @xmath143 for some @xmath144 and @xmath38 satisfies @xmath124 , then @xmath145 where @xmath146 with @xmath147 given in ( a ) . moreover , we have the estimate @xmath148 where @xmath149 is a constant depending only on @xmath150 and @xmath76 . here we discuss how to localize solutions of neumann - boundary problems . for given @xmath24 , let us define @xmath151 [ localization ] for given @xmath152 , suppose @xmath153 solves the following : @xmath154 if @xmath155 , then we have @xmath156 without loss of generality , let us set @xmath157 , @xmath158 , .and consider the function @xmath159 then @xmath160 , and @xmath161 on @xmath162 , @xmath163 on @xmath164 , and @xmath165 on @xmath166 . therefore by comparison principle ( theorem [ thm : comp ] ) we have @xmath167 similarly one can show that @xmath168 in @xmath169 . now we conclude by evaluating @xmath170 in the region @xmath171 . the following holds as a corollary of above lemma with @xmath172 : [ localization2 ] let @xmath173 satisfy the following in the viscosity sense : * @xmath174 , @xmath175 in @xmath176 . * @xmath177 on @xmath178 ; * @xmath179 on @xmath180 ; * @xmath181 . then we have @xmath182 from corollay [ localization2 ] , the following holds . [ general : cp ] suppose @xmath183 satisfies the following in the viscosity sense : * @xmath184 in @xmath185 ; * @xmath113 on @xmath108 , @xmath186 @xmath187 ; * @xmath114 on @xmath188 , where @xmath189 is continuous . then @xmath113 in @xmath185 . here we state the homogenization result obtained in @xcite . [ original_convergence ] let @xmath11 satisfy @xmath109 . then there exists a unique operator @xmath190 satisfying @xmath191 such that the following is true : let @xmath107 be a bounded domain in @xmath1 whose boundary is @xmath192 and consists of two nonempty parts @xmath108 and @xmath51 , and suppose that @xmath193 are continuous . if @xmath23 solves @xmath194 then @xmath23 uniformly converges to the solution @xmath38 of @xmath195 moreover there exists @xmath74 such that @xmath196 , the theorem is valid in the setting of the unbounded domain @xmath197 with @xmath198 , @xmath199 and with uniformly bounded @xmath23 , though in this setting we would not have the quantitative estimate . using above result we can apply compactness arguments to show the following . [ ext ] let @xmath12 be a positive constant and let @xmath200 be bounded and hlder continuous . for given @xmath24 , let @xmath201 be the unique bounded viscosity solution of @xmath202 then for any @xmath71 , there exists @xmath203 only depending on @xmath12 , the bound and hlder exponent of @xmath204 such that @xmath205 is the unique bounded viscosity solution of @xmath206 \1 . for fixed @xmath52 , this is a consequence of theorem [ original_convergence ] and lemma [ localization ] . let us fix @xmath12 and @xmath207 . to show that @xmath203 is independent of @xmath52 , suppose not . then for some @xmath71 there exists a sequence of directions @xmath208 and a sequence of uniformly bounded , uniformly hlder continuous functions @xmath209 such that @xmath210 , @xmath211 locally uniformly converges to @xmath212 and the the constants @xmath213 given by goes to infinity as @xmath214 . let us denote @xmath215 and @xmath216 by the corresponding solutions given in above theorem with @xmath217 . since @xmath218 is uniformly hlder , along a subsequence it converges to a function @xmath219 . using the stability properties of viscosity solutions as well as theorem [ original_convergence ] , one can check that @xmath219 solves @xmath220 now we choose @xmath203 so that holds for @xmath221 and @xmath222 , yielding a contradiction . now we are ready to study the averaging properties for solutions of @xmath223 ) , beginning with the strip domain : later we will approximate the general domain with these solutions . for given @xmath224 , let us define @xmath225 consider @xmath23 solving @xmath226 the main theorem in this section is given below : [ thm : planar ] for given @xmath224 and @xmath227 , let @xmath228 solve @xmath229 . then the following holds : * for irrational directions @xmath52 , there exists a unique constant @xmath230 $ ] independent on the choice of @xmath231 such that @xmath23 locally uniformly converges to the linear profile @xmath232 * for rational directions @xmath52 , if @xmath233 goes through the origin ( that is if @xmath234 ) , then the statement in ( i ) holds for @xmath52 as well . * let @xmath52 be an irrational direction . then for @xmath23 solving @xmath235 and @xmath38 as given in , we have the following estimate : for any @xmath74 , there exists a constant @xmath236 such that @xmath237 where @xmath238 is as given in lemma [ lemma - m ] ( ii ) . the proof of theorem [ thm : planar ] consists of several lemmas below : the outline largely follows that of the corresponding result ( theorem 2.4 ) in @xcite : for clarity we present the full proof below . due to the uniform hlder regularity of @xmath239 ( theorem [ thm : reg2](a ) ) , along subsequences @xmath240 in @xmath241 . note that there may be different limits along different subsequences @xmath242 . below we will show that if @xmath52 is an irrational direction then all subsequential limits of @xmath239 coincide . first we present a flatness " result : we show that , sufficiently ( @xmath243- ) away from the neumann boundary @xmath244 , @xmath228 is almost a constant on hyperplanes parallel to @xmath233 . [ claim 2 ] let us fix @xmath245 and @xmath246 , and let us denote @xmath247 . then for any @xmath248 with @xmath249 , and for any @xmath74 , there exists a constant @xmath250 such that for any @xmath251 @xmath252 with @xmath253 as given in lemma [ lemma - m ] ( ii ) . let @xmath254 . by lemma [ lemma - m ] ( ii ) , there exists @xmath255 such that @xmath256 @xmath257 and @xmath258 let us compare @xmath259 with @xmath260 in the domain @xmath261 . due to theorem [ thm : reg2 ] , and the fact that @xmath262 it follows that @xmath263 moreover , due to the hlder continuity of @xmath228 up to the dirichlet boundary @xmath264 we have @xmath265 lastly , due to the fact that @xmath266 , we have @xmath267 in @xmath268 . putting together above estimates , we can apply lemma [ localization ] to @xmath269 with @xmath270 and @xmath271 to obtain that @xmath272 now we can conclude , by observing @xmath273 where the second inequality is due to with @xmath274 and the third inequality is due to the fact that @xmath275 and theorem [ lemma - reg ] . next we define the average linear profile for @xmath23 : pick a point @xmath276 and for each @xmath277 let us define @xmath278 by @xmath279 such that @xmath280 by lemma [ claim 2 ] and by the comparison principle ( theorem [ thm : comp ] ) , we obtain that @xmath281 where @xmath282 and @xmath283 are linear functions which satisfies @xmath284 on @xmath285 and @xmath286 from and the definition of @xmath287 the following estimate holds : for @xmath288 , latexmath:[\[\label{c2 } that @xmath239 uniformly converges to linear profiles along subsequences . [ claim 4][theorem 3.1 , ( i ) ] if @xmath52 is an irrational direction , then there exists a unique @xmath290 which only depends on @xmath52 such that @xmath228 uniformly converges to a linear profile @xmath291 . let @xmath292 be sufficiently small , let @xmath293 , and let @xmath294 solve @xmath295 in @xmath296 and @xmath297 solve @xmath295 in @xmath298 . we will show that @xmath299 can be made arbitrarily small if we choose @xmath54 sufficiently small , independent of the choice of @xmath300 and @xmath301 . this concludes theorem 3.1 ( i ) . in the proof below we strongly use the fact that @xmath302 and thus that linear functions solve @xmath124 . let us re - scale @xmath303 and denote by @xmath304 and @xmath305 as the corresponding neumann boundaries of @xmath306 and @xmath307 . we first translate @xmath308 by @xmath309 such that @xmath308 and @xmath310 are close . by ( iii ) of lemma [ lemma - m ] , there exist @xmath311 such that @xmath312 where @xmath313 and @xmath314 . hence after translating @xmath315 by @xmath316 , we may suppose that @xmath317 and @xmath318 are defined , respectively , on the extended strips @xmath319 where @xmath320 and @xmath321 . without loss of generality , we may assume that @xmath322 . since @xmath323 and @xmath324 , we have @xmath325 regularity of @xmath326 ( see theorem [ thm : reg2 ] ) , we conclude that there exists a constant @xmath327 independent of @xmath54 and @xmath328 such that latexmath:[\[\label{boundary2 } let @xmath278 be given by . then by @xmath330 means that the slope of @xmath331 in the direction of @xmath52 ( i.e. @xmath332 ) is between @xmath333 on @xmath334 . now let us consider the linear profiles @xmath335 whose respective slopes are @xmath336 and @xmath337 . here @xmath338 and @xmath339 are chosen to match the dirichlet boundary data of @xmath340 . now we define @xmath341 and @xmath342 where @xmath343 and @xmath344 are constants satisfying @xmath345 ( see figure 2 . ) note that , due to , in @xmath346 we have @xmath347 and thus it follows that @xmath348 and @xmath349 satisfies , in the viscosity sense , @xmath350 \4 . let us define @xmath351 where @xmath352 is given in . then @xmath353 and @xmath354 respectively solves @xmath355 and @xmath356 since @xmath357 on @xmath358 , from the comparison principle for @xmath359 ( theorem [ general : cp ] it follows that @xmath360 hence we conclude @xmath361 where @xmath362 is the slope of @xmath363 , which is defined as in for @xmath364 . now we can conclude . the proof of the following lemma is immediate from lemma [ claim 4 ] and . [ irrational_error ] [ error estimate : theorem [ thm : planar ] ( iii ) ] for any irrational direction @xmath52 there is a unique homogenized slope @xmath365 and @xmath366 such that for @xmath367 the following holds : for any @xmath246 and @xmath228 solving @xmath368 in @xmath369 , @xmath370 [ claim 4 ] [ theorem [ thm : planar ] ( ii ) ] let @xmath52 be a rational direction . if the neumann boundary @xmath233 passes through @xmath371 , then there is a unique homogenized slope @xmath31 for which the result of lemma [ irrational_error ] holds with @xmath372 the proof is parallel to that of lemma [ claim 4 ] . let @xmath315 and @xmath340 be as given in the proof of lemma [ claim 4 ] . note that , since @xmath373 and @xmath374 have their neumann boundaries passing through the origin , @xmath375 without translation of the @xmath376 variable , and thus we do not need to use the properties of hyperplanes with an irrational normal ( lemma [ lemma - m](iii ) ) to estimate the error between the shifted neumann boundary datas . as mentioned in @xcite , if @xmath52 is a rational direction with @xmath377 , the values of @xmath378 on @xmath379 and @xmath380 may be very different under any translation , and thus the proof of lemma [ claim 4 ] fails . in this case @xmath228 may converge to solutions of different neumann boundary data depending on the subsequences . in the previous section we have shown that for an irrational direction @xmath381 , there is a unique homogenized slope @xmath31 for any solutions of @xmath368 in @xmath268 . in this section we investigate the continuity properties of @xmath31 as well as the mode of convergence for @xmath23 as the normal direction @xmath52 of the domain varies . for section 4 and 5 we assume the following additional condition on the homogenized operator @xmath40 as given in theorem [ original_convergence ] : @xmath382 as mentioned before , the condition is equivalent to saying that @xmath40 is rotation and reflection invariant . [ continuity ] let @xmath383 be as given in theorem 3 . , and suppose that @xmath40 satisfies . then @xmath384 has a continuous extension @xmath385 . more precisely for any @xmath386 and @xmath71 there exists @xmath387 such that the following holds : if @xmath388 and @xmath389 are irrational such that @xmath390 then we have * @xmath391 ; * for given @xmath246 , the solutions @xmath392 of @xmath393 in @xmath394 and the average slope @xmath395 given as in satisfies @xmath396 in the proof we indeed show that , for any directions @xmath388 and @xmath389 satisfying , the range of @xmath397 fluctuate only by @xmath207 , if @xmath54 is sufficiently small ( see ) . the fact that @xmath398 s are irrational is only used to guarantee that there is only one subsequential limit for @xmath395 , with @xmath399 . note that our statement uses @xmath52 as the reference direction and thus is a different type of estimate than those in theorem [ thm : planar ] ) . thanks to lemma [ irrational_error ] , it is enough to consider the case @xmath371 . the main idea in the proof of theorem [ continuity ] is to use @xmath52 as the reference direction and approximate @xmath45 by piecewise continuous functions , where each continuous parts are projections " of @xmath5 on @xmath400 ( see further description on the approximation below ) . for simplicity of the presentation , we will prove the theorem in @xmath401 : at the end of the proof will describe the modification required for @xmath1 . first let us describe the main ideas in the proof . we begin by introducing several notations . for notational simplicity and clarity in the proof , we assume that @xmath402 : we will explain in the paragraph below how to modify the notations and the proof for @xmath403 . let us define @xmath404 and for @xmath399 @xmath405 let us also define the family of functions @xmath406 + 1.\ ] ] ( see figure 3 ) . then @xmath407 is a @xmath408-periodic function with respect to @xmath409 . @xmath48_a remark for @xmath403 _ : in two dimensions , if @xmath52 is a rational direction different from @xmath410 , take the smallest @xmath411 such that @xmath412 mod @xmath413 . then we define @xmath414 , where @xmath415 , and @xmath407 is a @xmath416-periodic function . if @xmath52 is an irrational direction , take the smallest @xmath411 such that @xmath417 mod @xmath413 . then @xmath407 as defined above is almost @xmath416- periodic up to the order of @xmath207 with respect to @xmath418 . we point out that it does not make any difference in the proof divided in the following two subsections if we replace the periodicity of @xmath407 by the fact that @xmath407 s are periodic up to the order @xmath207 . @xmath48 * proof by heuristics * since the domains @xmath419 and @xmath420 point toward different directions @xmath388 and @xmath389 , we can not directly compare their boundary data , even if @xmath421 and @xmath422 cover most part of the unit cell in @xmath423 . to overcome this difficulty we perform a multi - scale homogenization as follows . first we consider the functions @xmath407 @xmath424 , whose profiles cover most values of @xmath5 up to the order of @xmath425 , where @xmath426 is the hlder exponent of @xmath5 . note that most values of @xmath5 are taken on @xmath421 and on @xmath422 since @xmath388 and @xmath389 are irrational directions . on the other hand , since @xmath388 and @xmath389 are very close to @xmath410 which is a rational direction , the averaging behavior of a solution @xmath228 in @xmath419 ( or @xmath420 ) would appear only after @xmath54 gets very small , as @xmath388 ( or @xmath389 ) approaches @xmath427 . let @xmath428 $ ] . if @xmath429 is chosen much smaller than @xmath207 , then we can say that the neumann data @xmath430 is ( almost ) repeated @xmath431 times on @xmath432 with period @xmath54 , up to the error @xmath433 . ( see figure 4 . ) similarly , on the next piece of the boundary , @xmath434 is ( almost ) repeated @xmath431 times and then @xmath435 is repeated @xmath431 times : this pattern will repeat with @xmath436 ( @xmath437 mod @xmath438 ) . if @xmath431 is sufficiently large , i.e. , if @xmath439 is sufficiently small compared to @xmath207 , the solution @xmath228 of @xmath440 in @xmath419 will exhibit averaging behavior , @xmath441-away from @xmath233 . more precisely , on the hyperplane @xmath442 located @xmath443-away from @xmath233 , @xmath228 would be homogenized by the repeating profiles of @xmath407 ( for some fixed @xmath444 ) with an error of @xmath433 . this is the first homogenization of @xmath228 near the boundary of @xmath419 : we denote , by @xmath445 , the corresponding values of the homogenized slopes of @xmath228 on @xmath442 in each @xmath441-segment . now more than @xmath441 away from @xmath421 , we obtain the second homogenization of @xmath228 , whose slope is determined by @xmath445 , @xmath446 . in the proof this second homogenization is divided into two parts , in the middle region which is @xmath441 to @xmath447-away from @xmath233 and then in the rest of the domain @xmath419 . the homogenization argument in the middle region is to ensure that the oscillation of the operator @xmath11 in @xmath376-variable does not alter the behavior of the solution too much as @xmath52 varies . note that , due to after rotation and reflection we may assume that the arrangement of @xmath448 is the same for @xmath388 and @xmath389 . therefore the second homogenization procedure applied to @xmath388 and @xmath389 yields that @xmath449 is small . below we will present a rigorous proof for theorem [ continuity ] , with above heuristics in mind . let @xmath71 be given . we will first prove a series of lemmas which analyzes localized versions of @xmath23 in the three areas of the domain ( near - strip , middle - strip , and inner - strip ) . based on the estimates , in section 4.3 we will then put together the localized barriers to construct appropriate test functions ( sub- and supersolutions ) of @xmath450 . the goal is , to compare these test functions with @xmath392 to show the following : for given @xmath71 and for @xmath398 and @xmath54 satisfying the assumptions given in theorem [ continuity ] , there exists a constant @xmath451 such that @xmath452 from above inequality theorem [ continuity ] follows . let @xmath453 and @xmath407 as given above . we continue with the notations . given @xmath71 , let @xmath203 be as given in theorem [ ext ] , for neumann boundary data @xmath454 with @xmath455 . let us choose irrational unit vectors @xmath456 such that @xmath457,\ ] ] where @xmath458 and @xmath459 . let us also define @xmath460 , \,\,\ , m = [ \dfrac{\delta}{\eta_0}].\ ] ] and lastly suppose @xmath461 the condition is to ensure that @xmath462 , i.e. we consider sufficiently small @xmath54 so that @xmath395 to approach the correct averaged slope @xmath463 . with above definition of @xmath54 and @xmath431 , consider the strip regions @xmath464 \times \rr \,\,\hbox { for } \,\ , k \in \zz.\ ] ] let @xmath465 $ ] denote @xmath466 in modulo @xmath438 , where @xmath467}+1 $ ] . note that , since @xmath468 , the function @xmath469 defined in is ( almost ) repeated @xmath431 times on @xmath470 , where @xmath471 . this fact and the hlder continuity of @xmath5 yield that @xmath472 similarly one can define @xmath473\times\r$ ] for @xmath474 to approximate @xmath5 on @xmath475 . let us first consider the averaging behavior of solutions near ( @xmath441- close to ) the neumann boundary . for the strip domain @xmath476 , let @xmath228 solve @xmath477 similarly we define @xmath478 in the strip domain @xmath479 first we replace the neumann data @xmath5 with the family of functions @xmath436 introduced in . let @xmath306 and @xmath480 be solutions in @xmath481 and @xmath482 , satisfying @xmath483 and @xmath484 next we localize the neumann boundary data : for given @xmath485 , let @xmath486 and @xmath487 solve @xmath488 and @xmath489 we will show that the profile of @xmath228 and @xmath490 does not change too much with above perturbation and localization of the neumann data . [ a ] there exists a dimensional constant @xmath85 such that the following holds : for each @xmath491 , @xmath492 and @xmath493 we will only prove the lemma for @xmath228 and @xmath331 . let us define @xmath494 \1 . since @xmath495 on @xmath496 , and @xmath497 on @xmath233 , we obtain that @xmath498 by adding linear profiles @xmath499 to @xmath228 , and to @xmath331 and then apply the comparison principle to get respective orders . next we compare @xmath331 and @xmath500 . let @xmath501 . observe that @xmath502 by the construction of @xmath436 and by the hlder continuity of @xmath5 . it follows that @xmath503 lastly note that , since @xmath266 , we have @xmath504 we would like to apply the re - scaled ( replacing @xmath441 as the unit scale ) version of lemma [ localization ] to @xmath331 and @xmath505 in @xmath506 with @xmath507 . we then obtain @xmath508 next we show that @xmath486 and @xmath509 are almost linear , and that they do not differ from each other very much . let @xmath510 be given as the approximating slope of @xmath486 as in , i.e. choose a point @xmath511 and define @xmath512 by @xmath513 similarly one can define @xmath514 and @xmath515 . [ aa ] let @xmath516 be as given in theorem [ lemma - interior ] , and let @xmath517 $ ] . then there exists a dimensional constant @xmath85 such that the following holds true : respectively on @xmath518 and @xmath519 , latexmath:[\[\label{flat } moreover latexmath:[\[\label{baby : continuity } \1 . we will prove the first inequality on @xmath486 . fix @xmath521 . due to the fact that @xmath522 , the following holds : for any @xmath523 , there exists @xmath524 such that @xmath525 consider the function @xmath526 which still solves @xmath527 in @xmath528 . we will compare @xmath486 and @xmath529 in the domain @xmath530 without loss of generality , we may assume @xmath531 where @xmath532 . observe that comparison with linear profiles yields that @xmath533 moreover , since @xmath534 is constant in @xmath535-variable , the last property of ( [ lp ] ) and theorem [ thm : reg2 ] yields that @xmath536 let @xmath537 . then by applying the comparison principle in @xmath107 we have @xmath538 since @xmath539 on @xmath540 , we get @xmath541 now we conclude that @xmath542 where the last inequality follows since @xmath543 . \3 . to prove the statement for @xmath480 , one can argue as above , by replacing @xmath544 by @xmath545 and @xmath481 by @xmath546 , and @xmath442 with @xmath547 . we proceed to prove . recall that @xmath548 . first we compare @xmath549 with @xmath550 , solving @xmath551 we will show that @xmath552 to show that @xmath550 and @xmath553 are close . due to , it is enough to argue in the @xmath441-neighborhood of the origin . note that the neumann boundary of @xmath550 and @xmath486 only differ by @xmath554 in @xmath555-neighborhood of the origin . therefore theorem [ thm : reg2 ] yields that latexmath:[\[\label{eqno1 } moreover , since @xmath266 , we have @xmath557 on @xmath558 . therefore yields that @xmath559 yields that @xmath560 let us take @xmath561 to conclude . the following corollary is immediate from lemma [ a ] and lemma [ aa ] as well as theorem [ thm : reg2 ] . [ coco ] denote @xmath562 , then for @xmath563 given as in lemma [ aa ] @xmath564 from the previous section we have seen that the solutions @xmath565 and @xmath565 average in the unit of @xmath441 , when the solution is @xmath441-away from the neumann boundary , and moreover that they average to a similar value . if the operator @xmath11 was homogeneous we could use this result and perform a second homogenization , and use and to conclude . however we have to be careful with the inhomogeneities of @xmath11 , such that different normal directions of the hyperplanes pointing toward @xmath388 and @xmath389 does not make a difference in the way the respective solutions average out . this is what we are going to analyze in this subsection . let @xmath566 , and let us consider the domain @xmath567 note that by @xmath568 . let @xmath569 as given in corollary [ coco ] , and let @xmath278 solve @xmath570 next let @xmath571 solve @xmath572 we will show that @xmath278 and @xmath571 are close to each other . [ lem35 ] there exists @xmath573 such that @xmath574 let us apply theorem [ ext ] with @xmath575 on @xmath576 ( or its interpolation so that @xmath577 is hlder continuous ) to the re - scaled function @xmath578 due to , @xmath579 , and thus theorem [ ext ] yields . latexmath:[\[\label{first } next we are going to show that @xmath571 is close to a linear profile away from @xmath442 . to see this , recall that the slopes @xmath581, ... ,@xmath582 are repeated @xmath12-times on the neumann boundary of @xmath107 . hence the homogenization arguments for the homogenized operator @xmath40 should apply in our setting , if @xmath12 is chosen sufficiently large . this is what we will show below in detail . let @xmath583 and @xmath584 be respectively the average slope of @xmath278 and @xmath571 respectively in @xmath107 as in . [ lem36 ] let @xmath76 given as in lemma [ aa ] . then there exists a dimensional constant @xmath85 such that @xmath585 choose a point @xmath586 . since @xmath571 has constant slope @xmath569 on @xmath576 with @xmath587 , and since the slopes @xmath581, ... ,@xmath582 are repeated on @xmath442 , for any @xmath588 such that @xmath589 there is a point @xmath590 such that @xmath591 and @xmath592 due to and theorem [ thm : comp ] we have @xmath593 . moreover @xmath594 where the inequality follows from the interior lipschitz regularity of @xmath595 ( theorem [ lemma - interior ] ) . with ( [ claim22 ] ) , we can conclude due to the @xmath596 regularity of @xmath595 . similarly , for the normal direction @xmath389 one can construct @xmath597 and @xmath598 accordingly in @xmath599 note that corollary [ coco ] implies latexmath:[\[\label{37 } recall that @xmath392 for @xmath399 solve @xmath450 in @xmath185 . with the estimates obtained in the previous section , we are now ready to construct our barrier for @xmath604 and @xmath605 . as mentioned before we will construct barriers in three separate regions and patch them up . let us construct it for the normal direction @xmath388 : let @xmath606 where @xmath76 is as given in . * in far - away region : * let us define @xmath607 where @xmath608 * in the middle strip : * next consider @xmath550 : the unique ( bounded ) viscosity solution of @xmath609 where @xmath610 is obtained by approximating @xmath569 such that @xmath611 and @xmath612 such @xmath613 satisfying - exists for @xmath614 since , due to , for any given @xmath615 , @xmath616 * in the near - boundary region : * last , let @xmath617 be the unique ( bounded ) viscosity solution of @xmath618 our goal is to show that the function @xmath619 is a supersolution of @xmath359 in @xmath620 . observe that , due to lemma [ lem35 ] and lemma [ lem36 ] , @xmath621 this means the patch - up of @xmath622 and @xmath550 together is a supersolution , i.e. , @xmath623 it remains to see if the same holds for the patch - up of @xmath550 and @xmath617 . note that @xmath550 is a constant on its dirichlet boundary @xmath624 . due to this fact and the small oscillation of @xmath613 given in , as well as theorem [ thm : reg2 ] and theorem [ ext ] applied to re - scaled versions of @xmath550 yields that @xmath625 has its @xmath626 norm of size @xmath627 in the scale of @xmath441 . more precisely , for any @xmath628 we have @xmath629 where @xmath630 let us fix a point @xmath521 . next we will invoke lemma [ localization ] for @xmath631 with @xmath632 and @xmath633 to obtain @xmath634 solves @xmath635 with the fixed boundary data @xmath550 replaced by the constant @xmath636 . suppose @xmath637 . due to corollary [ coco ] we have latexmath:[\[\label{kookoo22 } as well as yields that latexmath:[\[\label{kookoo3 } it follows from , and the @xmath596 regularity of @xmath550 and @xmath617 in @xmath441-scale that @xmath639 and thus the patch - up of @xmath550 and @xmath617 is a supersolution , i.e. @xmath640 summarizing , we have shown that @xmath641 is a supersolution of @xmath642 in @xmath643 . hence by comparison principle we obtain that @xmath644 in @xmath643 . in particular , @xmath645 similarly one can construct a subsolution @xmath646 , to show that @xmath647 above equation yields that @xmath648 similarly , @xmath649 moreover by we have @xmath650 this proves theorem [ continuity ] ( a ) . lastly observe that @xmath651 given in section 4.2.2 , has its only microscopic term in the neumann boundary data , which are constants in each segments @xmath544 of @xmath442 of size @xmath652 due to , with the constants repeating itself after @xmath653 pieces of @xmath544 : that is the neumann boundary data is periodic with period @xmath243 . since @xmath571 solves @xmath651 , @xmath584 converges uniformly to a limit @xmath451 as @xmath47 with the convergence rate only depending on @xmath207 and @xmath54 , as long as is satisfied . therefore , if necessary approximating the neumann data with functions of the form @xmath654 where @xmath204 is periodic and lipschitz , we conclude that latexmath:[\[\label{conclusion } if @xmath54 is sufficiently small . since @xmath71 is arbitrarily chosen and @xmath147 only depends on @xmath656 and @xmath17 , this proves the second claim of theorem [ continuity ] ( b ) . @xmath657 [ for dimensions higher than two ] in higher dimensions @xmath658 and for @xmath157 , one can define @xmath659 for @xmath660 $ ] . let us also define @xmath661 \times ... \times [ ( k_{n-1}-1)n\e , k_{n-1}n\e]\times \r.\ ] ] then parallel arguments in the previous two subsections apply to yield the corresponding result to theorem [ continuity ] . in this section we will use theorem [ continuity ] as well as stability properties of viscosity solutions to prove our main result . as given in the introduction , let @xmath0 be a bounded domain with @xmath37 boundary containing a unit ball @xmath662 . let us denote @xmath663 the outward normal vector of @xmath0 at @xmath10 . suppose that @xmath21 does not have any flat boundary parts in the following sense : for any @xmath664 and sufficiently small @xmath665 , there exists @xmath666 and @xmath667 such that latexmath:[\[\label{condition22 } [ main2 ] let @xmath0 and @xmath12 as given above , and let @xmath669 $ ] be as given in theorem [ continuity ] . consider @xmath228 solving @xmath670 and let @xmath40 be the homogenized operator given in theorem [ original_convergence ] with the assumption . then @xmath23 converges uniformly to the unique viscosity solution @xmath38 of @xmath671 let @xmath331 solve @xmath676 next let @xmath677 and let @xmath278 solve @xmath678 since @xmath21 is @xmath37 , we may assume that the hyperplane @xmath679 is contained in the @xmath680-neighborhood of @xmath681 in @xmath682 . after a translation , we may assume that @xmath371 . first note that @xmath331 and @xmath278 will oscillate at most of order @xmath687 in their respective domains @xmath688 and @xmath689 : this can be checked by comparison with linear profiles . let us consider the re - scaled functions @xmath690 then theorem [ thm : reg2 ] , the hlder continuity of @xmath5 , and the fact that @xmath691 and @xmath692 oscillates up to @xmath693 yields @xmath694 in their respective domains @xmath695 and @xmath696 , where @xmath74 is as given in theorem [ thm : reg2 ] . consequently we have @xmath697 let us choose @xmath466 sufficiently close to @xmath408 so that @xmath698 let us define @xmath699 so that @xmath700 and @xmath701 on @xmath442 , where @xmath442 is given in . now by , the comparison principle ( theorem [ thm : comp ] ) applies to @xmath702 and @xmath703 in the domain @xmath704 to yield we will only show ( a ) : parallel arguments apply to ( b ) . the proof follows the perturbed test function method introduced by evans @xcite , and consists of several round of perturbations of the test functions @xmath715 . it follows from previously known results ( see e.g. , @xcite,@xcite ) that @xmath716 moreover from comparison with linear functions it is straightforward to show that @xmath717 on @xmath12 . therefore if @xmath718 fails to be a subsolution of @xmath713 , then there exists a smooth function @xmath715 which touches @xmath718 from above at a boundary point @xmath664 and satisfies , for some @xmath71 , @xmath719 due to the regularity of @xmath715 and the continuity of @xmath41 , there exists @xmath720 such that @xmath721 let @xmath684 denote the outer normal of @xmath21 at @xmath722 . by adding @xmath723 to @xmath724 and restricting the domain to @xmath725 with @xmath726 , we may assume that @xmath727 except at @xmath274 in @xmath728 and @xmath729 where @xmath673 is the constant given in lemma [ close ] . this is possible due to our assumption , i.e. , since @xmath21 does not have any flat boundary parts . we now perturb @xmath715 and consider @xmath731 such that note that by ( b)-(c ) @xmath738 can have its minimum at @xmath588 in @xmath733 only if @xmath588 is more than @xmath739-away from @xmath722 . this step is necessary to exclude the possibility that @xmath740 is rational and the irrational normal direction near the minimum of @xmath738 is too close to @xmath740 compared to the size of @xmath54 to observe the averaging behavior . for convenience let @xmath741 , @xmath742 , and set @xmath743 . we will construct @xmath731 as @xmath744 , where @xmath745 and @xmath746 will be constructed below . let us first explicitly @xmath747 , which satisfies @xmath748 , @xmath749 as @xmath750 in @xmath0 , and @xmath751 let us define @xmath752 , and let us define @xmath5 by @xmath753 in @xmath754 , and @xmath755 where @xmath756 $ ] is a @xmath37 function which has support @xmath757 $ ] , @xmath758 for @xmath759 and @xmath760 . note that , since @xmath21 is @xmath37 , @xmath761 : in particular , for small @xmath762 , @xmath763 is empty in @xmath764 . it follows that @xmath5 is in @xmath765 . clearly @xmath5 goes to zero in @xmath766 as @xmath750 , and so it remains to check . straightforward computations yield @xmath767 from which follows . to ensure that the remaining term @xmath111 in the construction of @xmath731 would vanish as @xmath47 , we estimate the @xmath770 norm of @xmath771 . observe that , from the definition of @xmath5 , it follows that @xmath772 hence @xmath773 and in particular if @xmath774 , note that , due to , @xmath776 is uniformly bounded . therefore @xcite yields that @xmath777 are uniformly hlder continuous in @xmath778 ( proposition 2.1 and proposition 4.2 in @xcite ) . moreover , due to , @xmath779 vanishes as @xmath750 for @xmath780 . due to this fact and the equi - continuity of @xmath746 one can make use of the stability of @xmath770-viscosity solutions for @xmath781 for @xmath782 ( see @xcite ) and employ a compactness argument as in lemma 2.3 of @xcite to show that @xmath746 uniformly converges to zero in @xmath728 as @xmath750 . \4 . due to the definition of @xmath718 and ( a ) , for @xmath720 as given above we have @xmath786 on @xmath787 if @xmath54 is sufficiently small , along a subsequence of @xmath54 . moreover , @xmath738 has a local minimum at @xmath788 with @xmath789 as @xmath47 : we may add a constant to @xmath731 to assume that @xmath790 . let us denote @xmath791 . note that @xmath792 lies outside of @xmath793 due to the construction of @xmath731 , and due to the comparison principle applied to @xmath715 and @xmath38 , @xmath794 . this step is to extract only the normal component of @xmath731 at @xmath795 and make a new function @xmath796 with it , so that we do not have to worry about the tangential derivative of @xmath731 at @xmath792 . we also adjust @xmath228 accordingly . let us decompose @xmath731 into @xmath797 where @xmath798 then @xmath799 observe that , due to and the fact that @xmath800 is a linear function , @xmath801 still satisfies instead of @xmath715 . furthermore , since @xmath801 is smooth , we may choose @xmath54 sufficiently small to replace @xmath801 by a linear profile @xmath802 such that @xmath803 where @xmath149 depends on the @xmath37-norm of @xmath801 . lastly we will use theorem [ continuity ] to derive a contradiction to . let us choose @xmath820 and define @xmath821 so that @xmath822 solves @xmath823 in @xmath824 with @xmath825 . due to and theorem [ continuity ] , we can choose @xmath54 is sufficiently small so that @xmath826 then due to , the comparison principle applies to @xmath827 and @xmath822 to yield that @xmath828 after scaling back to @xmath278 , we get @xmath829 since @xmath830 for any @xmath76 due to theorem [ thm : reg2 ] , we choose @xmath831 to obtain @xmath832 if @xmath54 is sufficiently small : this yields a contradiction to , and we can conclude . due to proposition [ prop : main ] and the comparison principle ( theorem [ thm : comp ] ) , we have @xmath833 . but due to definition @xmath834 . therefore we conclude that @xmath835 : this is equivalent to the local uniform convergence of the sequence @xmath228 to @xmath836 , which is the unique viscosity solution of @xmath99 . abc m. arisawa , _ long - time averaged reflection force and homogenization of oscillating neumann boundary conditions , _ ann . poincare anal . lineaire * 20 * ( 2003 ) , pp . 293 - 332 . g. barles , f. da lio , _ local @xmath837 estimates for viscosity solutions of neumann - type boundary value problems , _ j. diff . eqns , * 225 * , no . 1(2006 ) , pp 202241 . g. barles , f. da lio , p. l. lions and p.e . souganidis , _ ergodic problems and periodic homogenization for fully nonlinear equations in half - type domains with neumann boundary conditions , _ indiana university mathematics journal * 57 * , no . 5 ( 2008 ) , pp 2355 - 2376 . a. bensoussan , j. l. lions , and g. papanicolaou . _ asymptotic analysis for periodic structures , _ volume * 5 * of studies in mathematics and its applications . north- holland publishing co. , amsterdam , 1978 . l. a. caffarelli and x. cabre , _ fully nonlinear elliptic equations , _ ams colloquium publications , vol . providence , ri . l. caffarelli , m.g . crandall , and m. kocan and a. swiech , _ viscosity solutions of fully nonlinear equations with measurable ingredients , _ cpam * 49 * ( 1997 ) , pp.365397 . l. a. caffarelli , p.e . , souganidis and l. wang , _ homogenization of fully nonlinear , uniformly elliptic and parabolic partial differential equations in stationary ergodic media _ , comm . applied math . * 58 * no . 3 ( 2005 ) pp . . s. choi , i. kim and lee , _ homogenization of neumann boundary data with fully nonlinear pdes , _ to appear in analysis & pde m. crandall , h. ishiii and p. l. lions , _ users guide to viscosity solutions of second order partial differential equations . * 27 * , no.1 ( 1992 ) pp . 167 . h. ishii , p.l . lions , _ viscosity solutions of fully nonlinear second - order elliptic partial differential equations , _ j. differential equations * 83 * ( 1 ) ( 1990 ) 2678 . k. a. lee and h. shagholian , _ homogenization of the boundary value for the dirichlet problem , _ submitted . h. tanaka , _ homogenization of diffusion processes with boundary conditions , _ stochastic analysis and applications , adv . related topics , vol . 7 , dekker , new york , 1984 , pp . 411 - 437 . a. swiech , _ @xmath838 interior estimates for solutions of fully nonlinear , uniformly elliptic equations , _ adv . differential equations . * 2 * no . 6 ( 1997 ) ,

transient diode laser absorption spectroscopy has been used to measure three strong vibronic bands in the near infrared spectrum of the ch , ethynyl , radical not previously observed in the gas phase . the radical was produced by ultraviolet excimer laser photolysis of either acetylene or ( 1,1,1)-trifluoropropyne in a slowly flowing sample of the precursor diluted in inert gas , and the spectral resolution was doppler - limited . the character of the upper states was determined from the rotational and fine structure in the observed spectra and assigned by measurement of ground state rotational combination differences . the upper states include a state at 6696 , a second state at 7088 , and a state at 7110 . by comparison with published calculations ( r. tarroni and s. carter , _ j. chem . phys _ * 119 * , 12878 ( 2003 ) and _ mol . phys_. * 102 * , 2167 ( 2004 ) ) , the vibronic character of these levels was also assigned . the observed states contain both and electronic character . several local rotational level perturbations were observed in the excited states . kinetic measurements of the time - evolution of the ground state populations following collisional relaxation and reactive loss of the radicals formed in a hot , non - thermal , population distribution were made using some of the strong rotational lines observed . the case of ch may be a good place to investigate the behavior at intermediate pressures of inert colliders , where the competition between relaxation and reaction can be tuned and observed to compare with master equation models , rather than deliberately suppressed to measure thermal rate constants .

1608.02624

the c@xmath0h radical ( ethynyl ) is an intermediate of significant importance in the chemistry of hydrocarbon combustion , particularly in acetylene - air mixtures.@xcite it is also related to the substituted ethynyl species which are the crucial intermediates in the hydrogen abstraction - acetylene addition ( haca ) mechanism that leads to the formation of polyaromatic ring species on the way to soot in fuel rich environments . @xcite recently , a haca - like mechanism has also been invoked to explain the growth of single - walled carbon nanotubes in low temperature catalyzed - chemical vapor deposition syntheses using acetylene and inert gas flowing over mildly heated metal - containing catalysts . in this model , the adsorbed c@xmath0h intermediate acts as a continuously regenerated `` catalyst '' in the process.@xcite the radical was one of the earliest molecules identified in interstellar space via its microwave spectrum . @xcite models of interstellar chemistry@xcite invoke reaction pathways involving c@xmath0h to explain the presence of polyaromatic hydrocarbons ( pahs ) in the interstellar and star - formation regions in space . finally , investigations of the atmosphere of titan , a moon of saturn , have shown that photochemistry in its upper atmosphere leads to significant concentrations of polyynes , assumed to be produced in similar reaction sequences to those involved in the terrestrial formation of pahs and soot , and including ethynyl - like intermediates in a chain reaction synthetic pathway . @xcite in support of the extensive reaction network modeling in multiple environments , rate constants have been measured for reactions of ethynyl radicals with various partners over a range of temperatures . @xcite following the astronomical observation of the ethynyl radical,@xcite the gas - phase microwave spectrum was recorded and the molecular structure determined in the 1980s.@xcite prior to this , laboratory observations had been limited to work in low temperature matrices.@xcite the infrared spectrum of the radical is complicated by the presence of a low - lying _ a _ @xmath3 excited electronic state which splits due to the renner - teller effect . the three resulting electronic surfaces are vibronically coupled , so the vibrationally excited levels of the ground _ x _ @xmath7 state all contain some @xmath8-state character . the curl@xcite group reported extensive work on the infrared spectrum in the 3@xmath9 m ( c - h stretching ) region and hirota@xcite recorded and analyzed the spectrum in the c - c stretching region and identified hot bands originating in the low frequency bending vibrational mode . all observed infrared spectra showed effects resulting from the strong vibronic interactions and local @xmath10-dependent perturbations due to background levels derived from combinations of lower frequency modes . many more spectroscopic studies were reported in subsequent years and pfelzer__et al.__@xcite survey the work up to the mid 1990 s . detailed understanding of the vibronic interactions leading to the observed spectra has relied upon _ ab initio _ methods to label the vibronic levels observed and to guide assignments , particularly at higher internal energies . peri _ _ et al.__@xcite made a state - of - the - art calculation of the vibronic states up to approximately 5000 @xmath2 in 1992 , and were able to identify the vibronic levels observed to that time . later , tarroni and carter@xcite reported more accurate calculations based on variational solutions of the nuclear motion on the coupled high level _ ab initio _ potential energy surfaces . importantly , the second of these papers reported absorption intensities to vibronically excited levels from the zero point level of the ground state thereby allowing vibronic state labels to be assigned to observed spectra with confidence in energy regions where the density of states is large . in low temperature matrices , forney _ _ et al.__@xcite had reported infrared spectra to much shorter wavelengths than previously seen , and hsu _ _ et al.__@xcite identified many higher lying vibronic states in laser - induced fluorescence ( lif ) spectra in the @xmath11 band system . the vibronic structure out to similar energy levels was also observed in an anion photodetachment experiment by zhou _ _ et al.__@xcite recently , sharp - williams _ et al . _ observed new gas phase spectra of c@xmath0h between 3600 and 4100 @xmath2 in a jet cooled sample,@xcite assigning the spectra by comparison to the tarroni and carter s results@xcite and identifying multiple local perturbations in the rotational structure of the upper vibronic levels . additional bands of c@xmath0h in the 4250@xmath124550 @xmath2 region have very recently been analyzed in a fourier transform emission spectrum , and a band near 6340 @xmath2 has been characterized by diode laser spectroscopy in a concentration modulated discharge source.@xcite both the calculations@xcite and matrix spectrum@xcite show that the most intense bands in the entire infrared and near - infrared spectra of the radical lie around 7100 @xmath2 or 1.4 @xmath9 m . the vibronic levels at this energy are strongly mixed , but derive much of their intensity from the near - vertical transition to the @xmath8 ( 0,0,2 ) state , in a region of spectrum where no gas phase spectra have yet been reported . we expect that the spectra in this wavelength region will be advantageous for kinetic or non - intrusive remote sensing applications . in addition to the strength of these near - infrared transitions , efficient spectroscopic sources and detectors are available at these wavelengths , and optical fibers are well suited for coupling probe radiation in and out of the sample . in this work , we report measurements of the c@xmath0h spectrum between approximately 6630 and 7135 @xmath2 by diode laser transient absorption spectroscopy . we have identified and rotationally assigned three band systems in this region , based on ground state combination differences @xcite and comparison to the predictions of tarroni and carter.@xcite in the course of our spectroscopic studies , we were initially puzzled to find that the decay rate of the vibronic ground state c@xmath0h radical signals was several times slower than expected , based on published room temperature rate constants @xcite for the reaction of c@xmath0h with our chosen photolytic precursor , cf@xmath13c@xmath6h , at the sample pressures we had chosen for convenient collection of radical spectra . a time - dependent spectral simplification observed in ar : cf@xmath13c@xmath6h mixtures that we attributed to vibrational thermalization was not observed in ar : c@xmath0h@xmath0 mixtures , where a dense set of both assigned and unassigned lines all decayed at comparable rates . the ground state radical lifetimes were also longer than anticipated in acetylene , based on published thermal rate constants and the sample composition . to investigate these effects further , we modified the detection system slightly for fixed frequency kinetic studies , and undertook some pressure and dilution studies , to see the effects of competitive reaction and relaxation on the time - dependent spectra . the kinetic observations and discussions are reported following the spectroscopic observations and analysis below . one important reason to characterize these strong bands of c@xmath0h in the 7000 @xmath2 region is for future use in kinetic investigations . in this preliminary study , the complex interplay of hot formation , vibronic relaxation and energy - dependent reactivity is evident but far from fully characterized . useful comparisons can be made with previous spectroscopic work where conditions reported for observing spectra of different hot or cold - band transitions in c@xmath0h had been optimized . additional comparisons can be made with previous kinetic spectroscopy studies of c@xmath0h reactions , where either complete or negligible vibronic relaxation prior to reaction is a recognized precondition for the measurement of reliable thermal rate constants . hirota and co - workers @xcite had used 193 nm photolysis of a 2:3 mixture of c@xmath0h@xmath0:ar at 2 torr total pressure to search for c@xmath0h transitions in the 1850 @xmath2 region . they reported many lines , but none that could be assigned to cold bands of c@xmath0h , unless they replaced the ar with h@xmath0 , d@xmath0 , co or ch@xmath14 as a buffer gas . he and n@xmath0 were also reported to be ineffective at vibrational relaxation . these observations are hard to understand given current understanding of the kinetics . the acetylene pressure was high enough to make the estimated lifetime of c@xmath0h shorter than the microsecond detector response , and adding hydrogen would seemingly only decrease the lifetime by means of an additional reactive loss channel.@xcite clear spectroscopic evidence of the @xmath15 fundamental of c@xmath0h was nevertheless obtained . hsu et al.@xcite used pump - probe delay time - dependent lif intensities to help assign a selection of bands originating in various vibrationally excited states of c@xmath0h ( @xmath16 ) . they reported successively longer lifetimes for lower energy states , but no resolved growth kinetics for any bands , following 193 nm excitation of c@xmath0h@xmath0/h@xmath0 mixtures . in a subsequent report,@xcite the substitution of sf@xmath17 for ar , he , or h@xmath0 in the c@xmath0h@xmath0 photolysis sample was found to increase the relative intensity of otherwise weak lif bands originating in low bending levels of c@xmath0h , consistent with enhanced vibrational relaxation by sf@xmath17 . no lif transitions originating from the ground vibrational state of c@xmath0h ( @xmath16 ) have been reported , presumably due to poor franck - condon factors . for kinetic spectroscopy studies , vibronic relaxation prior to reaction is typically tested by varying the pressure or choice of an inert buffer to verify insensitivity of the observed reaction rates.@xcite direct measurements of hot band relaxation@xcite are preferable , but not always feasible , to confirm the timescale of vibrational thermalization . using the newly assigned bands , we report the results of our investigation into pressure - dependent kinetics of state - resolved relaxation in competition reaction with two c@xmath0h precursors : c@xmath0h@xmath0 and cf@xmath18c@xmath0h . the vibronic relaxation appears to be slower than has often been claimed . the transient absorption spectra were recorded using a spectrometer very similar to that described previously . @xcite ethynyl radicals were formed by 193 nm arf excimer laser photolysis of a sample of cf@xmath19c@xmath20h ( trifluoromethyl acetylene , or 3,3,3 trifluoropropyne ) obtained from synquest laboratories , diluted in argon . a sacher laserteknik tunable diode laser , model tec 500 , with output centered at 1450 nm was the spectroscopic source . one portion of the laser power was sent to a wavemeter ( burleigh wa-1500 ) , and another portion was used as a reference beam for active noise reduction . the main beam was directed through a herriott - type absorption cell with 35 passes between mirrors separated by 2 meters . signal and reference beams were imaged onto an _ a.c._-coupled dual ingaas photoreceiver ( new focus 1617 ) . the excimer laser beam , with typical pulse energies of 30 mj in a 3 @xmath21 beam , passed through the center of the absorption cell and overlapped the probe beam in the central portion of the cell . we estimate the effective absorption path length through the photolysis volume was about 25 meters . mixtures of ar and cf@xmath19c@xmath20h in a 1:1 ratio were prepared in a large storage flask at pressures of approximately one atmosphere . this gas mixture flowed slowly at 5 sccm though the cell . total pressures used for the spectral scans were typically 1 torr . mixtures of acetylene in ar were also used for comparison in some spectral regions , producing similar but weaker spectra with generally faster decay rates . because of the stronger and longer - lasting signals , the systematic spectroscopy scans were carried out with the cf@xmath18c@xmath20h precursor . some additional exploratory measurements of the relaxation and reaction kinetics were made at fixed probe frequencies on selected ground state ethynyl radical transitions , as a function of precursor and inert buffer gas ( argon ) concentrations . these measurements used a _ d.c._-coupled balanced photoreceiver ( new focus 1817 ) , with sequential measurement and off - line subtraction of on- and off - resonant signals to remove the reproducible background associated with acoustic transients and increase the fidelity of the transient signals . tests with varying flow rates or a reduction of the excimer repetition rate below its typical value of 10 hz showed no sign of secondary photolysis or sample depletion . initial spectroscopic searches for the strong bands observed in the low temperature matrix spectrum @xcite and predicted by tarroni and carter@xcite were carried out by photolysis of a slowly flowing sample of cf@xmath18c@xmath20h in ar at 193 nm . transient absorption bands presumed to be due to c@xmath20h were detected throughout the search region . further support for the attribution of absorption lines to c@xmath20h was obtained by observing some of the same lines in spectra using c@xmath20h@xmath20 as the photolytic precursor . spectra recorded immediately following the photolysis pulse were much more complicated than spectra at longer delay times , which we attribute to collisional cooling of an initially formed hot distribution . the time development of signals in samples containing c@xmath6h@xmath6 was quite different to that observed with cf@xmath18c@xmath20h , as might be expected due to differences in absorption cross section , quantum yields for c@xmath6h , the initial state distributions , relaxation kinetics , and rates of reaction with the precursors . some further kinetic investigations are presented below , but for the purposes of the spectroscopic studies , cf@xmath18c@xmath20h was found to provide larger signals and simpler spectra after partial relaxation , when compared to c@xmath22h@xmath6 samples . therefore , all of the subsequent spectroscopic measurements were made using cf@xmath18c@xmath20h precursor in ar . examples of a section of the observed spectra near 7125 @xmath2 at short and long time delays with the cf@xmath18c@xmath20h precursor are shown in figure [ figure1 . ] . this spectrum was obtained by averaging 40 photolysis laser shots per wavelength step ( typically 0.005 @xmath2 ) in a 1:1 mixture of cf@xmath18c@xmath20h and argon at a total pressure of 1.0 torr . figure [ figure1.](a ) shows a complex c@xmath6h spectrum at zero time delay following the photolysis pulse with a time gate of 1.0 @xmath23s . after a 3.6 @xmath23s time delay , the congestion is substantially reduced , leaving a much simpler spectrum , as shown in figure [ figure1.](b ) . two branches , each with a bandhead , can be identified and assigned to _ _ r__@xmath24 and _ _ r__@xmath6 branches of a @xmath25 band . perturbations starting at _ _ r__@xmath6(15.5 ) and _ _ r__@xmath24(16.5 ) are also evident in figure [ figure1 . ] . the strong band observed in the ne matrix spectrum at 7064 @xmath2 @xcite has been attributed to a combination of two bands,@xcite a @xmath25 band calculated at 7106 @xmath2 and a @xmath26 band at 7099 @xmath2 , consistent with the sub - band identifications above . the @xmath1 lower state in both of these transitions is the ground state of the radical , and analysis of the cold spectra proceeded by searching for ground state combination differences based on the ground state constants of gottlieb et al.@xcite assuming vibronic bands of the type predicted . @xmath1 state energy levels are described by a rotational quantum number ( @xmath10 ) and a parity label ( + or @xmath12 ) with @xmath27 ( f@xmath28 ) or @xmath29 ( f@xmath0 ) , where @xmath30 is the total angular momentum excluding spin . the lowest level @xmath31 ( @xmath32 , f@xmath33 ) has ( + ) parity and the parity labeling alternates with @xmath30 for @xmath1 states . as a result , _ p- _ and _ r- _ branches are doubled . the doublet splitting is illustrated in figure [ figure2 . ] ( b ) for lines in the @xmath34 transition at 7111.4 @xmath2 and is due to the difference in the spin - rotation splittings between the ground and the excited states . the early time spectrum in figure [ figure2.](a ) shows additional unassigned , presumably hot band , lines that are gone at the time of the delayed spectrum shown in figure [ figure2.](b ) . a more detailed comparison of the time dependence of ground state and vibronically excited states for different sample mixtures and pressures is presented in section [ sec : kinetics ] below . the energy levels of the @xmath3 state are also labeled by rotational ( @xmath10 ) and parity quantum numbers . two sub - states @xmath35 and @xmath36 result from spin - orbit coupling . analysis below shows the spin - orbit constant , a , is small and negative , therefore the @xmath36 and @xmath35 energy levels map onto f@xmath28 and f@xmath0 , respectively . as a result , 6 main branches ( p@xmath28 , p@xmath0 , q@xmath28 , q@xmath0 , r@xmath28 , and r@xmath0 ) were observed . the four additional satellite branches , with f@xmath37f@xmath0 were not observed here . seeking further support for some of the @xmath25 assignments , we searched the 6700 @xmath2 region for hot band transitions coming from the _ x _ ( 0,1@xmath38,0 ) ( @xmath39 symmetry ) to the same upper level states . the fundamental bending vibrational frequency is 371.6034(4 ) @xmath2 ; rotational and fine structure parameters for this level were determined by kanamori et . @xcite we did not find these hot band transitions above our noise level at the calculated frequencies , likely due to poor franck - condon factors . instead , an additional @xmath26 band was identified in this region . figure [ figure3 . ] shows a portion of this spectrum at 0 @xmath9s and 3.6 @xmath9s , with some _ _ r__@xmath24 , _ _ r__@xmath22 , _ _ p__@xmath24 and _ _ p__@xmath6 branch lines labeled . the band origin of this @xmath26 is at 6695.678 @xmath40 . complete data files of the two spectral regions between 6630 and 7135 @xmath2 studied here , at three different times ( 0 , 2.6 and 3.6 @xmath23s ) are included as supplementary material.*(supplemental material ) * the @xmath1 rotational hamiltonian can be written as : @xmath41 leading to the energy level expression : @xmath42 the b@xmath43 , d@xmath43 , and @xmath44 rotational and spin - rotational ground state constants were fixed to the literature values @xcite . the energies for the @xmath3 state were modeled by including the spin - orbit interaction ( a ) , rotational constant ( b ) quartic centrifugal distortion correction ( d ) , the spin - rotation ( @xmath45 ) , and the @xmath46-doubling ( @xmath47 ) . in the hund s case ( a ) basis limit , the effective hamiltonian for the @xmath3 state is @xcite : @xmath48 in equation [ eq3 ] , @xmath49 and @xmath50 are the shift operators of @xmath51 , the total angular momentum in the absence of nuclear spin , and of @xmath52 , the total electron spin , and @xmath53 is the azimuthal coordinate of the unpaired electron . the energies of the @xmath3 state were determined by numerical diagonalization of the hamiltonian matrix representation of dimension 4x4 constructed in a sequentially coupled , non - parity conserving hund s case ( a ) basis set . the energy levels can be modeled using either the hund s case ( a ) basis set , which is a good approximation for the low - rotational levels or hund s case ( b ) basic set , which is better for higher rotational levels ( n @xmath54 11 ) . the calculated energy levels are the same in both hund s case ( a ) and case ( b ) limits . @xcite in the @xmath26 band at 7088 @xmath2 , 80 transition wavenumbers were identified via combinational differences using the ground state parameters constants of gottlieb et al.@xcite however , only 32 transition wavenumbers were entered directly into the fit to the upper state parameters , due to multiple perturbations in the upper state energy levels . deviations in excess of 0.015 @xmath2 , more than three times the measurement error , were considered evidence of local perturbations strong enough to exclude a measured transition from the fit . the measured line centers are presented in table [ table1 ] , along with the assignments and differences between the calculated and observed transition wavenumbers . similarly , 66 transition wavenumbers in the @xmath26 band at 6696 @xmath2 were identified by combination differences using the same ground state parameters , and 49 transition wavenumbers were used in the fit . the line centers are presented in table [ table2 ] , along with the assignments and differences between the calculated and observed transition wavenumbers . the transitions from @xmath34 bands at 6696 and 7088 @xmath2 were least squares fit to the simple energy expressions in equation [ eq2 ] , described above , to determine the excited state constants including b@xmath55 , d@xmath55 , @xmath56 and the origin for both upper @xmath57 states . the standard deviation of the fits were 0.004 @xmath58 and 0.009 @xmath58 for the bands at 6696 and 7088 @xmath2 respectively . the optimized parameters and associated errors are given in table [ table4 ] . in the @xmath25 band at 7110 @xmath2 , 69 transition wavenumbers were identified via combinational differences using the ground state parameters . of these , 57 were included in the least squares fit to determine the excited state parameters _ _ a__@xmath55 , _ _ b__@xmath55 , _ _ d__@xmath55 , @xmath59 , @xmath60 , and the band origin . the standard deviation of the fit is 0.010 @xmath58 . the optimized parameters and associated errors are given in table [ table4 ] . the transient absorption of c@xmath6h transitions for different photolytic precursors and buffer gas pressures offers a view of the competition between collisional thermalization and reactive loss . figure [ figure4 . ] illustrates the time - dependent absorption signals observed on the _ r_(4 ) line of the 7088 @xmath58 band at a constant cf@xmath13c@xmath6h precursor partial pressure of 50 mtorr with decreasing amounts of ar buffer gas . small corrections have been applied to normalize for changes in photolysis and probe laser intensity variations , so that the percent absorption signals are directly comparable . the rise rate becomes progressively slower as the ar pressure decreases , and the peak amplitude decreases . at 5 and 10 torr , the decay rate is about the same , and in good agreement with the published @xcite room temperature rate constant of 3.5@xmath6110@xmath62 @xmath63 molec@xmath64 s@xmath64 for the c@xmath6h + cf@xmath13c@xmath65h reaction , and the partial pressure of cf@xmath13c@xmath6h . the delayed maximum and slower decay at the lowest total pressure of 0.25 torr superficially resembles the expectations of slower vibrational relaxation . when the vibrational relaxation cascade is no longer rapid compared to the reactive loss , the observed decay rate will no longer be a measure of the reaction rate , as the relaxation will continue to provide a delayed source of radicals in the probed energy level . the absolute amplitudes tell an additional story , however . if the changing ar pressure only affected the thermalization rate , and the reaction rate depended only on the constant partial pressure of cf@xmath13c@xmath20h , the observed time dependence would be described by an exponentially decaying total population of c@xmath6h times a monotonically increasing , time - dependent fractional population in the detected state . a signal probing c@xmath6h ( 0,0,0 ) at low ar pressure could never exceed the signals at higher ar pressure under these conditions , contrary to our observations . the most likely explanation appears to be that the unrelaxed population in higher energy states does not react as rapidly as the thermalized c@xmath6h radicals , so that delayed relaxation extends the lifetime of the total c@xmath6h radical pool . a selection of r - branch lines probing rotational states from @xmath30=4 to @xmath30=15 in the 7088 @xmath2 band were measured to assess the time - dependent rotational distributions , and to test for the possibility of high rotational states affecting the kinetics . for a 1% mixture of cf@xmath18c@xmath0h in ar at 5 torr , the normalized time - dependent signals were superimposable throughout the rise and fall , as one would expect for rotational thermalization faster than the @xmath9s time scale . a plot of the data is included in the * supplementary material , figure s1*. we note that the apparent temperature of the stationary rotational distribution would be 270 @xmath66 10 k if the rotational line intensities are interpreted with unperturbed line strength factors . the apparently colder than ambient rotational temperature can be explained by an increasing rotational mixing with spectroscopically dark levels at higher @xmath10 . the upper level of this transition is a @xmath67 state whose @xmath68state character is predominantly @xmath69.@xcite this level will be mixed with the dark @xmath70 components of the same ( 030 ) level in the renner - teller mixed @xmath8 state , via a _ k_-type resonance@xcite that includes a @xmath71dependent mixing term . we have modeled this interaction and confirmed that a mixing coefficient defined as a fractional contribution of dark state wavefunction of @xmath72 , when combined with this @xmath73dependence , accounts the observed reduction in the intensity of the higher @xmath10 lines in the band . an analogous but contrasting set of pressure dependent measurements in argon / acetylene mixtures is illustrated in figure [ figure5 . ] . as with cf@xmath13c@xmath6h , increasing the ar partial pressure at fixed precursor pressure makes the signal rise more quickly to a larger maximum and a faster decay . higher pressures are required to reach a limiting decay rate , which at 10 torr , is still about 30% slower than expected , given the 50 mtorr partial pressure of c@xmath0h@xmath0 and a room temperature rate constant of 1.3 x 10@xmath74 @xmath75 molec@xmath76 s@xmath76 . @xmath77@xcite apparently even in the presence of ar at 10 torr the vibrational thermalization is not yet complete on the time scale of reaction with c@xmath0h@xmath0 . the differences between cf@xmath13c@xmath6h and c@xmath0h@xmath0 can be further illustrated by comparing the growth and decay of an absorption signal originating in the vibrationless state of c@xmath0h with an unassigned hot band line observed at 7111.470 @xmath2 with either precursor . this line was illustrated in figure [ figure2 . ] , and is typical of many , observed to decay quickly as the ground state lines grew in . figure [ figure6 . ] compares these signals for two different probed transitions on samples of the two photolytic precursors , each prepared as a 200:1 mixture of ar : precursor , measured at 10 torr total pressure . the large signals probe the @xmath78 level of the ground state , and illustrate the smaller signals and faster reaction for c@xmath0h@xmath0 than for cf@xmath13c@xmath6h . the smaller , faster signals , expanded in the inset figure contrast the hot band kinetics in the same two samples . without knowing the exact energy of the probed state , no quantitative conclusions can be drawn , but it can be seen that even a 200:1 ratio of ar buffer to precursor is insufficient open up a clear separation of time scales for vibrational relaxation and reactive loss of the vibrationless c@xmath0h in the case of c@xmath0h@xmath0 , since the decays of the vibrationless and hot band levels are similar , while it is adequate in the case of cf@xmath19c@xmath22h . this may be due to an initially higher extent of internal excitation in the acetylene photodissociation , perhaps also affected by differences in the intrinsic competition between reaction and relaxation for collisions of the hot radicals with the two different precursor molecules . the initial c@xmath0h internal energy distribution of the photofragments produced in acetylene photodissociation has been characterized by h atom rydberg time - of - flight measurements . @xcite the structured kinetic energy spectra indicate preferential excitation in high bending states , low rotational excitation , and a significant population at peaks near the energy of the @xmath8 state . the available energy for c@xmath0h fragments in the 193 nm photodissociation of cf@xmath18c@xmath0h is about 30% higher ( 198 kj mol@xmath76 vs. 151 kj mol @xmath76 ) , @xcite but the additional degrees of freedom in the cf@xmath18 co - fragment compared to h atoms provide statistical grounds to expect a cooler c@xmath0h internal energy distribution from cf@xmath18c@xmath0h than from c@xmath0h@xmath0 . no direct comparison has been reported , however , to our knowledge . a striking difference in the relative intensities of prompt and delayed spectra under the high concentration conditions initially used to investigate the spectroscopy was one of the observations that led us to the current kinetic measurements . in our preliminary spectroscopic measurements , a substantial spectral simplification was observed in the cf@xmath18c@xmath0h : ar samples during the first few microseconds , as illustrated in figures [ figure1.]-[figure3 . ] . selected short scans of c@xmath0h@xmath0:ar samples at the same 1:1 mixing ratio and total pressure of 1 torr showed the same transitions , different initial relative intensities , a more rapid decay , but no significant differences in the decay rates of the lines assignable to ground state levels and other unassigned lines that behaved like hot bands in the cf@xmath18c@xmath0h : ar sample . we now attribute this behavior to the dominance of the reaction of c@xmath0h with c@xmath0h@xmath0 in competition with less efficient vibrational relaxation by either ar or c@xmath0h@xmath0 . we believe most of the reactive loss seen under these conditions is occurring from vibrationally excited c@xmath0h . previously , there have been no reported gas phase measurements of c@xmath6h in the 6700 @xmath40 and 7100 @xmath58 regions studied here . however , the origins determined for all three bands identified and assigned are in good agreement with the calculated values from tarroni et al . @xcite compared to the earlier matrix work,@xcite a new @xmath26 band was identified and assigned to the transition from the ground state to the x(0,8@xmath79,2 ) state again based on tarroni and carter.@xcite this state has a very small fraction of @xmath80 electronic state mixing , and was assigned to the vibrational combination level of @xmath81 state . the standard deviation for the fit of this band is smaller than the other two observed bands in this region , possibly reflecting its less mixed character . we estimate the overall intensity ratio between the three bands at 6696 , 7088 and 7110 @xmath2 is about 1:4:2 , respectively , which is in good agreement with the calculated intensity ratios.@xcite neglecting small nuclear hyperfine coupling effects , the observed discrete @xmath10-dependent perturbations require perturbing levels of the same @xmath10 and parity to lie nearby in energy . by examining plots ( figure [ figure7 . ] ) of observed@xmath12calculated energies against @xmath82 , for the upper @xmath1 state of the band at 7088 @xmath2 , local perturbations at @xmath83 ( @xmath12 parity ) , @xmath84 ( + parity ) , and for all @xmath10 associated with @xmath85 can be seen . for the first two levels , only one @xmath10-component of a given @xmath30 is perturbed , implying that the perturbing level must belong to a state where the parity does not@xcite alternate with @xmath30 , _ i.e. _ probably a nearby @xmath86 state . however , the observed @xmath87 state at 7110 @xmath2 is _ not _ the perturber because @xmath83 ( @xmath88 parity ) of the @xmath1 state is pushed down by a state that has @xmath83 at higher energy , while the @xmath83 ( @xmath12 parity ) of the 7110 @xmath2 @xmath87 state has a lower energy , by approximately 0.5 @xmath2 , as shown in figure [ figure8 . ] . the observed @xmath3 state level in question is also perturbed and pushed down in energy , as can be seen in table iii . similarly , @xmath89 of the @xmath1 state lies well below the energy of the corresponding level in the @xmath3 state which therefore can not cause the observed @xmath1 perturbation either . we conclude the perturbing level is a spectroscopically dark state . supplementary data available@xcite for the published calculations list all @xmath67 and @xmath3 levels up to 10 000 @xmath2 in energy , but there is no other state which could plausibly lead to the low - j perturbations , implying it is actually a @xmath90 or higher angular momentum state . above @xmath91 , the 7088 @xmath2 @xmath1 state suffers a much larger perturbation as can be seen in figure [ figure7 . ] . this perturbation affects both @xmath92 and @xmath12 parity levels equally so is qualitatively different to that seen for the two lower rotational levels . however , at these @xmath10-values , all c@xmath0h states will be close to case(b ) , and so the arguments used to infer the character of the perturber above can not be applied . the @xmath93 level is calculated@xcite to lie some 40 @xmath2 below the 7088 @xmath2 experimental level , and this could plausibly be the source of the observed high-@xmath10 perturbations in 7088 @xmath2 band . the @xmath25 band at 7110 @xmath2 shows strong @xmath94 branches and much weaker _ p _ branches . for example at the position of @xmath95(12.5 ) and @xmath96(11.5 ) , the observed spectroscopic features are barely above the noise . this unusual intensity pattern is not unknown for vibronic transitions involving @xmath97 and @xmath1 states@xcite and occurs when @xmath981 perturbations are present . the @xmath99 branch intensity pattern for transitions terminating in the mixed level can be very complex . however , the @xmath981 perturbation always affects the intensity of @xmath100 = + 1 ( @xmath101-branch ) and @xmath881(@xmath102-branch ) in the opposite direction . if the perturbation is strong , a whole branch ( e.g @xmath101-branch ) can be missing while the other ( e.g. @xmath102-branch ) gets much higher intensity as was observed in nco by dixon @xcite . the numerous additional lines observed in the prompt spectrum of the recorded region are likely hot bands of c@xmath0h . from the generally reliable excited vibronic energies calculated by tarroni and carter@xcite and experimental rotational constants for many low - lying excited levels , we may hope to identify some of these additional bands using combination differences , although we have so far been unsuccessful in this attempt . clearly , the kinetic analysis of vibronic relaxation would be more compelling with secure spectral assignments . because many free radicals are formed with substantial internal energy , and undergo rapid reactions that can compete with relaxation , there is recent interest in improving theoretical treatments of non - thermal effects in combustion @xcite and atmospheric chemistry . @xcite energetic radicals or adducts may react completely , prior to the establishment of any local thermodynamic equilibrium of internal states , under conditions when the reaction rates and even branching ratios may be rapidly changing with internal energy . the case of c@xmath6h may be a good place to investigate the behavior at intermediate pressures of inert colliders , where the competition between relaxation and reaction can be tuned and observed , rather than deliberately suppressed . the vibronic mixing of the _ a _ @xmath3 and _ x _ @xmath1 states in c@xmath6h can make the energy dependent reactivity more complex than in other radicals with a simpler electronic structure . the time dependence of ground state absorption signals for a photolytically generated hot radical like c@xmath0h will depend on the initial state distribution , the vibronic relaxation cascade due to energy transfer collisions with both inert and unsuccessfully reactive collision partners , as well as the reactive loss of the radicals in their evolving , nonthermal energy distribution . the energy dependence of the reaction rate plays a well - known and central role in the conventionally measured temperature - dependent rate constants of thermalized radicals . the energy transfer characteristics are typically less easily measured under the high energy conditions relevant to chemical activation or photodissociation . it is , however , frequently possible to devise conditions when the subtle details of the energy transfer are unimportant , and internal energy relaxation eigenvalues are well separated from the chemically significant eigenvalues of the master equation that describes the coupled reaction and relaxation processes . @xcite previous studies have shown dramatically increased reactivity of unrelaxed @xcite and partially relaxed @xcite c@xmath0h radicals from the photodissociation of acetylene in collisions with h@xmath20 , for example . the thermal rate constant for this reaction near room temperature is 5.7 @xmath61 10@xmath103 @xmath104 molec@xmath76 sec@xmath76 , increasing sharply with temperature , following nonlinear arrhenius behavior.@xcite it may be important to distinguish between the temperature dependence of the thermal rate constant and the variable reactivity of internally excited states of the radical . theoretical calculations@xcite of possible vibrational mode - specific effects in this reaction and its reverse do not find vibrational enhancement of the reaction with stretching excitation of c@xmath0h and even find an inhibition with bending excitation . the potential energy surface@xcite used for the full - dimensional dynamics calculations@xcite does not , however , include the low energy _ a _ state , which has a barrierless path to h + c@xmath0h@xmath0 , according to the calculations of peeters et al.@xcite the low energy _ a _ @xmath3 state of c@xmath0h may contribute further complexity to an energy - dependent reactivity , as transition states and barriers to reactions may be quite different for _ a _ and _ x _ state interactions . the vibronic mixing of _ a _ and _ x _ states is strong enough , however , that they may not be kinetically distinct species , but rather an energy - dependent mixture , with a correspondingly mixed reactivity . @xcite even the lowest vibrationally excited states nominally belonging to the _ x _ @xmath7 state of c@xmath0h are calculated to have about 0.05 @xmath8-state character , and the zero - order vibrationless level of the @xmath8 state is quite evenly fragmented among five vibronic levels with energies between 3604 and 4593 @xmath2 , none of which have more than 40% @xmath8 state character.@xcite in contrast to the reaction with hydrogen , the thermal rate constant for the reaction of c@xmath0h with acetylene is much faster : 1.3 @xmath61 10@xmath74 @xmath75 molec@xmath76 s@xmath76 near 300 k , with little temperature dependence , as reviewed by van look and peeters.@xcite the trends in transient ground state population illustrated in figure [ figure5 . ] and even more so in figure [ figure4 . ] for cf@xmath18c@xmath0h suggest that slowing down the relaxation of the hot radicals at lower buffer pressure inhibits the reactive loss . such behavior can be expected for a barrierless association followed by a submerged barrier to fragmentation.@xcite a comparison of the transient spectra of hot and cold band lines following c@xmath0h@xmath0 photodissociation similar to our figure [ figure6 . ] was reported by yan , et al.@xcite this work has been widely cited @xcite as evidence that in the presence of 20 torr he , the vibrational thermalization of c@xmath0h will be complete in less than a microsecond . without fully reproducing the conditions of yan al . @xcite we can not be certain , but it appears likely that in the 3.5% mixture of acetylene and he used for this measurement , the disappearance rate of vibrationally excited c@xmath0h is dominated by reaction with c@xmath0h@xmath0 rather than vibrational relaxation by he . at the lower acetylene pressures typically used in subsequent kinetic experiments , the reactive loss is suppressed and vibrational excitation can be expected to persist to longer times . sf@xmath17 has often also been added to test gas mixtures to accelerate vibrational relaxation , although several workers have reported that it was found to have no effect on ground state kinetic spectroscopy of c@xmath0h and was therefore left out.@xcite until future experiments are able to provide a direct monitor of the vibrational relaxation , we can only echo a summary statement from laufer and fahr @xcite in their 2004 review of the reactions and kinetics of ethynyl radicals from acetylene photodissociation : the role of vibrational excitation in the subsequent chemistry is not clear . " as the computational machinery for _ ab initio _ kinetics becomes more robust and well tested , the ambition grows to improve the treatment of chemical activation problems , including multiwell branching in competition with relaxation.@xcite coupled electronic states add to the challenge . particularly for fast radical reactions that may react to completion faster than thermalization , a collection of thermal rate constants may not be sufficient to characterize a mechanism . an extended set of state - resolved c@xmath0h kinetic measurements could provide a challenging target for the development of next - generation master - equation - based kinetic models . the newly identified strong near - infrared bands can facilitate these measurements . three new near - infrared bands of c@xmath0h have been observed and analyzed in the gas phase by diode laser spectroscopy . transitions centered at 7110 , 7088 , and 6696 @xmath2 were found to be in good agreement with the calculated energies of vibronic levels of mixed @xmath4 and @xmath5 electronic character.@xcite the transitions are among the strongest of all those in the infrared and near infrared spectrum of the radical and provide a convenient spectroscopic tool for monitoring concentrations of the ground state radical in chemical mixtures . kinetic measurements show that the competition between reactive loss and vibrational relaxation of the initially hot radicals makes interpretation of the kinetic signals complicated , especially because of the presence of the low - lying excited electronic state which has different reactivity and whose wavefunction contaminates all vibrationally excited levels of the ground electronic state . these attributes combine to make c@xmath0h a particularly attractive vehicle for studying the effects of the presence of multiple potential surfaces on collisional processes in small free radicals where accurate calculations are now possible . see supplementary material for c@xmath6h spectral data in two regions between 6630 and 7135 @xmath2 obtained at 0 , 2.6 , and 3.6 @xmath105s time delay , using 1:1 mixtures of cf@xmath19c@xmath20h and ar at a total pressure of 1.0 torr . supplemental figure s1 and its caption , showing transient absorption signals for a selection of rotational levels in a 1% mixture of cf@xmath18c@xmath0h in ar at a total pressure of 5 torr and a boltzmann temperature analysis . work at brookhaven national laboratory was carried out under contract no . de - sc0012704 with the u.s . department of energy , office of science , and supported by its division of chemical sciences , geosciences and biosciences within the office of basic energy sciences . we thank stephen klippenstein for productive discussions and exploratory calculations on the submerged barriers in the c@xmath0h + cf@xmath18c@xmath0h reaction . 73ifxundefined [ 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 10.1002/kin.10141 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1002/anie.201404537 [ * * , ( ) ] link:\doibase 10.1021/jp402481y [ * * , ( ) ] link:\doibase 10.1021/jp5124162 [ * * , ( ) ] link:\doibase 10.1016/j.carbon.2014.01.020 [ * * , ( ) ] link:\doibase 10.1021/jp808346h [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/78/3/036501 [ * * ( ) , 10.1088/0034 - 4885/78/3/036501 ] link:\doibase 10.1086/181646 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1073/pnas.1113827108 [ * * , ( ) ] link:\doibase 10.1006/jmsp.2001.8459 [ * * , ( ) ] in @noop _ _ , ( , ) pp . , in link:\doibase 10.1016/s0273 - 1177(01)00059-x [ _ _ ] , , vol . , ( , ) pp . , link:\doibase 10.1021/cr030039x [ * * , ( ) ] link:\doibase 10.1021/j100306a044 [ * * , ( ) ] link:\doibase 10.1021/j100383a003 [ * * , ( ) ] link:\doibase 10.1021/j100168a029 [ * * , ( ) ] link:\doibase 10.1021/j100151a026 [ * * , ( ) ] link:\doibase 10.1021/j100135a009 [ * * , ( ) ] link:\doibase 10.1021/j100044a013 [ * * , ( ) ] link:\doibase 10.1021/jp960201i [ * * , ( ) ] link:\doibase 10.1021/jp9532677 [ * * , ( ) ] link:\doibase 10.1021/jp031162v [ * * , ( ) ] link:\doibase 10.1021/jp401354x [ * * , ( ) ] link:\doibase 10.1086/183706 [ * * , ( ) ] link:\doibase 10.1086/160647 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1016/0301 - 0104(75)87025-x [ * * , ( ) ] link:\doibase 10.1063/1.1724865 [ * * , ( ) ] link:\doibase 10.1063/1.443331 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1063/1.445112 [ * * , ( ) ] link:\doibase 10.1063/1.448927 [ * * , ( ) ] link:\doibase 10.1016/0022 - 2852(87)90294 - 3 [ * * , ( ) ] link:\doibase 10.1063/1.452163 [ * * , ( ) ] link:\doibase 10.1016/0022 - 2860(88)80269 - 2 [ * * , ( ) ] link:\doibase 10.1063/1.453571 [ * * , ( ) ] link:\doibase 10.1063/1.454877 [ * * , ( ) ] link:\doibase 10.1016/0022 - 2852(88)90106 - 3 [ * * , ( ) ] link:\doibase 10.1006/jmsp.1996.0058 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1039/c1cp21523j [ * * , ( ) ] link:\doibase 10.1063/1.3532088 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.jms.2015.01.007 [ * * , ( ) ] link:\doibase 10.1021/j100280a044 [ * * , ( ) ] link:\doibase 10.1139/v94 - 085 [ * * , ( ) ] link:\doibase 10.1021/jp1115965 [ * * , ( ) ] @noop _ _ , ed . ( , , ) @noop * * , ( ) link:\doibase 10.1063/1.470361 [ * * , ( ) ] link:\doibase 10.1063/1.475415 [ * * , ( ) ] @noop * * , ( ) @noop _ _ , ed . ( , ) link:\doibase 10.1098/rsta.1960.0003 [ * * , ( ) ] link:\doibase 10.1021/jp062693x [ * * , ( ) ] , \doibase http://dx.doi.org/10.1016/j.proci.2014.05.118 [ * * , ( ) ] link:\doibase 10.1126/science.1224106 [ * * , ( ) ] link:\doibase 10.1021/acs.jpca.5b00640 [ * * , ( ) ] link:\doibase 10.1021/acs.jpca.5b06025 [ * * , ( ) ] link:\doibase 10.1063/1.1436481 [ * * , ( ) ] link:\doibase 10.1063/1.468060 [ * * , ( ) ] link:\doibase { 10.1063/1.2206180 } [ * * , ( ) ] link:\doibase { 10.1063/1.3120607 } [ * * , ( ) ] link:\doibase 10.1021/jp407701z [ * * , ( ) ] link:\doibase 10.1021/acs.jpca.5b01127 [ * * , ( ) ] ] cdcdcdcdc n@xmath43 & r@xmath106 & obs@xmath88calc & r@xmath107 & obs@xmath88calc & p@xmath106 & obs@xmath88calc & p@xmath107 & obs@xmath88calc + 1 & 7093.035 * & -0.055 & 7093.035 * & -0.069 & & & & + 2 & 7095.640 & 0.016 & 7095.640 & -0.003 & & & & + 3 & 7098.034 & 0.003 & 7098.034 * & -0.022 & 7078.468 * & -0.056 & 7078.468 * & -0.066 + 4 & 7100.303 & -0.009 & 7100.303 * & -0.038 & 7075.245 & 0.013 & 7075.245 & -0.002 + 5 & 7102.500 * & 0.035 & 7102.500 & 0.001 & 7071.812 & -0.001 & 7071.812 * & -0.022 + 6 & 7104.475 & -0.014 & 7104.380 * & -0.148 & 7068.256 & -0.013 & 7068.256 * & -0.038 + 7 & 7106.485 * & 0.102 & 7106.419 & -0.009 & 7064.632 * & 0.035 & 7064.632 & 0.004 + 8 & 7108.153 & 0.006 & 7108.202 & 0.006 & 7060.786 & -0.012 & 7060.691 * & -0.143 + 9 & 7109.791 & 0.012 & 7109.843 & 0.010 & 7056.969 * & 0.098 & 7056.902 & -0.009 + 10 & 7111.303 * & 0.026 & 7111.354 * & 0.018 & 7052.817 & 0.003 & 7052.862 & 0.002 + 11 & 7112.634 & -0.007 & 7112.692 & -0.013 & 7048.638 & 0.011 & 7048.684 & 0.007 + 12 & 7113.888 * & 0.020 & 7113.950 & 0.013 & 7044.333 * & 0.024 & 7044.387 * & 0.023 + 13 & 7114.955 & -0.001 & 7115.023 & -0.007 & 7039.850 & -0.007 & 7039.907 & -0.010 + 14 & 7115.939 * & 0.034 & 7116.014 * & 0.030 & 7035.293 * & 0.022 & 7035.348 & 0.012 + 15 & 7116.899 * & 0.188 & 7116.966 * & 0.171 & 7030.554 & 0.005 & 7030.614 & -0.005 + 16 & 7118.899 * & 1.526 & 7118.987*@xmath108 & 1.525 & 7025.726 * & 0.037 & 7025.791 * & 0.027 + 17 & 7117.238 * & -0.650 & 7117.365 * & -0.617 & 7020.877 * & 0.187 & 7020.936 * & 0.167 + 18 & 7117.652 * & -0.602 & 7117.800 * & -0.553 & 7017.073 * & 1.524 & 7017.208 * & 1.575 + 19 & 7117.937 * & -0.531 & 7118.095 * & -0.477 & 7009.619 * & -0.645 & 7009.749 * & -0.605 + 20 & 7118.129 * & -0.400 & 7118.277 * & -0.360 & 7004.234 * & -0.599 & 7004.378 * & -0.550 + 21 & & & & & 6998.728 * & -0.526 & 6998.875 * & -0.479 + 22 & & & & & 6993.122 * & -0.402 & 6993.274 * & -0.355 + cdcdcdcdc n@xmath43 & r@xmath106 & obs@xmath88calc & r@xmath107 & obs@xmath88calc & p@xmath106 & obs@xmath88calc & p@xmath107 & obs@xmath88calc + 1 & 6701.285 & -0.008 & 6701.285 & -0.001 & & & & + 2 & 6703.991 & -0.002 & 6703.991 * & 0.009 & 6689.773 & -0.009 & 6689.773 & -0.001 + 3 & 6706.622 & 0.003 & 6706.612 & 0.007 & 6686.717 & -0.010 & 6686.717 & 0.001 + 4 & 6709.172 & 0.001 & 6709.151 & -0.002 & 6683.600 & -0.001 & 6683.589 & 0.003 + 5 & 6711.647 & 0.001 & 6711.626 & 0.001 & 6680.399 & -0.002 & 6680.383 & 0.000 + 6 & 6714.045 & 0.001 & 6714.023 & 0.004 & 6677.133 & 0.005 & 6677.108 & 0.002 + 7 & 6716.365 & 0.004 & 6716.334 & 0.001 & 6673.786 & 0.007 & 6673.757 & 0.003 + 8 & 6718.596 & 0.000 & 6718.564 & 0.000 & 6670.355 & 0.002 & 6670.329 & 0.005 + 9 & 6720.744 & -0.001 & 6720.705 & -0.004 & 6666.856 & 0.007 & 6666.818 & 0.002 + 10 & 6722.805 & -0.001 & 6722.757 & -0.009 & 6663.265 & 0.002 & 6663.227 & 0.000 + 11 & 6724.724 * & -0.050 & 6724.700 * & -0.032 & 6659.594 & 0.001 & 6659.547 & -0.006 + 12 & 6726.640 & -0.008 & 6726.603 & 0.002 & 6655.835 & -0.002 & 6655.788 & -0.005 + 13 & 6728.426 & 0.004 & 6728.378 & 0.006 & 6651.934 * & -0.057 & 6651.918 * & -0.026 + 14 & 6730.117 * & 0.024 & 6730.064 * & 0.025 & 6648.043 & -0.008 & 6647.997 & -0.004 + 15 & 6731.704 * & 0.049 & 6731.652 * & 0.054 & 6644.019 & 0.004 & 6643.966 & 0.005 + 16 & 6733.200 * & 0.095 & 6733.147 * & 0.103 & 6639.898 * & 0.021 & 6639.845 * & 0.026 + 17 & & & & & 6635.679 * & 0.045 & 6635.626 * & 0.054 + 18 & & & & & 6631.375 * & 0.094 & 6631.315 * & 0.099 + c d c d c d c d c d c d c n@xmath43 & r@xmath106 & obs@xmath88calc & r@xmath107 & obs@xmath88calc & q@xmath106 & obs@xmath88calc & q@xmath107 & obs@xmath88calc & p@xmath106 & obs@xmath88calc & p@xmath107 & obs@xmath88calc + 2 & & & & & 7106.633 & -0.003 & 7109.394 & 0.011 & & & & + 3 & & & 7118.985 * @xmath108 & -0.022 & 7106.287 * & -0.062 & 7108.538 & -0.015 & & & & + 4 & 7119.585 & -0.009 & 7120.878 & -0.005 & 7105.685 * & -0.125 & 7107.783 * & 0.087 & 7094.684 & 0.013 & & + 5 & 7121.538 * & -0.018 & 7122.627 & -0.003 & 7105.060 & 0.009 & 7106.970 * & 0.22 & 7091.194 * & -0.023 & 7092.767 * & -0.017 + 6 & 7123.318 & -0.001 & 7124.239 & 0.010 & 7104.025 * & -0.063 & 7105.673 & -0.009 & 7087.551 & 0.000 & 7088.835 & 0.000 + 7 & 7124.870 & -0.013 & 7125.672 & 0.001 & 7102.983 * & 0.040 & 7104.474 & -0.002 & 7083.682 & -0.007 & 7084.759 & 0.001 + 8 & 7126.259 & -0.001 & 7126.952 & 0.004 & 7101.729 * & 0.103 & 7103.134 & 0.013 & 7079.623 & -0.006 & 7080.539 & 0.005 + 9 & 7127.446 & -0.004 & 7128.061 & 0.005 & & & & & & & 7076.159 & 0.005 + 10 & 7128.450 & -0.006 & 7128.981 & -0.010 & & & 7099.934 & -0.006 & 7070.934 & 0.007 & 7071.617 & 0.006 + 13 & 7129.277 & -0.001 & 7130.755 & 0.006 & 7096.621 & 0.007 & 7098.101 & -0.006 & 7066.297 & -0.002 & 7066.903 & 0.003 + 11 & 7129.906 & -0.008 & 7129.748 & -0.004 & 7094.623 & 0.030 & & & 7061.482 & -0.006 & 7062.016 & -0.003 + 12 & 7130.378 & 0.012 & 7130.318 & -0.017 & & & 7093.971 & 0.028 & 7056.497 & 0.003 & & + 14 & 7130.642 & 0.011 & 7130.961 & 0.001 & 7090.024 & -0.002 & & & 7051.313 & -0.005 & 7051.720 & -0.014 + 15 & 7130.700 & -0.009 & 7131.016 & 0.016 & 7087.466 & -0.013 & 7089.087 & -0.015 & 7045.971 & 0.013 & 7046.324 & -0.014 + 16 & & & 7130.906 * & 0.052 & & & & & & & & + ccccc & x(0,0,0 ) & x(0,8@xmath109,2 ) & x(0,2@xmath109,3)a(0,3,0)@xmath110 & x(0,3@xmath38,3)a(0,0,2)@xmath38 + & @xmath111 & @xmath112 & @xmath112 & @xmath86 + a & & & & -4.29(2 ) + b & 1.4568256(2 ) & 1.42115(5 ) & 1.3941(2 ) & 1.3719(1 ) + dx10@xmath113 & 0.357(3 ) & 1.582(3 ) & 0.89(7 ) & 0.68(4 ) + @xmath45 & 0.0020883(1 ) & 0.0015(2 ) & -0.0071(5 ) & -0.0336(6 ) + p+2q & & & & -0.0925(8 ) + q & & & & -0.00447(3 ) + origin & & 6695.678(2 ) & 7087.645(7 ) & 7109.646(6 ) + & + origin & & 6696.7 & 7099.0 & 7106.7 + intensity & & 327.5 & 1384.1 & 744.2 + + +

we investigate inclusive prompt photon and semi - inclusive prompt photon - hadron production in high energy proton - nucleus collisions using the color glass condensate ( cgc ) formalism which incorporates non - linear dynamics of gluon saturation at small via balitsky - kovchegov equation with running coupling . for inclusive prompt photon production , we rewrite the cross - section in terms of direct and fragmentation contributions and show that the direct photon ( and isolated prompt photon ) production is more sensitive to gluon saturation effects . we then analyze azimuthal correlations in photon - hadron production in high energy proton - nucleus collisions and obtain a strong suppression of the away - side peak in photon - hadron correlations at forward rapidities , similar to the observed mono - jet production in deuteron - gold collisions at forward rapidity at rhic . we make predictions for the nuclear modification factor and photon - hadron azimuthal correlations in proton(deuteron)-nucleus collisions at rhic and the lhc at various rapidities .

1204.1319

the color glass condensate ( cgc ) formalism has been successfully applied to many processes in high energy collisions involving at least one hadron or nucleus in the initial state . examples are structure functions ( inclusive and diffractive ) in deeply inelastic scattering of electrons on protons or nuclei , and particle production in proton - proton , proton - nucleus and nucleus - nucleus collisions , for a recent review see ref.@xcite . the predicted suppression of @xmath2 for the single inclusive hadron production in deuteron - nucleus ( da ) collisions as well as the disappearance of the away side peak in di - hadron angular correlations in the forward rapidity region of rhic @xcite are two of most robust predictions of the formalism which have been confirmed @xcite , see also refs.@xcite . the cgc formalism has been also successful in providing predictions for the first lhc data @xcite in proton - proton ( pp ) and nucleus - nucleus collisions @xcite . nevertheless , there are more recent , alternative phenomenological approaches which combine nuclear shadowing , transverse momentum broadening and cold matter energy loss to describe the rhic data @xcite . therefore , one needs to consider other observables which may help clarify the underlying dynamics of forward rapidity particle production at small @xmath0 . inclusive prompt photon production @xcite and prompt photon - hadron angular correlations @xcite in the forward rapidity region are two such examples . furthermore , there are advantages to studying prompt photon production as compared to hadron production . it is theoretically cleaner ; one avoids the difficulties involved with description of hadronization of final state quarks and gluons , usually described by fragmentation functions valid at high transverse momentum . also , one does not have to worry about possible initial state - final state interference effects which may be present for hadron production . in case of photon - hadron vs. di - hadron angular correlations , again the underlying theoretical understanding is more robust . unlike di - hadron correlations which involve higher number of wilson lines @xcite , photon - hadron correlations depend only on the dipole cross section properties of which are well understood . both processes have been investigated previously , albeit not in detail and only in a limited kinematic range @xcite , see also refs.@xcite . in this work , we extend the existing results for inclusive prompt photon production by clearly separating the contribution of direct and fragmentation photons . we show that direct photons are more sensitive to gluon saturation effects in the kinematics regions considered . we then investigate the dependence of prompt photon - hadron azimuthal angular correlations on high gluon density effects and show that gluon saturation effects lead to disappearance of the away side peak . the effect is very similar to the disappearance of the away side peak in di - hadron correlations observed in the forward rapidity region of rhic in da collisions @xcite . therefore , a measurement of this correlation at rhic and the lhc would greatly help to clarify the role of cgc in the dynamics of particle production at high energy . the advantage of the cgc formalism over the more phenomenological models is that the cross section for many of these processes have the same common ingredient @xcite , the dipole total cross section ; the imaginary part of the forward scattering amplitude of a quark - antiquark dipole on a proton or nucleus target . its rapidity ( energy ) dependence is governed by the b - jimwlk / bk evolutions equations @xcite and is pretty well - understood . the most recent advances in our understanding of the rapidity dependence of the dipole cross section include the running coupling constant corrections and the full next - to - leading order corrections @xcite . the only input is the dipole profile ( dependence on the dipole size @xmath3 ) at the initial rapidity @xmath4 which is modeled , usually motivated by the mclerran - venugopalan ( mv ) model @xcite . the sensitivity to this initial condition is expected to go away at very large rapidities , see sec . iii and ref.@xcite . this paper is organized as follows ; we consider prompt photon - hadron production cross section in section iia and inclusive prompt photon production in section iib where we describe how to separate the contribution of direct and fragmentation photons . we then present our detailed numerical results and predictions at kinematics appropriate for rhic and the lhc experiments in section iii . we summarize our results in section iv . the cross section for production of a quark and a prompt photon with @xmath5-momenta @xmath6 and @xmath7 respectively ( both on - shell ) in scattering of a on - shell quark with @xmath5-momentum @xmath8 on a target ( either proton or nucleus ) in the cgc formalism has been calculated in @xcite and is given by d = ( 2 ) ( p^- - l^- -k^- ) _ d [ ] d^2 d^2 e^i ( + ) n_f ( b_t , r_t , x_g ) , [ cs_gen ] where @xmath9 $ ] is given by _ d [ ] = 8 [ ( p^-)^2 + ( l^-)^2 ] , which , after using the explicit forms of the momenta in the expression for the trace , can be written as , & & d^q(p ) t q(l)(k ) x d^2 dk_t^2 dl_t^2 dy _ dy_l d = e_q^2 _ em ( 2)^3 k^-k_t^2 1 + ( l^-p^-)^2 ^2 + & & + & & d^2 e^i ( + ) n_f ( b_t , r_t , x_g ) , [ cs ] where the symbol @xmath10 stands for a proton @xmath8 or a nucleus @xmath11 target , @xmath12 is the nucleon - nucleon center of mass energy and @xmath13 is the ratio of the incoming quark to nucleon energies such that @xmath14 . the outgoing photon and quark rapidities are defined via @xmath15 and @xmath16 whereas @xmath17 is angle between the final state quark and photon , @xmath18 . we note that this cross section was first computed in @xcite in coordinate space using the dipole formalism , the result of which agrees with the expression in after fourier transforming to momentum space . the imaginary part of of ( quark - antiquark ) dipole - target forward scattering amplitude @xmath19 satisfies the b - jimwlk equation and has all the multiple scattering and small @xmath0 evolution effects encoded . it is defined as n_f(b_t , r_t , x_g ) = 1n_c < tr [ 1 - v^ ( x_t ) v ( y_t ) ] > , [ cs_def ] where @xmath20 is the number of color . the vector @xmath21 is the impact parameter of the dipole from the target and @xmath22 denotes the dipole transverse vector . the matrix @xmath23 is a unitary matrix in fundamental representation of @xmath24 containing the interactions of a quark and the colored glass condensate target . the dipole scattering probability depends on bjorken @xmath25 via the b - jimwlk renormalization group equations . in the present case , it is related to the prompt photon and final state quark rapidities and transverse momenta via x_g = 1[k_t e^-y _ + l_t e^-y_l ] . [ x_g ] in order to relate the above partonic production cross - section to proton ( deuteron)-target collisions , one needs to convolute the partonic cross - section in with the quark and antiquark distribution functions of a proton ( deuteron ) and the quark - hadron fragmentation function : @xmath26 where @xmath27 is the transverse momentum of the produced hadron , and @xmath28 is the parton distribution function ( pdf ) of the incoming proton ( deuteron ) which depends on the light - cone momentum fraction @xmath13 and the hard scale @xmath29 . a summation over the quark and antiquark flavors in the above expression should be understood . the function @xmath30 is the quark - hadron fragmentation function ( ff ) where @xmath31 is the ratio of energies of the produced hadron and quark and pseudo - rapidity @xmath32 is the same . ] . note that due to the assumption of collinear fragmentation of a quark into a hadron , the angle @xmath17 is now the angle between the produced photon and hadron . the light - cone momentum fraction @xmath33 are related to the transverse momenta and rapidities of the produced hadron and prompt photon via ( details are given in the appendix ) @xmath34 the single inclusive prompt photon cross section can be readily obtained from by integrating over the momenta of the final state quark . integration over the quark energy @xmath35 is trivially done by using the delta function and leads to ( after shifting @xmath36 ) , @xmath37 \frac{1}{k^2_t } \int d^2 \vec{r_t } \,d^2 \vec{l_t } \frac{l_t^2}{[z\ , \vec l_t - \vec k_t ] ^2 } \ , e^{i \vec{l_t}\cdot \vec{r_t } } \ , n_f ( b_t , r_t , x_g ) , \end{aligned}\ ] ] where @xmath38 denotes the fraction of the projectile quark energy @xmath39 carried by the photon and @xmath40 . various limits of this expression have been studied in @xcite where it was shown that in the limit where photon has a large transverse momentum @xmath41 such that the collinear singularity is suppressed , one recovers the lo pqcd result for direct photon production process @xmath42 convoluted with the unintegrated gluon distribution function of the target . on the other hand , if one performs the @xmath43 integration above without any restriction , one recovers the lo pqcd expression for quark - photon fragmentation function convoluted with dipole scattering probability . in the limit where one can ignore multiple scattering of the quark on the target ( `` leading twist '' kinematics ) , this expression reduces to the pqcd one describing lo production of fragmentation photons . it is therefore useful to explicitly separate the contribution of fragmentation photons from that of the direct photons . to this end , we rewrite as @xmath44 \frac{1}{k^2_t } \int d^2 \vec{l_t}\ , l_t^2\big[\frac{1}{[z \vec l_t -\vec k_t]^2}-\frac{1}{k_t^2}\big ] n_f(x_g , b_t , l_t ) , \nonumber\\ & + & \frac{e_q^2 \alpha_{em}}{\pi ( 2\pi)^3 } z^2 [ 1 + ( 1 - z)^2]\frac{1}{k^4_t}\int d^2 \vec{l_t } \ , l_t^2 n_f(x_g , b_t , l_t ) , \end{aligned}\ ] ] where we have added and subtracted the second term . notice that we use the same notation for coordinate representation of the forward dipole - target scattering amplitude @xmath45 and its two - dimensional fourier transform . the second term in this expression describes production of direct photons whereas the first term gives the contribution of fragmentation photons . in order to see this more explicitly , we let @xmath46 in the first term and keep the most divergent piece of the @xmath43 integral to get @xmath47\frac{1}{k^4_t } \int^{k^2_t}d^2\vec{l_t}\,l_t^2\ , n_f(\bar{x}_g , b_t , l_t ) \end{aligned}\ ] ] where @xmath48 is the leading order quark - photon fragmentation function @xcite , @xmath49 is new and is our main result for single inclusive prompt photon production which includes contribution of both fragmentation ( first term ) and direct ( second term ) photons . in order to ensure that the divergence present in is properly removed , one needs to regulate it self - consistently . here we have done this separation by imposing a hard cutoff which would result in a mismatch between the finite corrections to our results and those that are included in parameterizations of photon fragmentation function , for example , using the @xmath50 scheme . however this mismatch is a higher order effect in the coupling constant and is therefore expected to be parametrically small . it should be noted that the separation between the direct and fragmentation contributions depends on the hard scale , chosen to be the photon transverse momentum , which is already well - known in pqcd . exhibits some interesting features ; the dipole scattering probability @xmath45 is probed at @xmath51 ( where @xmath52 is the external momentum ) in case of fragmentation photons whereas it depends on the internal momentum @xmath43 in case of direct photons . furthermore , in case of direct photons , the integrand is peaked at values of transverse momenta @xmath53 . this means that fragmentation photons should be much less sensitive to high gluon density effects than direct photons since they probe the target structure at higher transverse momenta . this will be verified numerically in the following sections . the forward dipole - target scattering amplitude appears in both semi - inclusive photon - hadron and inclusive prompt photon cross - section eq . ( [ qh - f],[pho4 ] ) and incorporates small - x dynamics which can be computed via first principle non - linear b - jimwlk equations @xcite in the cgc formalism . in the large @xmath20 limit , the coupled b - jimwlk equations are simplified to the balitsky - kovchegov ( bk ) equation @xcite , a closed - form equation for the evolution of the dipole amplitude in which both linear radiative processes and non - linear recombination effects are systematically incorporated . the running - coupling bk ( rcbk ) equation has the following simple form : @xmath59\ , , \label{bk1}\ ] ] where @xmath60 . the evolution kernel @xmath61 is given by balitsky s prescription @xcite with the running coupling . the explicit form of @xmath61 with details can be found in refs.@xcite . the only external input for the rcbk non - linear equation is the initial condition for the evolution which is taken to have the following form motivated by mclerran - venugopalan ( mv ) model @xcite , @xmath62\ , \label{mv}\ ] ] where @xmath63 gev @xcite . the initial saturation scale @xmath64 ( with @xmath65 for a proton and nuclear target ) at starting point of evolution ( at @xmath66 ) and the parameter @xmath67 , are free parameters which are determined from a fit to other experimental measurements at small - x . it was shown that inclusive single hadron data in pp collisions at rhic can be described with a initial saturation scale within @xmath68 @xcite . however , hera data on proton structure functions prefers the lower value for the proton initial saturation scale @xmath69 @xcite . one have also freedom to run @xmath67 as a free parameter in the @xmath70 minimization and obtain its preferred value in a fit to hera data . in order to investigate the uncertainties due to initial condition of the rcbk equation , we will consider the following three parameter sets which all provide an excellent fit to the hera data for proton targets @xcite : @xmath71 in the mv model @xcite , the parameter @xmath67 in is @xmath72 . however , it has been recently shown @xcite that the effective value of @xmath67 can be larger than one when the sub - leading corrections to the mv model are included @xcite . the parameter @xmath67 appears to be also important in order to correctly reproduce the single inclusive particle spectra , and a larger value @xmath73 is apparently preferable at large-@xmath52 @xcite . in our approach the difference between proton and nuclei originates from different initial saturation scales @xmath64 in the rcbk equation via eq.([mv ] ) . in the case of inclusive hadron production in proton - nucleus collisions , due to theoretical uncertainties and rather large errors of the experimental data , it is not possible to uniquely fix the initial value of @xmath74 . in the case of minimum - bias dau collisions , the initial nuclear ( gold ) saturation scale within @xmath75 is consistent with the rhic inclusive hadron production data @xcite . the extracted value of @xmath74 is also consistent with the dis data for nuclear targets @xcite . here , we will also consider the uncertainties due to the initial condition of the rcbk equation for a nuclear target . note that @xmath74 should be considered as an impact - parameter averaged value since it was extracted from the minimum - bias data . for the minimum - bias collisions , one may assume that the initial saturation scale of a nuclei with atomic mass number a , scales linearly with @xmath76 , namely we have @xmath77 where the parameter @xmath78 is fixed from a fit to data . in ref.@xcite , it was shown that nmc data can be described with @xmath79 . we will use the nlo mstw 2008 pdfs @xcite and the nlo kkp ffs @xcite . for the photon fragmentation function , we will use the full leading log parametrization @xcite . we assume the factorization scale @xmath29 in the ffs and the pdfs to be equal and its value is taken to be @xmath27 and @xmath52 for the semi - inclusive and inclusive prompt photon production , respectively . we start by considering direct and fragmentation photon production in pa collisions at rhic and the lhc . in nuclear collisions , nuclear effects on single particle production are usually evaluated in terms of ratios of particle yields in pa and pp collisions ( scaled with a proper normalization ) , the so - called nuclear modification factor @xmath80 . the nuclear modification factor @xmath1 is defined as @xmath81 where the yield @xmath82 can be calculated from the invariant cross - section given in . the normalization constant @xmath83 is the number of binary proton - nucleus collisions . we take @xmath84 and @xmath85 at @xmath86 and @xmath87 tev , respectively @xcite . in order to compare our predictions for @xmath88 with the experimental value , one should take into account possible discrepancy between our assumed normalization @xmath83 and the experimentally measured value for @xmath83 by rescaling our curves . again we expect that some of the theoretical uncertainties , such as sensitivity to @xmath89 factors ( which effectively incorporates the missing higher order corrections ) , will drop out in @xmath90 . in we show the nuclear modification factor for both pa and da collisions at rhic . this is to facilitate a comparison of and to distinguish between the genuine saturation effects in the target nucleus from isospin effects in the projectile deuteron . tev ) at @xmath91 . the curves are obtained from using the solution to rcbk equation with the initial saturation scale @xmath92 for proton and @xmath93 for a nucleus ( gold ) . , width=8 ] clearly there is a large difference between a proton and a deuteron projectile as far as prompt photon production is concerned . this difference is more pronounced in the forward rapidity region and at high transverse momentum where one probes the quark content of the projectile . this is due to difference between the up and down quark distributions of a proton ( note that nuclear effects in the wave function of a deuteron are ignored as they are known to be small ) . this is a well known effect , nevertheless , for sake of clarity and to illustrate the difference between inclusive hadron production and qed probe namely prompt photon production , we illustrate this in some detail . in case of photon production , the production cross section is weighed by the the charged squared of a given quark flavor . for example ( assuming only two flavor ) , for a proton projectile this is given by @xmath94 where @xmath95 denote the distribution functions of up and down quarks in a proton . ignoring nuclear effects in a deuteron , we assume a deuteron is a system of free proton and a neutron in which case the corresponding expression is @xmath96 where @xmath97 denote the distribution functions of up and down quarks in a neutron . assuming isospin symmetry gives @xmath98 and @xmath99 which leads to @xmath100 $ ] for a deuteron . comparing this expression with two times that of a proton , the relative contribution of up quarks in a deuteron ( @xmath101 ) is smaller than that of a twice a proton ( @xmath102 ) . since there are more up quarks than down quarks ( by a factor of @xmath103 in this kinematics ) in a proton , and their relative weight is smaller , this leads to a further reduction of @xmath2 as compared with @xmath80 in prompt photon production , see . we note that in the absence of the charged squared factor , which is the case for inclusive hadron production , one would get @xmath104 as one should since possible nuclear effects in a deuteron are ignored here . at the lhc the isospin effect is absent since the same projectile is used for the pp and pa collisions . this helps to understand the physics of qcd saturation more clearly , as the suppression of the signal will not be contaminated with isospin effect . tev ( right ) and the lhc @xmath105 tev ( left ) energy at various rapidities . the curves are the results obtained from and the solution to rcbk equation with the initial saturation scale @xmath92 for a proton and @xmath93 for a nucleus ( gold ) , corresponding to set ii in . , title="fig:",width=8 ] tev ( right ) and the lhc @xmath105 tev ( left ) energy at various rapidities . the curves are the results obtained from and the solution to rcbk equation with the initial saturation scale @xmath92 for a proton and @xmath93 for a nucleus ( gold ) , corresponding to set ii in . , title="fig:",width=8 ] tev ( right ) and the lhc @xmath105 tev energy ( left ) . the curves are the results obtained from and the solution to rcbk equation using different initial saturation scales for a proton @xmath106 and a nucleus @xmath74 . the band shows our theoretical uncertainties arising from allowing a variation of the initial saturation scale of the nucleus in a range consistent with previous studies of dis structure functions as well as particle production in minimum - bias pp , pa and aa collisions in the cgc formalism , see the text for the details.,title="fig:",width=8 ] tev ( right ) and the lhc @xmath105 tev energy ( left ) . the curves are the results obtained from and the solution to rcbk equation using different initial saturation scales for a proton @xmath106 and a nucleus @xmath74 . the band shows our theoretical uncertainties arising from allowing a variation of the initial saturation scale of the nucleus in a range consistent with previous studies of dis structure functions as well as particle production in minimum - bias pp , pa and aa collisions in the cgc formalism , see the text for the details.,title="fig:",width=8 ] tev energy . the band ( cgc - rcbk - av ) similar to corresponds to the results obtained from and the solutions to the rcbk evolution equation using different initial saturation scales for a proton @xmath106 and a nucleus @xmath74 , see the text for the details.,title="fig:",width=8 ] tev energy . the band ( cgc - rcbk - av ) similar to corresponds to the results obtained from and the solutions to the rcbk evolution equation using different initial saturation scales for a proton @xmath106 and a nucleus @xmath74 , see the text for the details.,title="fig:",width=8 ] in minimum - bias da @xmath107 tev ( rhic ) and pa @xmath108 tev ( lhc ) collisions . the curves are the results obtained from and the solution to rcbk equation with the initial saturation scale @xmath92 for a proton and @xmath93 for a nucleus . left : comparison of the inclusive prompt photon nuclear modification factor predictions from the cgc ( in this paper ) and the standard collinear factorization approach @xcite . the band cgc - rcbk - av is the same as in . , title="fig:",width=8 ] in minimum - bias da @xmath107 tev ( rhic ) and pa @xmath108 tev ( lhc ) collisions . the curves are the results obtained from and the solution to rcbk equation with the initial saturation scale @xmath92 for a proton and @xmath93 for a nucleus . left : comparison of the inclusive prompt photon nuclear modification factor predictions from the cgc ( in this paper ) and the standard collinear factorization approach @xcite . the band cgc - rcbk - av is the same as in . , title="fig:",width=8 ] in , we show the minimum - bias nuclear modification factor for the direct , fragmentation and the inclusive prompt photon production at rhic and the lhc energies @xmath109 tev at various rapidities @xmath32 obtained from eqs.([pho4 ] , [ rpa ] ) supplemented with rcbk solution with the initial saturation scale for proton @xmath69 and nuclei @xmath110 . it is seen that the nuclear modification @xmath90 for the fragmentation photon is bigger than the direct and inclusive prompt photon . this is what we expected in our picture since direct photon cross - section in probes the target structure function at lower transverse momentum @xmath52 ( and consequently lower x ) than the fragmentation part with transverse momentum @xmath51 and therefore is more sensitive to the suppression of structure function and the saturation effect . however , as we increase the energy the enhancement of the fragmentation photon @xmath90 at rhic will be also replaced with suppression at the lhc , see top panel . this is simply due to the fact that both the fragmentation and the direct part depend on the color dipole forward amplitude which encodes the small - x dynamics and at higher energy , the small - x evolution leads to suppression of @xmath90 . in a collider experiment such as the lhc , the secondary photons coming from the decays of hadrons , overwhelm the inclusive prompt photon measurements with order of magnitudes . in order to reject the background , isolation cuts are imposed @xcite . contribution of fragmentation prompt photon is reduced by imposing an isolation cut is the total transverse momentum of a fragmentation jet , the photon s energy is then @xmath111 and the total hadronic energy within the jet is @xmath112 . by isolation cut criterion , the hadronic energy does not have to be more than @xmath113 in the isolation cone . this gives the lower limit of @xmath114 ( or @xmath13 convolution ) in integral , namely @xmath115 . given that the integrand of fragmented part is proportional to @xmath116 and dominated at lower limit of integrand , we expect that the isolation cut reduces the fragmentation contribution more severely than the direct one . ] . a proper incorporation of the isolation cut criterion in our framework is beyond the scope of this paper . however , from it is seen that at higher energy at forward collisions , @xmath90 for direct and single inclusive prompt photon becomes remarkably similar , indicating that to a good approximation , one may assume that the nuclear modification factor for direct and isolated prompt photon are equal . in , we show the minimum - bias nuclear modification factor for the direct photon production at rhic and the lhc energies @xmath109 tev at various rapidities @xmath32 obtained from rcbk solutions with the initial proton saturation scale @xmath117 and @xmath118 corresponding to parameter sets i and ii in . for nuclear target in minimum - bias collisions , we take two initial saturation scales for nuclei ( gold and lead ) @xmath119 which are extracted from a fit to other experimental data on heavy nuclear target @xcite . for a proton target , we have checked that parameter sets ii and iii give similar results for @xmath90 with better than @xmath120 accuracy . therefore , in we only show results obtained from two parameter sets i and ii in . the band in shows our uncertainties arising from a variation of the initial saturation scale of the nucleus in a range consistent with previous studies of dis structure functions as well as particle production in minimum - bias pp , pa and aa collisions in the cgc formalism . one may therefore expect that the possible effects of fluctuations ( of nucleons in a nucleus ) on particle production is effectively contained in our error band . from , it is seen that the nuclear modification for direct photon production is very sensitive to the initial saturation scale in proton and nuclei . however , this uncertainties will be reduced for more forward collisions at higher energy at the lhc . the same effect has been observed for the inclusive hadron production in pa collisions @xcite . this clearly indicates that the nuclear modification in p(d)a collisions is a sensitive probe of saturation effects and @xmath90 measurements for direct photon and inclusive hadron provide crucial complementary information about initial saturation scale and small - x evolution dynamics . in , it is seen that at a fixed rapidity and energy for a fixed initial saturation scale for proton @xmath106 , a bigger initial saturation scale for nuclei @xmath74 leads to a bigger broadening and consequently enhances the cross - section and @xmath90 if @xmath83 is kept fixed . in , we show our predictions for @xmath80 for direct photon ( right ) and inclusive prompt photon ( left ) production in pa collisions at various rapidities a the lhc @xmath121 tev energy . the band ( cgc - rcbk - av ) similar to corresponds to the results obtained from with the solutions of the rcbk evolution equation ( [ bk1 ] ) . in ( right ) , we compare @xmath90 for direct photon at @xmath122 for rhic energy @xmath107 tev and the lhc energies @xmath123 tev . it is seen that the suppression of @xmath88 at the lhc is larger compared to @xmath2 at rhic and persists at higher transverse momentum . this larger suppression is even more impressive given that fact that a good amount of the observed suppression of @xmath2 at rhic is due to the projectile being a deuteron rather than a proton . in ( left ) , we compare the cgc prediction ( cgc - rcbk - av ) obtained here with the collinear factorization result ( eps09 ) @xcite for inclusive prompt photon @xmath88 at @xmath122 at the lhc . it is seen that the lhc measurements of the inclusive prompt photon at forward rapidities can discriminate between the collinear ( standard parton model ) and the cgc approach . some words of caution are in order here . strictly speaking our formalism is less reliable for collisions at around mid - rapidities and high transverse momenta . this is due to the fact that our formula is valid for asymmetric collisions like pa or pp collisions at forward rapidities when a projectile can be treated in the standard collinear approximation while for the target we systematically incorporated the small - x re - summation ( at the leading log approximation ) effects . note however , for the case of pp collisions ( our reference for @xmath124 ) at rhic , the saturation scale of target proton is rather small , and it is not clear that the cgc formulation will be applicable . moreover , our parameter sets in was obtained from a fit to hera data at small - x @xmath125 and for virtualities @xmath126 \,\text{gev}^2 $ ] @xcite . therefore , our predictions are less reliable at high-@xmath52 ( @xmath127 gev ) . defined in for minimum - bias p(d)a and pp collisions at rhic @xmath107 tev ( upper ) and the lhc @xmath105 tev energy ( lower ) obtained from the rcbk solutions with different initial saturation scales . , title="fig:",width=8 ] defined in for minimum - bias p(d)a and pp collisions at rhic @xmath107 tev ( upper ) and the lhc @xmath105 tev energy ( lower ) obtained from the rcbk solutions with different initial saturation scales . , title="fig:",width=8 ] defined in for minimum - bias p(d)a and pp collisions at rhic @xmath107 tev ( upper ) and the lhc @xmath105 tev energy ( lower ) obtained from the rcbk solutions with different initial saturation scales . , title="fig:",width=8 ] defined in for minimum - bias p(d)a and pp collisions at rhic @xmath107 tev ( upper ) and the lhc @xmath105 tev energy ( lower ) obtained from the rcbk solutions with different initial saturation scales . , title="fig:",width=8 ] we now focus on azimuthal angle @xmath17 correlations of the prompt photon - hadron spectrum , where the angle @xmath17 is the difference between the azimuthal angle of the measured hadron and single prompt photon . we present our predictions for semi - inclusive prompt photon - hadron ( for hadron we consider only neutral pion here ) production at rhic and the lhc in pp and p(d)a collisions in terms of @xmath128 defined as follows , @xmath129/{d\sigma^{p(d)\ , t \rightarrow h(q)\,\gamma(k)\ , x } \over d^2\vec{b_t}\ , dk_t^2\ , dq_t^2\ , dy_{\gamma}\ , dy_l\ , d\theta } [ \delta \theta= \delta \theta_c],\ ] ] where the prompt photon - hadron cross - section in above expression is given in . this definition has a simple meaning of the probability of , the single semi - inclusive prompt photon - hadron production at a certain kinematics and angle @xmath17 given the production with the same kinematics at a fixed reference angle @xmath130 . we take @xmath131 . as will show the @xmath128 defined in this way has a non - trivial structure and can probe the physics of small - x and gluon saturation . in principle , one is free to chose a different reference angle @xmath130 , however any value @xmath132 will only change the normalization rather than the main picture . the advantage of the above definition for the azimuthal correlations is that it is experimentally easier to measure as it does not require a different experimental setup and run for the trigger or reference . moreover , in da collisions at rhic , the isospin effect in @xmath128 will drop out via normalization contribution to the azimuthal correlation defined via . therefore , due to our particular definition of @xmath128 in , the differences between a deuteron and a proton projectile are negligible unlike the prompt photon production case . ] and this facilitates to single out the importance of the saturation effect at forward rapidities in contrast to the nuclear modification factor @xmath124 . defined in for minimum - bias da collisions at rhic @xmath107 tev for two different windows of kinematics of transverse momentum of produced prompt photon @xmath52 and hadron ( neutral pion ) @xmath27 at fixed rapidity @xmath133 . the curves are the results obtained from the rcbk equation solution with the initial saturation scale @xmath92 for proton and @xmath93 for a nuclei . , width=8 ] for minimum - bias p(d)a collisions at different energies @xmath134 tev for a fixed rapidity @xmath133 and transverse momentum of the produced prompt photon @xmath135 gev and hadron ( neutral pion ) @xmath136 gev . the curves are the results obtained from the rcbk equation solution with the initial saturation scale @xmath92 for proton and @xmath93 for a nuclei , width=8 ] in this approach a fast valence quark from the projectile proton radiates a photon before and after multiply scattering on the color glass condensate target , see . in this picture , the projectile is treated in the collinear factorization , and therefore the photon radiation from quark at this level has the standard features of pqcd , including the back - to - back correlation in the transverse momentum . as a result of multiple scatterings , the quark acquires a transverse momentum comparable with the saturation scale , the only relevant scale in the system , and the intrinsic angular correlations are washed way . in , we show @xmath128 at forward rapidity @xmath133 for @xmath137 gev at rhic and @xmath138 gev at the lhc for two different initial saturation scale for proton @xmath139 and nuclei @xmath119 . for such low @xmath140 s we are most likely probing the saturation region of the nuclear wave function due to the small values of @xmath25 . it is clear that the away - side prompt photon - hadron cross - section ( at @xmath141 ) is suppressed for the bigger saturation scale ( corresponding to a denser system ) . it is also seen from that @xmath128 is very sensitive to the initial saturation scale . unfortunately , this bring rather large theoretical uncertainties . however , as we will show in the following the suppression of the way - side correlations seems to be a robust feature of our results and it less depends on our theoretical uncertainties . in we show the relative azimuthal correlations obtained from the rcbk solution for a fixed initial saturation scale @xmath142 and @xmath143 at forward rapidities @xmath133 at rhic @xmath107 tev for two different kinematics windows of transverse momenta : we show in top panel , the results with a fix transverse momentum of prompt photon @xmath144 gev at different transverse momentum of produced hadron , and in down panel , with a fixed transverse momentum of hadron @xmath135 gev but at various transverse momentum of the produced prompt photon . it is seen , the relative azimuthal correlation is suppressed at @xmath145 as the transverse momentum of the produced hadron or prompt photon decreases and becomes comparable to the actual saturation scale of the system . this is the case regardless which of two transverse momenta of the hadron or prompt photon decreases . in lower panel , it is seen that when the prompt photon transverse momentum becomes comparable with the saturation effect the away - side azimuthal angular correlation of photon - hadron completely washes away . the same effect happens at lower transverse momentum of the produced hadron . this is simply because of fragmentation effect namely the transverse momentum of the produced parton ( that should be compared with the saturation scale ) is higher than the transverse momentum of the fragmented hadron . again , the suppression of away - side correlations is clearly due to the saturation effect since as we lower the transverse momentum of the produced particle , the system of hadron - photon become more sensitive to the small - x gluon saturation . note that the hadron - photon cross - section in has collinear singularity . therefore for a proper investigation of the correlations at @xmath146 , in principle , one should first extract the collinear singularity in a same fashion as demonstrated in section ii by introducing the quark - photon fragmentation function . therefore , our results at near - side @xmath147 should be less reliable . nevertheless , we expect that the sensitivity to the collinear singularity effect should drop out in the correlation defined in via normalization . we checked that contrary to the away - side correlations , the near - side peak is not sensitive to the saturation physics as the correlations does not change with varying the density of the system , see also . in order to further understand the relative sensitivity of the away side peak to saturation dynamics , in , @xmath128 we compare at various energies at rhic and the lhc for a fixed transverse momentum of the produced prompt photon @xmath135 gev and hadron @xmath136 gev at rapidity @xmath133 . it is clear that the away side peak goes away as one increases the energy . again this is due to the fact that as we increase the energy , the gluon density increases and non - linear gluon recombination or the saturation effect becomes important . from , it is obvious that at the lhc the away - side azimuthal correlations of photon - hadron will be strongly suppressed . we conclude that the suppression of the away - side azimuthal photon - hadron correlations defined via , with decreasing the transverse momentum of the produced prompt photon or hadron , or increasing the energy , or increasing the size / density of system , all uniquely can be explained within the universal picture of gluon saturation without invoking any new parameters or ingredients to our model . we have investigated prompt photon production and prompt photon - hadron azimuthal angular correlations in proton - proton and proton - nucleus collisions using the color glass condensate formalism . we have provided predictions in the kinematic regions appropriate to rhic and the lhc experiments . we have shown that single inclusive and direct prompt photon production cross section in p(d)a collisions at forward rapidities at both rhic and the lhc is suppressed , as compared to normalized production cross section in proton - proton collisions . at rhic , a good portion of the predicted suppression is due to the projectile being a deuteron rather than a proton . this suppression is larger at the lhc compared to rhic which is even more impressive given that the projectile at the lhc is a proton . we showed that direct photon production is most affected by gluon saturation effects in the target nucleus than the fragmentation photons . however , at the lhc energies at forward rapidities the nuclear modification suppression for direct , fragmentation and inclusive prompt photon production is rather similar . we showed that the nuclear modification factor @xmath90 for inclusive prompt photon production at rhic and the lhc is a sensitive probe of small - x dynamics . we note that our results based on gluon saturation dynamics and using the color glass condensate formalism are different from those coming from the collinear factorization @xcite . therefore , @xmath90 measurement at rhic and the lhc is a crucial test of different factorization schemes , see also refs.@xcite for other observables . we have also studied prompt photon - hadron azimuthal angular correlations in kinematic regions which can be probed by rhic and the lhc experiments . it is shown that the away side peak in photon - hadron angular correlation goes away as one lowers the final state particle s momenta , very similarly to the disappearance of the away side peak in di - hadron correlations in forward rapidity da collisions at rhic . at fixed transverse momenta , the suppression of the away side peak gets stronger as one goes to larger rapidities ( more forward ) or higher energy or denser system as expected , due to stronger saturation effects in the target nucleus . presently , we are not aware of any alternative approach which leads to this novel phenomenon . note that in contrast to the nuclear modification factor for prompt photon , the prompt photon - hadron azimuthal angular correlation defined via eq.([az ] ) is free from the isospin effect , and can be considered as a cleaner probe of saturation effect . finally , we emphasize that prompt photon - hadron azimuthal angular correlations suffers from much less theoretical uncertainties as compared to di - hadron azimuthal angular correlations and thus a measurement of this correlation would go a long way toward establishing the dominance of gluon saturation effects at small @xmath25 . it will be interesting to see what the predictions of pqcd - motivated models @xcite are for photon - hadron azimuthal angular correlations . in these models one usually needs to combine models of higher twist shadowing , the cronin effect and cold matter energy loss in order to describe the data on single inclusive hadron production and di - hadron azimuthal angular correlations . the advantage of the cgc formalism is that the same framework can be used to describe nuclear shadowing of structure functions @xcite at small @xmath0 and includes transverse momentum broadening ( the cronin effect ) @xcite . it does not however include cold matter energy loss due to longitudinal momentum transfer between the projectile and the target which may be important at the very forward rapidities . it is not clear at the moment how to calculate this effect from first principles qcd . even though this energy loss itself is small , due to steepness of the production cross section at forward rapidity , it can suppress the cross section significantly . the purpose of this appendix is to define the kinematics and derive the needed relations between various light - cone energy fractions which appear in the production cross sections used . this is slightly different from the standard relations used in production cross sections based on collinear factorization theorems of pqcd . we first consider scattering of a quark on the target where a photon and a quark are produced , depicted in fig . 1 , q(p ) a ( p_a)(k ) q(l ) x , where @xmath11 is a label for the multi - gluon state , described by a classical field representing a proton or nucleus target . in the standard pqcd ( leading twist ) kinematics , only one parton from the target interacts . this is not the case here since the target is described by a classical gluon field representing a multi - gluon state with intrinsic momentum rather than an individual gluon with a well defined energy fraction @xmath25 and zero transverse momentum . nevertheless , since most of the gluons in the target wave function have momentum of order @xmath148 , one can think of the state describing the target as being labeled by a ( four ) momentum @xmath149 . in this sense , the gluons in the target collectively carry fraction @xmath25 of the target energy and have intrinsic transverse momentum denoted @xmath150 . this also means that there is no integration over @xmath25 in our case unlike the collinearly factorized cross sections in pqcd ( this basically corresponds to setting @xmath25 equal to the lower limit of @xmath25 integration in pqcd cross sections ) . we thus have @xmath151 where @xmath152 are the momenta of the incoming projectile , target and the produced hadron respectively . ( pseudo)-rapidities of the produced quark and photon are related to their energies via @xmath153 imposing energy - momentum conservation at the partonic level via @xmath154 and using leads to @xmath155 the above relations and ( and the on mass shell condition ) can be used to derive the following expressions for the energy fractions @xmath156 . we obtain , @xmath157 where the final hadron transverse momentum and rapidity are denoted by @xmath27 and @xmath158 , and we used @xmath159 . note that light - cone momentum fraction @xmath25 appears in the dipole forward scattering amplitude @xmath19 whereas @xmath13 is the fraction of the projectile proton ( deuteron ) carried by the incident quark , see . to derive an expression for the lower limit of @xmath31 integration in , we note that @xmath160 . using the relation between the minus components of the four momenta given above , we get x_q = k^- + . the minimum value of @xmath31 occurs when @xmath13 is maximum , i.e. , @xmath161 . we then have z_f^min = q^- - k^- , which can be written in terms of the transverse momenta and rapidities of the final state hadron and photon as z_f^min = e^_h 1 - e^_. we now consider the kinematics of single inclusive photon production cross section . the cross section is obtained from by integrating over the final state quark momenta . this requires some care as we have now explicitly separated direct and fragmentation photons in . again using eqs.([a1 ] , [ a2 ] , [ a31 ] ) , we obtain the following relation , x_g=(k_te^-_+ ) , [ xg1 ] where opposite to , we avoided to introduce @xmath158 and @xmath31 . one can use the energy - momentum delta functions in to obtain the following relation , where the parameter @xmath114 is the fraction of energy of parton carried away by photon and it is defined as follows , z = e^_. in case of direct photons with transverse momentum @xmath52 , one should shift momentum @xmath36 in ( this is how we obtained the expersion ) . assuming that @xmath162 , we get [ a6 ] |x_g = , where we have now used @xmath58 to denote the light cone momentum fraction of the target carried by gluons for production of direct photons so as to distinguish it from the momentum fraction involved in production of fragmentation photons . in the later case , the integration variable @xmath43 has been shifted twice . implementing the shifts in and noting that the @xmath43 integration in the fragmentation photon production cross section is dominated by its singularity at @xmath163 we get , for fragmentation photons , x_g=. we are grateful to javier albacete for useful communications and for providing us with the latest tables for the numerical solution of rcbk equation . we would also like to thank william brooks and raju venugopalan for useful discussions . we acknowledge and thank barbara jacak , john lajoie and richard seto for helpful discussions about phenix detector capabilities . is supported in part by the doe office of nuclear physics through grant no . de - fg02 - 09er41620 , from the `` lab directed research and development '' grant ldrd 10 - 043 ( brookhaven national laboratory ) , and from the city university of new york through the psc - cuny research program , grant 64554 - 00 42 . the work of a.h.r is supported in part by fondecyt grants 1110781 . f. gelis , e. iancu , j. jalilian - marian and r. venugopalan , ann . nucl . part . * 60 * , 463 ( 2010 ) [ arxiv:1002.0333 ] ; j. jalilian - marian and y. v. kovchegov , prog . phys . * 56 * , 104 ( 2006 ) [ hep - ph/0505052 ] ; r. venugopalan , arxiv:1012.4699 . d. kharzeev , y. v. kovchegov and k. tuchin , phys . d * 68 * , 094013 ( 2003 ) [ hep - ph/0307037 ] ; j. l. albacete , n. armesto , a. kovner , c. a. salgado and u. a. wiedemann , phys . rev . lett . * 92 * , 082001 ( 2004 ) [ hep - ph/0307179 ] . c. marquet , nucl . a * 796 * , 41 ( 2007 ) [ arxiv:0708.0231 ] . i. arsene _ et al . * 93 * , 242303 ( 2004 ) [ nucl - ex/0403005 ] ; j. adams _ et al . lett . * 97 * , 152302 ( 2006 ) [ nucl - ex/0602011 ] ; a. adare _ et al . _ , arxiv:1105.5112 ; e. braidot , for the star collaboration , arxiv:1005.2378 . j. jalilian - marian , nucl . * a748 * , 664 ( 2005 ) [ nucl - th/0402080 ] ; k. tuchin , nucl . phys . * a798 * , 61 ( 2008 ) [ arxiv:0705.2193 ] ; v. p. goncalves , m. s. kugeratski , m. v. t. machado and f. s. navarra , phys . lett . b * 643 * , 273 ( 2006 ) [ hep - ph/0608063 ] ; a. stasto , b - w xiao , f. yuan , arxiv:1109.1817 . cms collaboration , phys . lett . * 105 * 022002 ( 2010 ) [ arxiv:1005.3299 ] ; alice collaboration , phys . lett . * 105 * , 252301 ( 2010 ) [ arxiv:1011.3916 ] ; cms collaboration , j. high energy phys . * 08 * , 141 ( 2011 ) [ arxiv:1107.4800 ] ; atlas collaboration , phys . b710 * , 363 ( 2012 ) [ arxiv:1108.6027 ] . e. levin and a. h. rezaeian , phys . * d82 * , 014022 ( 2010 ) [ arxiv:1005.0631 ] ; phys . rev . * d82 * , 054003 ( 2010 ) ; phys . rev . * d83 * , 114001 ( 2011 ) [ arxiv:1102.2385 ] ; arxiv:1011.3591 ; a. h. rezaeian , arxiv:1110.6642 . j. l. albacete and a. dumitru , arxiv:1011.5161 . d. kharzeev , e. levin and m. nardi , nucl . phys . * a747 * , 609 ( 2005 ) [ hep - ph/0408050 ] ; n. armesto , c. a. salgado , and u. a. wiedemann , phys . lett . * 94 * , 022002 ( 2005 ) [ hep - ph/0407018 ] . f. arleo , k. j. eskola , h. paukkunen and c. a. salgado , jhep * 1104 * , 055 ( 2011 ) [ arxiv:1103.1471 [ hep - ph ] ] ; d. de florian , r. sassot , p. zurita and m. stratmann , arxiv:1112.6324 [ hep - ph ] . z. -b . kang , i. vitev and h. xing , arxiv:1112.6021 [ hep - ph ] ; r. b. neufeld , i. vitev , b. -w . zhang , [ arxiv:1010.3708 ] . f. gelis and j. jalilian - marian , phys . d * 66 * , 014021 ( 2002 ) [ hep - ph/0205037 ] ; j. jalilian - marian , nucl . a * 753 * , 307 ( 2005 ) [ hep - ph/0501222 ] ; r. baier , a. h. mueller and d. schiff , nucl . a * 741 * , 358 ( 2004 ) [ hep - ph/0403201 ] . j. jalilian - marian , nucl . a * 770 * , 210 ( 2006 ) [ hep - ph/0509338 ] , eur . j. c * 61 * , 789 ( 2009 ) [ arxiv:0808.2769 ] . a. dumitru and j. jalilian - marian , phys . d * 82 * , 074023 ( 2010 ) [ arxiv:1008.0480 ] , phys . d * 81 * , 094015 ( 2010 ) [ arxiv:1001.4820 ] ; j. jalilian - marian , phys . d * 85 * , 014037 ( 2012 ) [ arxiv:1111.3936 ] . b. z. kopeliovich , a. v. tarasov and a. schafer , phys . c * 59 * , 1609 ( 1999 ) [ hep - ph/9808378 ] ; b. z. kopeliovich , j. raufeisen and a. v. tarasov , phys . b * 503 * , 91 ( 2001 ) [ hep - ph/0012035 ] . b. z. kopeliovich , a. h. rezaeian , h. j. pirner and i. schmidt , phys . lett . * b653 * , 210 ( 2007 ) [ arxiv:0704.0642 ] ; phys . rev . * d77 * , 034011 ( 2008)[arxiv:0711.3010 ] ; a. h. rezaeian _ et al _ , arxiv:0707.2040 ; b. z. kopeliovich , a. h. rezaeian and i. schmidt , nucl . a807 * , 61 ( 2008 ) [ arxiv:0712.2829 ] ; m. a. betemps and v. p. goncalves , jhep * 0809 * , 019 ( 2008 ) ; b. z. kopeliovich , e. levin , a. h. rezaeian and i. schmidt , phys . lett . * b675 * , 190 ( 2009 ) [ arxiv:0902.4287 ] ; m. v. t. machado and c. b. mariotto , eur . j. * c61 * , 871 ( 2009 ) ; a. h. rezaeian and a. schaefer , phys . rev . * d81 * , 114032 ( 2010 ) [ arxiv:0908.3695 ] ; m. a. betemps and m. v. t. machado ; phys . rev . * d82 * , 094025 ( 2010 ) ; s. abreu _ et . al . _ , j. phys . * g35 * , 054001 ( 2008 ) [ arxiv:0711.0974 ] . j. l. albacete and c. marquet , phys . lett . * 105 * , 162301 ( 2010 ) [ arxiv:1005.4065 ] . a. dumitru , a. hayashigaki , j. jalilian - marian , nucl . phys . * a765 * , 464 ( 2006 ) [ hep - ph/0506308 ] , a. dumitru , a. hayashigaki , j. jalilian - marian , nucl . * a770 * , 57 ( 2006 ) [ hep - ph/0512129 ] ; t. altinoluk and a. kovner , phys . rev . * d83 * , 105004 ( 2011 ) [ arxiv:1102.5327 ] . b. z. kopeliovich and a. h. rezaeian , int . j. mod e18 * , 1629 ( 2009 ) [ arxiv:0811.2024 ] ; f. dominguez and c. marquet , b - w xiao and f. yuan , phys . rev . * d83 * , 105005 ( 2011 ) ; e. avsar , arxiv:1203.1916 . j. jalilian - marian , a. kovner , l. d. mclerran and h. weigert , phys . rev . * d55 * , 5414 ( 1997 ) ; j. jalilian - marian , a. kovner , a. leonidov and h. weigert , nucl . * b504 * , 415 ( 1997 ) ; phys . rev . * d59 * , 014014 ( 1999 ) ; phys . rev . * d59 * , 014015 ( 1999 ) ; phys . rev . * d59 * , 034007 ( 1999 ) [ erratum - ibid . * d59 * , 099903 ( 1999 ) ] ; a. kovner , j. g. milhano and h. weigert , phys . * d62 * , 114005 ( 2000 ) ; a. kovner and j. g. milhano , phys . rev . * d61 * , 014012 ( 2000 ) ; e. iancu , a. leonidov and l. d. mclerran , nucl . * a692 * , 583 ( 2001 ) , phys . * b510 * , 133 ( 2001 ) ; e. ferreiro , e. iancu , a. leonidov and l. mclerran , nucl . * a703 * , 489 ( 2002 ) ; h. weigert , nucl . a * 703 * , 823 ( 2002 ) [ hep - ph/0004044 ] . i. balitsky , nucl . * b463 * , 99 ( 1996 ) [ hep - ph/9509348 ] ; y. v. kovchegov , phys . rev . * d60 * , 034008 ( 1999 ) ; phys . rev . * d61 * , 074018 ( 2000 ) . i. balitsky , phys . rev . * d75 * , 014001 ( 2007 ) ; i. balitsky , g. a. chirilli , phys . rev . * d77 * , 014019 ( 2008 ) ; e. gardi , j. kuokkanen , k. rummukainen and h. weigert , nucl . phys . * a784 * , 282 ( 2007 ) ; y. v. kovchegov and h. weigert , nucl . phys . * a784 * , 188 ( 2007 ) . e. avsar , a. m. stasto , d. n. triantafyllopoulos and d. zaslavsky , jhep * 1110 * , 138 ( 2011 ) [ arxiv:1107.1252 ] . l. d. mclerran and r. venugopalan , phys . d * 49 * , 2233 ( 1994 ) [ hep - ph/9309289 ] ; j. jalilian - marian , a. kovner , l. d. mclerran and h. weigert , phys . rev . d * 55 * , 5414 ( 1997 ) [ hep - ph/9606337 ] . j. l. albacete , n. armesto , j. g. milhano and c. a. salgado , phys . rev . * d80 * , 034031 ( 2009 ) [ arxiv:0902.1112 ] ; j. l. albacete , n. armesto , j.g . milhano , p. quiroga arias and c. a. salgado , eur . . j. * c71 * , 1705 ( 2011 ) [ arxiv:1012.4408 ] . k. dusling , f. gelis , t. lappi and r. venugopalan , nucl . * a836 * , 159 ( 2010 ) [ arxiv:0911.2720 ] . a. dumitru and e. petreska , arxiv:1112.4760 . a. dumitru , j. jalilian - marian and e. petreska , phys . d * 84 * , 014018 ( 2011 ) [ arxiv:1105.4155 ] ; s. jeon and r. venugopalan , phys . d * 70 * , 105012 ( 2004 ) [ hep - ph/0406169 ] , s. jeon and r. venugopalan , phys . d * 71 * , 125003 ( 2005 ) [ hep - ph/0503219 ] . a. d. martin , w. j. stirling , r. s. thorne and g. watt , phys . lett . * b652 * , 292 ( 2007 ) ; a. d. martin , w. j. stirling , r. s. thorne and g. watt , eur . j. * c63 * , 189 ( 2009 ) . e. l. berger and j - w qiu , phys . rev . * d44 * , 2002 , ( 1991 ) ; s. catani , m. fontannaz , j. p. guillet and e. pilon , jhep * 0205 * , 028 ( 2002 ) ; p. aurenche , m. fontannaz , j - p . guillet , e. pilon and m. werlen , phys . rev . * d73 * , 094007 ( 2006 ) ; d. denterria and j. rojo , nucl b860 * , 311 ( 2012 ) . h. kowalski , t. lappi , c. marquet and r. venugopalan , phys . c * 78 * , 045201 ( 2008 ) [ arxiv:0805.4071 ] ; h. kowalski , t. lappi and r. venugopalan , phys . lett . * 100 * , 022303 ( 2008 ) [ arxiv:0705.3047 ] ; j. jalilian - marian and x. -n . wang , phys . d * 63 * , 096001 ( 2001 ) [ hep - ph/0005071 ] , phys . d * 60 * , 054016 ( 1999 ) [ hep - ph/9902411 ] ; j. jalilian - marian , phys . rev . c * 80 * , 054904 ( 2009 ) [ arxiv:0906.5175 ] . j. jalilian - marian , y. nara and r. venugopalan , phys . b * 577 * , 54 ( 2003 ) [ nucl - th/0307022 ] .

many problems of theoretical and practical interest involve finding an optimum over a family of convex functions . for instance , finding the projection on the convex functions in , and optimizing functionals arising from some problems in economics . in the continuous setting and assuming smoothness , the convexity constraints may be given locally by asking the hessian matrix to be positive semidefinite , but in making discrete approximations two difficulties arise : the continuous solutions may be not smooth , and functions with positive semidefinite discrete hessian need not be convex in a discrete sense . previous work has concentrated on non - local descriptions of convexity , making the number of constraints to grow super - linearly with the number of nodes even in dimension 2 , and these descriptions are very difficult to extend to higher dimensions . in this paper we propose a finite difference approximation using positive semidefinite programs and discrete hessians , and prove convergence under very general conditions , even when the continuous solution is not smooth , working on any dimension , and requiring a linear number of constraints in the number of nodes . using semidefinite programming codes , we show concrete examples of approximations to problems in two and three dimensions .

0804.1693

convex and concave functions appear naturally in many branches of science such as biology ( growth ) , medicine ( dose - response ) , or economics ( utility , production or costs ) , spurring in turn the interest of other areas , for example , statistics . it is no surprise , then , that many problems of theoretical and practical interest involve the optimization of a functional over a family of convex functions . a particularly important case is when the functional to be minimized is the norm of some normed function space @xmath1 , i.e. , given @xmath2 find @xmath3 where @xmath4 is a family of convex functions in @xmath1 . typical choices are the spaces @xmath5 or @xmath6 , and , in general , the @xmath7 spaces , where @xmath8 is a convex domain in @xmath9 . sometimes the convexity is a reasonable shape assumption on the model , which could be replaced by or added to other shape constraints such as radial symmetry , harmonicity or upper and lower bounds . this is the case , for example , of newton s problem of minimal resistance ( see , e.g. , @xcite ) . more surprisingly perhaps , the convexity constraint may be a consequence of the model , as in the design of some mechanisms in economics @xcite . actually , our interest in the subject arose from one of these problems , which we will call the _ monopolist problem _ , and is described in some detail in section [ sec : economics ] . in this problem we wish to find @xmath10 where @xmath11^d$ ] , @xmath12 is a non - negative probability density function over @xmath13 , and @xmath4 is a family of convex functions on @xmath13 with some further properties . the monopolist problem is numerically very challenging since , unlike the problems coming from physics , the space dimension @xmath14 may be much higher than @xmath15 or @xmath16 . there is a big qualitative jump going from one to more dimensions when dealing with convexity constraints . this can be appreciated readily by looking at the statistics literature , where the one dimensional convex regression or density problem has been considered both theoretically and numerically , and using different measuring functionals ( see , e.g. , @xcite ) , but very little has been done numerically in two or more dimensions . in this regard , it is worth mentioning the work by shih , chen and kim @xcite , who use appropriate splines in the mars ( multivariate adaptive regression splines ) algorithm . one of the main difficulties in obtaining discrete approximations to optimization problems over convex functions lies in giving a local and finite description of convex functions if @xmath17 . though this could be done for smooth functions of continuous variables by asking the hessian matrix to be positive semidefinite at all points , there is no similar characterization for discrete functions on meshes . as a matter of fact , we show in examples [ example : hessnoconv ] and [ example : convnohess ] that this can not be done easily . it goes without saying that there is a lot of work done on convex functions of continuous variables , as exemplified by the classical book by rockafellar @xcite . traditionally , this work leans more on properties satisfied by convex functions rather than on properties that imply convexity . in particular , very little has been done on local properties guaranteeing convexity . for discrete variables , there is a rather large work done by the discrete mathematics community for fixed lattices , and a number of definitions for discrete convex functions have been proposed ( see , e.g. , @xcite ) . again , usually these definitions are non - local . as far as we know , there exists very few literature on optimization problems on convex functions in dimensions two or more , either theoretically or numerically . besides those already mentioned ( and the references therein ) , let us cite three more which are prominent in the context of our work . * carlier and lachand - robert @xcite obtained the @xmath18 regularity of a variant of the monopolist problem , substituting the functional in for @xmath19 under some restrictions on the domain @xmath8 and the density @xmath12 . they obtained also @xmath18 regularity for convex minimizers of functionals of the form @xmath20 * carlier , lachand - robert and maury @xcite proposed a numerical scheme for minimizers of functionals of the type @xmath21 on closed convex subsets of convex functions of @xmath6 or @xmath5 , where @xmath22 is a quadratic function of @xmath23 and @xmath24 , and @xmath25 . we will discuss their approach in the next section ( after the inequalities ) . + as carlier et al . @xcite point out , their work encompasses the problem of finding @xmath26 for given @xmath27 , i.e. , a @xmath28-norm projection , and this problem is equivalent to that of finding the convex envelope @xmath29 of @xmath12 . thus , minimizing over convex functions and finding the convex envelope of a function are two quite related tasks . * being one of the main problems in computational geometry , there are a number of well established codes for finding the convex hull of a set of points in @xmath9 , which are very efficient in low dimensions . hence , it is natural to try to use these codes to approximate optimization problems on convex functions , an approach which lachand - robert and oudet @xcite applied to several problems . + it would be very interesting to see whether these ideas could be carried over , since the convex hull codes are quite fast in low dimensions . our approach , based on semidefinite programming , takes a different direction from those mentioned . let us recall that a semidefinite program is an optimization problem of the form @xmath30 where @xmath31 , @xmath32 are symmetric @xmath33 matrices , and @xmath34 indicates that the symmetric matrix @xmath35 is positive semidefinite . by letting the matrices @xmath36 be diagonal , we see that the program is a generalization of linear programming ( and includes it strictly ) . thus , in a semidefinite program the constraints can be a mixture of linear inequalities and positive semidefinite requirements . in this paper we give a theoretical framework for approximating many optimization problems on convex functions using a finite differences scheme which imposes a positive semidefinite constraint on a discretization of the hessian matrix . although not linear , our approach seems very natural and has many advantages . being of a local nature , the number of constraints grows only linearly with the number of nodes , and it works for any dimension of the underlying space . notwithstanding the already mentioned counterexamples ( [ example : hessnoconv ] and [ example : convnohess ] ) of the relation between convexity and positive semidefinite hessian , we will show that for many problems we obtain convergence to a continuous optimum . this convergence holds even for non - smooth optima , as those arising when projecting in the @xmath37 norm or in some problems of the type given by equation . in practice , our definition can be used to advantage by using the existing efficient semidefinite codes and , furthermore , it is very simple to program in higher dimensions . as a final remark , we think that our approach might be a first step to deal with other problems where convexity is a consequence of the model , such as transportation problems where the cost is quadratic , i.e. , for solving the monge - ampre equations , a prime example of fully non - linear differential equations . this article is organized as follows . in section [ sec : review ] we summarize some techniques that could be used to deal with the numerical approximation of optimal convex functions , showing their strengths and weaknesses , and introduce the discrete hessian . in section [ sec : convex : cont ] we give different characterizations of smooth convex functions of continuous variables , which allow us to extend the definition of the hessian to @xmath18 functions in terms of averages . the heart of the paper is section [ sec : approxs ] : in order to have a good definition of discrete convexity , we need to show that any convex function of continuous variables may be approximated by its discrete counterparts , and conversely , that a converging sequence ( in a suitable norm ) of discrete convex functions will do so to a convex function of continuous variables . we show in this section that the approximation of the discrete functions to the limit @xmath23 is uniform in @xmath23 and its derivatives if @xmath23 is smooth , and uniform in @xmath23 over compact subsets of @xmath8 , if @xmath23 is merely convex and bounded ( and hence continuous ) . although our definition is stated using the hessian , if a sequence of discrete functions converges to a function which is not @xmath38 or even @xmath18 , still convexity of the limit function is guaranteed . in section [ sec : functionals ] we pose a general structure for optimization problems on convex functions which fits the results of the previous sections , and may be applied to several important problems . in addition , error bounds may be obtained if the continuous optimal solution is smooth . the results of sections [ sec : convex : cont][sec : functionals ] are somewhat independent of semidefinite programming . however , as explained in section [ sec : review ] , for many problems of interest the functional and constraints involved may be expressed as a semidefinite program , and we present in section [ sec : numerical ] numerical examples showing that the current codes make it feasible to use our scheme on them . we begin section [ sec : numerical ] by considering the monopolist problem in two and three dimensions , comparing our results to the analytical solution when @xmath39 in . we continue by showing some norm projections , exploring the behavior under different functionals of an example given by carlier , lachand - robert and maury @xcite . moreover , we exhibit an explicit example in which the @xmath37 projection is not unique . we finish the section on numerical examples by showing how our scheme could be used to fit a ( discrete ) convex function to noisy data given on the nodes of a regular mesh , using the @xmath28 norm ( i.e. , a least squares approach ) , and also the @xmath40 and @xmath37 norms , which are not often seen . in this section we present some ideas and techniques that could be used to approximate numerically an optimal convex function . for the sake of simplicity , we will work on the unit @xmath14-dimensional cube in @xmath9 , @xmath11^d$ ] . if @xmath41 and @xmath42 is a discretization of @xmath43 $ ] , then a convex function @xmath23 defined on @xmath43 $ ] satisfies the inequalities @xmath44 conversely , if a function @xmath23 defined only in the mesh points @xmath45 satisfies the inequalities , we can always extend it linearly in the intervals @xmath46 $ ] , @xmath47 , so that the resulting piecewise linear function will be convex on @xmath43 $ ] . thus , the inequalities may be used to define discrete convexity in one dimension . it is clear that via the piecewise linear extension we can approximate any convex function in @xmath43 $ ] . on the other hand , as we show in lemma [ lemma:1d : convergence:1 ] , under some adequate assumptions if a sequence of discrete convex functions ( satisfying for each corresponding subdivision ) converges pointwise , then the limit defines a unique convex function in @xmath43 $ ] . let @xmath48 be given , @xmath49 , and suppose @xmath50 is a subdivision of @xmath43 $ ] satisfying and such that @xmath51 let @xmath23 be defined on @xmath52 and satisfy . for a given @xmath53 , @xmath54 , let us consider @xmath55 from we see that @xmath56 for all @xmath57 such that @xmath58 , @xmath59 . therefore , if @xmath60 we have @xmath61 where @xmath62 and @xmath63 are constants depending only on @xmath53 and @xmath64 ( but not on @xmath23 or @xmath52 ) . in other words , if @xmath23 is extended as a piecewise linear function to all of @xmath43 $ ] , the resulting function is uniformly lipschitz on compact subsets of @xmath65 . hence , [ lemma:1d : convergence:1 ] let @xmath66 be a sequence of meshes in @xmath43 $ ] , @xmath67 , satisfying for each @xmath48 , and such that @xmath68 for @xmath69 . suppose @xmath70 is defined in @xmath71 for each @xmath48 , and the sequence @xmath72 is such that for all @xmath73 , @xmath74 exists and is finite , and let @xmath75 then @xmath23 may be extended to all of @xmath43 $ ] as a continuous convex function in a unique way . moreover , the convergence of @xmath70 ( extended piecewise linearly ) to @xmath23 is uniform in compact subsets of @xmath65 . also , by the arzela - ascoli theorem , [ lemma:1d : convergence:2 ] let @xmath66 be a sequence of meshes as in the previous lemma . suppose @xmath76 is given , and that for each @xmath48 a function @xmath70 is defined in @xmath71 so that @xmath77 and assume @xmath70 is extended to all of @xmath43 $ ] piecewise linearly . then , there exists a subsequence of @xmath78 of @xmath79 converging to a continuous convex function @xmath23 defined on @xmath43 $ ] . moreover , the convergence of the subsequence @xmath78 to @xmath23 is uniform in compact subsets of @xmath65 . thus , from a theoretical point of view , the discrete functions satisfying are well understood . from the numerical point of view , when used in a discretization of an optimization problem , the constraints coming from the inequalities are linear , and solving the resulting discrete optimization problem with them is usually not much harder than solving it without them . hence , the case @xmath41 poses no major trouble . for @xmath17 it will be more convenient to work only with regular meshes ( or grids ) on @xmath11^d$ ] . thus , for fixed @xmath80 ( @xmath81 for some @xmath82 ) , the mesh @xmath71 will consist of all points @xmath83 of the form @xmath84 with @xmath85 . denoting by @xmath86 the interior of @xmath13 , we set @xmath87 and denote by @xmath88 the set of real valued functions defined on @xmath71 . a first simple idea to extend the inequalities to more dimensions , is to consider the set of functions @xmath89 satisfying the convexity constraints @xmath90 where @xmath91 denotes the @xmath92-th vector of the canonical basis of @xmath9 . this set of discrete functions is very appealing because the convexity is modelled by using only @xmath93(number of mesh points ) linear constraints , as in the case @xmath41 . as we have seen , this approach is exact in one dimension , and gives satisfactory results in many cases in more dimensions ( in particular for some of the specific examples treated later in section [ sec : economics ] ) , but there is no guarantee of convexity in the limit function if @xmath17 . in fact , by taking the interpolant , with this set we can certainly approximate any function of continuous variables which is convex , but we can also approximate other functions . for example , we may approximate @xmath94 which is linear and thus convex in each coordinate direction , but not convex in @xmath95 . this shows that the definition of discrete convexity should be done with more care : when @xmath17 , we have to take into account every possible direction . it is reasonable , then , to say that a discrete function is convex if @xmath96 since as @xmath48 goes to @xmath97 the possible directions become dense , it is possible ( under some conditions ) to regain convexity in the limit as we had for one dimension . in a way , this is the approach followed by carlier , lachand - robert and maury @xcite . in a two dimensional setting , they consider discrete convex functions which are restrictions to a mesh of convex functions of continuous variables , and show that this definition is equivalent to an intrinsic one , stated only in terms of the value of the function at the grid points , similar to . the problem with this description is that it is non - local , and the number of constraints needed in two dimensions ( after pruning ) reportedly grows approximately as @xmath98 , where @xmath99 is the number of nodes in the grid . moreover , this approach is very difficult to extend to higher dimensions . in order to keep the definition of discrete convexity local , another possibility is to consider discretizations of the hessian matrix , as we do in this work . for @xmath100 and @xmath101 , we define the ( forward ) first order finite differences by @xmath102 and the second order finite differences by @xmath103 or , if @xmath104 , by @xmath105 clearly , not all of these finite differences are defined for points in @xmath106 . therefore , when mentioning @xmath107 or @xmath108 for @xmath101 , we will implicitly assume that @xmath109 and @xmath92 ( and eventually @xmath22 ) are such that the corresponding finite difference is well defined at @xmath109 . finally , we define the discrete hessian of @xmath100 at @xmath110 , as the symmetric matrix @xmath111 whose @xmath112 entry is @xmath113 . we are faced now with two issues : how can a discrete optimization problem with positive semidefinite discrete hessian constraints be solved ? once we found a discrete solution , how does it approximate the continuous solution ? as mentioned in the introduction , for the first issue we will use the semidefinite programming model , where the objective is linear and the constraints are positive semidefinite . thus , if the objective of the continuous problem is not linear , we must rewrite it as a constraint . however , it is worth mentioning that not every functional may be readily modelled as a positive semidefinite constraint , an example of such functionals being @xmath114 arising in newton s problem of minimal resistance . although this method may not be used for _ all _ functionals , it can be used in _ many _ cases as we now illustrate . the interested reader is referred to the article by vandenberghe and boyd @xcite for other possibilities . in what follows we associate with each mesh node @xmath115 , @xmath116 , the unknown value @xmath117 , the discrete hessian @xmath118 , and a @xmath14-dimensional cube @xmath119 centered at @xmath115 of measure @xmath120 , so that @xmath121 . the program @xmath122 subject to @xmath23 convex , is modeled by adding @xmath99 more unknowns @xmath123 ( for a total of @xmath124 ) as @xmath125 in turn , the constraints of the form @xmath126 are lumped as @xmath127 , and finally modeled as the linear inequalities @xmath128 this is analogous to the @xmath40 projection : @xmath129 and the constraints of the form @xmath130 are written in the form @xmath131 the @xmath132 projection is handled in a similar way . this is analogous to the other projections , except that we only need to add one more variable @xmath133 ( instead of @xmath99 ) , obtaining the discrete program @xmath134 the next issue is to see how well the discrete solutions of the semidefinite program approximate the continuous convex solution . our first hope is that @xmath135 will be equivalent to some version of convexity of discrete functions , for example to the one given by the inequalities . the next examples show that this is not true . the first one shows that a non - convex function may have a positive semidefinite discrete hessian , and the second one shows that the discrete hessian may not be positive semidefinite for some convex functions . [ example : hessnoconv ] let us consider @xmath136 , and @xmath23 defined on the 2-dimensional @xmath137 grid @xmath138 by @xmath139 we may check that ( with @xmath140 ) , @xmath141 whose eigenvalues are @xmath15 and @xmath97 ( with eigenvectors @xmath142 and @xmath143 ) , and is thus positive semidefinite . however , the restriction to the diagonal @xmath144 is not convex : @xmath145 notwithstanding these examples , in section [ sec : approxs ] we will show that , under suitable conditions , by asking for positive semidefinite discrete hessians we may conveniently approximate continuous convex functions , and obtain variants of lemmas [ lemma:1d : convergence:1 ] and [ lemma:1d : convergence:2 ] . however , we will need some previous results which are tackled in the following section . in this section we deal with variants of the hessian matrix , initially defined for @xmath38 functions , which allow us to characterize convexity for @xmath18 functions locally . let us start by introducing some more precise notation : the distance from a point @xmath109 to a set @xmath35 will be denoted by @xmath146 . if @xmath147 and @xmath80 , @xmath148 the gradient of @xmath23 is denoted by @xmath149 , and we write @xmath150 for @xmath151 . if @xmath152 is an open set , we denote by @xmath153 the set of functions having all derivatives up to order @xmath154 continuous on @xmath8 , and by @xmath155 the set of functions having all derivatives up to order @xmath154 uniformly continuous in @xmath8 . if @xmath8 is omitted , @xmath156 . let @xmath157 be a non - negative , real valued function in @xmath158 , vanishing outside @xmath159 , and such that @xmath160 and for @xmath161 define @xmath162 the function @xmath163 defined for functions @xmath23 and points @xmath109 whenever the right hand side makes sense , has many interesting well known properties , and we state some of them without proof , referring the reader to , e.g. , the book by ziemer ( * ? ? ? * theorem 1.6.1 and remark 1.6.2 ) . [ theorem : ziemer ] if @xmath164 , then for every @xmath161 , @xmath165 and @xmath166 for every multi - index @xmath167 , where for @xmath168 , @xmath169 . if @xmath170 then @xmath171 is defined for @xmath172 and @xmath173 . @xmath174 whenever @xmath109 is a lebesgue point of @xmath23 . if @xmath175 , and @xmath176 is a compact subset of @xmath8 , then @xmath177 converges uniformly to @xmath23 on @xmath176 as @xmath178 . the functions @xmath177 , also called regularizations or mollifiers of @xmath23 , will make us work often with the sets @xmath179 where @xmath180 , @xmath181 . our first result is a simple characterization of continuous convex functions using the regularizations @xmath177 , and we omit its proof . [ lemma : cont:1 ] suppose @xmath182 and @xmath177 is defined as in . we have , if @xmath23 is convex , then @xmath177 is convex in @xmath183 for all @xmath161 . conversely , if for a sequence of @xmath53 s converging to @xmath97 , @xmath177 is convex in @xmath183 , then @xmath23 is convex . the hessian of @xmath184 at @xmath109 is the symmetric matrix @xmath185 whose @xmath186 entry is @xmath187 , and convex functions in @xmath38 are characterized by the positive semidefinite condition @xmath188 also , for @xmath184 , and @xmath189 , the average @xmath190 defined for @xmath191 , converges to @xmath192 for all @xmath193 , as @xmath194 . thus the condition @xmath195 implies the convexity of @xmath23 . actually , since the average of positive semidefinite matrices is positive semidefinite , is equivalent to convexity for @xmath196 . the definitions of @xmath197 and @xmath198 involve second order derivatives which are not defined if @xmath23 is not smooth enough . however , we may express @xmath198 merely in terms of @xmath23 and @xmath24 integrating by parts , and the resulting formula will make sense for @xmath199 . thus , for @xmath200 and @xmath191 let us define @xmath201 as the symmetric @xmath202 matrix whose diagonal entries are [ equs : cont:2 ] @xmath203 and , if @xmath104 , @xmath204 - \int_{f^{-}_{\delta , i}(x)\cap f^{+}_{\delta , j}(x ) } u(y)\,{\mathrm{d}}{y } + \int_{f^{-}_{\delta , i}(x)\cap f^{-}_{\delta , j}(x ) } u(y)\,{\mathrm{d}}{y } \bigr),\end{gathered}\ ] ] where @xmath205 denotes the @xmath206 dimensional face of the cube @xmath207 having outward normal @xmath208 . of course , since the equations were obtained integrating by parts , we have : [ lemma : cont:2 ] if @xmath184 then @xmath209 . as we show now , the extension @xmath210 of @xmath211 to functions in @xmath18 , still gives a local characterization of convexity . [ theorem : cont:1 ] the following are valid if @xmath200 : if @xmath23 is convex , then @xmath212 for all @xmath189 and @xmath191 . if for a sequence @xmath213 , @xmath214 for all @xmath215 , then @xmath23 is convex . suppose @xmath200 , and consider the regularization @xmath177 defined in . since we can interchange derivatives and integrals with the convolution , @xmath216 if @xmath23 is convex , by the first part of lemma [ lemma : cont:1 ] we know that @xmath177 is convex in @xmath217 , and since @xmath218 , for @xmath219 we have @xmath220 . thus , by lemma [ lemma : cont:2 ] , @xmath221 using theorem [ theorem : ziemer ] , since @xmath177 and their derivatives converge uniformly to @xmath23 in compact subsets of @xmath222 , we must have @xmath223 for @xmath109 in compact subsets of @xmath224 , and therefore in the whole of @xmath224 . on the other hand , if @xmath225 for @xmath191 , since an integral mean of positive semidefinite matrices is positive semidefinite , using and lemma [ lemma : cont:2 ] , we see that @xmath226 which implies that @xmath177 is convex in @xmath227 . letting @xmath228 go to @xmath97 while keeping @xmath53 fixed , we see that @xmath177 is convex in @xmath183 , and by the second part of lemma [ lemma : cont:1 ] , @xmath23 must be convex . there are two main issues when defining the set of discrete approximants to be used : we want it to be rich enough to approximate every convex function , and we want this set to be not too large , to avoid convergence to non - convex functions . the first point is very natural , and necessary to approximate the solution to the problem . the second point might look artificial at first sight , but if not enforced , then we could be approximating a non - convex function . this is the case , for instance , if we only require convexity along the coordinate axes , as we stated in section [ sec : review ] . our discrete approximations will be a subset of functions having positive semidefinite discrete hessians , and the main purpose of this section is to address the two issues mentioned above . we will show in theorem [ theorem : discrete : approx ] that , despite the examples [ example : hessnoconv ] and [ example : convnohess ] , the discrete hessian @xmath229 may be used to obtain very good approximations to convex functions of continuous variables . this is not surprising since we can approximate , say , @xmath230 , by discrete functions whose finite differences up to order @xmath15 converge to the derivatives up to order @xmath15 of @xmath23 , and then add a small perturbation to bring up the eigenvalues of the discrete hessians . this is the main idea of the proof , and in its course , we will use the following part of the hoffman - wielandt theorem @xcite . [ theorem : hoffman ] there exists a positive constant @xmath231 , depending only on the dimension @xmath14 , such that if @xmath232 $ ] and @xmath233 $ ] are symmetric @xmath234 matrices , and @xmath235 and @xmath236 are their minimum eigenvalues , then @xmath237 [ theorem : discrete : approx ] let @xmath230 be convex . then , for any @xmath161 , there exists @xmath238 such that for any @xmath48 , @xmath239 , there exists a function @xmath70 defined on @xmath71 satisfying for all @xmath101 , @xmath240 and @xmath241 let us recall that in the inequality , for @xmath242 we consider only the finite differences that are defined at that @xmath109 . if @xmath23 is convex and smooth , given @xmath243 there exists @xmath244 such that for all @xmath245 and @xmath246 , @xmath247 where @xmath231 is the constant in theorem [ theorem : hoffman ] . hence , since @xmath23 is convex and therefore @xmath248 , the minimum eigenvalue of @xmath249 is uniformly bounded below by @xmath250 . now consider the function @xmath251 for which , if @xmath48 small enough , @xmath252 where @xmath253 is the identity matrix in @xmath254 . if @xmath255 , the function @xmath256 defined on @xmath71 , satisfies @xmath257 and @xmath258 thus , for some constant @xmath259 depending only on @xmath14 , the inequality holds with @xmath53 replaced by @xmath260 . the result follows now by taking @xmath261 and @xmath262 appropriately . the main implication of theorem [ theorem : discrete : approx ] is that any smooth convex function is a limit of a sequence of functions with positive semidefinite discrete hessians , giving an affirmative answer to the first issue mentioned at the beginning of this section . moreover , an application of lemma [ lemma : cont:1 ] implies this result also for non - smooth convex functions , with convergence in @xmath263 on compact subsets of @xmath8 ( see definitions below ) . we would like to show next that our set of approximants also solves the second issue . that is , if we have a convergent ( in certain norm ) sequence of functions with positive semidefinite discrete hessians , then the limit is convex . on the other hand , if the sequence is not convergent , we would like also to understand under which further conditions on the sequence we may extract a subsequence converging to a convex function . in what follows , we will work with sequences of functions defined on finer and finer meshes , that is , sequences @xmath264 with @xmath265 for every @xmath266 , such that @xmath267 and @xmath268 decreases to @xmath97 . we will denote by @xmath269 the set of all such sequences , and use the notation @xmath270 so as not to clutter even more the notation , usually we will drop the index @xmath266 , writing @xmath271 for @xmath272 , when this does not lead to confusion . if @xmath273 and @xmath274 , we will say that @xmath275 if for any @xmath161 , there exists @xmath276 , so that @xmath277 for all @xmath278 and @xmath279 , that is , @xmath280 as @xmath281 . in this case we will write , with a little abuse of notation , @xmath282 if in addition @xmath199 , we write , similarly , @xmath283 to indicate that @xmath284 and @xmath285 uniformly for all @xmath286 for which the finite differences make sense and @xmath287 . [ lemma : discrete : convergence ] suppose @xmath288 and @xmath199 are such that @xmath289 and @xmath290 then @xmath23 is convex . we follow closely what was done in section [ sec : convex : cont ] , using a variant of the divergence theorem for discrete variables , so that only approximations to the function or its derivatives but not the second derivatives are needed . for @xmath291 and @xmath292 , let us define @xmath293 where the sum is over all @xmath294 such that @xmath295 . since the sum of positive semidefinite matrices is a positive semidefinite matrix , we have @xmath296 in one dimension we have just one entry in @xmath297 , involving a term of the form @xmath298 that is , if @xmath41 then @xmath299 if @xmath136 and @xmath300 , the diagonal entries are similar to the one dimensional case . for instance , @xmath301 where both sums are on @xmath154 s such that @xmath302 . on the other hand , the off - diagonal terms are of the form @xmath303 where @xmath304 for @xmath305 we get similar expressions to those of equations and for the diagonal terms , and for the off - diagonal terms , except that as when going from to they must be summed over mesh points on @xmath206 dimensional surfaces perpendicular to the ( say ) @xmath92-th direction , or @xmath306 dimensional surfaces perpendicular to the ( say ) @xmath186 plane . this was also the case in the equations for the continuous case , and so we can think of these sums as approximations to those integrals , where small @xmath14-dimensional cubes of side length @xmath48 centered at mesh points have been used . in order to do this we must multiply the entries by @xmath307 so as to obtain the correct dimensions . in fact , since we are assuming @xmath308 and @xmath309 involves either no finite differences ( off the diagonal ) or only first order differences ( in the diagonal ) which converge uniformly to ( respectively ) @xmath23 or its first derivatives , @xmath310 where @xmath311 since the limit of positive semidefinite matrices is positive semidefinite . on the other hand , due to the convergence of @xmath70 to @xmath23 in the @xmath312 norm , we can see that , as @xmath67 , @xmath313 & \qquad - \int_{f^{-}_{\delta , i}(x)\cap f^{+}_{\delta , j}(x ) } u(y)\,{\mathrm{d}}{y } + \int_{f^{-}_{\delta , i}(x)\cap f^{-}_{\delta , j}(x ) } u(y)\,{\mathrm{d}}{y } \bigr),\end{aligned}\ ] ] if @xmath314 . thus , @xmath315 which implies @xmath316 for all @xmath291 . the result now follows from theorem [ theorem : cont:1 ] . the convergence in @xmath317 should be guaranteed by the definition of the discrete problem and is problem dependent . we make now a general assumption on an algorithm to ensure convergence in @xmath317 . let us denote by @xmath318 the space of lipschitz continuous functions defined on @xmath13 , recalling that @xmath319 , the space of continuous functions having derivatives in the weak sense up to order one bounded . if @xmath320 , we let @xmath321 be the set of functions @xmath322 such that both @xmath323 for all @xmath245 and @xmath324 for all @xmath287 and @xmath325 whenever @xmath326 . of course , the condition @xmath324 is equivalent to @xmath327 . similarly , let us denote by @xmath328 the space of functions whose first derivatives are in @xmath318 , and by @xmath329 the set of all functions in @xmath330 which have all weak derivatives up to order @xmath15 bounded by @xmath331 . in particular , we have @xmath332 . by analogy to the continuous case , let us define for @xmath48 and @xmath331 positive the following spaces of discrete functions : @xmath333 with the understanding that on the right hand sides we take @xmath334 or @xmath335 for all @xmath336 and @xmath22 where it makes sense . the following result is a version of the arzela - ascoli theorem , and we omit its proof . [ theorem : arzela ] if @xmath273 for which @xmath337 for all @xmath80 , then there exists @xmath338 and a subsequence @xmath339 of @xmath340 such that @xmath341 combining the previous results , we have : [ theorem : discrete : converge ] if @xmath273 is such that @xmath342 for all @xmath343 and @xmath80 , then there exists @xmath344 , and a subsequence @xmath339 of @xmath340 such that @xmath345 applying theorem [ theorem : arzela ] to the functions @xmath346 , perhaps on some smaller meshes @xmath347 with @xmath348 , we may find functions @xmath349 and a subsequence of @xmath340 , @xmath339 , such that @xmath350 to show that there exists @xmath23 such that @xmath351 ( in the classical sense ) , we define for @xmath352 , @xmath353 since for @xmath287 , @xmath354 is bounded ( by @xmath331 ) and continuous , @xmath23 is well defined and continuous . using that @xmath355 converge uniformly to @xmath354 , for @xmath356 we may write @xmath357 but , for a given @xmath358 and all @xmath359 we have @xmath360 and therefore @xmath361 arguing as in equation and taking limits , we may also verify the `` independence of path '' , that is , for @xmath362 and @xmath363 we may write , @xmath364 using the continuity of @xmath23 , we see that the last equation is valid for all @xmath193 and @xmath358 small enough , and , using once more the continuity of @xmath23 , that @xmath365 for all @xmath193 and all @xmath189 , and therefore @xmath366 the convexity of @xmath23 follows from lemma [ lemma : discrete : convergence ] . [ remark : noconvergenceof2 ] if @xmath367 is a sequence of convex functions converging pointwise to @xmath23 , then @xmath23 must be convex . however it is not true in general that the second derivatives or the hessian of @xmath367 will converge to that of @xmath23 , even if we have uniform convergence of @xmath367 and its derivatives to those of @xmath23 . in this sense , theorem [ theorem : discrete : converge ] can not be bettered too much . for example , consider in @xmath41 the functions @xmath368 and , with @xmath81 for @xmath82 , @xmath369 we have , @xmath370 0 \le u_h'(x ) = u'(x ) - \dfrac{\sin ( \pi n x)}{n\pi } \le 1 , & \lim_{h\to 0}\,u'_h(x ) = u'(x ) , \end{array}\ ] ] where the limits are uniform in @xmath109 . also , @xmath371 taking @xmath372 , @xmath373 , for @xmath374 we obtain @xmath375 and so the second order differences at dyadic points converge to this constant value . however , also at these points , @xmath376 . on the other hand , we may weaken the conditions on convergence , following what was done for the one - dimensional case in section [ sec : review ] . if @xmath70 is defined in @xmath71 and @xmath377 for @xmath101 , its restriction to the one - dimensional line obtained by fixing all coordinates except the @xmath92-th one , satisfies @xmath378 for @xmath287 , since these are coefficients in the main diagonal of @xmath379 . that is , the inequalities are satisfied , and hence if for some @xmath76 , @xmath380 then for any given @xmath161 we may find a constant @xmath381 , depending on @xmath64 and @xmath53 but not on @xmath23 , such that @xmath382 therefore , by applying theorem [ theorem : arzela ] to @xmath14-dimensional cubes contained in @xmath222 , we may strengthen lemma [ lemma : discrete : convergence ] to obtain a variant of lemma [ lemma:1d : convergence:1 ] : [ corollary : discrete : convergence ] let @xmath383 such that @xmath384 and suppose @xmath23 is a ( finite ) function defined on @xmath385 satisfying @xmath386 then , the convergence of @xmath70 to @xmath23 is uniform on compact subsets of @xmath222 , and @xmath23 may be uniquely extended to a convex function defined on all of @xmath13 . similarly , we may weaken the conditions of theorem [ theorem : arzela ] to obtain the following version of lemma [ lemma:1d : convergence:2 ] : [ corollary : arzela ] if @xmath76 and @xmath273 are such that @xmath387 then there exists a continuous convex function @xmath23 defined on @xmath13 and a subsequence @xmath339 of @xmath340 such that @xmath388 moreover , the convergence is uniform on compact subsets of @xmath222 . we conclude this section with some comments . [ remark : bdy ] if @xmath184 , @xmath187 is defined initially for @xmath193 , but the very definition of @xmath38 as the set of those functions in @xmath389 having second order derivatives uniformly continuous on @xmath222 , makes it possible to define @xmath187 for @xmath390 by a limiting argument . and the same goes for lower order derivatives , and even @xmath192 . the situation is different for discrete functions defined on @xmath71 for fixed @xmath80 , since we can not take limits . however , the particular geometry of @xmath13 makes it possible ( as we did ) to consider for @xmath242 as many finite differences as we can . for instance , @xmath391 is well defined for @xmath242 as long as @xmath392 . pushing the definitions a little further , we may define the discrete hessian @xmath249 for @xmath393 , by including as many second derivatives as we can . for example , if @xmath394 and @xmath140 , we can define @xmath395 \operatorname{\delta}^2_{h,12 } u(x ) & \operatorname{\delta}^2_{h,22 } u(x)\\[2pt ] \end{bmatrix } & \quad & \text{if $ x = ( 1/2,1/2,0)$},\\[14pt ] h_h u(x ) = \begin{bmatrix } \operatorname{\delta}^2_{h,11 } u(x ) \end{bmatrix } & & \text{if $ x = ( 1/2,0,0)$ } , \end{array}\ ] ] leaving @xmath249 undefined if @xmath396 . as the reader may verify , our previous results involving @xmath229 remain valid with this interpretation of the discrete hessian . we are in position now to use finite difference approximations of a wide class of optimization problems on convex functions . let us describe this technique by assuming , for instance , that @xmath397 is a banach space of real valued functions on @xmath13 , the functional @xmath398 is defined and continuous on @xmath1 , and we are interested in the optimization problem @xmath399 where @xmath4 is a family of convex functions , @xmath400 . if the functions in @xmath4 may be approximated by convex functions in @xmath401 , then using theorem [ theorem : discrete : approx ] it may be not too difficult to define for each @xmath80 ( or a sequence converging to @xmath97 of such @xmath48 s ) , a family @xmath402 , @xmath403 , and a functional @xmath404 defined on @xmath402 , such that : @xmath405 for all @xmath406 and @xmath110 , for any @xmath407 and any @xmath161 , there exists @xmath80 and @xmath408 such that @xmath409 . condition 2 immediately implies that @xmath410 to prove the converse , we observe that @xmath411 where @xmath412 . keeping @xmath320 fixed , we may find a sequence @xmath413 , with @xmath414 such that @xmath415 and using theorem [ theorem : discrete : converge ] , a subsequence @xmath416 and a convex function @xmath417 with @xmath418 converging to @xmath97 . if @xmath419 and @xmath404 are such that @xmath420 letting @xmath421 we will have @xmath422 putting together the inequalities and , we will have discrete approximations of the problem . let us give some more concrete examples . suppose , for instance that @xmath423 is the set of functions @xmath424 with finite norm @xmath425 and suppose @xmath4 is the set of all convex functions in @xmath132 . given @xmath2 we would like to find its projection on @xmath4 , that is , find @xmath426 such that @xmath427 and , getting rid of the square roots , we may set @xmath428 in this example we actually have a unique minimum , since the norm is strictly convex . we may consider then @xmath402 as the set of discrete functions @xmath429 with @xmath430 for all @xmath110 . assuming for simplicity that @xmath431 , for @xmath429 we define @xmath432 then , it is easy to see that and hold . in fact , the convex functions of @xmath1 may be approximated by @xmath433 convex functions , so that , given @xmath161 there exists @xmath434 such that @xmath435 and use theorem [ theorem : discrete : approx ] to find @xmath48 small enough and @xmath406 such that @xmath436 thus we will have @xmath437 for some @xmath406 . in this section we illustrate the behavior of the numerical scheme by applying it to the problems mentioned in the introduction , namely , the monopolist problem , norm projections on the set of convex functions , and fitness of data by discrete convex functions . in all of the following examples , we associate with each mesh node @xmath115 , @xmath116 , the unknown value @xmath117 , and the square @xmath438 , of area @xmath120 . we also consider the discrete hessian @xmath118 as discussed in the remark [ remark : bdy ] , i.e. , imposing convex constraints on the boundary whenever they make sense . even though theorem [ theorem : discrete : converge ] requires us to impose upper bounds on the second order differences in order to ensure convergence , the following numerical experiments were carried on without this requirement . convergence was nevertheless observed at optimal rates . the times reported correspond to the experiments being run on a pc , with a 2.8ghz pentium iv processor and 2 gb of ram , running _ linux_. the matrices were assembled using _ octave _ @xcite and the semidefinite program was solved using _ csdp _ 5 @xcite with the default parameters . the graphics were obtained using _ mathematica _ @xcite . since this problem is not widely known in the mathematics community , let us start by giving a brief description of it following the one given in @xcite , where it is referred to as the _ revenue maximization in a multiple - good monopoly _ problem . [ prob : monopolist ] a seller with @xmath14 different objects faces a single buyer , whose preferences over consumption and money transfers are given by @xmath439 , where @xmath440^d$ ] is the vector of the buyer s valuations , @xmath441 is the vector of quantities consumed for each good , and @xmath442 is the monetary transfer made to the seller . the valuations @xmath109 are only observed by the buyer , and a density function @xmath443 represents the seller s belief on the buyer s private information @xmath109 . the seller s problem is to design a revenue maximizing mechanism to carry out the sale , and it is enough to consider only direct revelation mechanisms : the buyer must prefer to reveal its information truthfully ( incentive compatibility ) and to participate voluntarily ( individual rationality ) . under these conditions , it can be proved that the seller s problem may be written as @xmath444 where @xmath11^d$ ] , @xmath12 is a non - negative probability density function over @xmath13 , and @xmath4 is the set of functions @xmath23 satisfying @xmath23 is convex , @xmath445 for all @xmath446 ( the gradient taken in the weak sense and the inequalities componentwise ) , and @xmath447 . the functional to be maximized is the seller s expected revenue . for a buyer of type @xmath109 , the solution @xmath23 to this optimization problem represents the utility received by her , and the @xmath92-th component of @xmath24 denotes the probability that she will obtain good @xmath92 . the restriction of convexity stems from incentive compatibility , and the condition @xmath448 from individual rationality . we show now some numerical results for the problem [ prob : monopolist ] for the special case @xmath39 , for which analytic solutions in 2 and 3 dimensions are known , allowing us to judge the behavior of the discrete approximations . that is , the functional to be minimized is @xmath449 in both 2 and 3 dimensions , the solutions are piecewise linear convex functions whose partial derivatives are either @xmath97 or @xmath450 . for example , for @xmath136 the solution is @xmath451 where @xmath452 and the value at the optimum is @xmath453 in figure [ fig : econo:2d ] we show the contour lines obtained . in ( a ) the analytic solution , and then the discrete solution for @xmath454 , @xmath455 and @xmath456 in , respectively , ( b ) , ( c ) and ( d ) , with the contours of the analytic solution shown in a lighter gray . the contours are @xmath457 ( the csdp solution is always positive due to the way it is solved ) . ( b ) , @xmath458 ( c ) , @xmath459 ( d ) . , title="fig : " ] + ( a ) ( b ) , @xmath458 ( c ) , @xmath459 ( d ) . , title="fig : " ] + ( b ) + ( b ) , @xmath458 ( c ) , @xmath459 ( d ) . , title="fig : " ] + ( c ) ( b ) , @xmath458 ( c ) , @xmath459 ( d ) . , title="fig : " ] + ( d ) we observe that the scheme introduces quite a bit of diffusion at lower resolutions , and , in particular , the heights at the point @xmath143 increase to the analytic solution at that point as @xmath48 decreases . however , the jump in the gradients is well captured , especially at higher resolutions , even though we are asking for conditions on the discrete hessians which are unbounded from above . in table [ tab : econo:2d ] we give some comparative results at different mesh sizes . in this table we indicate by @xmath460 the discrete functional evaluated at @xmath461 , the interpolant in @xmath71 of the exact solution , and the error column refers to @xmath462 ( @xmath37 error ) . it is interesting to notice that even though the solution is not smooth ( @xmath23 is only lipschitz ) the error in the @xmath37 norm is smaller or approximately equal to @xmath48 . in the last column of the table we show the quantity @xmath463 which is approximately @xmath464 for all the values of @xmath48 reported , showing that even under such low regularity assumptions on @xmath23 , the error in the functional behaves as @xmath465 . rlllcrc @xmath266 & @xmath81 & @xmath466 & @xmath460 & error ( @xmath37 ) & & @xmath467 + ' '' '' 8 & 0.125 & 0.5319 & 0.5444 & 0.0769 & 0.190s & 0.14 + 16 & 0.0625 & 0.5404 & 0.5478 & 0.0300 & 0.990s & 0.14 + 32 & 0.03125 & 0.5449 & 0.5488 & 0.0336 & 17.000s & 0.14 + 64 & 0.015625 & 0.5470 & 0.5491 & 0.0174 & 751.100s & 0.14 + @xmath468 & 0 & 0.5492 in three dimensions there is also a solution with partial derivatives which are either @xmath97 or @xmath450 , of the form @xmath469 this time the coefficients can not be expressed in a simple form , since they involve roots of polynomials of high degree , and we just give numerical approximations : @xmath470 so that @xmath471 , and the value of the functional is @xmath472 in figure [ fig : econo3d ] we show the regions where @xmath473 is @xmath474 or @xmath16 in , different shades of gray . we illustrate the regions with wire frames viewed from the positive octant in ( a ) , and exploded views of the solid regions from the positive octant in ( b ) and the negative octant in ( c ) . + ( a ) + ( b ) + ( c ) in table [ tab : econo:3d ] we give some comparative results at different mesh sizes . it is interesting to notice here that the @xmath37 error is not converging to zero with order @xmath465 . nevertheless , the quantity @xmath463 ( shown in the last column ) decreases slowly as @xmath48 goes to @xmath97 , meaning that the error @xmath475 in the functional behaves as @xmath465 in this range , exhibiting a similar behavior to that of the two dimensional case . rlcccrc @xmath266 & @xmath81 & @xmath466 & @xmath460 & error ( @xmath37 ) & & @xmath467 + ' '' '' 4 & 0.25 & 0.8195 & 0.8449 & 0.1356 & 0.21s & 0.20 + 8 & 0.125 & 0.8484 & 0.8605 & 0.1281 & 3.87s & 0.16 + 12 & 0.0833 & 0.8578 & 0.8647 & 0.1130 & 10.29s & 0.13 + 16 & 0.0625 & 0.8622 & 0.8661 & 0.1135 & 7 m 20s & 0.10 + 20 & 0.0500 & 0.8648 & 0.8671 & 0.1177 & 50 m 23s & 0.07 + @xmath468 & 0 & 0.8684 theoretically , semidefinite programs are polynomial time solvable . however , as the tables [ tab : econo:2d ] and [ tab : econo:3d ] show , in practice we can not go too far with the number of unknowns . a reasonable size for the two dimensional problems we considered is a mesh of about @xmath476 subdivisions . carlier , lachand - robert and maury @xcite gave several examples of @xmath132 and @xmath477 projections , and in this section we consider one of the functions they considered , namely , @xmath478 we show the graph of the original function in figure [ fig : fig2-f ] , and that of the resulting @xmath40 , @xmath28 , @xmath37 , @xmath132 and @xmath477 projections in figure [ fig : fig2 ] . as in the original article , these graphs are shown upside down . the interested reader may observe that our results are qualitatively different from those in @xcite . } $ ] , shown upside down . ] @xmath40 projection , shown upside down . ] , shown upside down . ] + @xmath28 projection , shown upside down . ] , shown upside down . ] + @xmath37 projection , shown upside down . ] , shown upside down . ] + @xmath132 projection , shown upside down . ] , shown upside down . ] + @xmath477 projection , shown upside down . ] , shown upside down . ] using a similar function , but with more symmetries , @xmath479 we show in figure [ fig : nonu ] that the @xmath37 projection need not be unique . the solution in ( b ) is the one obtained by the semidefinite programming code . projection . ( a ) original function , ( b ) and ( c ) two optimal solutions.,title="fig : " ] + ( a ) projection . ( a ) original function , ( b ) and ( c ) two optimal solutions.,title="fig : " ] + ( b ) projection . ( a ) original function , ( b ) and ( c ) two optimal solutions.,title="fig : " ] + ( c ) in all these examples , we used a mesh with @xmath480 nodes . the times ranged from about @xmath481 seconds for the @xmath37 projections to about @xmath482 seconds for the @xmath28 projection . the @xmath132 and @xmath477 projections took about the same time , near @xmath483 seconds . as mentioned in the introduction , many problems in science are modelled via convex functions , raising the question of how measured data ( usually non convex ) can be approximated by a convex function . often the fitness to data is done parametrically . for example , by assuming that the underlying function is a linear combination of some given polynomials , and minimizing over all possible parameters in a convenient norm . even in this case , approximating the data by a linear combination which is also convex but otherwise arbitrary may be challenging . in this section we show how our numerical scheme could be used for fitting data on a regular mesh by discrete functions having a positive semidefinite discrete hessian . though the resulting discrete functions may not be extended to a convex function of continuous variables , as shown in example [ example : hessnoconv ] , the underlying `` true '' convex function might be well captured by the discrete function . in the tests we show , we perturbed with random noise the values of the function @xmath484 on a regular mesh of @xmath480 nodes . the original function and the perturbed data are represented in figure [ fig : noise ] ( a ) and ( b ) , respectively . projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( a ) projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( b ) + projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( c ) projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( d ) + projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( e ) projection ( c ) , level curves of the @xmath28 projection ( d ) , level curves of the @xmath37 projection ( e ) , 3d graphs of the unperturbed function and @xmath37 projection ( f).,title="fig : " ] + ( f ) we chose to use a uniformly distributed noise between @xmath485 and @xmath53 for each node , where @xmath486 ( @xmath48 being the mesh size ) . in this way we simulate some intrinsic measurement noise whose distribution is presumably known , and that the mesh has been chosen finer than the measurement noise . needless to say , our choice of noise is quite arbitrary , and there are many other possibilities to choose from . in figure [ fig : noise ] ( c ) , ( d ) and ( e ) we show the resulting level curves for the @xmath40 , @xmath28 and @xmath37 discrete projections ( respectively ) , with the contours of the original unperturbed function in a lighter gray . the @xmath28 discrete projection may be considered as a variant of a least squares approximation , but the others are not seen as often . the times spent by the semidefinite code were similar to those of section [ sec : projections:2 ] , about @xmath487 , @xmath482 and @xmath481 seconds ( respectively ) . as is to be suspected , for the same mesh but smaller noise ( e.g. , taking @xmath488 ) , the results are better and the times smaller . comparing the level curves with those of the unperturbed function ( figure [ fig : noise ] ( c ) , ( d ) and ( e ) ) , it is apparent that the @xmath37 discrete projection seems to give the closest fit . notice that this projection is also faster than the @xmath40 or @xmath28 discrete projections by a factor of @xmath489 to @xmath490 . for the run shown , the maximum absolute value of the differences between the perturbed data and the original function on the mesh is about @xmath491 ( @xmath492 times the mesh size , as designed ) , whereas the maximum absolute value of the differences between the @xmath37 projection and the original function is @xmath493 , attained at @xmath65 . however , as can be seen in figure [ fig : noise ] ( e ) , the @xmath37 projection gives a good fit to the unperturbed function in the interior nodes , and is actually between @xmath494 ( somewhat smaller than the mesh size ) for nodes of the square @xmath495 , but deteriorates near the boundary ( as do the other projections ) . this is perhaps better appreciated in figure [ fig : noise ] ( f ) , where the graph of the @xmath37 discrete projection is compared to that of the unperturbed original function . the `` overshooting '' at the boundary is a typical and known phenomenon when fitting data by convex functions . see , for instance , the article by meyer @xcite , where the one dimensional case is discussed . * our gratitude to a. manelli for bringing the monopolist problem [ prob : monopolist ] to our attention , and for his many enlightening comments on the subject . * our thanks to the anonymous referees for their comments which made the presentation more clear , and for suggesting the applications in section [ sec : fit ] . nstor e. aguilera : : : consejo nacional de investigaciones cientficas y tcnicas and universidad nacional del litoral , argentina . + e - mail : [email protected] pedro morin ( corresponding author ) : : : consejo nacional de investigaciones cientficas y tcnicas and universidad nacional del litoral , argentina . + e - mail : [email protected]

spectral approximation by polynomials on the unit ball is studied in the frame of the sobolev spaces , . the main results give sharp estimates on the order of approximation by polynomials in the sobolev spaces and explicit construction of approximating polynomials . one major effort lies in understanding the structure of orthogonal polynomials with respect to an inner product of the sobolev space . as an application , a direct and efficient spectral - galerkin method based on our orthogonal polynomials is proposed for the second and the fourth order elliptic equations on the unit ball , its optimal error estimates are explicitly derived for both procedures in the sobolev spaces and , finally , numerical examples are presented to illustrate the theoretic results .

1310.2283

spectral methods have been used recently for solving partial differential equations on the unit disk , unit ball , or other domains with cylindrical or spherical geometry . their increasing popularity on these domains lies partially in various applications in earth sciences , disk or sphere shaped mirrors and lenses , fluid flow in a pipe or rotating cylinder , accretion disks in planetary astronomy , to name a few . in @xcite , poisson equation on an axisymmetric domain is transformed into a system of two - dimensional problems by the polar transformation , and the axisymmetric problems are then approximated by an appropriate spectral - galerkin method . fast spectral - galerkin methods for helmholtz equations on a disk or a cylinder are proposed in @xcite , using the polar transformation with essential pole conditions and the chebyshev or legendre polynomial bases in the radial direction . subsequently , these types of spectral galerkin methods have been extended to other domains with spherical geometries , including the 3-dimensional ball @xcite . meanwhile , mixed jacobi - fourier spectral method are presented for elliptic equations on a disk @xcite and a mixed jacobi - harmonic spectral approximation is proposed in @xcite for a navier - stokes equation in a ball . see @xcite for a comprehensive review of spectral methods and their special treatments in polar / spherical coordinates . moreover , an alternative approach for solving differential equations in a smooth domain is to map the domain into the unit ball and then apply a spectral method @xcite . however , apparent disadvantages and difficulties in such a type of spectral - galerkin method lie in : ( i ) . the polar transformation introduces variable coefficients in the reduced problem in polar / spherical coordinates , which increases the bandwidth of the discrete linear system ; ( ii ) . the enforcement of pole conditions induces artificial singularities at the pole either explicitly or implicitly , which brings about extra difficulties to both the design of efficient approximation schemes and the corresponding numerical analysis @xcite ; ( iii ) . as the numerical analysis is concerned , the approximation errors are measured and estimated in anisotropic , instead of genuine , sobolev norms . initiated by direct and efficient spectral method of atkinson and his collaborators that uses orthogonal polynomials to solve on linear elliptic equations on the disk with both homogeneous dirichlet and neumann boundary conditions @xcite . the second author of this paper studied sobolev orthogonal polynomials on a unit ball @xcite , which leads to a tri - diagonal linear system for second order elliptic equation with constant coefficients , as can be seen below . it this also give challenges to researchers on spectral approximations on the unit ball and its applications in direct and efficient spectral methods on the unit ball . one of the challenging problems in the spectral methods on the unit ball is to measure and estimate the errors of approximation in genuine , instead of anisotropic , sobolev norms . such estimates were established for the product domain in @xcite but has been lacking in most of the works on the unit disk or the unit ball . the problem of characterizing best approximation by the smoothness of functions is intensively studied in approximation theory . the two problems are closely related but not exactly the same as we shall explain below . the purpose of this paper is to conduct a comprehensive study for the spectral approximation on the unit ball @xmath3 of @xmath4 , making use of recent advances in both approximation theory , orthogonal polynomials , and spectral methods . spectral approximation for solving an elliptic equation on @xmath3 looks for approximate solutions that are polynomials written in terms of certain orthogonal basis on the ball and their coefficients are determined by the galerkin method . to understand the convergence of such an approximation process , it is necessary to study polynomial approximation in the sobolev space @xmath2 , where @xmath5 is a positive integer , that consists of functions whose derivatives up to @xmath5-th order are all in @xmath6 . in some literatures , the space @xmath2 is called @xmath7 . for @xmath8 , let @xmath9 denote its best polynomial approximation of degree at most @xmath10 . for spectral approximation , the desired estimate for @xmath11 is of the form @xmath12 where @xmath13 denotes the norm of @xmath14 defined by @xmath15 one of the main result of this paper is to establish this estimate and , more generally , establish its analogue in the space @xmath16 for @xmath17 . the difficulty of quantifying the error of polynomial approximation on the unit ball lies in the strong influence of the boundary of the ball on the approximation behavior . this is well documented for approximation on a closed interval on the real line . a complete characterization of best approximation on the unit ball is only carried out recently . in @xcite , two moduli of smoothness and their equivalent @xmath18-functionals were introduced and used to establish both direct and inverse theorems that characterize the behavior of best approximation on the unit ball . in @xcite , approximation in the sobolev space was studied and estimate for @xmath19 was established , more generally for @xmath20 ( ( * ? ? ? * corollary 5.4 ) ) , and the derivative estimates were established for angular derivatives , which however do not imply . what we can prove relatively effortless ( see theorem [ eq : approx - weight ] below ) is the following estimate @xmath21 where @xmath22 vanishes on the boundary sphere @xmath23 of @xmath24 and @xmath25 is the partial sum of the fourier orthogonal expansion in @xmath6 . this estimate , however , is weaker than because of the power of @xmath26 in its left hand side . it tuns out that what we need for proving is the orthogonal structure of the soblolev space @xmath2 , not the orthogonal structure of @xmath6 . an essential step in our study is to study orthogonal polynomials for with respect to the inner product @xmath27 of @xmath2 , which we call the sobolev orthogonal polynomials . initially motivated by direct and efficient spectral method of atkinson and his collaborators that uses orthogonal polynomials to solve linear elliptic equations on the disk @xcite , the sobolev orthogonal polynomials on the ball with respect to @xmath28 were studied in @xcite and those with respect to @xmath29 were studied in @xcite . in these works , sobolev orthogonal bases were constructed in terms of the orthogonal polynomials for @xmath30 with @xmath31 and @xmath32 , respectively , where the weight function @xmath33 , which are given explicitly in terms of spherical harmonics and the jacobi polynomials @xmath34 that are orthogonal polynomials with respect to @xmath35 on @xmath36 $ ] . for larger @xmath5 , however , the orthogonal structure is more complicated , and we need to extend the orthogonal basis for @xmath30 to allow @xmath37 to be negative integers , which in turn requires us to use extensions of the jacobi polynomials with negative indexes . this is prompted by the realization that the sobolev orthogonal polynomials for @xmath38 and @xmath39 in @xcite can be expressed in terms of orthogonal polynomials for @xmath30 with @xmath40 and @xmath41 , and , heuristically , the negative weight could cancel out the @xmath42 term in . the jacobi polynomials with negative indexes have been used in spectral approximation on other domains in @xcite . one of our main results is an explicitly constructed mutually orthogonal polynomial basis for @xmath43 , which could be used as the building blocks for the spectral - galerkin method . for @xmath44 , its @xmath10-th best polynomial approximation is given by the @xmath10-th partial sum of its fourier orthogonal expansion on the ball . for @xmath45 , we shall prove that the best approximating polynomials to @xmath46 are @xmath47 , the partial sums of the fourier orthogonal expansion in @xmath14 equipped with the inner product @xmath48 , which can be expressed explicitly in terms of the mutually orthogonal polynomials that we constructed . for the @xmath16 with @xmath49 , the best approximating polynomial is not explicitly known , but we are able to show that a near - best approximating polynomial , denoted by @xmath50 and defined via a smooth cut - off function @xmath51 , satisfies our sharp estimate in @xmath16 . both @xmath47 and @xmath52 are given by explicit formulas that can be easily computed numerically ( see section 4 ) . our main result on approximation in the sobolev space is the following : let @xmath53 . for any @xmath54 , @xmath55 , @xmath56 , there is a constant @xmath57 independent of @xmath46 and @xmath10 , such that @xmath58 where @xmath52 can be taken as @xmath59 for @xmath60 . more precise results of this nature are stated in section 4.1 below . to illustrate the application of this result in the spectral approximation , we will consider two examples , the helmholtz equation and the biharmonic equation on the unit ball , and demonstrate how our results on approximation in the sobolev space can be used to error estimates in the spectral - galerkin method . furthermore , we provide numerical examples for these equations for @xmath61 and @xmath62 , which further illustrate our findings . the paper is written with readers in both approximation theory community and spectral method community in mind . the problem of is originated and studied in the spectral method , which is closely tied to the problem of characterizing best approximation that has been a central theme and studied intensely in approximation theory . our approach uses a mixed bag of tools , developed in both approximation theory and spectral methods . it is our hope that this paper will stimulate further collaboration between the two communities . the paper is organized as follows . in the next section we present background materials , orthogonal polynomials on the unit ball , fourier orthogonal expansions , and recent results on approximation on the unit ball . the orthogonal structure of the sobolev space is developed in section 3 . the main results on approximation by polynomials in the sobolev space are stated and proved in section 4 . finally , in section 5 , we discuss applications of our main results in the spectral - galerkin methods and present our numerical examples . to keep the presentation fluent , we leave technical details of extending orthogonal bases to negative indexes and proving equivalence of norms in the sobolev space to appendix a and appendix b , respectively . for @xmath63 , we use the usual notation of @xmath64 and @xmath65 to denote the euclidean norm of @xmath66 and the dot product of @xmath67 . the unit ball and the unit sphere in @xmath68 are denoted , respectively , by @xmath69 throughout this paper , we let @xmath70 denote the @xmath71-th partial derivative operator , let @xmath72 be the gradient and let @xmath73 be the usual laplace operator . we denote by @xmath57 a constant that depends only on @xmath74 , @xmath5 or other fixed parameters , its value may change from line to line . we follow the notation in @xcite . let @xmath75 denote the space of homogeneous polynomials of degree @xmath10 in @xmath76 variables . it is known that @xmath77 harmonic polynomials of @xmath76-variables are polynomials in @xmath75 that satisfy the laplace equation @xmath78 . spherical harmonics are the restriction of harmonic polynomials on the unit sphere . let @xmath79 denote the space of spherical harmonic polynomials of degree @xmath10 . it is well known that @xmath80 if @xmath81 , then @xmath82 in spherical polar coordinates @xmath83 . we call @xmath84 a solid spherical harmonic . evidently , @xmath85 is uniquely determined by its restriction on the sphere . we shall also use @xmath86 to denote the space of solid spherical harmonics . the spherical harmonics of different degrees are orthogonal with respect to the inner product @xmath87 where @xmath88 is the surface measure and @xmath89 is the surface area ; the inner product is normalized so that @xmath90 . in spherical polar coordinates , the laplace operator can be written as @xmath91 where @xmath92 and @xmath93 , the spherical part of @xmath94 , is the laplace - beltrami operator that has spherical harmonics as eigenfunctions ; more precisely , for @xmath95 , @xmath96 let @xmath97 and let @xmath98 be an orthonormal basis of @xmath86 such that @xmath99 . the spherical harmonic expansion of @xmath46 is defined by @xmath100 we define the partial sum of the harmonic expansion and the projection operator @xmath101 by @xmath102 respectively . the projection operator is independent of the choice of orthonormal basis of @xmath86 . furthermore , since @xmath103 is homogeneous , we can extend its definition to the unit ball by @xmath104 for @xmath105 . we extend @xmath106 accordingly . if @xmath107 is a harmonic function on the unit ball , then @xmath108 is the best approximation to @xmath107 in @xmath6 . our basic reference in this section is @xcite . for @xmath109 , let @xmath110 be the weight function defined by @xmath111 the classical orthogonal polynomials on the unit ball are orthogonal with respect to the inner product @xmath112 where @xmath113 is the normalization constant such that @xmath114 . for clarity , we write @xmath115 . let @xmath116 denote the space of polynomials in @xmath76 real variables . for @xmath117 let @xmath118 denote the linear space of polynomials in @xmath76 variables of ( total ) degree at most @xmath10 . a polynomial @xmath119 is called orthogonal with respect to @xmath110 on the ball if @xmath120 for all @xmath121 . let @xmath122 denote the space of orthogonal polynomials of total degree @xmath10 with respect to @xmath110 . it is well known that @xmath123 the space of @xmath124 has many different bases . let @xmath125 denote a basis of @xmath122 , then @xmath126 if @xmath127 . the basis is called mutually orthogonal if @xmath128 whenever @xmath129 , and it is called orthonormal if @xmath130 in addition . let @xmath131 be the pochhammer symbol . we use the standard multi index notation that , for @xmath132 , @xmath133 one basis of @xmath122 is given in terms of the jacobi polynomials and spherical harmonics . let @xmath134 denote the usual jacobi orthogonal polynomial of degree @xmath135 with respect to weight function @xmath136 on @xmath36 $ ] . for @xmath137 and @xmath138 , let @xmath139 be an orthonormal basis for @xmath140 . define @xmath141 then the set @xmath142 is a mutually orthogonal basis of @xmath122 whenever @xmath143 . more precisely , @xmath144 where @xmath145 is given by @xmath146 this is a standard mutually orthogonal basis on the unit ball ; see @xcite . we include a constant in the definition of @xmath147 in order to extend this definition to the case of @xmath148 , which is explained in appendix [ gball ] . it is known that orthogonal polynomials with respect to @xmath110 are eigenfunctions of a second order differential operator @xmath149 . more precisely , we have @xmath150 where @xmath151.\ ] ] in term of the mutually orthogonal basis @xmath152 , the fourier orthogonal expansion of @xmath153 is defined by @xmath154 we define the partial sum of the orthogonal expansion and the projection operator @xmath155 by @xmath156 respectively . by definition , @xmath157 is the orthogonal projection of @xmath158 onto @xmath118 ; that is , @xmath159 if @xmath160 and @xmath161 for @xmath162 , let @xmath163 denote the @xmath164 norm @xmath165 and let @xmath166 be the uniform norm for @xmath167 . furthermore , for @xmath162 , let @xmath168 denote the @xmath169 norm @xmath170 and let @xmath171 be the uniform norm for @xmath172 . in the case of @xmath173 , we shall denote the norm by @xmath174 . in the remaining of this subsection , we write @xmath175 ( resp . @xmath176 ) for either @xmath177 or @xmath164 ( resp . @xmath168 or @xmath178 ) , and write @xmath179 and @xmath180 for either @xmath181 and @xmath106 defined in or @xmath182 and @xmath183 defined in . when the setting is on @xmath23 , @xmath184 . [ defn : bestapp ] let @xmath185 if @xmath162 and @xmath186 if @xmath187 . for @xmath188 , the error of the best approximation to @xmath46 by polynomials of degree at most @xmath10 is defined by @xmath189 with norm specified , we shall write @xmath190 , @xmath191 and @xmath192 . the standard hilbert space theory shows that @xmath180 is the best @xmath193 approximation to @xmath46 ; that is , @xmath194 for @xmath49 , we no longer know the polynomial of best approximation explicitly , but a near best approximation is known ( see , for example , @xcite ) . [ defn : lnf ] a @xmath195-function @xmath51 on @xmath196 is called an admissible cut - off function if @xmath197 for @xmath198 and @xmath199 for @xmath200 . if @xmath51 is such a function , define @xmath201 when @xmath202 is specified , we will write @xmath203 and @xmath204 accordingly . since @xmath51 is supported on @xmath205 , the summation in @xmath206 can be terminated at @xmath207 , so that @xmath206 is a polynomial of degree at most @xmath208 . it approximates @xmath46 as well as the best approximation polynomial of degree @xmath10 . [ thm : near - best ] let @xmath185 if @xmath162 and @xmath209 if @xmath187 . then 1 . @xmath210 and @xmath211 for @xmath160 . 2 . for @xmath212 , @xmath213 . 3 . for @xmath212 , there is a constant @xmath214 , independent of @xmath46 , such that @xmath215 the quantity @xmath192 and @xmath190 can be characterized by the smoothness of the function @xmath46 ; see @xcite and section 4 below . we recall result on the characterization of best approximation by polynomials in @xmath164 in terms of the smoothness of the functions . in approximation theory , smoothness of a function is usually measured by the modulus of smoothness and its equivalent @xmath18-functional . since we are primarily interested in functions in sobolev spaces , we shall state the result only in terms of @xmath18-functional . for @xmath216 and @xmath162 , we define the sobolev space @xmath217 to be the space of functions whose spherical / angular derivatives up to @xmath5-th order are all in @xmath164 . for @xmath218 , we replace @xmath175 space by the space @xmath219 of continuous functions on @xmath23 . the norm and semi - norm of @xmath217 can be defined by @xmath220 where @xmath221 are angular differential operators . in polar coordinates on the plane @xmath222 , @xmath223 , which explains their name ; see ( * ? ? ? * section 1.8 ) for further properties of these operators . [ def : k - funcs ] let @xmath224 if @xmath225 and @xmath226 if @xmath227 . for @xmath228 an @xmath229 , define the k - functional @xmath230 this definition and the characterization of best approximation below were established in @xcite , where an equivalent modulus of smoothness was also defined . [ thm : jacksons ] let @xmath231 and let @xmath232 if @xmath162 , and @xmath226 if @xmath227 . then @xmath233 and @xmath234 the estimate is usually called direct , or jackson , inequality , while is usually called inverse inequality . if @xmath235 , then we can choose @xmath236 in the infimum of @xmath18-functional , which gives the following corollary . [ cor : jacksons ] let @xmath237 and let @xmath235 if @xmath162 , and @xmath226 if @xmath227 . then @xmath238 we will also need an estimate in the fractional order sobolev space @xmath239 , where @xmath240 , which is defined as the interpolation space @xmath241 ; see appendix b. [ th : approxharm ] if @xmath242 for @xmath243 , @xmath244 and @xmath1 , then @xmath245 since @xmath246 maps @xmath86 to itself for @xmath247 ( * ? ? ? * lemma1.8.3 ) , it follows readily that @xmath248 . as a result , @xmath249 and @xmath250 . thus by and , @xmath251 for @xmath252 , which gives for @xmath253 . consequently , it follows that @xmath254 which implies that @xmath255 for any @xmath55 , where @xmath256 denotes the norm of the operator from @xmath257 . it then follows from that @xmath258 this completes the proof of . we recall result on best approximation by polynomials in @xmath259 . we define the sobolev space @xmath260 to be the space of functions whose derivatives up to the @xmath5-th order are all in @xmath261 . for @xmath218 , we replace @xmath175 space by the space @xmath262 of continuous functions on @xmath3 . the norm of @xmath263 is defined by @xmath264 when @xmath265 , we write @xmath266 . a @xmath18-functional on the unit ball ( and its equivalent modulus of smoothness ) is defined in @xcite and used to characterize the best approximation in @xmath267 . throughout this paper , we define @xmath268 [ defn : k - funcb2 ] let @xmath269 if @xmath162 and @xmath270 if @xmath187 . for @xmath271 and @xmath272 , define @xmath273 where @xmath274 both direct and inverse theorems were established for @xmath275 in ( * ? ? ? * theorem 6.6 ) . they are analogues of theorem [ thm : jacksons ] . we will only state the corollary that is an analogue of corollary [ cor : jacksons ] and only for @xmath173 , where we write @xmath276 . [ cor : jackson ] let @xmath237 and let @xmath277 if @xmath162 , and @xmath278 if @xmath227 . then @xmath279 it should be mentioned that @xcite contains another @xmath18-functional that differs from @xmath280 in its last term , which can also be used to estimate @xmath190 . several results on approximation in the sobolev spaces and lipschitz spaces were established in @xcite , which contains , for example , the estimates @xmath281 for the spectral approximation , however , we are more interested in the derivatives @xmath282 instead of the angular derivatives . one result in this direction can be derived with the help of the following lemma . [ lem : diff - sn - mu ] for @xmath283 and @xmath284 , @xmath285 by the definition of fourie orthogonal expansion , @xmath286 and @xmath287 . by lemma [ lm : diffv ] , it follows that @xmath289 for all @xmath290 . consequently , @xmath291 . since @xmath292 reproduces polynomials of degree at most @xmath293 , @xmath294 , which implies that @xmath295 this proves the first identity in . since @xmath296 , it follows that @xmath297 from which the second identity in follows immediately . if @xmath298 for @xmath162 , or @xmath299 for @xmath187 , then for @xmath300 , @xmath301 this follows immediately from lemma [ lem : diff - sn - mu ] , theorem [ thm : near - best ] and corollary [ cor : jackson ] . as explained in the introduction , the estimate with @xmath265 and @xmath302 is weaker than the desired estimate because of the factor @xmath303 that appears in its left hand side . in this section , we consider orthogonal structure in the sobolev space @xmath304 . let @xmath305 [ def : ipd - s ] for @xmath306 , we define a bilinear form on the space @xmath307 by @xmath308 where @xmath309 , are positive constants . it is easy to see that this defines an inner product for @xmath2 . we denote the space of orthogonal polynomials of degree @xmath10 with respect to this inner product by @xmath310 . the reason that we use the negative index to denote such an inner product will become clear momentarily . for our purpose , we need to extend the definition of orthogonal polynomials @xmath311 defined in so that @xmath37 can be negative . the extension is carried out in appendix [ gball ] . below we shall use @xmath312 for @xmath313 , and what we essentially need is the following lemma proved in appendix [ gball ] . [ deltap ] let @xmath109 , @xmath314 and @xmath315 . then for @xmath316 , @xmath317 furthermore , make the convention @xmath318 if @xmath319 or @xmath320 ; and define @xmath321 if @xmath322 , and @xmath323 otherwise . then for @xmath324 , @xmath325 where @xmath326 , in particular , @xmath327 if @xmath328 . for @xmath329 , the inner product becomes @xmath330 [ thm : opnabla ] a mutually orthogonal basis of @xmath331 is given by @xmath332 with @xmath333 in particular , the space @xmath331 can be decomposed as @xmath334 we first assume @xmath335 . by green s identity , , and , @xmath336 for @xmath337 , @xmath338 , so that @xmath339 which finally completes the proof . this theorem was first established in @xcite , where the polynomials @xmath340 for @xmath341 are written in the form @xmath342 which follows from . recall that polynomials in @xmath122 are eigenfunctions of a second order differential operator @xmath149 for @xmath283 . it turns out that polynomials in @xmath331 are eigenfunctions of @xmath343 , which explains our notation of @xmath344 . for @xmath345 , however , @xmath346 is closely related , but not exactly , the space of eigenfunctions of @xmath347 ; see the discussion in @xcite . in the case of @xmath348 , the inner product becomes @xmath349 [ thm : opdelta ] a mutually orthogonal basis of mutually orthogonal basis for @xmath346 is given by @xmath350 where where @xmath351 is an orthonormal basis of @xmath352 . in particular , the space @xmath346 satisfies a decomposition @xmath353 furthermore , @xmath354 the decomposition in the theorem is established in @xcite , from which the mutually orthogonal basis follows from results in @xcite . the polynomials @xmath355 are closely related to @xmath356 as can be seen by . indeed , @xmath357 except when @xmath358 , in which case @xmath359 and @xmath360 for @xmath361 , where @xmath362 is a constant that can be obtained by comparing the leading coefficients . the orthogonal structure of @xmath2 for @xmath363 is more complicated . as we shall see below , the analogue of no longer holds if @xmath364 . we start with a lemma . [ lem : ynj ] let @xmath365 and @xmath10 be nonnegative integers . for @xmath366 , @xmath367 \big\vert_{x=\xi } = 4^k ( -j)_k ( -k)_{j - k } \frac{(n+\frac{d}{2})_{k}}{(n+\frac d 2)_j } y(\xi).\ ] ] it suffices to prove for all @xmath368 , @xmath369 . by , , and , we obtain @xmath370 & = \delta^k [ ( 1-\|x\|^2)^j p^{j , n}_{0,\ell}(x ) ] = \frac{(-j)_j } { ( n+j+\tfrac{d}{2})_j } \delta^{k}p^{-j , n+2j}_{j,\ell}(x ) \\ & = 4^k ( -j)_j \frac{(n+2j+\tfrac{d}{2}-2k)_{2k}}{(n+j+\tfrac{d}{2})_j}p^{2k - j , n+2j-2k}_{j - k,\ell}(x),\end{aligned}\ ] ] which is equal to zero if @xmath371 and its restriction on @xmath23 is zero if @xmath372 by . it is easy to see that the right hand side of is also zero for @xmath135 in these ranges . in the remaining case @xmath373 , we use and to derive @xmath374 and simplify the constant by @xmath375 then is established . for @xmath376 , consider the system of linear equations @xmath377 the system has a unique solution , since the matrix of the system is tridiagonal with nonzero diagonal elements . in fact , @xmath378 , @xmath379 and the rest @xmath380 can be deduced recursively starting from @xmath381 . [ def : ynj ] for any @xmath382 , let @xmath383 , @xmath384 , be the unique solution of the linear system . if @xmath385 , define @xmath386 and , if @xmath387 , define @xmath388 then for @xmath389 , @xmath390 and @xmath391 , @xmath392 the second identity of follows directly from the definition of @xmath393 and the lemma [ lem : ynj ] . to prove the first identity , we use the spherical - polar coordinates @xmath394 and derive from and that @xmath395 = \ , & q(\rho^2 ) \rho^{n-2 } \delta_0 y^n_{\ell}(\xi ) + \big(\partial_\rho^2 + \tfrac{d-1}{\rho}\partial_\rho \big ) q(\rho^2 ) \rho^n y^n_{\ell}(\xi ) \\ = \ , & dq(\|x\|^2 ) y^n_\ell ( x ) , \notag\end{aligned}\ ] ] where @xmath396 is defined by @xmath397.\ ] ] if @xmath398 is a polynomial of degree @xmath135 , then @xmath396 is a polynomial of degree @xmath399 . in particular , this shows that @xmath400 . consequently , each @xmath401 is the solution of the following elliptic equation @xmath402 which admits a unique solution by the posedness and regularity theory of the elliptic equation ( * ? ? ? * thoerem 5.5.2 , pp.390 - 391 ) . on the other hand , it is easy to see that @xmath403 is also a solution of . by the uniqueness of the solution , we must have @xmath404 , which completes the proof . we did not find a closed - form formula for of @xmath383 . here are the first three @xmath405 : @xmath406 with the help of @xmath407 , we can now define a mutually orthogonal basis for @xmath310 . [ lm : defq ] for @xmath314 , @xmath315 , @xmath408 and @xmath409 , define @xmath410 where @xmath391 . then , for @xmath411 and @xmath409 , 1 . @xmath412 for @xmath413 ; 2 . if @xmath5 is even , then @xmath414 ; 3 . if @xmath5 is odd , then @xmath415 for @xmath416 and @xmath417 . for @xmath418 , it follows from that @xmath419 , which instantly gives item 2 and item 3 for @xmath420 . further , for @xmath421 , one derives from that @xmath422 owing to @xmath423 . hence , item 1 follows if @xmath420 . a combination of and implies that @xmath424 is a constant independent of @xmath425 . hence , @xmath426 is defined in such a way that @xmath427 for @xmath428 , as can be easily verified using . furthermore , by and , we obtain @xmath429 by and @xmath430 for odd @xmath5 , @xmath431 ; thus @xmath432 , which proves item 2 and item 3 for @xmath428 . finally , if @xmath433 , all three items follow directly from . [ thm : hjn - rho ] the polynomials in @xmath434 form a mutually orthogonal basis of @xmath435 . more precisely , @xmath436 for @xmath43 defined in , where @xmath437 from item 1 of lemma [ lm : defq ] and the orthonormality of @xmath438 , it follows immediately that @xmath439 and the left hand side is equal to zero if @xmath440 or @xmath441 . thus , we remain to consider @xmath442 . if @xmath5 is odd , we temporarily denote @xmath443 if @xmath444 and @xmath445 . from item 3 of lemma [ lm : defq ] , we obtain that @xmath446 which is equal to zero if @xmath447 or @xmath448 , whereas it can be seen from with @xmath449 that @xmath450 \delta_{n , n ' } \delta_{j , j'}\delta_{\ell,\ell'}\end{aligned}\ ] ] for @xmath451 . as a result , this completes the proof of for odd @xmath5 . if @xmath5 is even , we obtain from item 2 of lemma [ lm : defq ] that @xmath452 it is obvious that @xmath453 if @xmath454 or @xmath455 . for @xmath456 , it follows from that @xmath457 ^ 2 } { n+\frac{d}{2}-s } \delta_{n , n ' } \delta_{j , j'}\delta_{\ell,\ell ' } , \end{aligned}\ ] ] which proves for even @xmath5 . the proof is completed . just as the projection operator defined for @xmath283 , we define the orthogonal projection operator @xmath458 , where @xmath5 is a positive integer , by @xmath459 let @xmath460 and @xmath231 . for @xmath461 , @xmath462 form lemma [ lm : defq ] , theorem [ thm : hjn - rho ] and the orthonormality of @xmath463 , it follows that @xmath464 for @xmath465 . in the remaining case of @xmath5 is odd and @xmath466 , it follows from lemma [ lm : defq ] , theorem [ thm : hjn - rho ] and that @xmath467 y^{n-2k}_{\ell } ( \xi ) \\ = & \frac{1 } { d(n-2k)+\lambda_k } \sum_{\ell=1}^{a^d_{n-2k } } \big [ d \la \delta^k f , \partial_{\sn } y^{n-2k}_{\ell } \ra_{\sph } + \lambda_k \la \delta^k f , y^{n-2k}_{\ell } \ra_{\sph } \big ] y^{n-2k}_{\ell } ( \xi ) \\ = & \sum_{\ell=1}^{a^d_{n-2k } } \la \delta^k f , y^{n-2k}_{\ell } \ra_{\sph } y^{n-2k}_{\ell } ( \xi ) = \proj_{n-2k}^{\ch } \delta^kf(\xi),\end{aligned}\ ] ] where the third equality sign is derived using green s formula . this follows essentially from the item 1 of lemma [ lm : defq ] , which implies immediately that @xmath468 for all @xmath5 and @xmath469 so that holds . in the remaining case that @xmath5 is odd and @xmath466 , applying green s identity shows that @xmath470 , using the fact that @xmath471 . the same argument shows that @xmath472 , which proves that holds in this case as well . [ lem : proj - f=0 g ] assume that @xmath460 satisfies @xmath473 . then for @xmath474 , @xmath475 the fact of @xmath476 in @xmath46 implies immediately from that @xmath477 if @xmath478 , so that @xmath479 . by and , @xmath480 for @xmath411 . applying green s identity repeatedly and using , we then obtain @xmath481 which is zero if @xmath482 ; while for @xmath418 , it follows from , , and that @xmath483 which finally proves . this section contains our main results on approximation in the sobolev space on the ball and their proofs . to facilitate readers who are mainly interested in the results , we state our main theorems in the first subsection and give their proofs in subsequent subsections . let @xmath484 be an admissible cut - off function . with respect to the inner product @xmath43 of @xmath485 , we define @xmath486 by definition , @xmath487 and @xmath488 , and it is obvious that @xmath489 our first result is on approximation in the space of @xmath490 . [ 1stapprox ] let @xmath491 . if @xmath492 and @xmath493 , then @xmath494 where @xmath495 can be replaced by @xmath496 if @xmath60 . for @xmath497 , @xmath498 , the errors are not directly bounded by @xmath191 but we still have the order of convergence . [ nthapprox ] let @xmath499 and @xmath500 . if @xmath492 and @xmath493 , then , for @xmath501 , @xmath502 where @xmath503 can be replaced by @xmath504 if @xmath60 . we denote by @xmath505 the subspace of @xmath16 defined by @xmath506 where @xmath507 denote the normal derivative of @xmath23 . for @xmath306 and @xmath508 , we define a semi - norm of @xmath16 by @xmath509 if @xmath510 , then it can be shown ( see the end of appendix b ) that , for @xmath511 , @xmath512 [ nthapprox = wcirc ] let @xmath513 and @xmath514 . if @xmath515 with @xmath55 and @xmath56 , then , for @xmath501 , @xmath516 where @xmath503 can be replaced by @xmath504 if @xmath60 . it should be pointed out that if @xmath517 , then @xmath473 for some @xmath518 and , by lemma [ lem : proj - f=0 g ] , @xmath519 these results will be proved in the following subsections , where the following observation will be useful . for @xmath460 , it follows immediately from applying the hlder inequality on @xmath24 and @xmath23 that @xmath520 using lemma [ lem : equivnorm ] and the inequality , we obtain the following lemma . for @xmath460 and @xmath521 , with @xmath522 and @xmath56 , @xmath523 according to theorem [ thm : opnabla ] , if @xmath525 , then we can decompose @xmath46 into two parts @xmath526 then it is readily checked by lemma [ lem : proj - f=0 g ] that @xmath527 for @xmath528 , let @xmath529 be defined as in . the following lemma is essential for the proof of theorem [ 1stapprox ] . [ main - lemma ] for @xmath284 and @xmath530 , @xmath531 since @xmath532 , it suffices to prove @xmath533 for the first identity of . from the fourier expansion of @xmath46 , @xmath534 it follows that @xmath535 if @xmath341 , by integration by part and , we obtain that , for @xmath536 , @xmath537 if @xmath337 , then @xmath538 . since @xmath539 is a homogenous polynomial of degree @xmath540 , and @xmath541 , it follows that @xmath542 , so that @xmath543 as a result , holds , which proves the first identity of . since @xmath544 , the second identity of follows readily . _ proof of theorem [ 1stapprox]_. for @xmath284 , by , theorem [ thm : near - best ] and , we obtain @xmath545 which proves . furthermore , by and , @xmath546 putting these together then applying lemma [ equivnorm1 ] with @xmath547 and @xmath548 , lemma [ equivnorm1 ] with @xmath549 , and @xmath550 , we obtain @xmath551 \le c n^{-r+1 } \|f\|_{w_p^{r}}. \end{aligned}\ ] ] to prove , we use the aubin - nitsche duality argument . we define @xmath552 then @xmath553 , @xmath554 and @xmath555 with @xmath556 . consider the following auxiliary elliptic boundary value problem @xmath557it admits a unique solution @xmath558 such that @xmath559 ( ( * ? ? ? * theorem 5.5.2 , pp.390 - 391 ) ) . let @xmath560 denote the inner product with @xmath561 . the equivalent variational form reads @xmath562 since @xmath563 reproduces polynomials of degree @xmath10 , it follows that @xmath564 consequently , by with @xmath565 , the hlder inequality , and with @xmath566 , we obtain @xmath567 we then apply @xmath568 and again to obtain @xmath569 this completes the proof of . by definition , @xmath47 is a polynomial in @xmath118 and @xmath52 is a polynomial in @xmath570 . we need an analogue of theorem [ main - lemma ] . [ main - lemma3 ] for @xmath571 and @xmath501 , @xmath572 by the definition , we recall that @xmath573 we first assume @xmath574 . by item 2 of lemma [ lm : defq ] , it follows that @xmath575 which is equal to zero if @xmath576 . by item 1 of lemma [ lm : defq ] , @xmath577 holds for @xmath578 and @xmath579 , it then follows from , and that @xmath580 for any @xmath578 . since @xmath581 is a mutually orthogonal basis of @xmath582 , we conclude that @xmath583 which leads to for even @xmath5 . now we assume @xmath584 . the same argument of using lemma [ lm : defq ] , and shows that @xmath585 for @xmath586 and @xmath587 \\ = & \left . \frac { \la \delta^m f , p^{-1,n-2m}_{j - m,\ell}\ra_{-1 } } { h^{-1}_{j - m , n-2m}}\right|_{\lambda_0=\lambda_m } p^{-1,n-2m}_{j - m,\ell}(x ) , \quad j\ge m.\end{aligned}\ ] ] this implies @xmath588 by theorem [ thm : opnabla ] . thus , also holds for odd @xmath5 . _ proof of theorem [ nthapprox ] . _ if @xmath589 , it follows from theorem [ main - lemma3 ] , theorem [ thm : near - best ] and corollary [ cor : jackson ] that @xmath590 whereas , if @xmath584 , we obtain form theorem [ main - lemma3 ] , theorem [ 1stapprox ] , and corollary [ cor : jackson ] that @xmath591 moreover , for @xmath592 , we obtain , by and , @xmath593 putting these together and applying lemma [ lem : equivnorm ] and , we obtain @xmath594 \\ & \le c n^{-r+s } \|f\|_{w_p^r(\ball ) } , \notag \end{aligned}\ ] ] which establishes for the case @xmath595 . next we consider the estimate for @xmath596 . this requires the following formula : for @xmath597 @xmath598 where @xmath599 denote the derivative in the radius direction . for @xmath329 , the first term in the right hand side is taken to be zero and is the classical green s identity . for @xmath600 , we apply green s identity repeatedly . we need the following auxiliary partial differential equations with boundary values , @xmath601 let @xmath602 be defined as in with all @xmath603 . using with @xmath604 shows that @xmath605 for @xmath606 , where @xmath607 for @xmath1 . if @xmath608 , then with @xmath609 shows that @xmath610 , which implies that @xmath611 . this shows that the homogenous problem of has a unique solution @xmath612 . hence , by theorem 5.4.4/2 and theorem 5.5.1 in @xcite , @xmath613 is an isomorphic mapping from the space @xmath614 onto @xmath615 , where @xmath616 as in the proof of theorem [ 1stapprox ] , we use duality arguments and define @xmath617 by the isomorphism property of @xmath613 , with @xmath618 admits a unique solution that satisifes @xmath619 ( * ? ? ? * theorem 5.5.1 ) . since @xmath50 reproduces polynomials of degree @xmath10 , it follows that @xmath620 as a result , we derive from with @xmath618 , and with @xmath621 that @xmath622 we then apply @xmath623 and obtain , using again , that @xmath624 by the definition of @xmath625 , this establishes for @xmath626 . finally , the case @xmath627 of follows from lemma [ lem : interd ] . if @xmath628 or if @xmath46 is a radial function , then @xmath629 . as a result , by corollary [ cor : jackson ] , can be replaced by @xmath630 the proof of the first inequality of now follows exactly the proof of theorem [ nthapprox ] . the second inequality of follows from @xmath631 which completes the proof . to illustrate our results , we consider numerical solutions of two elliptic equations of the second and fourth order , respectively , on the unit ball , and we choose the spectral - galerkin method using orthogonal polynomials on the ball . we will carry out a convergence analyses of the approximation scheme in the hilbert space and present numerical examples that illustrate our theorems . we consider the non - homogenous boundary problem of the helmholtz equation on the unit ball , @xmath632 where the constant @xmath633 and @xmath634 . let @xmath635 + \lambda \la u , v\ra_{\ball } = \la u , v\ra_{-1}+ \lambda \la u , v\ra_{\ball}\ ] ] in the variational formulation , solving is equivalent to find @xmath636 such that @xmath637 which , by the lax - milgram lemma @xcite , admits a unique solution that satisfies @xmath638 the spectral - galerkin approximation to amounts to find @xmath639 such that @xmath640 which has a unique solution that satisfies with @xmath641 in place of @xmath642 . by theorem [ thm : opnabla ] , the orthogonal expansions of @xmath643 can be written as @xmath644 substituting this expression into and setting @xmath645 , we obtain a linear system of equations on @xmath646 . the matrix @xmath647 $ ] contains two parts . the first part , called stiff matrix , has been computed in . to evaluate the second part , called the mass matrix , we use the definition of @xmath648 given in . by and , @xmath649 \big(\frac{t-1}{2}\big)^k \\ & \overset{\eqref{jacobip1}}= ( j+\b+1)_j { \widehat}{p}^{(0,\b)}_j(x ) - ( j+\b)_{j-1 } { \widehat}{p}^{(0,\b)}_{j-1}(x),\end{aligned}\ ] ] which implies it is not difficult to check that @xmath650 hence , using , we obtain that @xmath651 thus , the stiff matrix is diagonal and the mass matrix is tridiagonal when the coefficients are arranged appropriately . the convergence of this approximation scheme is given in the following theorem . let @xmath642 and @xmath641 be the solutions of and , respectively . if @xmath652 with @xmath653 , then @xmath654 recall that @xmath655 denote the @xmath10th partial sum of orthogonal expansion with respect to @xmath28 . by , and , @xmath656 taking @xmath657 , we obtain @xmath658 which implies that @xmath659 thus , by lemma [ equivnorm1 ] and , for @xmath313 , @xmath660 furthermore , by a standard aubin - nitsche argument , we also have @xmath661 where we omit the details . together , the last two displayed inequalities complete the proof . we consider the following fourth order elliptic equation on the unit ball , @xmath662 where the constants @xmath663 and , for simplicity , we consider homogeneous boundary . in the variational formulation , solving is equivalent to find @xmath664 such that @xmath665 let the approximation space be @xmath666 . the spectral galerkin approximation scheme for amounts to find @xmath667 such that @xmath668 which has a unique and stable solution by the lax - milgram lemma @xcite . the convergence of this approximation scheme is given in the following theorem . [ convbi ] let @xmath642 and @xmath641 be the solutions of and , respectively . if @xmath652 with @xmath345 , then @xmath669 from and , it follows that @xmath670 . since @xmath671 , by , we obtain @xmath672 taking @xmath673 in , the above inequality shows that @xmath674 which leads to , by , @xmath675 consequently , the estimate for @xmath676 of follows from in theorem [ nthapprox = wcirc ] . a standard dual argument can then be used to derive the error estimate in the case of @xmath626 an @xmath677 of , we omit details . we consider examples for the equations and . [ exam1 ] for the helmholtz equation , we give numerical results in two and three dimensions with @xmath678 in the following settings : * @xmath61 , @xmath679 and @xmath680 such that @xmath681 ; * @xmath62 , @xmath682 and @xmath683 such that @xmath684 . let @xmath685 be the zeros and the corresponding christoffel numbers of the jacobi polynomials @xmath686 . set @xmath687 , @xmath688 for @xmath689 and @xmath690 for @xmath691 . we report the discrete maximum error @xmath692 and the discrete @xmath193error @xmath693 , defined by @xmath694 with the measuring points @xmath695 in two dimensions and by @xmath696 with @xmath697 in three dimensions . theoretically , the spectral - galerkin approximation with any @xmath698 recover the exact solution of example [ exam1 ] ( a ) . figure [ fig1 ] shows the maximum and the @xmath193errors between the exact solution and the approximation solution of . it is easy to see from figure [ fig1 ] ( a ) that all the errors plotted are close to the machine precision , and an exponential oder of convergence is found in figure [ fig1 ] ( b ) . these conclusions match our theoretical results . - ( solid ) and maximum ( dashed ) errors of example [ exam1 ] . left : @xmath61 , @xmath699 ; right : @xmath62 , @xmath700 . , title="fig:",scaledwidth=45.0%]- ( solid ) and maximum ( dashed ) errors of example [ exam1 ] . left : @xmath61 , @xmath699 ; right : @xmath62 , @xmath700 . , title="fig:",scaledwidth=45.0% ] ( a)(b ) @xmath701 [ exam2 ] for the biharmonic equation , we consider an example with @xmath702 , @xmath703 and the exact solution @xmath704 . the function @xmath46 in the right hand side is determined by . figure [ fig2 ] shows the maximum and the @xmath193- errors of the approximation scheme . an exponential order of convergence is observed in this plot , which is in agreement with theorem [ convbi ] . - ( solid ) and maximum ( dash ) errors of example [ exam2].,scaledwidth=49.0% ] for @xmath705 , the jacobi polynomials are defined by @xmath706 which are orthogonal to each other with respect to the weight function @xmath707 on @xmath36 $ ] , @xmath708 as it is shown in @xcite , writing the @xmath709 in the following explicit form @xmath710 extends the definition of @xmath711 for all negative values of @xmath712 and/or @xmath713 in literature . however , if @xmath714 , then a reduction of the degree of @xmath715 occurs . to avoid the degree reduction , we define the generalized jacobi polynomials by @xmath716 where @xmath717 if @xmath714 and @xmath718 otherwise . we also define that @xmath719 whenever @xmath135 is a negative integer . by the definition , it is evident that @xmath720 the lemma below contains several properties of these polynomials , which are well known properties of the jacobi polynomials if @xmath323 . for @xmath721 , @xmath722 by comparison of the corresponding powers of @xmath723 , both and follow from , and is an immediate consequences of . we now extend the definition of the orthogonal polynomials on the unit ball to negative @xmath37 , let @xmath109 . for @xmath137 and @xmath138 , let @xmath139 be an orthonormal basis for @xmath140 . define @xmath724 we now prove lemma [ deltap ] , which we restate below . . _ let @xmath314 and @xmath725 . then for @xmath316 , @xmath726 where @xmath727 , @xmath728 and @xmath318 if @xmath319 or @xmath320 . in particular , @xmath729 if @xmath730 . _ the identity is an immediate consequence of and . let @xmath398 be a polynomial of degree @xmath135 . recall that , by , @xmath731= dq(\|x\|^2 ) y^n_\ell ( x)\ ] ] with @xmath732 $ ] . let @xmath733 , @xmath734 and @xmath735 . it follows from and that @xmath736 { \widehat}{p}_j^{(-s,\b)}(t ) \\ & = 4\sum_{k = j_0}^j \frac { ( k - s+1)_{j - k } k } { ( j - k)!k ! ( j - s+\b+k+1)_{j - k } } \big [ ( \b + k)\frac{t-1}{2 } + k-1 \big ] \big(\frac{t-1}{2}\big)^{k-2 } \\ & = 4(j+\b ) \sum_{k = j_0}^j \frac { ( k - s+2)_{j - k } } { ( j - k)!(k-1 ) ! ( j - s+\b+k+1)_{j - k } } \big(\frac{t-1}{2}\big)^{k-1 } \\ & \quad + \frac{4 ( j_0-s+1)_{j - j_0 } } { ( j - j_0)!(j_0 - 2 ) ! ( j - s+\b+j_0 + 1)_{j - j_0 } } \big(\frac{t-1}{2}\big)^{j_0 - 2 } \\ & = 4(j+\b ) { \widehat}{p}_{j-1}^{(-s+2,\b)}(t ) + p_{j_0 - 2}(t),\end{aligned}\ ] ] where @xmath737 , and the last equal sign is derived using the fact that @xmath738 if @xmath739 . putting above computations together , it follows that @xmath740 \\ & = 4(n - j+\tfrac{d}{2}-1)_{j+1 } { \widehat}{p}^{(-s+2,n-2j+\tfrac{d}{2}-1)}_{j-1}(2\|x\|^2 - 1)y^{n-2j}_{\ell}(x ) + q_{j_0 - 2}(x)\\ & = 4(n+\tfrac{d}{2}-1)(n+\tfrac{d}{2}-2)p^{-s+2,n-2}_{j-1,\ell}(x)+ q_{j_0 - 2}(x),\end{aligned}\ ] ] where @xmath741 . using this identity recursively , we derive @xmath742 where @xmath743 for certain @xmath744 . specifically , @xmath745 if @xmath730 , in return , @xmath746 and @xmath747 . this completes the proof . monic orthogonal polynomials in @xmath748 are defined by ( cf . @xcite ) @xmath749 since @xmath750 if @xmath751 , @xmath752 is a polynomial of degree @xmath753 ; in fact , @xmath754 . moreover , @xmath755 is a basis of @xmath748 . moreover , @xmath756 is well defined if and only if @xmath757 for all @xmath758 . if there exists @xmath759 such that @xmath760 and @xmath761 , we use a truncated series for @xmath762 , @xmath763 which removes the lower order terms in @xmath764 . [ lm : diffv]for @xmath765 and @xmath109 with @xmath766 , @xmath767 taking derivative @xmath70 on or , and using the fact that @xmath768 , it is easy to see that @xmath769 which , when used recursively , leads to . in this appendix we discuss equivalent norms of the sobolev space @xmath16 . several results that we shall need hold for fairly general domain @xmath770 in @xmath4 . we state only their simplified version for @xmath771 and/or @xmath772 . in this appendix , we adopt the convention that @xmath773 means @xmath774 for some constants @xmath775 . for @xmath306 and @xmath508 , we define a semi - norm of @xmath16 by @xmath776 [ lem : interd ] let @xmath777 , let @xmath778 , and let @xmath135 and @xmath5 be integers with @xmath779 . there exists a constant @xmath780 such that for every @xmath781 , @xmath782 as a consequence of this lemma , it follows that @xmath783 where the last equivalence signs are derived from ( * ? ? ? * thoerem 4.2.4 , p.316 ) . we need another lemma on equivalent norms in @xmath784 . * theorem 1.1.16 ) [ equivnorm1 ] let @xmath313 and let @xmath785 be a continuous seminorm in @xmath0 such that @xmath786 for any nonzero polynomial @xmath787 . then @xmath788 a combination of with lemma [ equivnorm1 ] leads to @xmath789 \\ \sim & \sum_{|\a|=s } \|\partial^{\a } f\|_{w^m_p(\ball ) } + \cf(f ) . \end{split } \end{aligned}\ ] ] recall that the angular differential operators are defined by @xmath790 . we define a semi - norms of @xmath217 with @xmath231 and @xmath508 by @xmath791 where @xmath792 for any @xmath793 analogous to , @xmath794 has the following equivalent norms @xmath795 we need the fractional order sobolev space on @xmath23 , which is defined via the interpolation space . for @xmath796 with @xmath797 , we define the @xmath18-functional @xmath798 for @xmath799 and @xmath800 , the fractional order sobolev space @xmath801 is defined as the interpolation space @xmath241 via the @xmath18-functional @xcite , @xmath802 where the norm is defined by @xmath803^p \frac{dt}{t}\right)^{1/p}.\end{aligned}\ ] ] it follows that @xmath804 and , furthermore @xcite , @xmath805 it is worth to point out that the interpolator @xmath806 is of type @xmath807 . let @xmath808 , @xmath809 for @xmath810 and assume @xmath811 , @xmath812 . it follows then that @xmath813 and @xmath814 recall that @xmath815 and @xmath816 . we define @xmath817 and define @xmath818 using @xmath819 accordingly . the following lemma is needed in the proof of main theorems . [ lem : equivnorm ] for @xmath54 and @xmath820 , @xmath821 by thoerem 5.5.2 in @xcite , @xmath822 is an isomorphic mapping from @xmath823 onto @xmath824 , which means that @xmath825 it then follows from recursive reduction that , for any @xmath826 , @xmath827 which proves for @xmath828 . furthermore , assuming @xmath829 and taking @xmath830 and @xmath547 in lemma [ equivnorm1 ] , we can then deduce that @xmath831 which proves for @xmath829 . the proof is completed . in particular , for @xmath510 , together with the inequality @xmath832 for @xmath833 imply the inequality . assume @xmath834 . the multi - laplace operator @xmath835 is properly elliptic in the sense that @xmath836 and for any couple of linearly independent vectors @xmath837 the polynomial @xmath838 in the complex variable @xmath839 has exactly @xmath5 roots @xmath840 , with positive imaginary part . one set @xmath841 . let @xmath842 be @xmath5 distinct nonnegative integers . the system @xmath843 , @xmath844 is said to be a normal system if for each ( normal ) vector @xmath425 on @xmath23 , it holds @xmath845 further , @xmath843 is said to be a complemented system with respect to @xmath846 if for each @xmath847 and each tangential vector @xmath848 in the point @xmath425 with respect to @xmath23 , the identity @xmath849 holds for a certain polynomial @xmath850 if and only if the numbers @xmath851 , where @xmath852 s are polynomials in the variable @xmath839 , @xmath853 j. p. boyd and f. yu , comparing seven spectral methods for interpolation and for solving the poisson equation in a disk : zernike polynomials , logan shepp ridge polynomials , chebyshev fourier series , cylindrical robert functions , bessel fourier expansions , square - to - disk conformal mapping and radial basis functions , _ journal of computational physics _ , * 230 * , ( 2011),14081438 p. d. lax and a. n. milgram , parabolic equations . _ contributions to the theory of partial differential equations _ , pp . 167190 , annals of mathematics studies , no . princeton university press , princeton , n. j. , 1954 .

the statistics of strongly lensed arcs in samples of galaxy clusters provide information on cluster structure that is complementary to that from individual clusters . however , samples of clusters that have been analyzed to date have been either small , heterogeneous , or observed with limited angular resolution . we measure the lensed - arc statistics of 97 clusters imaged at high angular resolution with the hubble space telescope , identifying lensed arcs using two automated arc detection algorithms . the sample includes similar numbers of x - ray selected ( macs ) and optically selected ( rcs ) clusters , and spans cluster redshifts in the range . we compile a catalogue of arcs in the x - ray selected subsample and arcs in the optical subsample . all but five of these arcs are reported here for the first time . at , the x - ray selected clusters have a significantly higher mean frequency of arcs , per cluster , versus in the optical sample . the strikingly different lensing efficiencies indicate that x - ray clusters trace much larger mass concentrations , despite the similar optical luminosities of the x - ray and optical clusters . the mass difference is supported also by the lower space density of the x - ray clusters , and by the small einstein radii of the few arcs in the optical sample . higher - order effects , such as differences in concentration or substructure , may also contribute . [ firstpage ] gravitational lensing galaxies : clusters : general

1004.2048

galaxy clusters are natural laboratories for studying a variety of astrophysical processes and for testing cosmological models . in particular , the masses and mass profiles of clusters have proved to be useful for constraining cosmological parameters ( e.g. bridle et al . 1999 ; reiprich & b " ohringer 2002 ; voit 2005 ; allen et al . 2008 ; vikhlinin et al . gravitational lensing is frequently used to map the evolution of cluster mass profiles , ellipticities , and substructure . one approach is to perform detailed modeling of individual clusters using strong and weak lensing ( e.g. , abdelsalam et al . 1998 ; broadhurst et al . 2005 ; leonard et al . 2007 ; limousin et al . 2007 ; richard et al . 2007 ) . however , since this kind of approach requires deep data for individual clusters that exhibit numerous lensed images , the results may not be representative of the vast majority of clusters . a complementary approach is to measure the statistics of lensed arcs in large samples of clusters . lensing statistics thus provide another means to study clusters as a population . for the past decade there has been debate concerning theoretical lensing statistics predictions and their confrontation with observations . bartelmann et al . ( 1998 ; b98 ) performed lensing simulations using artificial sources at redshift @xmath6 by ray tracing through the five most massive clusters formed in a cosmological n - body dark matter simulation ( kauffmann et al . the observed number of giant arcs , with length - to - width ratio @xmath7 and @xmath8 mag , present over the whole sky was estimated by extrapolating from observations of a subsample of x - ray selected clusters from the einstein extended medium sensitivity survey ( emss ) , and compared to the theoretical calculation . b98 found that the estimated number of observed arcs is larger by almost an order of magnitude than the number predicted by the now - standard @xmath9 model . later estimates of lensed arcs statistics in clusters from both the las campanas distant cluster survey ( zaritsky & gonzales 2003 ; arcs with @xmath7 and @xmath8 mag ) and the red - sequence cluster survey ( rcs ; gladders et al . 2003 ) confirmed the estimates of the observed number of arcs derived by b98 . most recently , hennawi et al . ( 2008 ) analyzed a sample of 240 clusters , optically selected from the sloan digital sky survey ( sdss ) , and found that @xmath10 of them are strong lenses , similar to the findings of gladders et al . the largest catalogue of arcs to date was compiled by sand et al . ( 2005 ) who found @xmath11 arcs in @xmath12 clusters . however , their systematic search for arcs was performed on a largely heterogeneous cluster sample . the apparent overproduction of arcs by real clusters has stimulated further theoretical studies of arc statistics . meneghetti et al . ( 2000 ) studied numerically the effect of the masses of the individual cluster galaxies on a cluster s lensing cross section , and found it to be negligible , as also found in a study by flores , maller , & primack ( 2000 ) . however , the increase in lensing cross section due to the central cluster cd galaxy may be as high as @xmath13 ( meneghetti , bartelmann , & moscardini 2003 ) and the increase in cross section due to the intra - cluster gas could perhaps be by a factor of a few ( puchwein et al . 2005 , rozo et al . oguri , lee , & suto ( 2003 ) argued that halo triaxiality could also play an important role in increasing cluster lensing cross sections . torri et al . ( 2004 ) raised the possibility that x - ray selection of clusters may favor merging systems , which may be more efficient lenses . wambsganss , bode , & ostriker ( 2004 ) pointed out that since lensing cross section is a steep function of source redshift , the conflict between theory and observations could be the result of the assumed source redshifts in the simulations . similarly , dalal , holder , & hennawi ( 2004 ) performed a lensing simulation using artificial background sources at different redshifts and a large sample of simulated clusters . they found that their prediction for the number of lensed arcs was consistent with an observed number that they derived from a sample of x - ray selected emss clusters . the difference between this result and that of b98 was explained by the combination of three main effects : the inclusion of sources at different redshifts ; the use of a higher source density in the dalal et al . simulation ; and an observed cluster number density lower than the one used by b98 for estimating the all - sky number of arcs . a more observationally oriented approach to lensing statistics simulations was introduced by horesh et al . ( 2005 ; h05 ) in order to test specifically the lensing efficiency of individual clusters , independent of the separate question of the number density of clusters . h05 repeated the b98 simulations using the same simulated clusters , but using background sources from the hubble deep field ( hdf ) , each at a redshift based on its actual photometric redshift . observational effects including background , photon noise , and the light of cluster galaxies were added to the simulated lensed images . a mass - matched sample of @xmath14 x - ray - selected clusters ( smith et al . 2005 ) observed at high angular resolution with the hubble space telescope ( hst ) was used for comparison with the simulated sample . finally , an automated objective arc - detection algorithm was applied to both the observed and the simulated samples . this procedure permitted measuring and comparing the frequency of arcs over a larger range in magnitudes ( down to @xmath15 mag ) . h05 found that the lensing efficiency of their simulated clusters at @xmath16 was consistent , to within poisson errors , with that of their observed sample . while the analysis suggested that the observed clusters could be somewhat more efficient lenses by up to a factor of two , this conclusion was limited by the small size of both the observed and the simulated samples , as well as the parameters assumed in the simulations . indeed , an important parameter that affects all theoretical studies of arc statistics is @xmath17 , the overdensity within an @xmath18 radius comoving sphere . past simulations have used diverse values : 0.9 ( b98 ; dalal et al . 2004 ; h05 ) or 0.95 ( wambsganss et al . 2004 ; hennawi et al . fedeli et al . ( 2008 ) have recently analyzed the effect of @xmath17 on the arc statistics question , and pointed out that the most recent values of @xmath17 from wmap5 ( @xmath19 ; dunkley et al . 2009 ) revive and reinforce the discrepancy between theory and observations of arc statistics . a possibly related debate has emerged recently on the subject of the size of the einstein radius in clusters . broadhurst and barkana ( 2008 ) calculated the distribution of einstein radii in clusters with a spherical navarro , frenk , & white ( nfw ; 1996 ) profile , and with a concentration distribution according to neto et al . they compared their prediction with the observed einstein radii of three clusters , among them abell 1689 , and found that the observed radii are significantly larger than the theoretical expectation . yet another cluster with a large einstein radius was recently reported by zitrin et al . sadeh & rephaeli ( 2008 ) have calculated the concentration distribution of clusters based on the distribution of cluster formation times . they too find a discrepancy , albeit weak , between the observed einstein radius of abell 1689 and its expected value . oguri & blandford ( 2009 ) , however , argue that the einstein radius they obtain using a generalized triaxial form of the nfw profile ( jing & suto 2002 ) is consistent with that observed in abell 1689 . in addition , they provide a prediction for the distribution of einstein radii , which can be tested with a large statistical cluster sample . clearly , resolution of these problems requires , on the theoretical side , improved simulations , incorporating the most realistic cosmological parameters , source parameters , and observational effects ; and from the observational perspective , large , well - understood samples of clusters at various redshifts , selected by diverse methods and uniformly observed at the high depth and resolution needed for the clear detection of large arcs . in this paper , we address this observational perspective . we explore the observed statistical properties of 97 galaxy clusters imaged with hst . this cluster sample is large enough to be separated into several subsamples based on redshift and selection type . we apply two different arc detection algorithms to the clusters , and compile a high - resolution arc catalogue . we then study the arc statistics in the various subsamples . in a forthcoming publication , we will compare the observed statistics of this sample to new , improved , calculations of matched simulated samples . throughout this paper we adopt a @xmath20 cosmology with parameters @xmath21 , @xmath22 , and @xmath23 . magnitudes are in the vega system . we have compiled from the hst archive a sample of clusters observed with the advanced camera for surveys ( acs ) . the acs has a field of view @xmath24 , a pixel scale of @xmath25 , and a point - spread function full width at half maximum of @xmath26 . among the clusters in our sample , @xmath27 are from the massive cluster survey ( macs ; ebeling , edge , & henry , 2001 ) , and @xmath28 are from the red - sequence cluster survey ( rcs ; gladders & yee , 2005 ) . to these we add the @xmath14 clusters of smith et al . ( 2005 ) , observed with wfpc2 , and already analyzed in h05 , for a total of @xmath29 clusters . each of the three wfpc2 wf ccds had a fov of @xmath30 and a pixel scale of @xmath31 . we begin with a brief summary of the relevant details of each of these surveys . lccrc cluster & ra & dec & @xmath32 ( ref ) & bcg + & & & & @xmath33 + macsj0035.4@xmath342015 & 00:35:26.2 & @xmath35 & ... ( 1 ) & 19.55 + macsj0916.1@xmath340023 & 09:16:11.5 & @xmath36 & ... ( 1 ) & - + macsj0140.0@xmath340555 & 01:40:00.9 & @xmath37 & ... ( 1 ) & 20.25 + macsj0140.0@xmath343410 & 01:40:05.6 & @xmath38 & ... ( 1 ) & 19.91 + macsj0152.5@xmath342852 & 01:52:34.4 & @xmath39 & ... ( 1 ) & 20.32 + macsj0451.9@xmath400006 & 04:51:54.7 & @xmath41 & 0.430 ( 2 ) & 20.16 + macsj0520.7@xmath341328 & 05:20:42.0 & @xmath42 & ... ( 1 ) & 19.29 + macsj0712.3@xmath405931 & 07:12:20.4 & @xmath43 & 0.328 ( 2 ) & 19.00 + macsj0845.4@xmath400327 & 08:45:27.8 & @xmath44 & ... ( 1 ) & 19.26 + macsj0947.2@xmath407623 & 09:47:13.2 & @xmath45 & 0.345 ( 3 ) & 18.89 + macsj0949.8@xmath401708 & 09:49:51.8 & @xmath46 & ... ( 1 ) & 19.81 + macsj1006.9@xmath403200 & 10:06:54.7 & @xmath47 & ... ( 1 ) & 19.36 + macsj1115.2@xmath405320 & 11:15:14.8 & @xmath48 & ... ( 1 ) & 19.68 + macsj1115.8@xmath400129 & 11:15:51.9 & @xmath49 & 0.355 ( 3 ) & 19.52 + macsj1133.2@xmath405008 & 11:33:13.3 & @xmath50 & 0.389 ( 4 ) & 19.64 + macsj1206.2@xmath340847 & 12:06:12.2 & @xmath51 & 0.440 ( 5 ) & 19.92 + macsj1236.9@xmath406311 & 12:36:58.8 & @xmath52 & 0.302 ( 4 ) & 18.91 + macsj1258.0@xmath404702 & 12:58:02.1 & @xmath53 & ... ( 1 ) & 19.64 + macsj1319.9@xmath407003 & 13:20:08.5 & @xmath54 & ... ( 1 ) & 19.07 + macsj1354.6@xmath407715 & 13:54:30.6 & @xmath55 & 0.396 ( 4 ) & - + macsj1652.3@xmath405534 & 16:52:18.8 & @xmath56 & ... ( 1 ) & 19.27 + macsj2135.2@xmath340102 & 21:35:12.1 & @xmath57 & 0.33 ( 6 ) & 19.24 + macsj2243.3@xmath340935 & 22:43:20.2 & @xmath58 & ... ( 1 ) & - + notes - last column gives the magnitude of the brightest cluster galaxy . redshift references : ( 1 ) ebeling et al . , in preparation ; ( 2 ) stott et al . ( 2007 ) ; ( 3 ) allen et al . ( 2008 ) ; ( 4 ) edge et al . ( 2003 ) ; ( 5 ) balestra et al . ( 2007 ) ; ( 6 ) smail et al . ( 2007 ) . lcccccc cluster & ra & dec & @xmath32 & @xmath59 & @xmath60 & bcg + & ( j2000 ) & ( j2000 ) & & [ @xmath61 & [ @xmath62 & @xmath63 + macsj0018.5@xmath401626 & 00:18:33.8 & @xmath64 & 0.546 & 19.6 & 3.3 & 19.6 + macsj0025.4@xmath341222 & 00:25:29.4 & @xmath65 & 0.584 & 8.8 & 1.8 & - + macsj0257.1@xmath342325 & 02:57:08.8 & @xmath66 & 0.505 & 13.7 & 2.5 & 18.3 + macsj0454.1@xmath340300 & 04:54:11.1 & @xmath67 & 0.538 & 16.8 & 2.9 & 18.9 + macsj0647.7@xmath407015 & 06:47:50.1 & @xmath68 & 0.591 & 15.9 & 2.8 & 18.9 + macsj0717.5@xmath403745 & 07:17:32.9 & @xmath69 & 0.546 & 24.6 & 3.9 & - + macsj0744.8@xmath403927 & 07:44:52.8 & @xmath70 & 0.698 & 22.9 & 3.7 & 19.1 + macsj0911.2@xmath401746 & 09:11:11.2 & @xmath71 & 0.505 & 7.8 & 1.6 & 18.8 + macsj1149.5@xmath402223 & 11:49:35.5 & @xmath72 & 0.544 & 17.6 & 3.0 & 18.9 + macsj1423.8@xmath402404 & 14:23:48.6 & @xmath73 & 0.543 & 16.5 & 2.9 & 18.1 + macsj2129.4@xmath340741 & 21:29:26.3 & @xmath74 & 0.589 & 15.7 & 2.6 & 19.5 + macsj2214.9@xmath341359 & 22:14:57.3 & @xmath75 & 0.503 & 14.1 & 2.5 & 18.2 + note - redshifts and x - ray luminosities are from ebeling et al . @xmath60 are based on the @xmath76 relation of reiprich & b " ohringer ( 2002 ) . last column gives the magnitude of the brightest cluster galaxy . macs ( ebeling et al . , 2001 ) has provided a statistically complete , x - ray selected sample of the most x - ray luminous galaxy clusters at @xmath77 . based on sources detected in the rntgen satellit ( rosat ) all - sky survey ( rass , voges et al . 1999 ) , macs covers 22,735 deg@xmath78 of extragalactic sky ( @xmath79 deg ) ; the present macs sample , estimated to be at least 90% complete , comprises 124 clusters all of which have spectroscopic redshifts . owing to the high x - ray flux limit of the rass and the lower redshift limit of @xmath80 , macs clusters feature x - ray luminosities of , typically , 510@xmath81 erg s@xmath82 in the 0.12.4 kev band ( ebeling et al . macs thus probes the high end of the cluster mass function , including some of the most powerful gravitational lenses ( smith et al . 2009 ; zitrin et al . 2009 ) ; see also smail et al . ( 2007 ) for a spectacular case of galaxy - galaxy lensing in the field of a macs cluster . macs clusters have been used for a wide range of cosmological and astrophysical applications , e.g. , in cosmological studies ( allen et al . 2008 ; mantz et al . 2008 , 2009a , b ) , investigations of large - scale structure ( ebeling , barrett , & donovan 2004 ; kartaltepe et al . 2008 ) , and studies of the galaxy content and gas properties of individual clusters ( e.g. , ma et al . 2008 , 2009 ) . here we use images of 35 macs clusters observed with the acs ( go-09722 , go-10491 , go-10875 , pi ebeling ) . we divide these clusters into two subsamples according to redshift , @xmath83 , and @xmath84 , which consist of 23 and 12 clusters , respectively . the low - redshift sample was observed with _ hst _ in snapshot mode , meaning the telescope schedulers chose a fraction of the targets from the full macs sample , based solely on their scheduling convenience . thus , the clusters we analyse are an unbiased , representative selection from the entire macs sample . the medium - redshift sample consists of a complete set of 12 macs clusters in this redshift range that are visible from hawaii . strong - lensing mass reconstructions of the clusters in this subsample have been recently presented by zitrin et al . ( 2010 ) . the low - redshift clusters were observed through the @xmath85 filter ( mean wavelength @xmath86 ) with exposure times of @xmath87 s , while the medium - redshift sample was observed through the @xmath88 filter ( mean wavelength @xmath89 ) with exposure times of @xmath90 s. applying the @xmath91 relation is the mass enclosed within @xmath92 , the radius within which the average density is equal to @xmath93 times the critical cosmological density at the observed redshift . ] of reiprich & b " ohringer ( 2002 ) yields a cluster mass range of @xmath94 , and @xmath95 for the low- and medium - redshift samples , respectively . the cluster properties are listed in tables 1 and 2 . the sample of smith et al . ( 2005 ) , as analyzed in h05 , consists of 10 galaxy clusters from the x - ray brightest abell - type clusters of galaxies ( xbacs ) catalogue ( ebeling et al . 1996 ) , with @xmath96 . the @xmath97 kev flux limit of @xmath98 erg @xmath99 s@xmath82 applied to this redshift range implies x - ray luminosities @xmath100 erg s@xmath82 , i.e. similar to the macs clusters at their higher redshifts . details of this sample and its properties can be found in table 1 of h05 . the rcs survey was conducted using the canada - france - hawaii telescope ( cfht ) through the @xmath101 and @xmath102 filters . gladders & yee ( 2005 ) applied a red - sequencing technique to an area of @xmath103 , and a catalogue of @xmath104 clusters at @xmath105 was compiled . the survey is complete to @xmath106 magnitude limits of @xmath107 and @xmath108 in @xmath102 and @xmath101 , respectively . like macs clusters , rcs clusters have also been used in many applications , e.g. , studying the scaling relations between different cluster properties ( hicks et al . 2008 ) , and exploring the evolution of the red - sequence galaxy luminosity function ( gilbank et al . 2008 ) . among the rcs clusters , a subset of @xmath109 clusters was proposed for hst observation , again in snapshot mode , out of which @xmath28 were selected by hst schedulers based on scheduling conveneience , and imaged using acs , ( go-10626 , pi loh ) . contrary to the macs and xbacs clusters that we analyse here , which were chosen in an unbiased way from among complete samples , we do not know what were the criteria , if any , for selecting the 150 rcs clusters to be potential hst snapshot targets . we suspect that there may have been some bias toward including clusters that already had evidence of strong lensing , based on previous ground - based imaging . however , it is highly unlikely that the 150 clusters were chosen , intentionally or unintentionally , in a way that would _ avoid _ systems with strong lensing ( and it is also hard to imagine a logical reason for such a choice ) . the main result of our study will be that the rcs clusters observed by hst are inefficient as lenses , when compared to the truly unbiased sample of x - ray selected clusters . this conclusion , applied to the rcs clusters as a whole , will therefore only be strengthened , if the 150 rcs clusters were pre - selected to favor strong lenses . our results will thus provide a firm and useful upper limit on the rcs lensing fraction . the clusters were imaged through the @xmath88 filter with exposure times of @xmath110 s. luminosities and mass estimates of the rcs clusters have not been published to date . in @xmath111 below , we show that the rcs clusters and the x - ray selected clusters above have similar optical luminosities . as with the x - ray selected clusters above , we divide the rcs clusters into redshift bins : the same low ( @xmath83 ) and medium ( @xmath112 ) redshift subsamples which were defined above , and a third , high - redshift , subsample at @xmath113 . the three redshift subsamples consist of @xmath114 , @xmath114 , and @xmath115 clusters , respectively . the properties of the @xmath28 rcs clusters are listed in tables 3 , 4 , and 5 . lcccc cluster & ra & dec & @xmath32 & bcg + & & & & @xmath63 + [email protected] & 02:24:03.4 & @xmath116 & 0.314 & - + [email protected] & 03:51:39.5 & @xmath117 & 0.334 & 17.3 + [email protected] & 04:44:06.4 & @xmath118 & 0.437 & 18.0 + [email protected] & 05:15:37.0 & @xmath119 & 0.44 & 18.3 + [email protected] & 05:18:35.2 & @xmath120 & 0.475 & 18.6 + [email protected] & 09:28:22.3 & @xmath121 & 0.356 & 18.2 + [email protected] & 11:02:33.5 & @xmath122 & 0.423 & 17.6 + [email protected] & 11:02:59.2 & @xmath123 & 0.395 & 18.4 + [email protected] & 11:03:40.7 & @xmath124 & 0.492 & 19.3 + [email protected] & 13:19:12.7 & @xmath125 & 0.354 & - + [email protected] & 14:52:27.3 & @xmath126 & 0.325 & 18.1 + rcs145900.4@xmath40102336 & 14:59:00.8 & @xmath127 & 0.395 & 18.4 + rcs151110.7@xmath40100203 & 15:11:11.1 & @xmath128 & 0.455 & 18.1 + rcs151306.9@xmath40061124 & 15:13:06.5 & @xmath129 & 0.325 & 16.4 + [email protected] & 21:15:20.3 & @xmath130 & 0.331 & 17.6 + [email protected] & 21:21:35.0 & @xmath131 & 0.351 & 17.2 + rcs215609.1@xmath40012319 & 21:56:09.3 & @xmath132 & 0.335 & 16.9 + [email protected] & 22:39:52.8 & @xmath133 & 0.429 & 18.7 + lcccc cluster & ra & dec & @xmath32 & bcg + & & & & @xmath63 + [email protected] & 03:34:14.5 & @xmath134 & 0.668 & - + [email protected] & 03:50:27.4 & @xmath135 & 0.584 & 19.0 + [email protected] & 04:42:08.1 & @xmath136 & 0.522 & 18.9 + [email protected] & 05:11:27.8 & @xmath137 & 0.518 & 18.3 + [email protected] & 05:18:55.0 & @xmath138 & 0.544 & 19.0 + [email protected] & 05:19:19.8 & @xmath139 & 0.603 & 19.9 + [email protected] & 11:01:04.7 & @xmath140 & 0.639 & - + [email protected] & 11:07:33.5 & @xmath141 & 0.597 & 18.7 + [email protected] & 11:07:53.0 & @xmath142 & 0.579 & 19.4 + [email protected] & 11:08:14.5 & @xmath143 & 0.638 & - + [email protected] & 13:17:22.8 & @xmath144 & 0.535 & 18.2 + [email protected] & 13:23:35.5 & @xmath145 & 0.538 & 18.1 + [email protected] & 14:19:10.3 & @xmath146 & 0.647 & 19.4 + rcs151840.1@xmath40084500 & 15:18:40.3 & @xmath147 & 0.515 & 18.7 + [email protected] & 16:15:47.5 & @xmath148 & 0.514 & 18.8 + [email protected] & 21:52:23.2 & @xmath149 & 0.545 & 18.7 + [email protected] & 23:16:54.8 & @xmath150 & 0.56 & 19.6 + [email protected] & 23:47:17.4 & @xmath151 & 0.537 & 18.8 + lccc cluster & ra & dec & @xmath32 + [email protected] & 02:24:53.6 & @xmath152 & 0.906 + rcs025242.5@xmath34150024 & 02:52:42.7 & @xmath153 & 0.995 + [email protected] & 04:39:34.2 & @xmath154 & 0.786 + [email protected] & 05:19:40.3 & @xmath155 & 0.913 + [email protected] & 11:04:40.3 & @xmath156 & 0.715 + [email protected] & 11:06:52.2 & @xmath157 & 0.768 + [email protected] & 11:07:23.8 & @xmath158 & 0.794 + [email protected] & 11:22:25.5 & @xmath159 & 0.799 + [email protected] & 13:29:39.8 & @xmath160 & 0.901 + [email protected] & 14:50:40.2 & @xmath161 & 0.769 + [email protected] & 16:20:09.2 & @xmath162 & 0.797 + [email protected] & 21:18:52.6 & @xmath163 & 0.786 + [email protected] & 21:22:38.3 & @xmath164 & 0.856 + [email protected] & 21:52:49.2 & @xmath165 & 0.704 + [email protected] & 23:18:31.8 & @xmath166 & 0.809 + [email protected] & 23:42:20.4 & @xmath167 & 0.802 + in h05 , we introduced the use of an automated arc detection algorithm to arc statistics studies . automated arc detection is important for an objective , quantitative , and fair comparison of arc statistics in observed and simulated data . in the meantime , a number of other arc - detection algorithms have been published , by lenzen et al . ( 2004 ) , alard ( 2006 ) , and seidel & bartelmann ( 2007 ; sb07 ) . in the present work , we subject all of the images to two of these algorithms , h05 and sb07 . the h05 arc - detection algorithm is based on application of the sextractor ( bertin & arnouts 1996 ) object identification software . the output of repeated sextractor calls , using different detection parameters each time , is filtered using some threshold of object elongation . the final sextractor call is executed on an image combined from the filtered `` segmentation image '' outputs of the previous calls . the arc candidates detected in that last call are included in the final arc catalogue if they meet the required detection parameters defined by the user . the sb07 algorithm is based on light moments . the image is divided into small cells which are iteratively moved to their local light centres . then , for each cell , an ellipticity vector is calculated using light moments . adjacent cells with similarly oriented ellipticity vectors are joined together and considered as part of an arc candidate , whose outer boundary is determined by an active contour method . candidates are accepted if they conform to specified parameters . in the present work , we apply an acceptance criterion on arc length - to - width ratio of @xmath168 . we also use a magnitude limit of @xmath169 as another acceptance criterion which , given the exposure times of our sample , results in the detection of arcs with signal - to - noise @xmath170 . our magnitude limit is higher than most of the magnitude limits used in previous studies , such as b98 and zaritsky & gonzales 2003 , allowing us to include fainter arcs in our analysis . nevertheless , our acceptance threshold for arc detections is brighter than the arc detection limits of all the images , with their range of exposure times and filters , thus permitting a meaningful comparison of arc statistics among the various subsamples . this holds also for the wfpc2 images of the xbacs sample . although wfpc2 was less sensitive than acs , the wfpc2 exposure times were longer , typically 7000 s , leading to similar depths . furthermore , the somewhat lower angular resolution of wfpc2 , due to its larger pixels ( @xmath171 ) , is not important , since the arcs we consider are always much larger , and all the arcs we find below in acs images would have been detected in long wfpc2 exposures as well . we note that we use total - magnitude limit for arcs , rather than considering surface brightness , which could also plausibly be used . we do this to conform with previous observational and theoretical studies , but also because arcs , especially at hst resolution , display rich structure and unresolved clumps , and hence it is not clear that mean surface brightness would be a more relevant observable . due to the varying position of the cluster centres within the fov , the cluster coverage area varies . we therefore also limit our search to a @xmath172 radius from the cluster centre . the automated arc detection results were visually inspected in order to remove false positives such as spikes from saturated stars , galaxy spiral arms , and edge - on galaxies . while most of the arcs in our sample are detected by both programs , a few unmistakable lensed arcs are picked out by only one or the other . the sb07 arcfinder is more sucessful than the h05 arcfinder in detecting arcs that are superimposed on the light of cluster galaxies . on the other hand , the h05 arcfinder produce a better `` segmentation '' compared to the sb07 arcfinder , which sometimes breaks arcs into smaller arclets , which then do not qualify as giant arcs . we defer a more detailed comparison of these and other arcfinders to a future study . figure 1 shows the acs images of the clusters in which arcs are detected , and figure 2 provides zoom - ins on the individual arc features . table 6 lists the properties of the detected arcs , which we discuss in more detail below . 0.2 cm 0.2 cm 0.2 cm 0.2 cm 0.2 cm 0.2 cm 0.2 cm 0.2 cm 0.2 cm j8qu11020_arcs2(0.42 ) lcccccc cluster & arc i d & length & width & @xmath173 & @xmath174/ & radial separation + & & & & & @xmath175 & from cluster centre + & & [ arcsec ] & [ arcsec ] & & [ mag ] & [ arcsec ] + macsj0025.4@xmath341222 & a & 3.3 & 0.2 & 13.9 & 23.6 & - + & b & 2.7 & 0.3 & 9.64 & 23.7 & - + macsj0140.0@xmath340555 & a & 5.1 & 0.6 & 8.3 & 21.8 & 13.7 + macsj0257.1@xmath342325 & a & 3.0 & 0.3 & 10.6 & 23.7 & 11.6 + & b & 3.5 & 2.8 & 12.7 & 23.5 & 27.8 + [email protected] & a & 4.3 & 0.2 & 18.5 & 23.0 & 4.5 + macsj0451.9@xmath400006 & a & 20.2 & 0.7 & 29.3 & 20.5 & 38.3 + macsj0454.1@xmath340300 & a & 3.7 & 0.4 & 10.4 & 22.3 & 21.8 + macsj0520.7@xmath341328 & a & 6.1 & 0.5 & 12.2 & 22.2 & 30.9 + & b1 & 2.3 & 0.3 & 8.6 & 23.7 & 24.1 + & b2 & 2.6 & 0.3 & 8.8 & 23.5 & 27.4 + macsj0712.3@xmath405931 & a & 5.2 & 0.3 & 17.9 & 22.8 & 23.2 + macsj0717.5@xmath403745 & a & 4.2 & 0.3 & 15.4 & 23.0 & - + & b & 3.0 & 0.4 & 8.5 & 22.7 & - + & c & 4.6 & 0.2 & 19.0 & 23.1 & - + macsj0744.8@xmath403927 & a & 6.1 & 0.5 & 12.3 & 20.5 & 23.3 + & b & 5.1 & 0.2 & 27.3 & 23.9 & 35.0 + macsj0947.2@xmath407623 & a & 6.4 & 0.4 & 14.3 & 22.3 & 13.5 + & b & 2.9 & 0.3 & 14.7 & 22.6 & 41.7 + & c & 6.5 & 0.3 & 23.0 & 23.4 & 19.1 + macsj0949.8@xmath401708 & a & 3.3 & 0.4 & 8.8 & 22.8 & 37.6 + macsj1115.2@xmath405320 & a & 3.2 & 0.3 & 10.2 & 23.6 & 31.6 + & b1 & 3.4 & 0.3 & 10.3 & 22.9 & 56.2 + & b2 & 3.5 & 0.4 & 8.5 & 23.5 & 57.3 + & c & 4.4 & 0.5 & 8.5 & 22.3 & 34.5 + macsj1115.8@xmath400129 & a@xmath176 & 4.8 & 0.3 & 15.1 & 23.2 & 11.2 + & b & 5.1 & 0.2 & 26.7 & 23.7 & 37.2 + macsj1133.2@xmath405008 & a & 11.7 & 0.8 & 14.6 & 21.0 & 10.5 + & b & 2.5 & 0.2 & 10.2 & 23.5 & 17.0 + macsj1149.5@xmath402223 & a & 4.0 & 0.5 & 8.4 & 22.2 & 26.0 + macsj1206.2@xmath340847 & a@xmath78 & 13.9 & 0.5 & 26.8 & 21.1 & 20.7 + & b@xmath78 & 4.8 & 0.5 & 8.8 & 22.4 & 23.8 + & c & 3.7 & 0.5 & 8.2 & 23.3 & 59.3 + macsj1236.9@xmath406311 & a & 2.9 & 0.3 & 10.6 & 23.8 & 38.0 + macsj1354.6@xmath407715 & a & 7.7 & 0.6 & 12.9 & 21.7 & - + & b & 7.6 & 0.9 & 8.8 & 21.1 & - + & c & 3.4 & 0.2 & 14.6 & 23.8 & - + & d & 4.6 & 0.3 & 17.6 & 23.4 & - + [email protected] & a & 2.6 & 0.3 & 10.1 & 23.0 & 48.4 + [email protected] & a@xmath177 & 10.5 & 0.6 & 17.3 & 20.2 & 10.0 + & b@xmath177 & 3.8 & 0.3 & 11.3 & 22.4 & 17.2 + macsj1423.8@xmath402404 & a & 4.2 & 0.4 & 10.3 & 22.7 & 19.7 + & b & 3.6 & 0.2 & 14.5 & 23.3 & 20.6 + rcs151840.1@xmath40084500 & a & 3.8 & 0.3 & 11.9 & 21.6 & 11.2 + [email protected] & a & 3.8 & 0.4 & 10.4 & 21.6 & 3.4 + macsj2129.4@xmath340741 & a & 5.4 & 0.3 & 17.3 & 22.1 & 31.5 + rcs215609.1@xmath40012319 & a & 3.7 & 0.3 & 11.2 & 22.8 & 18.3 + macsj2214.9@xmath341359 & a & 4.0 & 0.3 & 12.4 & 23.3 & 14.3 + & b & 4.9 & 0.3 & 17.9 & 23.0 & 20.1 + notes : arcs previously reported by : @xmath176sand et al . ( 2005 ) ; @xmath78ebeling et al . ( 2009 ) ; @xmath177gladders et al . ( 2003 ) . in the macs sample we identify a total of @xmath178 arcs in @xmath179 out of the @xmath180 low - redshift clusters , and a total of @xmath115 arcs in @xmath181 out of the @xmath179 medium - redshift clusters . all but 3 of these arcs ( in two clusters ) have not been previously reported ( see table 6 ) . the arcs span a magnitude range of @xmath182 and a @xmath173 ratio range of @xmath183 . as shown in figure 3 , over half of the cluster lenses , in both the low- and medium - redshift macs subsamples , produce multiple arcs . a similar result was found in the xbacs sample of h05 , in which 17 arcs ( with @xmath168 ) , in 7 out of the 10 clusters at @xmath16 , were detected . in terms of the distribution of the angular separation of arcs from the cluster centres , in the low - redshift macs subsample , as shown in fig . 4 , the lensed arcs are uniformly distributed at separation angles of @xmath184 . in the medium - redshift macs sample , the arcs are distributed slightly closer to the cluster centres , but both distributions are consistent , given the small numbers per bin . there are no arcs in this sample beyond @xmath185 . since there is an uncertainty concerning the centre position of the cluster macsj1354.6@xmath407715 , as discussed below , we do not include its arcs in the above analysis . in addition , each of the apparently merging arc pairs macsj0520.7@xmath341328 b1/b2 and macsj1115.2@xmath405320 b1/b2 , are treated as one arc . we also exclude the arcs in macsj0717.5@xmath403745 from this analysis , since this cluster is highly disturbed ( ma , ebeling , & barrett ; 2009 ) and therefore its centre can not be easily determined . only two arcs are detected in the low - redshift rcs cluster subsample . while both arcs have @xmath7 , they are still relatively short ( @xmath186 ) compared to some of the arcs found in the macs sample , which can be as long as @xmath187 . in the medium - redshift optical subsample , @xmath188 arcs are found in @xmath189 out of the @xmath114 clusters . two of these arcs ( in one cluster , see table 6 ) have been previously reported . no arcs are detected among the @xmath115 clusters of the high - redshift ( @xmath190 ) optical subsample . as seen in fig . 4 and table 6 , compared to the x - ray sample , the arcs in the rcs sample occur at significantly smaller separations , generally @xmath191 , and sometimes only @xmath192 . the only exception is [email protected] , whose arc appears @xmath193 from the cluster centre . however , as seen in figs . 1 and 2 , this arc may actually be a small - separation image produced by the local mass concentration traced by the galaxies near the arc . since arcs occur near critical curves , the small separations suggest significantly smaller einstein radii , and hence masses , for the rcs clusters . table @xmath2 summarizes the arc statistics of our various cluster subsamples . as noted above , only two arcs are detected in the rcs low - redshift subsample , compared to the @xmath178 arcs detected in the low - redshift macs subsample . the arc production efficiencies are , therefore , @xmath194 , and @xmath195 arcs per cluster for the rcs and macs subsamples , respectively , where we cite a @xmath196 confidence interval assuming poisson statistics . in the medium - redshift bin , the macs clusters are also more efficient lenses than the rcs clusters , with efficiencies of @xmath197 , and @xmath198 arcs per cluster , respectively . with zero detected arcs , the high - redshift rcs sample has an arc production efficiency of @xmath199 arcs per cluster ( @xmath200 confidence ) , which is consistent with the rcs efficiencies at lower @xmath32 . as the arc occurrence frequency is consistent among different redshift bins , we tabulate also the total frequency in the x - ray versus the optical subsamples . the frequencies differ at the @xmath106 level . we also derive the arc production efficiencies for arcs with @xmath201 , for comparison with previous studies in which this @xmath173 ratio was used to define giant arcs . all of the rcs arcs have @xmath7 and therefore the rcs cluster efficiencies remain unchanged . however , the macs cluster production efficiencies of arcs with @xmath7 are somewhat lowered to @xmath202 , and @xmath203 arcs per cluster for the low- , and medium - redshift subsamples , respectively . even so , both the low- and medium - redshift macs clusters are significantly more efficient lenses ( @xmath204 ) than their rcs counterparts . lcccccc subsample & @xmath205 & @xmath206 & & + & & & ( @xmath168 ) & ( @xmath7 ) & ( @xmath168 ) & ( @xmath7 ) + + xbacs ( @xmath207 ) & 10 & 7 & 17 & 12 & @xmath208 & @xmath209 + macs ( @xmath210 ) & 23 & 12 & 26 & 17 & @xmath195 & @xmath202 + macs ( @xmath211 ) & 12 & 9 & 16 & 13 & @xmath197 & @xmath203 + macs ( @xmath212 ) & 35 & 21 & 42 & 30 & @xmath213 & @xmath214 + total ( @xmath215 ) & 45 & 28 & 59 & 42 & @xmath216 & @xmath217 + + rcs ( @xmath210 ) & 18 & 2 & 2 & 2 & @xmath194 & @xmath194 + rcs ( @xmath211 ) & 18 & 4 & 5 & 5 & @xmath198 & @xmath198 + rcs ( @xmath218 ) & 16 & 0 & 0 & 0 & @xmath219 & @xmath219 + rcs ( @xmath212 ) & 36 & 6 & 7 & 7 & @xmath220 & @xmath220 + total ( @xmath221 ) & 52 & 6 & 7 & 7 & @xmath222 & @xmath222 + in addition to the automatic detection results , we have visually inspected all of the clusters in our sample . we find that there are several clusters that show signs of strong lensing but in which no arc was detected algorithmically , using our detection thresholds . in some cases , the arcs are too faint , while in others the arcs may be bright but have a length - to - width ratio below our threshold . sometimes , an arc is projected close to another galaxy , making its detection difficult . we find six such clusters with signatures of strong lensing that are not included in our arc catalogue . two of the clusters ( macsj1319.9@xmath407003 and macsj2135.2@xmath340102 ) , are found in the low - redshift macs subsample , two ( [email protected] and [email protected] ) are in the medium - redshift rcs subsample , the fifth cluster , macsj0647.7@xmath407015 , is in the medium - redshift macs subsample , and the sixth cluster , rcs rcs025242.5@xmath34150024 , is in the high - redshift rcs subsample . figure 5 shows each of these cases . 0.2 cm inspection of the image of macsj1354.6@xmath407715 suggests the existence of two separate galaxy concentrations . arcs a and b ( see fig . 1 ) seem to straddle one centre . about @xmath223 west of that centre there seems to be another mass concentration enclosed by arcs c and d. we note that south of arc c there is an additional arc which is not included in our arc catalogue due to its small @xmath173 ratio . the two galaxies at the centres of the two concentrations have magnitudes of 19.8 mag ( east clump ) , and 19.2 mag ( west clump ) . this cluster may be during some stage of a merger , but with still a considerable amount of substructure . although the two clumps may be chance projections of two clusters at different redshifts , this is unlikely given the rarity of such massive lensing clusters . moreover , the optical colors of the early - type galaxies across the field are also consistent with a single redshift . existing rosat data show that the x - ray emission is centred on the system , but the entire hst / acs field shown in fig . 6 spans only a few rosat resolution elements , making it impossible to say anything about the x - ray flux distribution relative to the two mass and optical - light concentrations . higher resolution x - ray imaging ( already approved with _ chandra _ ) and optical spectroscopy are needed to select among these alternatives . close inspection of the central area of the cluster macsj2129.4@xmath340741 reveals two objects that appear to be radially distorted ( see fig . the image parity of each of these objects seems to be flipped , as expected in lensing . an alternative explanation for these objects is tidal tails due to physical interaction between galaxies . again , optical spectroscopy is needed to resolve the issue . we have found an extraordinarily large arc ( @xmath224 ) in the cluster rcs025242.5@xmath34150024 ( fig . this arc is found near a galaxy which is too bright ( @xmath225 ) to belong to this high redshift ( @xmath226 ) cluster . it seems that the arc is produced by the gravitational field of this foreground galaxy , and therefore we do not include it in our arc catalogue . the results of our arc survey , presented above , can serve as a new and improved observational basis for future arc statistic studies . however , our survey also shows clearly that the arc - production efficiency of x - ray - selected clusters such as macs and xbacs is higher by a factor of @xmath227 than that of rcs clusters . in this section , we carry out additional analysis and discussion of the meaning of this result . at a given redshift , the cross section for lensed arc formation depends primarily on mass , although mass profile , ellipticity and substructure are also important . the mass dependence weakens towards the high mass end at @xmath228 ( dalal et al . 2004 ; hennawi et al . the stark difference in the arc frequency between the x - ray selected and optically selected clusters immediately raises the possibility that they probe different mass ranges . based on their x - ray luminosities , the x - ray selected clusters have masses of @xmath229 . unfortunately , there is scant information of the x - ray properties of the rcs clusters , and hence on their masses . for example , hicks et al . ( 2008 ) recently observed with _ a sample of 13 rcs clusters , of which detailed analysis was possible for nine . they found significant differences in the mass - temperature - luminosity relations of x - ray selected and rcs clusters , x - ray underluminosity in some rcs clusters , and evidence that rcs clusters have a larger fraction of their baryons in stars . nevertheless , since optical flux is one of the few observables we do have available for the rcs clusters , we begin by comparing the optical luminosities of the macs and rcs subsamples . the hst / acs field of view covers only the central core regions of the clusters , and therefore we examine several proxies for the optical luminosity . as a first proxy for optical luminosity , we examine the luminosities of the brightest cluster galaxies ( bcgs ) of the rcs and macs samples . in sdss clusters , hansen et al . ( 2005 ) have found a correlation between cluster mass and bcg luminosity . the bcg magnitudes were measured using sextractor by including the light from pixels which belong to the bcg and are above the detection threshold . since the low - redshift subsamples are observed through different filters , we first calculate the f606w@xmath34f814w color for each rcs cluster redshift using an elliptical galaxy spectral template from kinney et al . ( 1996 ) , and convert the rcs cluster bcg f814w magnitudes to f606w . in this comparison we exclude the following clusters ( four macs and five rcs ) due to the uncertainty in determining their centres and in identifying the dominant bcgs : macsj0916.1@xmath340023 , macsj1354.6@xmath407715 , macsj2243.3@xmath340935 , macsj0257.1@xmath342325 , [email protected] , [email protected] , [email protected] , [email protected] , and [email protected] . we find that the bcg magnitudes are more uniformly distributed in the rcs subsample than in the macs subsample , and the bcgs span a wider magnitude range . nevertheless , in the low - redshift macs and rcs subsamples , the median bcg absolute magnitudes are , @xmath230 and @xmath231 , respectively . likewise , in the the medium - redshift macs and rcs subsamples the median bcg magnitudes are @xmath232 , and @xmath233 , respectively . a kolmogorov - smirnov ( ks ) test indicates that for both the low- and the medium - redshift subsamples , the null hypothesis that both the rcs and macs bcg magnitudes are derived from the same parent distribution can not be confidently rejected ( probabilities of 0.12 and 0.35 , respectively , for the null hypothesis ) . these numbers are summarized in table 8 . for a second comparison of optical luminosities , we measure integrated optical luminosity of the brightest galaxies within the cluster cores . we measure the total light of galaxies inside a physical aperture of radius 270 kpc ( at low @xmath32 ) and 370 kpc ( at medium @xmath32 ) . the contribution of foreground and background galaxies to the light is determined statistically in annuli of @xmath234 kpc and @xmath235 kpc , for the low- and medium - redshift subsamples , respectively , and subtracted from the core light . the area in which we measure the `` background '' is still well within the cluster , and hence , our cluster core luminosities are underestimated due to background over - subtraction . nevertheless , barring large profile differences ( see below ) , these biased estimates of cluster luminosity can still be compared meaningfully between the x - ray and optical samples . we include only the light from objects with magnitudes fainter than the cluster bcg magnitude , but brighter than 24 mag . we convert the macs low - redshift subsample s f606w luminosities to f814w luminosities assuming , again , the kinney et al . ( 1996 ) elliptical galaxy template and the filter transmission curves for the two bands . the resultant optical luminosity distributions ( fig . 9 ) of both the rcs and macs cluster are consistent with being drawn from the same parent distribution ( @xmath236 and @xmath237 probabilities for the null hypothesis , see table 8) . image sections of two radially distorted objects ( marked as @xmath238 and @xmath239 ) in the cluster macsj2129.4@xmath340741.,width=302 ] image section of a large arc found in the field of view of the high - redshift cluster rcs025242.5@xmath34150024 , around a foreground galaxy.,width=302 ] finally , as a third method of comparing optical luminosities , we simply count the light from _ * all * _ the pixels inside the above apertures and annuli ( but still leaving out the light from objects brighter than the bcg ) . this method takes into account the light from all the stars in the cluster cores , including stars in galaxies below the detection limit and diffuse intracluster light . as in the previous method , the core luminosity is underestimated due to background over - subtraction , but in a consistent way for the x - ray and optical clusters . in contrast to the two previous methods , where the dominant galaxies in the cluster core are early types , in this case the color correction , applied in order to convert f606w fluxes to f814w , is less clear - cut , since fainter and undetectable dwarf galaxies may well be blue . for the range in color terms from early - type to late - type galaxies , the luminosity distributions are either consistent with being drawn from the same parent distribution ( 0.05 probability for the null hypothesis , `` blue '' color correction ) , to marginally consistent ( 0.01 probability for the null hypothesis , `` red '' color correction ) . at most ( in the case of the low - redshift subsamples , and assuming the reddest color correction ) the medians of these two distributions differ only by a factor of @xmath240 . overall , as shown in table 8 , the medians of the rcs and macs cluster luminosity distributions and the results of the ks tests we applied to these distributions suggest that the rcs and macs cluster samples have similar optical luminosities . we note also that the macs clusters that actually display arcs ( shaded histograms in fig . 9 ) are not necessarily the most luminous ones , and that there is a large overlap of their luminosities with those of rcs clusters that are much less efficient arc producers . we have also measured and compared the optical light profiles of the two samples and , within the limited range of the cluster cores covered by the acs data , we find no significant differences . lcc subsample & optical luminosity measure & ks probability + + macs ( @xmath210 ) & @xmath241 & + rcs ( @xmath210 ) & @xmath242 & + macs ( @xmath211 ) & @xmath243 & + rcs ( @xmath211 ) & @xmath244 & + + macs ( @xmath210 ) & @xmath245 & + rcs ( @xmath210 ) & @xmath246 & + macs ( @xmath211 ) & @xmath247 & + rcs ( @xmath211 ) & @xmath248 & + + macs ( @xmath210 ) & @xmath249 & + rcs ( @xmath210 ) & @xmath250 & + macs ( @xmath211 ) & @xmath251 & + rcs ( @xmath211 ) & @xmath252 & + note- optical luminosity measure indicates f606w or f814w absolute magnitudes for the bcgs , and luminosities in units of @xmath253 for the two measures of core optical luminosity . probability is for the null hypothesis that a pair of distributions are not different . however , stars , let alone the small fraction of the stars that dominate the optical luminosity , are a tiny component of the total cluster mass , and it is therefore plausible that the masses of the two samples are very different , despite the similar optical luminosities , with masses significantly below @xmath254 for the rcs clusters . a strong argument for such a mass difference is the difference in the space densities of the two samples . from the numbers of clusters and the area surveyed ( see 2 ) , the projected density of macs clusters is @xmath255 . assuming the cluster mass function is probed correctly by x - ray surveys , only about one macs - like massive cluster is expected in the @xmath256 search area of the rcs survey . based on the cluster mass function ( e.g. reiprich & b " ohringer 2002 ) , the @xmath104 clusters found in the rcs search area imply that the vast majority of these clusters have masses of @xmath257 , an order of magnitude lower than macs clusters . this picture is further supported by the distributions of arc separations from the cluster centres . since arcs occur near critical curves , the separations can roughly represent the einstein radii of the clusters . as noted in 3 and seen in fig . 4 , the macs clusters have arcs at @xmath184 , with a median at @xmath258 , while the rcs arcs are generally much closer in , with a median of @xmath224 . the small einstein radii of most of the rcs clusters with arcs are similar to those of rich groups . a puzzling corollary of the above arguments , however , is the fact that in the small subsample of 52 rcs clusters imaged with acs , which constitute just 5 per cent of the full rcs sample , there are as many as two clusters ( rcs [email protected] and rcs 215609.1@xmath40012319 ) with arcs at separations implying @xmath187 einstein radii , and hence macs - like masses , in contrast to the expectation that of order just one such cluster exists in the _ entire _ rcs survey . furthermore , despite their mass , only about half of the macs clusters display arcs in our survey , because a galaxy in a suitable position in the source plane is required in order to produce an arc . the two large - separation rcs clusters in the hst sample would thus imply about 4 massive rcs clusters in the hst sample , and @xmath259 in the full rcs sample , as opposed to the @xmath260 expected from the x - ray - derived mass function . a possible explanation is that the hst rcs sample is not a fully representative selection of the rcs . indeed , the wide arcs of the cluster rcs [email protected] above were already noted in ground - based images by gladders et al . ( 2003 ) , and it may have been included in the hst sample for this reason . thus , the hst rcs sample could be a representative subsample of the rcs , _ plus _ a few of the most massive rcs clusters , and would thus be pre - biased in favor of lensing . since , despite this bias , the x - ray - selected clusters are still much more efficient lenses , the observed arc occurrence frequency in the rcs clusters imaged with hst provides an upper limit on the rcs arc frequency as a whole . on the other hand , gladders et al . ( 2003 ) discussed eight potential lenses out of the full rcs sample of about 1000 clusters . with random selection , one would expect 1.1 of these eight lenses to be included among the 150 rcs clusters in the hst snapshot sample , out of which the actually observed targets were chosen by hst schedulers . in fact , there are two of the eight gladders et al . ( 2003 ) potential lenses among the 150 hst targets . this could be the result of some slight bias in favor of known lenses , as described above , but could of course be due just to chance . we reiterate that , if the rcs sample is unbiased , our conclusion about the relatively low lensing inefficiency of rcs clusters hold . if the rcs sample was pre - biased , this conclusion is only strengthened . the simplest explanation for our measurement of a low lensing efficiency among rcs clusters , compared to x - ray - selected clusters , is a typical rcs cluster mass that is lower by an order of magnitude . this leaves open the question of what stands behind the similarity of x - ray and rcs clusters , in terms of stellar luminosity , optical profiles , numbers of galaxies , and general optical appearance . these similarities can not be due to chance line - of - sight projections ( rcs clusters are chosen based on redshifts of the early - type galaxies that characterise dense environments , so they are , in fact , real associations ) , nor due to projection effects along large - scale structure `` filaments '' simulations have shown that large scale structure may contribute only about 10 per cent to the cluster surface mass density ( wambsganss , bode , & ostriker 2005 ; hilbert et al . instead , there is a real and large variation in the total - mass - to - optical - light ratio among clusters . the low mass - to - light ratio of rcs cluster cores may be caused by a bias in favour of line - of - sight mergers in the optical selection process , a prominent and spectacular example of which is cl0024 + 24 ( czoske et al . 2002 ) . indeed , extensive spectroscopic follow - up of rcs clusters has uncovered several cases of close projection effects of possibly physically associated systems as well as line - of - sight substructure ( gilbank et al . 2007 ; cain et al . 2008 ) . a further effect to consider is the question of whether x - ray selection may favour the inclusion of clusters that are in the process of merging . torri et al . ( 2004 ) have found that , during a merger , the lensing cross section is increased by a factor of @xmath227 for a duration of a couple of hundred million years , while the x - ray luminosities of merging clusters are increased by a factors of @xmath261 . a similar conclusion regarding the x - ray luminosity of clusters during mergers was reached by randall , sarazin , & ricker ( 2002 ) . if x - ray - selected cluster samples indeed have a larger fraction of merging clusters , one could thus expect a larger fraction of highly efficient lenses in those samples . thus , the masses of the x - ray - selected clusters may be systematically overestimated as well . an interesting question is whether comparable optically and x - ray - selected cluster samples at @xmath262 also differ in their arc production efficiencies . we did not find giant arcs in any of the high - redshift rcs clusters we analyzed , even though their optical luminosities are comparable to those of the rcs clusters at low and medium redshifts . this contrasts with the results of gladders et al . ( 2003 ) who found rcs clusters to be more efficient lenses at high redshift . finally , the many arcs found in the macs low- and medium - redshift subsamples provide a statistically improved handle on the angular distribution of arcs in clusters . our results show that arcs do form at large angular separations from cluster centres , at up to @xmath172 , in some cases . thus , the large einstein radius of abell 1689 is probably not unique . we have conducted an algorithmically based search for lensed arcs in @xmath259 clusters observed with hst . our cluster sample includes an x - ray selected subsample ( xbacs ; macs ) and an optically selected subsample ( rcs ) , each in a range of redshifts . our search for giant arcs has produced @xmath179 , @xmath263 , and @xmath264 arcs ( @xmath265 ) in the xbacs , macs low - redshift , and macs medium - redshift subsamples , respectively . only @xmath239 , @xmath188 , and zero arcs were found in the low- , medium- , and high - redshift rcs subsamples . the arc production efficiency of the macs clusters is therefore higher by a factor of @xmath227 than that of the rcs clusters . the typical einstein radii of macs clusters are several times larger than those of the relatively few rcs cluster that do display strong lensing . if , as we suspect , the hst sample of rcs clusters was pre - selected in a way that favored strong lenses , then these conclusions would only be strengthened . these results constitute direct evidence , based on strong - lensing statistics , that optically selected rcs clusters are an order of magnitude less massive than x - ray selected clusters , despite their similar optical properties . this conclusion is supported by the factor-100 higher space density of rcs clusters . in the arc statistics literature to date , the observed statistics from x - ray and optical clusters have often been discussed together and interchangeably . we have demonstrated that x - ray and optically selected clusters likely probe distinct parts of the cluster mass function , and should therefore not be mixed in this way . in a forthcoming paper , we will address arc statistics from a theoretical point of view . we will present strong lensing statistics predictions using clusters from several of the latest cosmological simulations , calculated specifically for comparison with the observed samples analyzed here . we expect that the observational database we have presented , compared to these improved new simulations , will elucidate some of the contradictions that have been encountered to date in this field . 0.5 cm we thank eran ofek and dovi poznanski for their help with synthetic photometry and keren sahron for useful discussions . we thank the anonymous referee for constructive comments . d.m . acknowledges support by the israel science foundation and by the dfg through german - israeli project cooperation grant ste1869/1 - 1.ge625/15 - 1 . this research has made use of nasa s astrophysics data system ( ads ) bibliographic services , as well as the nasa / ipac extragalactic database ( ned ) . this work is based on observations made with the nasa / esa hubble space telescope , obtained from the data archive at the space telescope science institute . stsci is operated by the association of universities for research in astronomy , inc . under nasa contract nas 5 - 26555 . he gratefully acknowledges financial support from stsci grants go-09722 , go-10491 , and go-10875 . mb was supported by the transregio - sonderforschungsbereich tr 33 of the deutsche forschungsgemeinschaft .

we address the possible scenario that the large hadron collider ( lhc ) discovers only a higgs boson after 10 fb of operation , and attempt to identify this higgs boson as that of the standard model ( sm ) , the minimal universal extra dimension model ( mued ) , the littlest higgs model with-parity ( lht ) , or the minimal supersymmetric standard model ( mssm ) , using only the measurement of the product of gluon - fusion production cross section and the di - photon branching ratio . in mued , by decoupling any new physics sufficiently to evade the discovery reach at the lhc , the deviation of the signal from the sm is not statistically significant . however , in lht and mssm , it is possible to have a significant deviation in the signal that is consistent with this `` lone higgs scenario '' , and , in the case of a very large suppression , we can distinguish mssm and lht before the discovery of any new resonances . starting with the lone higgs scenario and the deviation in this measurement from the standard model prediction ( whether or not statistically significant ) , we offer tests that may discriminate the models and search strategies of discovering new physics signatures with increasing integrated luminosity .

0806.2608

the stability of the electroweak scale has driven the high energy physics community , both theorists and experimentalists alike , for nearly the past two decades . with the advent of the large hadron collider ( lhc ) , we can finally probe the mechanism of electroweak symmetry breaking ( ewsb ) and possibly new physics at the tev scale that stabilizes the electroweak scale . however , as such new physics is still a mystery , we need to be prepared for all the possibilities . in addition , with the multitude of models of new physics and the possible associated experimental signatures , we are also faced with the ` inverse problem ' of distinguishing models of new physics using the experimental data . in this work , we investigate one of the possible scenarios at the lhc , and attempt to disentangle three generic models of new physics based on experimental measurements of a higgs boson . we suppose that , after the first few years of operation with an integrated luminosity of 10 @xmath2 , the lhc has only discovered a lone scalar boson with couplings to the @xmath3- and @xmath4-bosons that are of the same magnitude as predicted in the standard model ( sm ) . while discovering only a higgs boson at the lhc ( with an integrated luminosity of several 100 @xmath2 ) has been dubbed a `` nightmare scenario @xcite , '' here we are only assuming no new physics , other than this higgs boson , is seen at this stage of operation of the lhc , and leave open the possibility that new physics may be uncovered with further operation time . indeed , one of the main goals of works of this type is to optimize further search strategies based on the information we have at hand from the discovered higgs boson . let us denote @xmath5 as the product of the higgs boson production cross section @xmath6 and the di - photon decay branching ratio @xmath7 . the main question that we attempt to answer in this work is : from the measurement of @xmath5 and its deviation from the sm prediction , can we identify this scalar boson as _ the _ higgs boson in the sm , minimal universal extra dimensions ( mued ) , littlest higgs with @xmath1-parity ( lht ) , or the lightest @xmath8-even boson in the minimal supersymmetric standard model ( mssm ) ? if not , we investigate whether we can use this measurement as a hint or bias , and devise further search strategies of new physics based on its deviation from the sm , regardless whether such deviation is statistically significant . questions of this type are in spirit similar to the lhc inverse problem @xcite , but with an emphasis on distinguishing the models rather than mapping the regions of parameter spaces of a particular model from the data . while such lhc inverse problems have been studied in the literature , they attempt to distinguish models through properties , such as spin , of the new resonances discovered at the lhc . our work here is also of similar spirit to mantry et al . @xcite @xcite and randall @xcite , where they discuss how the properties of the higgs boson can be modified due to states that are not directly observable at the lhc . our work is organized as follows . in section [ sec : precision ] , we discuss the precision to which the signal @xmath5 can be measured at the lhc after 10 @xmath2 of data . in section [ sec : hgg - model ] , we discuss the general pattern of deviations of the signal @xmath5 in the parameter spaces of the models , and roughly map out regions in parameter spaces that such deviation can be significant . we also apply the results of lhc reaches in these models to map out regions of parameter spaces that can be consistent with the aforementioned lone higgs scenario . in section [ sec : hgg - lone ] , we apply the lone higgs scenario as constraints on the parameter spaces , and see how the signal is affected . in particular , we find that in the lone higgs scenario with a large deviation in @xmath5 , we can potentially rule out mued , and , in some cases , distinguish between the mssm and lht . also in section [ sec : hgg - lone ] , we propose some parameter - independent tests that can also be used to distinguish these models . we conclude in section [ sec : conc ] with a summary of our results and offer outlook for projects of this type . in this section we present a brief overview of the detection of the higgs boson and the measurement of its properties at the lhc . more details can be found at the atlas technical design report ( tdr ) @xcite and cms tdr @xcite , and references therein . for reference , we show the production cross section of the higgs boson via gluon - gluon fusion at the lhc @xmath9 at the next - to - leading order in qcd , with the renormalization and factorization scales set to the mass of the higgs boson ( @xmath10 ) , using the latest parton distribution functions ( pdf ) , cteq 6.6 m @xcite , in the top plot of fig . [ fig : cteq66ggh ] . the uncertainties of this cross section , both the pdf - induced uncertainty as well as the relative difference with an earlier version of the pdf ( cteq6.1 ) , are of the order of a few percent as shown in the lower plot of fig . [ fig : cteq66ggh ] . the uncertainty in the luminosity will be on the order of 20% at the start of the lhc . however , the uncertainties in the measurements of the cross sections due to the uncertainty in the luminosity can be reduced partially by taking ratios of these cross sections to measured `` standard candle '' cross sections , such as @xmath11 and @xmath12 @xcite . at the lhc using the latest pdf , cteq6.6 m @xcite ( top plot ) and its pdf - induced uncertainty ( bottom plot , boxed points ) and relative difference with previous version of pdf , cteq6.1 ( triangle points ) . , title="fig:",width=384 ] at the lhc using the latest pdf , cteq6.6 m @xcite ( top plot ) and its pdf - induced uncertainty ( bottom plot , boxed points ) and relative difference with previous version of pdf , cteq6.1 ( triangle points ) . , title="fig:",width=384 ] the detection channels of the higgs boson depend significantly on its mass . although the higgs boson couples most strongly to the massive gauge bosons @xmath13 and the top quark , for higgs mass significantly lighter than the @xmath14 threshold ( @xmath15 gev ) , the decays @xmath16 are kinematically inaccessible , and the dominant decay channel of the higgs boson is @xmath17 . unfortunately , the di - jet background at the lhc is expected to overwhelm this signal , and the most promising channel of detecting the higgs boson is through its ( loop suppressed ) di - photon decay , @xmath18 , with a branching ratio of about 0.2% . the di - photon channel offers a very clean signature of higgs boson and enables a precise measurement of its mass . for @xmath19 gev , the decay channels @xmath20 and @xmath21 , and ( for even heavier @xmath10 ) @xmath22 become dominant . at the same time , the di - photon branching ratio decreases significantly . as we are assuming that the discovered higgs boson is ( maybe only one of several higgs bosons ) responsible for electroweak symmetry breaking , the most general renormalizable operators involving this higgs boson would be those in the sm . the loop contributions to these couplings from new physics will typically be of the percent level that are too small to be probed at the early stages of the lhc , and we may ignore the loop corrections to these couplings . on the other hand , the couplings @xmath23 and @xmath24 of dimension - five di - gluon and di - photon operators that characterize the gluon - fusion production rate and di - photon width of the higgs boson @xmath25 are one - loop at leading order in the sm , and loop contributions from new physics may be competitive . thus , to study this discovered boson in a bottom - up approach , we consider an effective lagrangian that includes all the renormalizable gauge and yukawa operators as in the sm , but with arbitrary coefficients , and only consider the leading - order effects of these operators . in addition , we include the two dimension - five operators with arbitrary coefficients that parameterizes the leading effects of the yet - undiscovered new physics on the higgs boson , and the measurement of @xmath5 essentially measures the product of @xmath26 ( which is proportional to the production cross section @xmath27 ) and the decay branching ratio @xmath7 . as the di - photon branching ratio is only significant for @xmath28 gev , we will only consider a higgs boson with a mass within this range . since the pdf - induced uncertainty is of the order of 5% , the precision to which these couplings can be measured depends crucially on the uncertainty in the luminosity at the lhc . the precision of which the couplings in our effective lagrangian can be measured at the lhc has been extensively studied @xcite @xcite . from zeppenfeld @xcite , we see that with 100 @xmath2 from both atlas @xcite and cms @xcite , the cross section @xmath5 can be measured to about 10% . this uncertainty is defined as @xmath29 where @xmath30 is the number of signal ( background ) events . we refer the readers to the reference for the numbers of events , backgrounds , and the significance of the signal . with only 10 @xmath2 of data , we naively scale our error by a factor of @xmath31 , and use 30% as the accuracy to which @xmath5 can be measured . thus , the measurement of @xmath5 with 10 @xmath2 at lhc can only distinguish models of new physics from the sm only if it deviates by more than 30% from the sm prediction , and we will see that this can often places stringent constraints on the parameter spaces of new physics models , independent of the lone higgs scenario . the minimal supersymmetric standard model ( mssm ) is the most widely - studied model of new physics both in theory and experiment @xcite . it extends the sm with superpartners that differ in spin by 1/2 from their sm counterparts , and the electroweak scale is stabilized by the presence of these superpartners if they have masses of about 1 tev . it is remarkable that in the mssm the lightest @xmath8-even boson has a mass that is bounded by about 125 gev if the mssm is to solve the hierarchy problem . as decay to @xmath32 and @xmath33 pairs are now kinematically suppressed , the di - photon channel is now the golden channel to search for the lightest @xmath8-even higgs boson . unfortunately , the mssm comes with 105 parameters @xcite and it is impossible to scan through such vast parameter space . we will make some simplifying assumptions that the first two generations of sfermions have mass matrices that are diagonal at the weak scale and the phases of all susy - breaking contributions are zero . these assumptions are consistent with the various flavor - changing experimental constraints , and leave us with a reduced parameter space . with this reduced parameter space , the signal @xmath5 can still vary greatly . for example , in fig . [ fig : mssmplot ] , we use ` hdecay ` @xcite ( which includes the ` feynhiggs ` package @xcite ) to show the deviation of @xmath5 from the sm values for various values of @xmath34 ( the mass of the @xmath8-odd higgs boson in the mssm ) , scanning over the s - top sector parameters @xmath35 where @xmath36 and @xmath37 are respectively the susy - breaking masses of the left- and right - handed s - tops ( the superpartners the top quark ) , and @xmath38 ( where @xmath39 is the top yukawa coupling ) is the coefficient of the trilinear interaction @xmath40 . we note the following points regarding the parameters in the s - top sector ( also see fig . [ fig : s - top - param ] ) . * the scanned range of @xmath41 includes the regions that give the largest mass for the lightest , @xmath8-even higgs boson , which occurs for @xmath42 . while @xmath42 leads a large higgs mass , the mixing in the s - top sector is @xmath43 , so we can have a large higgs mass without having a light s - top with @xmath44 . * to avoid s - tops with negative squared - mass , @xmath41 must satisfy ( for large @xmath45 ) @xmath46 for small @xmath36 and @xmath37 , it may not be possible to scan in the full region between @xmath47 . for small @xmath36 and @xmath37 , our scanned range in @xmath41 is limited requiring having two s - tops with positive masses . * with large @xmath36 and @xmath37 , even though positive s - top masses may allow @xmath48 to be large relative to @xmath49 , large @xmath41 can lead to a negative squared mass for the lightest , @xmath8-even higgs boson . this occurs when we have @xmath50 where @xmath51 is the electroweak scale ( i.e. the vacuum expectation value ) , and our scanned range of @xmath41 does not include such large @xmath41 . for simplicity , we hold all other parameters fixed as @xmath52 where @xmath53 is any slepton ( first two generations squark ) soft mass , @xmath54 the gluino mass , @xmath55 the wino mass , and @xmath56 the chargino mass . the bino mass @xmath57 is determined assuming unified gaugino mass at the grand unification scale @xmath58 gev . as a function of higgs mass in mssm for various values of @xmath34=200 ( top left ) , 300 ( top right ) , 500 ( bottom left ) , and 1000 ( bottom right ) gev . the s - top soft masses @xmath36 and @xmath37 are scanned from 300 gev to 1.5 tev , with @xmath41 scanned in the range of @xmath59 . the other susy - breaking values are fixed as @xmath60 gev , @xmath61 gev , and @xmath62 gev . , title="fig:",width=259 ] as a function of higgs mass in mssm for various values of @xmath34=200 ( top left ) , 300 ( top right ) , 500 ( bottom left ) , and 1000 ( bottom right ) gev . the s - top soft masses @xmath36 and @xmath37 are scanned from 300 gev to 1.5 tev , with @xmath41 scanned in the range of @xmath59 . the other susy - breaking values are fixed as @xmath60 gev , @xmath61 gev , and @xmath62 gev . , title="fig:",width=259 ] + as a function of higgs mass in mssm for various values of @xmath34=200 ( top left ) , 300 ( top right ) , 500 ( bottom left ) , and 1000 ( bottom right ) gev . the s - top soft masses @xmath36 and @xmath37 are scanned from 300 gev to 1.5 tev , with @xmath41 scanned in the range of @xmath59 . the other susy - breaking values are fixed as @xmath60 gev , @xmath61 gev , and @xmath62 gev . , title="fig:",width=259 ] as a function of higgs mass in mssm for various values of @xmath34=200 ( top left ) , 300 ( top right ) , 500 ( bottom left ) , and 1000 ( bottom right ) gev . the s - top soft masses @xmath36 and @xmath37 are scanned from 300 gev to 1.5 tev , with @xmath41 scanned in the range of @xmath59 . the other susy - breaking values are fixed as @xmath60 gev , @xmath61 gev , and @xmath62 gev . , title="fig:",width=259 ] denotes the line with @xmath63 , and the mass of the lightest , @xmath8-even higgs boson is maximized at 1-loop . having too large @xmath41 can lead to a negative s - top mass ( horizontally hashed region ) . however , even if @xmath41 is not large enough to lead to a negative s - top mass , it can be large enough to lead to a negative mass for the lightest , @xmath8-even higgs boson ( vertically hashed region ) . , width=336 ] though there is a general trend of suppression in @xmath5 in fig . [ fig : mssmplot ] , we see that the specific amount of suppression fluctuates with the parameters in the s - top sector as well as @xmath34 . ( we discuss the region where @xmath5 is enhanced in the lower - right plot of fig . [ fig : mssmplot ] at the end of this subsection . ) however , as we eventually will be interested in the lone higgs scenario , the s - fermions and the gauginos should be heavy enough to evade discovery . in particular , having large s - top masses and large @xmath64 leads to a large higgs mass in addition to heavy s - top masses to evade discovery . in fig . [ fig : mssmplot ] , we see that , as @xmath10 increases , the fluctuation in @xmath5 decreases , signaling the decoupling of the s - top sector . we show this more explicitly in fig . [ fig : runst ] where we scan over @xmath65 in the range of @xmath66 holding @xmath36 and @xmath37 fixed . we also fix other parameters as eq . ( [ eq : scannedset ] ) , and fix @xmath67 and @xmath34=1 tev . with @xmath36 and @xmath37 fixed as 1 tev ( 2 tev ) , the fractional deviation in @xmath5 changes by 11% ( 3% ) as we scan over @xmath41 . the plot shows that with larger @xmath68 , the deviation in @xmath5 is less sensitive to the variation in @xmath41 as both the s - tops decouple . we therefore consider a limited lone higgs scenario where the s - top soft masses are large enough that , in addition to evade direct discovery at the lhc , the s - top contributions to the gluon - gluon fusion and di - photon decay amplitudes decouple regardless of the value of @xmath41 . as the s - top contributions decouple , we attribute the suppression in @xmath5 to @xmath34 and @xmath45 in this limited lone higgs scenario . with @xmath41 scanned over the range @xmath69 with @xmath70 ( triangle points ) , and @xmath71 ( square points ) . with larger @xmath68 , the deviation in @xmath5 is less sensitive to the variations in @xmath41 because the s - top sector decouples . here we fix @xmath67 and @xmath72 tev in addition to those labelled in eq . ( [ eq : scannedset ] ) . for each set of s - top sector parameters , there are two arcs on the plot because positive values of @xmath41 gives a larger higgs mass than negative values of @xmath41 at two - loop . , width=408 ] as a function of @xmath34 for @xmath73 , and 30 . the soft masses are fixed to evade lhc discovery as @xmath74 tev , @xmath75 tev , @xmath76 , and @xmath77 . for @xmath78 gev , there is significant deviation in @xmath5 from the sm . , width=408 ] in this work we will restrict our attention to the range @xmath79 , where both the top and bottom yukawa couplings are perturbative and there is no danger for light or negative s - tau or s - bottom masses from @xmath45-enhanced mixing when @xmath80 , where @xmath81 and @xmath82 are respectively the s - bottom and s - tau . to study the effects of @xmath34 and @xmath45 in @xmath5 in the decoupling limit of heavy sfermions and gauginos , in fig . [ fig : mssmtplot ] we show the dependence in @xmath5 as a function of @xmath34 for @xmath83 and 30 , with other parameters fixed as @xmath84 this set of parameters differs from those in eq . [ eq : scannedset ] because we will eventually be interested in the mssm lone higgs scenarios . as we will see in the next subsection , these parameters give sufficiently heavy sfermions and gauginos that are out of the reach at the lhc with 10 fb@xmath0 of integrated luminosity . we note that the dependence in @xmath45 on @xmath5 is small in this range of @xmath45 . in particular , @xmath5 can be suppressed by more than 30% for @xmath85 gev . in the next subsection we will investigate whether this region can be consistent with a lone higgs scenario given the expected lhc reach for the heavy higgs bosons . we conclude this subsection with a brief discussion of the possibility of having an enhancement in @xmath5 in the mssm . in the lower - right plot of fig [ fig : mssmplot ] , in the region with small higgs mass ( @xmath86 100 gev ) we have enhanced @xmath5 that is contrary to the typical trend of suppression we note earlier . this is because the enhancement in the production @xmath87 ( due to light s - tops with small mixing ) compensates for the small suppression in the branching ratio @xmath7 ( due to large @xmath72 tev ) . on the other hand , in the lone higgs scenario we consider here with two heavy s - tops , the production cross - section @xmath87 is only enhanced slightly relative to the sm @xcite , and the suppression in @xmath5 is dominantly due to a suppression in @xmath7 resulting from @xmath45-enhanced @xmath88 and @xmath89 couplings , leading to a larger total decay width of the higgs boson . ( we also note that it is possible to have a suppression in the production cross - section @xmath87 in the mssm , leading to a so - called gluo - phobic higgs , when there is a large hierarchy between the s - tops @xmath90 and a significant mixing in the s - top sector @xcite . ) -@xmath91 plane at cms with an integrated luminosity of 10 @xmath2 , except for the higgs search , which assumes 2 @xmath2.,width=480 ] to investigate the viable lone higgs scenarios in the mssm , we here briefly summarize the known results in the literature regarding the lhc reach of the mssm particles . the summary of mssm discovery potential at the cms experiment at the lhc for the constrained mssm ( cmssm ) is shown in the cms tdr @xcite , and we include it in fig . [ fig : susy_cms ] . we here focus on reaches of several experimental signatures that allow us to place general bounds on the mssm parameter space , and we will simplify the mssm parameter space in terms of four general types of resonances in addition to the higgs boson . these four types are : ( i ) colored superpartners ( gluino and squarks ) , ( ii ) sleptons , ( iii ) neutralinos and charginos , and ( iv ) heavy higgs bosons . we discuss the discovery reach of each in turn . [ [ gluino - and - squarks ] ] gluino and squarks + + + + + + + + + + + + + + + + + + @xmath92 in the cmssm parameterization of the mssm parameters by the set @xmath93 , the colored superpartners are generically heavier than the non - colored superpartners , and the search for susy involves tracking down the cascade decays of pair - produced squarks and/or gluinos . while the nature of the cascades depends on the details of the mssm spectra ( see cms tdr @xcite for the various possibilities ) , the initial step of the cascade always involves emitting at least a quark and all cascades always end with the stable , lightest superpartner ( lsp ) that leaves the detector as missing energy ( assuming that @xmath94-parity is conserved ) . thus , the signature of jets plus missing energy provides the best reach of the mssm space ( line labelled jet+met in fig . [ fig : susy_cms ] ) , and the reach is roughly @xmath95 900 gev , corresponding to a gluino mass of @xmath96 2.4 tev . we also note that this reach is only mildly dependent on @xmath97 , and the squark masses are less constrained by this reach . nevertheless , we make a reasonable simplifying assumption that the reach for squarks is also @xmath982.4 tev . [ [ sleptons ] ] sleptons + + + + + + + + @xmath92moving away from the cmssm parameterization , if a large hierarchy existed between the colored superpartners and the non - colored superpartners such that at the lhc the gluino and squarks can not be directly produced at the lhc , then susy searches involve the production channels of non - colored superpartners , such as chargino - neutralino associated production and slepton pair - productions . while the right - handed slepton ( superpartner of the right - handed lepton ) can only decay to the lsp directly in cmssm , the left - handed slepton can decay to @xmath99 and @xmath100 if kinematically allowed ( depending on @xmath91 ) . in both cases of direct slepton pair production and indirect slepton production ( through the decays of neutralinos and charginos from associated chargino - neutralino associated production ) , there are always at least two leptons from the decays of the sleptons . furthermore , in the case of direct slepton pair - production , the two leptons must have opposite sign . the signature of the direct slepton pair production and indirect production then involves two leptons , missing energy , and jet veto , and the exclusion plot in the reference is reproduced in fig . [ fig : slepton - bound ] ( see chapter 13 , section 15 of the cms tdr @xcite for details on the cuts and significance of the reach ) . for large @xmath45 , the largest slepton mass is roughly @xmath101 , and using @xmath102 gev from fig . [ fig : slepton - bound ] , we obtain @xmath103 gev . for further references , there are detailed studies of slepton searches in the literature @xcite@xcite@xcite . -@xmath91 plane with @xmath67 , @xmath104 , and the sign @xmath56 is positive . the experimental signatures of the events are two leptons , missing energy , and no jets . , width=384 ] [ [ neutralinos - and - charginos ] ] neutralinos and charginos + + + + + + + + + + + + + + + + + + + + + + + + + @xmath92the reach of neutralino(@xmath105)/chargino(@xmath106 ) sector involves a tri - lepton signature that arises from @xmath107 with @xmath108 ( through a slepton @xmath109 , which may be real or virtual ) and @xmath110 . the lepton pair from the decay of @xmath111 will be of the same flavor , opposite sign ( sfos ) , while the lepton from @xmath112 can be any flavor . in addition to the three leptons and missing energy , no jets participate in this process and a jet veto can be used ( see chapter 13 , section 14 of the cms tdr @xcite for details ) , and the reach is presented in the bottom blue curve in fig . [ fig : susy_cms ] denoted as `` trilept '' . as with the case of the gluino / squark reach , we see that the curve is only mildly dependent on @xmath113 and has a value of @xmath114 gev for @xmath115 gev , and @xmath91 fluctuates between 100 gev and 200 gev for @xmath116 gev . the wino ( bino ) mass is related to @xmath97 in cmssm by @xmath117 ( @xmath118 ) . thus , taking the conservative reach of @xmath95 200 gev gives @xmath119 gev and @xmath120 gev . to obtain the full spectra in the neutralino / chargino sector requires the knowledge of the higgsino mass @xmath56 . however , assuming that @xmath121 and that @xmath56 is not nearly degenerate with @xmath122 , we do not expect large mixing in the neutralino / chargino sector , and we can approximate @xmath123 as the masses of @xmath124 [ [ higgs - bosons ] ] higgs bosons + + + + + + + + + + + + @xmath92because the variation in @xmath5 in the mssm has strong dependence on @xmath34 in the decoupling limit of heavy s - fermions and gauginos , the lhc reaches of the heavy higgs bosons play important roles in the determination of whether we have a lone higgs scenario with a large deviation in @xmath5 . the reach of the mssm heavy higgs bosons has been studied in the literature @xcite@xcite@xcite , and we have taken figures from these references in figs . [ fig : chargehiggs ] and [ fig : neutralhiggs ] . although in this work we focus on the lone higgs scenario with 10 fb@xmath0 of lhc data , we could only find discovery reach for the heavy higgs bosons with 30 fb@xmath0 of data . we will assume that the reach of the mssm heavy higgs bosons are further with 30 fb@xmath0 of data than with 10 fb@xmath0 , and conservatively apply the discovery reach of the heavy mssm higgs bosons at 30 fb@xmath0 to our 10 fb@xmath0 lone higgs scenario . the discovery reach of the charged higgs bosons @xmath125 is presented fig . [ fig : chargehiggs ] , with @xmath126 , where @xmath125 is either produced through the decay of a top quark @xmath127 produced in @xmath22 production if @xmath128 , or , if @xmath129 , produced through a virtual @xmath130 via @xmath131 in a @xmath132 production . the charged higgs boson then decays via @xmath133 , and as the coupling of @xmath134 is @xmath45-enhanced , stronger bounds can be placed with larger @xmath45 . however , for @xmath135 , the reach on @xmath136 is less than @xmath137 gev with cms , while atlas does not set a bound on @xmath34 for @xmath138 . thus , the discovery reach for the charged higgs bosons does not impose a severe constraint on @xmath34 . the discovery reaches of the neutral higgs bosons @xmath139 , and @xmath140 at cms and atlas are presented fig . [ fig : neutralhiggs ] using various processes as indicated on the plots . as with the charged higgs , these searches depend on @xmath45-enhanced couplings . for a given value of @xmath34 , these heavy higgs boson discovery reaches limit the lone higgs scenario with an upper bound on @xmath45 . the strongest discovery reach for the neutral higgs boson comes from the atlas search involving @xmath141 as shown in fig . [ fig : neutralhiggs](c ) , which places an upper bound of @xmath142 if we are to have @xmath143 in a lone higgs scenario . for larger values of @xmath45 , we would need larger values of @xmath34 to have viable lone higgs scenarios , and it is difficult to distinguish the lone higgs scenario from the sm by the suppression in @xmath5 . with 300 fb@xmath0 of integrated luminosity , the lhc can also discover the @xmath140 if @xmath45 is small ( @xmath144 ) , as shown in fig . [ fig : neutralhiggs](c ) , through the process @xmath145 , whose rate is significantly larger than that of the sm higgs boson @xmath146 . as the bounds from atlas are more stringent , we will mainly use fig . [ fig : neutralhiggs](c ) to set bounds on the lone higgs scenario in the mssm in the next section . ( for details , see the atlas tdr@xcite , the cms tdr @xcite , gennai et al . @xcite , and references therein . ) ( a ) is taken from the cms tdr @xcite , fig . [ fig : neutralhiggs](b ) from gennai et al . @xcite , and figs . [ fig : neutralhiggs](c ) and ( d ) from atlas tdr @xcite . these plots summarizes the cms and atlas reach of heavy neutral higgs bosons in various channels , as indicated on the plots . [ fig : neutralhiggs](c ) sets the strongest constraints on having a lone higgs scenario with large deviations in @xmath5 . , title="fig:",width=300 ] ( a ) is taken from the cms tdr @xcite , fig . [ fig : neutralhiggs](b ) from gennai et al . @xcite , and figs . [ fig : neutralhiggs](c ) and ( d ) from atlas tdr @xcite . these plots summarizes the cms and atlas reach of heavy neutral higgs bosons in various channels , as indicated on the plots . [ fig : neutralhiggs](c ) sets the strongest constraints on having a lone higgs scenario with large deviations in @xmath5 . , title="fig:",width=300 ] ( a ) is taken from the cms tdr @xcite , fig . [ fig : neutralhiggs](b ) from gennai et al . @xcite , and figs . [ fig : neutralhiggs](c ) and ( d ) from atlas tdr @xcite . these plots summarizes the cms and atlas reach of heavy neutral higgs bosons in various channels , as indicated on the plots . [ fig : neutralhiggs](c ) sets the strongest constraints on having a lone higgs scenario with large deviations in @xmath5 . , title="fig:",width=288 ] ( a ) is taken from the cms tdr @xcite , fig . [ fig : neutralhiggs](b ) from gennai et al . @xcite , and figs . [ fig : neutralhiggs](c ) and ( d ) from atlas tdr @xcite . these plots summarizes the cms and atlas reach of heavy neutral higgs bosons in various channels , as indicated on the plots . [ fig : neutralhiggs](c ) sets the strongest constraints on having a lone higgs scenario with large deviations in @xmath5 . , title="fig:",width=288 ] [ [ possibilities - of - a - lone - higgs - scenario - in - mssm ] ] possibilities of a lone higgs scenario in mssm + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @xmath92 .summary of lhc reach for different types of superpartners of the mssm with 10 @xmath2 of integrated luminosity . the reach of the heavy higgs bosons involve 30 @xmath2 of integrated luminosity . the assumptions we make are in parenthesis . [ cols="<,<,<",options="header " , ] stepping towards the lone higgs scenario , we impose the conditions that the higgs is lighter than 130 gev and no additional new resonances has been directly seen at the lhc after 10 @xmath2 of operation . this is presented in table [ tb : first - step ] . as noted earlier , decoupling the kk tops and gauge bosons in mued such that they are not directly accessible at the lhc after 10 @xmath2 necessarily implies that the resulting higgs boson would have a signal that is similar to the sm . in lht with @xmath147 gev , the deviation of @xmath5 is only more than 30% in the limited ranges of @xmath148 560 gev and @xmath149 gev , and this region of parameter space leads to a viable lone higgs scenario only for @xmath150 . the mssm can also lead to a viable lone higgs scenario with large deviation in @xmath5 , for example , with @xmath151 gev and @xmath67 . although we noted earlier that , relative to the sm , the signal @xmath5 is always enhanced in mued and suppressed in lht , in a lone higgs scenario we would not expect mued to show a significant enhancement . if we see a suppression in the signal , we have to distinguish between lht and the mssm . before we attempt to distinguish between the lht and the mssm based on large deviations in @xmath5 alone , we first address what would happen if the lhc observes a higgs boson with a deviation in @xmath5 that is less than 30% . suppose that the lhc finds a higgs boson after 10 @xmath2 whose @xmath5 measurement deviates less than 30% from the expected sm value . though we can not directly distinguish between the three new models of physics from the sm , our results may nonetheless be useful in devising further tests to distinguish the models , for example , with consistency checks . broadly speaking , if the deviation of the signal is more than the eventual accuracy of 10% , then it may be worthwhile to pursue this deviation as both a lead for consistency checks and a bias for planning search strategies / signals that are more optimized to one particular model of new physics . that is , we are placing more confidence in the measurement than its uncertainty warrants , in the hope that subsequent measurements would improve the uncertainty without a great shift in the measured value . for example , suppose that with 10 @xmath2 of lhc data , the measured value of @xmath5 shows a positive 20% deviation from the sm with a 30% uncertainty . while this is not a significant deviation from the sm and this deviation is likely to fluctuate ( and may even change its sign ) with subsequent measurements , we can use this result to favor mued over lht and the mssm , since neither of these models can give an enhancement in @xmath5 . as such a deviation corresponds to @xmath152 in the range of 800 to 900 gev and the masses of the kk tops and kk gluons of about 1 tev , we may use this deviation as a consistency check to rule out the mued model if none of the heavy kk particle states is found , and seek other explanations for the enhancement in the higgs signal . as another example , suppose that we see a 20% suppression of the signal from the sm with 10 @xmath2 of lhc data , we can then favor lht and the mssm over mued , and optimize our search strategies in order to both find additional expected resonances in these models as well as devise tests that may distinguish these two models when we find these resonances . to be more concrete , let us also suppose that the higgs mass is measured to be 125 gev . such higgs mass is rather large in the scope of the mssm , and is consistent with heavy s - tops , with perhaps significant mixing that evade direct discovery . if we favor the assumption nature is supersymmetric , from fig . [ fig : mssmtplot ] , we see that a suppression of 20% in @xmath5 corresponds to @xmath153 gev . on the other hand , in the context of lht , from the fig . [ fig : lhtplot ] , we see that such deviation correspond to a rather low value of @xmath154 gev . these pieces of information ( @xmath153 gev or @xmath154 gev ) can then be used to devise search strategies that can differentiate mssm from lht . if the lhc finds a lone higgs boson with @xmath5 that deviates from the sm value by more than 30% , then the prospects are very exciting . first , in the case of an enhancement in the measurement , we can strongly disfavor all three models of new physics under considerations here , and seek alternative explanations for such an enhancement . neither the mssm nor the lht can give a enhancement in @xmath5 relative to the sm in a lone higgs scenario , and this large enhancement can not be consistent with mued because in that case some light kk resonances must be produced and detected at the lhc . ( by definition , lone higgs scenario requires that no other resonances being detected . ) in the case of a suppression , we would strongly favor mssm and lht over mued and need to distinguish these two models based on the measured value of @xmath5 . as a consistency check with both mssm and lht , the higgs mass should be around @xmath155 gev to be consistent with both the mssm ( because the s - tops need to be heavy to evade discovery ) and lht ( because of the large deviation in @xmath5 ) . a possible first discriminating test between mssm and lht is the amount of deviation in @xmath5 . from fig . [ fig : lhtplot ] , the lht with @xmath156 gev can not give a suppression in @xmath5 of more than 40% , while the mssm with @xmath157 can accommodate a suppression larger than 40% with @xmath158 gev as shown in fig . [ fig : mssmtplot ] . therefore , a lone higgs scenario with a suppression in @xmath5 of more than 40% as compared to the sm can only be consistent with the mssm . ( as discussed in section [ sec : mssm - hgg ] , if we impose minimal flavor violation , the 3@xmath159 uncertainties in the measurement of @xmath160 imposes @xmath161 gev , and the suppression in @xmath5 can not be more than 55% relative to the sm . ) while the suppression between 30% and 40% in @xmath5 relative to the sm prediction can be consistent with both the mssm and the lht , we propose possible tests that can distinguish these two models with more integrated luminosity in the next subsection . given a lone higgs scenario with a measured suppression in @xmath5 that is greater than 30% relative to the sm , we note in the previous subsection that * if the suppression is greater than 40% relative to the sm , then mssm is strongly favored , and , * if the suppression in @xmath5 relative to the sm is between 30% and 40% , then both the mssm and the lht are viable . in this subsection , we discuss how we may distinguish the mssm from the lht when the suppression in @xmath5 relative to the sm is between 30% and 40% with further lhc operations and seeking new resonances beyond 10 fb@xmath0 first we consider the case that the lht is responsible for the observed suppression in @xmath5 relative to the sm . in this case , a suppression of 30% or more in @xmath5 relative to the sm prediction corresponds to @xmath162 gev . for @xmath163 gev , the viable range of @xmath164 is @xmath165 , and this viable range of @xmath164 becomes even smaller with smaller @xmath166 . for example , with @xmath167 gev , the viable range of @xmath168 is @xmath169 . with more luminosity , the lhc should discover either @xmath170 and/or @xmath171 because this viable range on @xmath168 disappears , and we no longer have a lone higgs scenario . for example , with @xmath163 gev and 100 fb@xmath0 of data , the lhc can discover @xmath171 if @xmath172 ( see fig . [ fig : lht - potential](b ) ) and discover @xmath170 if @xmath173 ( see fig . [ fig : lhtf ] ) . these two constraints of @xmath164 overlap , so at least one of @xmath171 or @xmath170 should be discovered at 100 fb@xmath0 , and if @xmath168 is between @xmath174 , then the lhc should discover both @xmath175 and @xmath170 with 100 fb@xmath0 . if we assume that the mssm is the model of new physics that underlies the observed lone higgs scenario with a suppression in @xmath5 between 30% and 40% , then fig . [ fig : mssmtplot ] tells us that @xmath176 , and fig . [ fig : neutralhiggs](c ) gives us the viable region on @xmath177 plane for a lone higgs scenario . from fig . [ fig : neutralhiggs](c ) , we also see , in some of the viable region , with higher luminosity the lhc has the potential to discover the heavy mssm higgs bosons @xmath178 in the @xmath179 mode . unfortunately , unlike the case of the lht , we are not guaranteed to discover additional resonances at the lhc with more luminosity . for example , with @xmath67 , @xmath151 gev , and 10 fb@xmath0 of lhc data , we have a lone higgs scenario with a significant suppression in @xmath5 compared to the sm prediction , but we are outside the lhc reach for the heavy mssm bosons @xmath178 even at 300 fb@xmath0 of integrated luminosity . we can also use the discovery of the heavy higgs bosons @xmath178 as a consistency check of mssm for a lone higgs scenario with a suppression in @xmath5 of more than 40% relative to the sm . suppose we have a lone higgs scenario with a measurement of @xmath5 that is suppressed by more than 40% relative to the sm . from fig . [ fig : mssmtplot ] , we see that such suppression corresponds to @xmath180 gev if @xmath181 . in fig . [ fig : lhs - mssm - bigdev ] , we superimpose onto fig . [ fig : neutralhiggs ] with the contour that corresponds to having a 40% suppression in @xmath5 relative to the sm and mark the viable region of parameter space . we see that in most of this parameter space , we have the potential to discover the heavy mssm neutral higgs boson @xmath178 when the integrated luminosity is increased to 100 fb@xmath0 if @xmath45 is sufficiently large . and with 300 fb@xmath0 of integrated luminosity , most of this parameter space is covered . in the atlas tdr ( fig . [ fig : neutralhiggs](c ) ) with the upper bounds on @xmath34 from a suppression of 40% in @xmath5 ( the negatively - sloped line ) . the hashed region below the intersection is the viable lone higgs scenario region with a suppression in @xmath5 of at least 40% . we assume @xmath182 throughout our analysis . , width=288 ] so far , we have attempted to use the lone higgs scenario and a large deviation in the measurement of @xmath5 to discriminate between mued , lht , and the mssm . for some states in these models of new physics , such as the s - leptons of the mssm , we can obtain viable lone higgs scenarios by simply raising the masses of the new states large enough to evade discovery without affecting significantly the phenomenology of the discovered boson . it is possible that these states are discovered and identified only after 10 @xmath2 of integrated luminosity , and these situations can certainty help us to further discriminate the models of new physics . in this section we give some more tests of this type that can discriminate between mued , lht , and the mssm , independent of the lone higgs scenario . ( some of these tests have been stated in the literature . ) since the mssm is the most extensively studied model of new physics , and it has arguably the most complicated collider phenomenology of the three models that we study in this work , we present tests that may separate lht and mued from the mssm rather than identify mssm directly . in the mssm , the cancellation of quadratic divergence introduces new states that differ in spin and must be pair - produced due to @xmath94-parity . in lht , the new physics that modifies the properties of the higgs boson come from the top - partners ( @xmath1-parity even or odd ) and the @xmath1-odd gauge bosons ( @xmath184 ) . the discovery potentials of the @xmath1-odd gauge bosons and top - partners were summarized in section [ sec : lht - lh ] , and it was pointed out that , while the masses of the top - partners depend on additional inputs @xmath185 ( with the top quark mass as one constraint ) , the top - partners contributions to @xmath5 is independent of these parameters . although we could raise the masses of both top - partners to evade lhc discovery at 10 @xmath2 , naturalness arguments favor light top - partners and it is not implausible that we discover the top - partners after 10 @xmath2 of operation . in naturalness arguments , the role of @xmath183 in lht is played by the s - tops @xmath186 in the mssm to cancel the quadratic divergences in the higgs boson self energy , and , assuming we see new resonances related to an extended top - sector , we would like to distinguish @xmath183 and @xmath186 . one potential signature that can be used to distinguish lht model from both the mssm and mued is the single - production of the @xmath1-even top - partner ( @xmath183 ) @xcite . the production mechanism of such signature is similar to the sm single - top production , with the top quark replaced by @xmath183 . as new colored states in both the mssm and mued must be pair produced , this can be used to distinguish the lht from the mssm and mued . furthermore , in the lht the single - top production rate is suppressed as compared to the sm rate , and a precise measurement in single - top production rate can give us information about the top sector in lht @xcite . as stated earlier , the additional particle content in the mued is very similar to that of the mssm : the kk - modes are partners to the sm particle content in mued just as the super - partners in the mssm . where as the gauginos obtain susy - breaking mass in the mssm , in mued models , the @xmath187 kk mode of the gauge bosons ( @xmath188 ) become massive through a ` geometric ' higgs mechanism : they eat linear combinations of ( dominantly ) the @xmath187 fifth - dimensional component of the gauge bosons ( @xmath189 ) and ( sub - dominantly ) the @xmath187 mode of the higgs boson @xmath190 . at the first kk level , there are then four physical scalar bosons from the un - eaten combinations that are dominantly the first kk mode of the higgs boson : @xmath191 , @xmath192 $ ] , and @xmath193 $ ] . furthermore , in the limit that @xmath194 and if we assume no significant radiative corrections to the higgs masses , these higgs bosons will have similar masses with fractional degeneracies of the order @xmath195 . effectively , the mued also contains a two higgs doublets just as the mssm : the zeroth and first kk - level higgs doublets . the higgs sector in the first kk level of mued models is similar to the mssm heavy higgs sector in many aspects : * they both contain two electrically neutral ( one @xmath8-even and the other @xmath8-odd ) higgs bosons , and a @xmath196 charged higgs boson , * these states are nearly degenerate : the fractional degeneracies in the masses are of the order @xmath197 in the mssm , and @xmath195 in mued . however , there is one crucial difference : the heavy higgs bosons in mssm are @xmath94-even and can be singly produced at the lhc , whereas these kk higgs are @xmath198-odd and must be pair produced . unfortunately , as the lhc discovery potential for the singly - produced heavy higgs bosons in the mssm do not place stringent constraints on parameter space ( this allows us to have a lone higgs scenario with a large deviation in @xmath5 ) , discovering new physics through pair - produced higgs bosons is much more difficult . nevertheless , we point out this crucial difference in the higgs sector in the mssm and mued since , as far as we know , it had not been explicitly stated in the literature . furthermore , it could generate new signal event signatures outside the context of lone higgs scenario . in this work we attempt to distill traces of physics beyond the sm in the scenario that the lhc only discovers a higgs boson after its initial years of operation with 10 @xmath2 of data . we focus on the scenario of light higgs boson and used the signal @xmath5 to distinguish between lht , mued , and the mssm . for simplicity , in the mssm we consider a limited lone higgs scenario where the s - top soft masses are large enough that , in addition to evade direct discovery at the lhc , the s - top contributions to the gluon - gluon fusion and di - photon decay amplitudes decouple regardless of the value of @xmath41 . given the expected accuracies of the measurement of the higgs - di - gluon and higgs - di - photon couplings at this stage of operation of lhc , and the implications of lone higgs scenario on the spectra of new physics , in mued it is difficult to have significant deviations from the sm . in the cases of the lht and the mssm , however , it is possible to have a significant suppression in the signal while discovering a lone higgs boson , and such lone higgs scenario may give rise to a new resonance ( @xmath178 in the mssm , and @xmath199 or @xmath170 in the lht ) with more luminosity . in the case where the measured deviation is very large , we offer tests that may potentially separate lht from the mssm even before a new resonance is discovered . in case where the deviation is small , our work may nevertheless be useful and we point out that the deviations can be used to bias a class of model from another , and as such can be useful in devising search strategies of the favored new physics . although we have only worked out detailed strategies for 10 fb@xmath0 integrated luminosity in the case of lone higgs scenario , similar kinds of considerations could also apply to various stages of lhc running . we hope to have demonstrated that even with the lone higgs scenario , many insights can still be learned before the end of the lhc era . this work is supported by the us national science foundation under grants phy-0354226 , phy-0555545 , and phy-0354838 . we thank chuan - ren chen for carefully proofreading this preprint . kh would like to thank r. sekhar chivukula , elizabeth h. simmons and neil d. christensen for stimulating discussions . we would also like to thank michael spira for his assistance with ` hdecay ` . we thank qing - hong cao , debajyoti choudhury , and sven heinemeyer for granting us permission to use the figures of their works . n. arkani - hamed , g. l. kane , j. thaler and l. t. wang , jhep * 0608 * , 070 ( 2006 ) [ arxiv : hep - ph/0512190 ] . s. mantry , m. j. ramsey - musolf and m. trott , phys . b * 660 * , 54 ( 2008 ) [ arxiv:0707.3152 [ hep - ph ] ] . s. mantry , m. trott and m. b. wise , phys . d * 77 * , 013006 ( 2008 ) [ arxiv:0709.1505 [ hep - ph ] ] . l. randall , jhep * 0802 * , 084 ( 2008 ) [ arxiv:0711.4360 [ hep - ph ] ] . atlas detector and physics performance . technical design report . vol . 2 . cern / lhcc99 - 15 ( 1999 ) . g. l. bayatian _ et al . _ [ cms collaboration ] , j. phys . g * 34 * , 995 ( 2007 ) . m. nadolsky _ et al . _ , arxiv:0802.0007 [ hep - ph ] . d. zeppenfeld , r. kinnunen , a. nikitenko and e. richter - was , phys . d * 62 * , 013009 ( 2000 ) [ arxiv : hep - ph/0002036 ] . m. duhrssen , s. heinemeyer , h. logan , d. rainwater , g. weiglein and d. zeppenfeld , phys . rev . d * 70 * , 113009 ( 2004 ) [ arxiv : hep - ph/0406323 ] . s. p. martin , arxiv : hep - ph/9709356 . a. djouadi , j. kalinowski and m. spira , comput . commun . * 108 * , 56 ( 1998 ) [ arxiv : hep - ph/9704448 ] . s. heinemeyer , w. hollik and g. weiglein , comput . commun . * 124 * , 76 ( 2000 ) [ arxiv : hep - ph/9812320 ] . r. v. harlander and m. steinhauser , jhep * 0409 * , 066 ( 2004 ) [ arxiv : hep - ph/0409010 ] . f. del aguila and l. ametller , phys . b * 261 * , 326 ( 1991 ) . h. baer , c. h. chen , f. paige and x. tata , phys . d * 49 * , 3283 ( 1994 ) [ arxiv : hep - ph/9311248 ] . h. baer , c. h. chen , f. paige and x. tata , phys . d * 52 * , 2746 ( 1995 ) [ arxiv : hep - ph/9503271 ] . s. gennai , s. heinemeyer , a. kalinowski , r. kinnunen , s. lehti , a. nikitenko and g. weiglein , eur . j. c * 52 * , 383 ( 2007 ) [ arxiv:0704.0619 [ hep - ph ] ] . e. barberio _ et al . _ [ heavy flavor averaging group ( hfag ) ] , arxiv : hep - ex/0603003 . g. dambrosio , g. f. giudice , g. isidori and a. strumia , nucl . b * 645 * , 155 ( 2002 ) [ arxiv : hep - ph/0207036 ] . g. degrassi , p. gambino and p. slavich , arxiv:0712.3265 [ hep - ph ] . b. c. allanach , comput . commun . * 143 * , 305 ( 2002 ) [ arxiv : hep - ph/0104145 ] . n. arkani - hamed , a. g. cohen , e. katz and a. e. nelson , jhep * 0207 * , 034 ( 2002 ) [ arxiv : hep - ph/0206021 ] . h. c. cheng and i. low , jhep * 0309 * , 051 ( 2003 ) [ arxiv : hep - ph/0308199 ] . h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ arxiv : hep - ph/0405243 ] . i. low , jhep * 0410 * , 067 ( 2004 ) [ arxiv : hep - ph/0409025 ] . n. arkani - hamed , a. g. cohen , e. katz , a. e. nelson , t. gregoire and j. g. wacker , jhep * 0208 * , 021 ( 2002 ) [ arxiv : hep - ph/0206020 ] . for reviews of little higgs models , see refs . @xcite and @xcite , and references therein . t. han , h. e. logan , b. mcelrath and l. t. wang , phys . d * 67 * , 095004 ( 2003 ) [ arxiv : hep - ph/0301040 ] . m. schmaltz and d. tucker - smith , ann . nucl . part . sci . * 55 * , 229 ( 2005 ) [ arxiv : hep - ph/0502182 ] . j. hubisz , p. meade , a. noble and m. perelstein , jhep * 0601 * , 135 ( 2006 ) [ arxiv : hep - ph/0506042 ] . c. r. chen , k. tobe and c. p. yuan , phys . b * 640 * , 263 ( 2006 ) [ arxiv : hep - ph/0602211 ] . q. h. cao and c. r. chen , phys . d * 76 * , 075007 ( 2007 ) [ arxiv:0707.0877 [ hep - ph ] ] . j. hubisz and p. meade , phys . d * 71 * , 035016 ( 2005 ) [ arxiv : hep - ph/0411264 ] . m. s. carena , j. hubisz , m. perelstein and p. verdier , phys . d * 75 * , 091701 ( 2007 ) [ arxiv : hep - ph/0610156 ] . d. choudhury and d. k. ghosh , jhep * 0708 * , 084 ( 2007 ) [ arxiv : hep - ph/0612299 ] . s. matsumoto , m. m. nojiri and d. nomura , phys . d * 75 * , 055006 ( 2007 ) [ arxiv : hep - ph/0612249 ] . t. appelquist , h. c. cheng and b. a. dobrescu , phys . d * 64 * , 035002 ( 2001 ) [ arxiv : hep - ph/0012100 ] . i. antoniadis , phys . b * 246 * , 377 ( 1990 ) . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 436 * , 257 ( 1998 ) [ arxiv : hep - ph/9804398 ] . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * , 263 ( 1998 ) [ arxiv : hep - ph/9803315 ] . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . d * 59 * , 086004 ( 1999 ) [ arxiv : hep - ph/9807344 ] . h. georgi , a. k. grant and g. hailu , phys . b * 506 * , 207 ( 2001 ) [ arxiv : hep - ph/0012379 ] . h. c. cheng , k. t. matchev and m. schmaltz , phys . d * 66 * , 036005 ( 2002 ) [ arxiv : hep - ph/0204342 ] . h. c. cheng , j. l. feng and k. t. matchev , phys . lett . * 89 * , 211301 ( 2002 ) [ arxiv : hep - ph/0207125 ] . g. servant and t. m. p. tait , nucl . b * 650 * , 391 ( 2003 ) [ arxiv : hep - ph/0206071 ] . g. servant and t. m. p. tait , new j. phys . * 4 * , 99 ( 2002 ) [ arxiv : hep - ph/0209262 ] . k. kong and k. t. matchev , jhep * 0601 * , 038 ( 2006 ) [ arxiv : hep - ph/0509119 ] . f. burnell and g. d. kribs , phys . d * 73 * , 015001 ( 2006 ) [ arxiv : hep - ph/0509118 ] . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 170 * , 377 ( 2007 ) [ arxiv : astro - ph/0603449 ] . d. hooper and s. profumo , phys . rept . * 453 * , 29 ( 2007 ) [ arxiv : hep - ph/0701197 ] . f. j. petriello , jhep * 0205 * , 003 ( 2002 ) [ arxiv : hep - ph/0204067 ] . q. h. cao , c. s. li and c. p. yuan , arxiv : hep - ph/0612243 .

we consider polynomials orthogonal with respect to the weight on , where is the bessel function of order . asheim and huybrechs considered these polynomials in connection with complex gaussian quadrature for oscillatory integrals . they observed that the zeros are complex and accumulate as near the vertical line . we prove this fact for the case from strong asymptotic formulas that we derive for the polynomials in the complex plane . our main tool is the riemann - hilbert problem for orthogonal polynomials , suitably modified to cover the present situation , and the deift - zhou steepest descent method . a major part of the work is devoted to the construction of a local parametrix at the origin , for which we give an existence proof that only works for .

1406.0969

in this paper we are interested in the polynomials @xmath0 that are orthogonal with respect to the weight function @xmath1 on @xmath2 , where @xmath1 is the bessel function of order @xmath8 . the bessel function is oscillatory with an amplitude that decays like @xmath9 as @xmath10 , and therefore the moments @xmath11 do not exist . it follows that the polynomials @xmath0 can not be defined by the usual orthogonality property @xmath12 asheim and huybrechs @xcite introduced the polynomials @xmath0 via a regularization of the weight with an exponential factor . for each @xmath13 , they consider the monic polynomial @xmath14 of degree @xmath15 that is orthogonal with respect to the weight function @xmath16 , in the following sense : @xmath17 and they take the limit @xmath18 provided that the limit exists . since the weight function @xmath19 changes sign on the positive real axis , there is actually no guarantee for existence or uniqueness of @xmath14 . for the limit we therefore also have to assume that @xmath14 exists and is unique for @xmath15 large enough . the polynomials @xmath0 can alternatively be defined by the moments , since the limiting moments for the bessel function of order @xmath8 are known , namely @xmath20 see @xcite . thus we have the determinantal formula ( which is familiar from the general theory of orthogonal polynomials ) @xmath21 with a hankel determinant @xmath22_{i , j=0}^{n-1}$ ] . the polynomial @xmath0 thus exists if and only if @xmath23 . asheim and huybrechs @xcite analyze gaussian quadrature rules with oscillatory weight functions , such as complex exponentials , airy and bessel functions . the nodes for the gaussian quadrature rule are the zeros of the orthogonal polynomials . since the weight is not real and positive on the interval of orthogonality there is a problem of existence and uniqueness of the orthogonal polynomials . in addition , even when the orthogonal polynomial exists , its zeros may not be real , and they may distribute themselves on some curve or union of curves in the complex plane as the degree tends to infinity . examples of this kind of behavior are known in the literature , for instance with laguerre or jacobi polynomials with non standard parameters , see @xcite , @xcite and @xcite , and for complex exponentials @xcite . nu_025.pdf nu_05.pdf nu_08.pdf nu_13.pdf in the present case , with orthogonality defined as , it was shown in ( * ? ? ? * theorem 3.5 ) that the zeros of @xmath0 are on the imaginary axis in case @xmath24 and @xmath15 is even . namely , if @xmath25 are the zeros of the orthogonal polynomial of degree @xmath26 ( where @xmath15 is even ) with respect to the positive weight @xmath27 on @xmath2 , then the zeros of @xmath0 are @xmath28 . here @xmath29 is the modified bessel function of the second kind . for @xmath30 the zeros of @xmath0 are not on the imaginary axis , as is clear from the illustrations given in @xcite , see also the figures [ fig : points_plot_small ] and [ fig : points_plot_big ] . the computations have been carried out in maple , using extended precision . from these numerical experiments asheim and huybrechs @xcite concluded that the zeros seem to cluster along the vertical line @xmath5 . more precisely , for @xmath31 , one sees in figure [ fig : points_plot_small ] that the vast majority of zeros are near a vertical line , which is indeed at @xmath5 . for @xmath32 one sees in figure [ fig : points_plot_big ] that the zeros with large imaginary part are close to the vertical line @xmath5 , although they are not as close to the vertical line as the zeros in figure [ fig : points_plot_small ] . we were intrigued by these figures and the aim of this paper is to give a partial explanation of the observed behavior of zeros . we are able to analyze the polynomials @xmath0 when @xmath33 in the large @xmath15 limit by means of a riemann - hilbert analysis . the result is that we indeed find that the real parts of most of the zeros tend to @xmath34 as @xmath4 . we are not able to handle the case @xmath32 , since in this case our method to construct a local parametrix at the origin fails . this difficulty may very well be related to the different behavior of the zeros in the case @xmath32 . it would be very interesting to analyze this case as well . from the figures it seems that there is a limiting curve for the scaled zeros , if we divide the imaginary parts of the zeros by @xmath15 and keeping the real parts fixed . this limiting curve is a vertical line segment if @xmath31 ( this will follow from our results ) , but we do not know the nature of this curve if @xmath32 . our first result is about the weak limit of zeros . [ th0 ] let @xmath35 . then the polynomials @xmath0 exist for @xmath15 large enough . in addition , the zeros of @xmath36 all tend to the interval @xmath37 $ ] and have the limiting density @xmath38.\ ] ] the convergence of zeros to the limiting density is in the sense of weak convergence of normalized zero counting measures . this means that if @xmath39 denote the @xmath15 zeros of @xmath0 , then @xmath40 in the sense of weak@xmath41 convergence of probability measures . equivalently , we have @xmath42 for every function @xmath43 that is defined and continuous in a neighborhood of @xmath37 $ ] in the complex plane . the weak limit of zeros , if we rescale them by a factor @xmath44 , exists and does not depend on the value of @xmath3 . theorem [ th0 ] is known to hold for @xmath45 , and we believe that it also holds true for @xmath32 . regarding the real parts of the zeros of @xmath0 as @xmath46 , we have the following result . [ th2 ] let @xmath47 , and let @xmath48 be fixed . then there exist @xmath49 and @xmath50 such that for @xmath51 , every zero @xmath52 of @xmath0 outside the disks @xmath53 and @xmath54 satisfies @xmath55 where @xmath56 for each fixed @xmath57 there are approximately @xmath58 zeros of @xmath0 in the disks @xmath53 and @xmath59 as @xmath15 is large , where @xmath60 this is a consequence of the weak convergence of zeros , see theorem [ th0 ] . clearly , @xmath61 as @xmath62 , and so it follows from theorem [ th2 ] by taking @xmath63 arbitrarily small that for all but @xmath64 zeros one has that the real part tends to @xmath34 as @xmath4 . we do not have information about the zeros in the disk @xmath65 . in our riemann - hilbert analysis we prove the existence of a local parametrix around the origin , but we do not have an explicit construction with special functions . therefore we can not analyze the zeros near the origin . on the other hand , we do have potential access to the extreme zeros in the disks @xmath66 since the asymptotics of the polynomials @xmath36 is given in terms of airy functions . from the figures it seems that the result also holds for the extreme zeros , but we omit this asymptotic result in theorem [ th1 ] , since it does not follow clearly from the construction of the local parametrices in this case . theorems [ th0 ] and [ th2 ] follow from strong asymptotic formulas for the rescaled polynomials @xmath67 these polynomials are orthogonal polynomials on the real line , but with a complex weight function . [ prop : pntildeorthogonal ] let @xmath68 . then the polynomial @xmath69 is the monic orthogonal polynomial of degree @xmath15 for the weight @xmath70 on the real line . that is , @xmath71 the function @xmath72 in is the modified bessel function of second kind of order @xmath3 . proposition [ prop : pntildeorthogonal ] is proved in section [ subsec : second ] . since @xmath73 as @xmath74 , see for instance @xcite , the condition @xmath75 is necessary for the convergence of the integral with @xmath76 . in case @xmath45 then is the real and positive weight function @xmath77 . then @xmath69 has all its zeros on the real line , and consequently the zeros of @xmath0 are on the imaginary axis . this way we recover the result of @xcite . for @xmath78 , the modified bessel function reduces to an elementary function and the weight function is @xmath79 the weight has three components : * an exponential varying weight @xmath80 with a potential function @xmath81 that is convex but non - smooth at the origin . * a square root singularity at the origin @xmath82 . * a complex phase factor @xmath83 with a jump discontinuity at the origin . the exponential varying weight determines the limiting density . indeed we have that @xmath84 is the minimizer of the logarithmic energy in external field @xmath85 among probability measures on the real line , see @xcite , and as is well - known , the zeros of the orthogonal polynomials with varying weight function @xmath80 have @xmath86 as limiting density . this continues to be the case for the weights as is claimed by theorem [ th0 ] . a riemann - hilbert analysis for the weight @xmath80 , and other freud weights , is in @xcite . the square root singularity and the jump discontinuity are known as fisher - hartwig singularities in the theory of toeplitz determinants . there is much recent progress in the understanding of toeplitz and hankel determinants with such singularities @xcite . this is also related to the asymptotics of the corresponding orthogonal polynomials , whose local behavior near a fisher - hartwig singularity is described with the aid of confluent hypergeometric functions , see the works of deift , its and krasovsky @xcite and also @xcite . we are facing the complication that the fisher - hartwig singularity is combined with a logarithmic divergence of the density @xmath86 at the origin , see . in our riemann - hilbert analysis we were not able to construct a local parametrix with special functions , and we had to resort to an existence proof , where we used ideas from @xcite and @xcite , although our proof is at the technical level different from either of these papers . away from the region where the zeros of @xmath87 lie , the asymptotic behavior is governed by the @xmath88 function associated with the limiting density @xmath86 , that is , @xmath89 where the density @xmath86 is given by . then @xmath88 is defined and analytic for @xmath90 $ ] . we prove the following asymptotic behavior of @xmath0 in the region away from the zeros . we continue to use @xmath91 as defined in . [ th1 ] let @xmath47 . then the polynomial @xmath0 exists and is unique for sufficiently large @xmath15 . moreover , the polynomial @xmath69 given by has the following behavior as @xmath46 : @xmath92 uniformly for @xmath93 in compact subsets of @xmath94 $ ] . here the branch of the function @xmath95 is taken which is analytic in @xmath96 $ ] and positive for real @xmath97 . in a neighborhood of @xmath98 we find oscillatory behavior of the polynomials @xmath69 as @xmath46 . we state the asymptotic formula for @xmath99 only . there is an analogous formula for @xmath100 . this follows from the fact that the polynomial @xmath0 has real coefficients , as all the moments in the determinantal formula are real . thus @xmath101 , and so @xmath102 to describe the behavior near the interval , we need the analytic continuation of the density , which we also denote by @xmath86 , @xmath103 which is defined and analytic in @xmath104 . for @xmath105 with @xmath106 we also define @xmath107 [ th3 ] let @xmath108 . there is an open neighborhood @xmath109 of @xmath98 such that for @xmath110 with @xmath99 we have @xmath111 \end{gathered}\ ] ] as @xmath4 , with @xmath86 and @xmath112 given by and . the asymptotic expansion is uniform for @xmath113 with @xmath99 and @xmath114 , @xmath115 , for every @xmath57 . the two terms @xmath116 and @xmath117 in describe the oscillatory behavior near the interval as well as the leading order behavior of the zeros . zeros can only happen when these two terms are of comparable absolute value so that cancellations can take place . when @xmath24 this happens for real @xmath113 . however , for @xmath30 this does not happen for real @xmath93 , but near the line @xmath118 , as we will show in section [ section44 ] . this leads to theorem [ th2 ] . the structure of the rest of the paper is as follows . in section [ section_rh ] we state the riemann hilbert problem @xmath119 for @xmath14 with @xmath13 , and we make an initial transformation @xmath120 in the rh problem for @xmath121 we can take the limit @xmath122 which leads to a rh problem for @xmath123 , that characterizes the polynomial @xmath124 . then we carry out the further transformations @xmath125 of the deift zhou nonlinear steepest descent method @xcite . the step @xmath126 is rotation and scaling , to translate the problem to the interval @xmath37 $ ] . this leads to the polynomials @xmath69 and the proof of proposition [ prop : pntildeorthogonal ] . the normalization at @xmath127 in the @xmath128 step is carried out using an equilibrium problem with a freud weight @xmath129 , where @xmath130 is the pointwise limit as @xmath46 of the varying weight @xmath131 the construction of the global parametrix @xmath132 on the interval @xmath37 $ ] involves two szeg functions @xmath133 and @xmath134 , that correspond respectively to an algebraic singularity of the weight function at the origin and to a complex phase factor . the local parametrices near the endpoints @xmath135 involve airy functions , since the density @xmath136 in behaves like a square root in a neighborhood of these endpoints . the main difficulty of the analysis is the construction of a local parametrix in a neighborhood of the origin , and the reason is the lack of analyticity of the weight function @xmath137 in that neighborhood . in this paper , we reduce the jump matrices in that local analysis to almost constant in a disk around @xmath138 and then use a small norm argument in @xmath139 to prove existence of a solution to this local rh problem . in this respect , the analysis is similar to the one presented by kriecherbauer and mclaughlin in @xcite . also , the same limiting potential @xmath140 appears in the work of bleher and bothner in @xcite . another example of non analytic weight function was considered in the work of foulqui , martnez finkelshtein and sousa , see @xcite and @xcite , although in this case the local parametrix at the origin is explicitly given in terms of confluent hypergeometric functions . finally , in section [ proofs ] we follow the transformations both outside and inside the lens , but away from the origin , to get the asymptotic information about @xmath87 and its zeros . this proves theorem [ th1 ] and [ th3 ] . theorem [ th0 ] follows from theorem [ th1 ] and theorem [ th3 ] is a consequence of [ th2 ] . we let @xmath30 and @xmath13 . orthogonal polynomials are characterized by a matrix valued riemann - hilbert problem as was first shown by fokas , its , and kitaev @xcite , see also @xcite . this characterization does not use the fact that the orthogonality weight is non - negative , and it therefore also applies to oscillatory weights . thus the polynomial @xmath14 satisfying is characterized by the following riemann - hilbert problem : [ rhfory ] @xmath141 is a @xmath142 matrix valued function that satisfies : * @xmath119 is analytic in @xmath143 . * @xmath119 satisfies the jump condition @xmath144 * as @xmath145 , @xmath146 where @xmath147 denotes the @xmath148 identity matrix . * @xmath149 remains bounded as @xmath150 . the polynomial @xmath14 exists and is unique if and only if the rh problem has a unique solution . in that case we have @xmath151 in the first transformation we use the following connection formula between @xmath1 and the modified bessel function @xmath72 of the second kind : @xmath152 we define the new matrix valued function @xmath121 by @xmath153 then @xmath121 has an analytic continuation across the positive real axis , due to the factorization . thus @xmath121 is defined and analytic in the complex plane except for the imaginary axis , and it satisfies the following rh problem : [ rhforx0 ] * @xmath121 is analytic in @xmath154 . * @xmath121 satisfies the jump condition ( the imaginary axis is oriented from bottom to top ) @xmath155 * as @xmath156 , @xmath157 * @xmath158 remains bounded as @xmath150 with @xmath100 , and @xmath159 the asymptotic condition follows from , the definition and the fact that @xmath160 terms in appear because of the behavior @xmath161 as @xmath150 for @xmath162 , see for instance ( * ? ? ? * formula 10.30.2 ) . note that by and @xmath163 in the rh problem for @xmath121 we can take @xmath122 . indeed , after setting @xmath164 in , the off - diagonal entries in the jump matrices still tend to @xmath138 as @xmath165 because of . we put @xmath164 and we consider the following rh problem . [ rhforx ] we seek a function @xmath166 satisfying : * @xmath123 is analytic in @xmath154 . * @xmath123 satisfies the jump condition ( the imaginary axis is oriented from bottom to top ) @xmath167 * as @xmath156 , @xmath168 * @xmath169 remains bounded as @xmath150 with @xmath100 , and @xmath170 if there is a unique solution then the @xmath171-entry is a monic polynomial of degree @xmath15 , say @xmath0 , and @xmath172 see . thus @xmath0 is the polynomial that we are interested in . we introduce a scaling and rotation @xmath173 and our main interest is in the rescaled polynomials @xmath174 whose zeros will accumulate on the interval @xmath37 $ ] as @xmath175 . more precisely , we define @xmath176 as @xmath177 from and the rh problem [ rhforx ] , we immediately obtain the following rh problem for @xmath178 : [ rhforu ] * @xmath176 is analytic in @xmath179 . * @xmath176 satisfies the jump condition @xmath180 * as @xmath156 , @xmath181 * @xmath178 remains bounded as @xmath182 with @xmath183 , and @xmath184 note that by , , and @xmath185 which is a monic polynomial of degree @xmath15 . the zeros of @xmath186 are obtained from the zeros of @xmath0 by rotation over @xmath187 degrees in the clockwise direction and by dividing by a factor @xmath188 . we can now prove proposition [ prop : pntildeorthogonal ] . the rh problem for @xmath176 is the rh problem for orthogonal polynomials on the real line for the varying weight function @xmath189 for @xmath190 , see @xcite . because of the @xmath191 factor , the weight function is not real on the real line , and it has a singularity at the origin because of the behavior of the @xmath72 function near @xmath138 . the singularity is integrable since @xmath75 , and so @xmath192 is the monic polynomial of degree @xmath15 satisfying . in order to normalize the rh problem at infinity we make use of an equilibrium problem with external field @xmath130 . the equilibrium measure @xmath193 minimizes the energy functional @xmath194 among all probability measures on @xmath195 . the minimizer is supported on @xmath37 $ ] . it is absolutely continuous with respect to the lebesgue measure , @xmath196 , and has density @xmath197 which corresponds to the case @xmath198 in @xcite . the integral can be evaluated explicitly and it gives the formula . note that @xmath136 grows like a logarithm at @xmath199 . the @xmath88 function is defined in . the boundary values @xmath200 and @xmath201 on the real axis satisfy @xmath202 the euler - lagrange equations for the equilibrium problem imply that we have ( see e.g. @xcite or @xcite ) @xmath203 , \\ < \ell , & \quad x\in(-\infty,-1)\cup(1,\infty ) . \end{cases}\ ] ] with the constant @xmath204 ( see theorem iv.5.1 in @xcite or formula ( 3.5 ) in @xcite ) @xmath205 a related function is @xmath206 where @xmath207 the @xmath208-function is analytic in @xmath209\cup i\mathbb{r}\right)$ ] . for @xmath210 $ ] we have from the variational equation @xmath211 thus @xmath212 gives an analytic extension of @xmath213 from @xmath37 $ ] into the upper half plane minus the imaginary axis , and of @xmath214 into the lower half plane minus the imaginary axis . note that @xmath215 is purely imaginary on @xmath37 $ ] , because of . on the imaginary axis , the function @xmath216 is not analytic because of the discontinuity in @xmath217 . the boundary values of this weight function satisfy @xmath218 and as a consequence , @xmath219 here we take the orientation of the imaginary axis from bottom to top . now we are ready for the third transformation of the rh problem and we define the matrix valued function @xmath220 where @xmath221 is the third pauli matrix . we also write @xmath222 then from the above definitions and properties and from the rh problem [ rhforu ] for @xmath176 we find that @xmath223 satisfies the following riemann hilbert problem . [ rhfort ] * @xmath223 is analytic in @xmath179 . * @xmath223 satisfies the jump conditions @xmath224 where @xmath225 is given in . * as @xmath156 , @xmath226 * @xmath227 remains bounded as @xmath150 with @xmath183 , and @xmath228 the off diagonal elements in the jump matrices on @xmath229 and @xmath230 tend to @xmath138 at an exponential rate , because of the euler lagrange condition . the jump matrix on the interval @xmath231 has a factorization @xmath232 while the jump matrix on @xmath233 factorizes as @xmath234 in order to open the lens around @xmath98 , we need the analytic extension of the function @xmath225 from to @xmath235 , which we also denote by @xmath225 , @xmath236 note that as @xmath4 , see and , @xmath237 which explains the factor @xmath238 that we introduced in and . $ ] , and contour @xmath239 consisting of @xmath240 , the segment @xmath241 and the real line . ] next , we fix a number @xmath242 and we open a lens around @xmath37 $ ] , which defines contours @xmath243 , @xmath244 and domains @xmath245 , @xmath244 as indicated in figure [ fig_lens2 ] . in the fourth transformation we define the matrix valued function @xmath246 : @xmath247 using the analytic extension for the function @xmath248 in each region , and @xmath216 defined in . in order to divide by @xmath248 we need to be careful with possible zeros of this function in the complex plane . following the general theory in @xcite , the bessel function @xmath249 is free from zeros in the half plane @xmath250 . using , we can conclude that @xmath251 . from the rh problem [ rhfort ] and we find that that @xmath246 is the solution of the following rh problem : [ rhfors ] * @xmath252 is analytic in @xmath253 , where @xmath239 is depicted in figure [ fig_lens2 ] . * @xmath252 satisfies the jump conditions @xmath254 where @xmath255 + here @xmath256 and @xmath257 using the appropriate values of @xmath258 and @xmath259 in each case . the imaginary axis is oriented upwards , and so for @xmath260 , we have that @xmath261 and @xmath262 ( @xmath263 and @xmath264 ) denote the limiting value from the left ( right ) half - plane . * as @xmath156 , @xmath265 * @xmath246 remains bounded as @xmath150 with @xmath183 , and @xmath266 note that as a consequence of the definition of @xmath216 in and formula , @xmath267 is decreasing on @xmath37 $ ] . because of the cauchy riemann equations , @xmath268 as we move away from the interval . we may and do assume that the lens is small enough such that @xmath269 on the lips of the lens . then it follows from and that the jump matrix @xmath270 on the lips of the lens tends to @xmath147 at an exponential rate as @xmath4 , if we stay away from the endpoints @xmath135 . also the jump matrix on @xmath229 and @xmath271 tends to the identity matrix . thus for any @xmath57 , there is a constant @xmath272 such that @xmath273 \cup [ -i \rho , i\rho ] \cup d(\pm 1 , \delta)).\ ] ] the condition needs some explanation , since and at first sight lead to the behavior @xmath274 as @xmath150 with @xmath275 . however , a cancellation takes place for the entries in the first column , as can be checked from the jump conditions for @xmath252 , see on the intervals @xmath231 and @xmath233 . since @xmath252 remains bounded as @xmath150 with @xmath183 , and @xmath276 one finds . if we ignore the jump matrices in the rh problem for @xmath252 except for the one on the interval @xmath37 $ ] , we arrive at the following rh problem for a @xmath142 matrix valued function @xmath132 : [ rhforngeneral ] * @xmath132 is analytic in @xmath94 $ ] . * @xmath132 satisfies the jump conditions @xmath277 * as @xmath156 , @xmath278 we solve the rh problem for @xmath132 by means of two szeg functions @xmath279 and @xmath280 , see also @xcite , that are associated with @xmath225 and @xmath281 , respectively . the first szeg function @xmath282 is defined by @xmath283 which is defined and analytic for @xmath284 $ ] . it satisfies @xmath285 it follows from that @xmath279 has no zeros in @xmath286 $ ] and @xmath287 in what follows we are not going to indicate the @xmath15-dependence in the notation for @xmath279 and @xmath288 , since the dependence on @xmath15 is only mildly . indeed , because of we have that @xmath279 tends to the szeg function for the weight @xmath82 with a rate as given in the following lemma . [ lem : d1nlimit ] we have @xmath289 as @xmath4 , with @xmath290-term that is uniform for @xmath291 \cup d(0 , \delta)\cup d(\pm 1,\delta))$ ] for any @xmath57 . the szeg function for @xmath82 is @xmath292 and so @xmath293 because of there exist @xmath294 @xmath295 then also for some @xmath296 , @xmath297 it follows that @xmath298 ) n } \int_{c_0/n}^1 \frac{1}{x\sqrt{1-x^2 } } dx \\ & \leq \frac{c_3}{\operatorname{dist}(z , [ -1,1 ] ) } \frac{\log n}{n } \end{aligned}\ ] ] with a constant @xmath299 that is independent of @xmath15 and @xmath93 . by deforming the integration path into the complex plane in such a way that it stays at a certain distance from @xmath93 , and applying similar estimates we find @xmath300 with a constant that is independent of @xmath291 \cup d(0 , \delta)\cup d(\pm 1,\delta))$ ] . similarly @xmath301 near @xmath302 we use and to find a @xmath303 such that @xmath304 the upper bound follows from the fact that @xmath305 if @xmath306 and @xmath13 and the explicit formula for @xmath307 see @xcite . then @xmath308 and @xmath309 for some new constant @xmath310 . combining the estimates , , and , we get @xmath311 with a @xmath290 term that is uniform for @xmath115 @xmath312 , and so by @xmath313 as claimed in . since is uniform for @xmath115 , @xmath312 , we can let @xmath145 , and obtain . the second szeg function @xmath280 corresponds to the weight @xmath314 , and is defined as @xmath315,\ ] ] with the branch of the square root that is positive for real @xmath97 . it is not difficult to check that @xmath316 is the conformal mapping from @xmath286 $ ] onto the sector @xmath317 that maps @xmath318 to @xmath319 , @xmath320 to @xmath321 , @xmath322 to @xmath323 and @xmath324 to @xmath325 . the szeg function @xmath280 is related to the function @xmath86 from . we have @xmath326 this follows from and by straightforward calculation . it follows from that @xmath280 satisfies @xmath327 and , since @xmath328 as @xmath150 , @xmath329 having @xmath330 and @xmath280 we seek @xmath132 in the form @xmath331 then @xmath132 satisfies the rh problem [ rhforngeneral ] if and only if @xmath332 satisfies the following standard rh problem : [ rhforn0general ] * @xmath332 is analytic in @xmath94 $ ] . * @xmath332 satisfies the jump conditions @xmath333 * @xmath334 as @xmath156 . the rh problem for @xmath332 has the explicit solution ( see for instance ( * ? ? ? * section 7.3 ) ) : @xmath335 for @xmath284 $ ] , and we take the branch of the fourth root that is analytic in @xmath286 $ ] and that is real and positive for @xmath97 . note that we can also write @xmath336 where @xmath337 is the conformal map from @xmath286 $ ] to the exterior of the unit disk . around the endpoints @xmath338 we build airy parametrices @xmath339 in the usual way . we take @xmath57 sufficiently small , and @xmath339 is defined and analytic in @xmath340 such that it has the same jumps as @xmath252 on @xmath341 , and such that @xmath342 as @xmath4 . we refer the reader for instance to the monograph by deift @xcite for details . in the fifth transformation we put @xmath343 then @xmath344 is defined and analytic outside of a contour consisting of @xmath239 and two circles around @xmath135 . the construction of the airy parametrix is such that it has the same jump as @xmath252 inside the circles . as a result @xmath344 is analytic inside the two disks . also @xmath252 and @xmath132 have the same jump on @xmath98 and it follows that @xmath344 is analytic across @xmath98 . therefore @xmath344 is analytic in @xmath345 where @xmath346 consists of two circles around @xmath135 , the parts of @xmath347 , @xmath243 , @xmath348 and @xmath271 outside of these circles , and the segment @xmath349 on the imaginary axis . see figure [ figq ] . ] from the rh problem [ rhfors ] for @xmath252 and it then follows that @xmath344 solves the following rh problem . [ rhforq ] * @xmath350 is analytic . * @xmath344 satisfies the jump condition @xmath351 on @xmath346 where @xmath352 here @xmath353 and @xmath354 are given by and . * as @xmath156 , @xmath355 * @xmath356 as @xmath150 . in the behavior around @xmath138 there is no longer a distinction between the upper and lower half planes , and @xmath344 remains bounded in all directions . we note that @xmath357 because of the matching property . we also note that @xmath358 ) \ ] ] because of , , and lemma [ lem : d1nlimit ] . the jump matrix @xmath359 on the imaginary axis can be rewritten as ( we use ): @xmath360 with @xmath353 on @xmath361 , and @xmath354 on @xmath362 . the entry @xmath363 in depends on @xmath15 , and tends to @xmath138 as @xmath364 for every @xmath365 , but not in a uniform way . hence , further analysis is needed in the next section . a similar situation is studied in ( * ? ? ? * section 5 ) , where the jump on the imaginary axis has the same structure and approaches the identity matrix at a rate @xmath366 as @xmath175 . in that case no local parametrix near the origin is needed . the construction of a local parametrix in a neighborhood of the origin follows the idea exposed in @xcite . we take @xmath368 with @xmath369 and we build a local parametrix @xmath370 defined in a neighborhood @xmath371 of @xmath138 . we use a cut - off function @xmath372 on @xmath373 such that 1 . @xmath374 is a @xmath375 function , 2 . @xmath376 for all @xmath377 , 3 . @xmath378 for @xmath379 , 4 . @xmath380 for @xmath381 . then we modify @xmath359 by multiplying the off - diagonal entry in the middle factor of by @xmath372 , and in addition we use this as a jump matrix in the full imaginary axis . thus @xmath382 with @xmath353 on @xmath383 and @xmath353 on @xmath384 . then the rh problem for the local parametrix @xmath370 at the origin is : [ rhforp ] * @xmath385 is analytic . * @xmath370 satisfies the jump condition @xmath386 where @xmath387 is given by . * @xmath388 as @xmath4 uniformly for @xmath389 with @xmath91 given by . [ propo8 ] the rh problem [ rhforp ] has a solution for @xmath15 large enough . the rest of this subsection is devoted to the proof of proposition [ propo8 ] . it takes a number of steps and it is the most technical part of the paper . we introduce a matrix @xmath391 in the following way : @xmath392 d_{\infty}^{\sigma_3 } n_0(z ) \begin{pmatrix } 0 & -1 \\ 1 & 0 \end{pmatrix } \widehat p(z ) \begin{pmatrix } 0 & 1 \\ -1 & 0 \end{pmatrix } n_0(z)^{-1 } d_{\infty}^{-\sigma_3 } , & \text{for } { \operatorname{im}}z > 0 . \end{cases}\ ] ] the extra factors in for @xmath183 are introduced in order to compensate the jumps of @xmath332 on @xmath37 $ ] . then @xmath370 satisfies the jump condition in the rh problem [ rhforp ] if and only if @xmath393 , where the jump is @xmath394 \begin{pmatrix } 1 & 0 \\ j_2(z ) ( d_1(z ) d_2(z))^2 \chi(z ) & 1 \end{pmatrix } , & \text { for } z \in i \mathbb r^- . \end{cases}\ ] ] note the difference in the triangularity structure . so , we look for @xmath390 that solves the following rh problem : [ rhforphat ] * @xmath395 is analytic . * @xmath390 satisfies the jump conditions @xmath396 where @xmath397 is given by . * @xmath398 as @xmath399 . our aim is to show that the rh problem for @xmath390 has a solution for @xmath15 sufficiently large , and that this solution satisfies in addition * @xmath400 as @xmath4 , uniformly for @xmath389 . having @xmath390 we define @xmath370 by in terms of @xmath390 , and it will satisfy the requirements of the rh problem [ rhforp ] . we prove the following result : [ lem : phat ] if @xmath401 , then for @xmath15 large enough there exists @xmath391 that solves the rh problem [ rhforphat ] , and as @xmath46 , @xmath402 in @xmath403 $ ] , and @xmath404 $ ] . it follows from lemma [ lem : phat ] that @xmath400 as @xmath4 , uniformly for @xmath389 , and because of , the same holds for @xmath405 . in the proof of this lemma we will need the following steps : 1 . we write the jump conditions for @xmath391 componentwise , and in terms of two integral operators @xmath406 and @xmath407 . we estimate the operator norms @xmath408 and @xmath409 as @xmath46 . this requires estimates for the functions @xmath410 , @xmath411 , @xmath133 and @xmath134 , which are uniform as @xmath46 for @xmath412 in a fixed interval around the origin on the imaginary axis . we show that the operators @xmath413 and @xmath414 are invertible for @xmath15 large enough , and this gives the existence and asymptotics of @xmath415 . finally , the estimates for @xmath391 are used to prove that the matrix @xmath416 , which will be defined in section [ finaltrans ] and which solves the riemann hilbert problem [ rhforr ] , is close to the identity matrix as @xmath46 . let us write @xmath417 these functions depend on @xmath15 , since @xmath353 , @xmath354 and @xmath330 depend on @xmath15 . note , however , that @xmath280 and @xmath418 do not depend on @xmath15 . the jump condition - yields that for @xmath419 , @xmath420 since @xmath421 for @xmath422 , we find that @xmath423 is analytic in @xmath424 $ ] , and @xmath425 is analytic in @xmath426 $ ] . then by the sokhotski - plemelj formula and the asymptotic condition @xmath427 as @xmath145 , we get @xmath428 we can write the equations in operator form if we introduce two operators @xmath429 ) \to l^2([-2i \varepsilon,0 ] ) \qquad \text { and } \qquad k_2 : l^2([-2i \varepsilon,0 ] ) \to l^2([0,2i \varepsilon])\ ] ] by @xmath430 ) , \\ ( k_2 g)(z ) & = \frac{1}{2\pi i } \int_{-2i \varepsilon}^0 \frac{\eta_2(s ) g(s)}{s - z } ds , \qquad g \in l^2([-2i \varepsilon,0]).\end{aligned}\ ] ] then @xmath431 , @xmath432 should solve @xmath433 and @xmath434 , @xmath435 should solve @xmath436 both @xmath406 and @xmath407 are integral operators between hilbert spaces with operator norms @xmath437 the @xmath438-integrals can be done explicitly . this leads to the estimates ( we also change to a real integration variable by putting @xmath439 ) @xmath440 the next step is to show that both integrals are finite ( so that @xmath406 and @xmath407 are well - defined bounded operators ) and that @xmath441 and @xmath442 tend to @xmath138 as @xmath4 . to this end , we need to control the functions @xmath443 and @xmath444 , defined in . the functions @xmath443 and @xmath444 are defined in terms of @xmath353 , @xmath354 , @xmath330 and @xmath280 , see . in this section we obtain estimates for all these functions for large @xmath15 . first we write the functions @xmath410 and @xmath411 in terms of bessel functions . because of the property @xmath447 for real @xmath3 , see @xcite , if we consider the positive imaginary axis and we write @xmath448 , with @xmath449 , then the function @xmath225 ( recall ) can be written as @xmath450 so @xmath451 . similarly , on the negative imaginary axis , @xmath452 so again @xmath453 . additionally , we have @xmath454,\\ \frac{n \pi^2}{2}\left[j_{\nu}(n\pi y)^2+y_{\nu}(n\pi y)^2\right ] , \end{aligned}\ ] ] in terms of hankel functions , see @xcite . we have the following auxiliary result : [ lem : j1_j2_bessel ] for @xmath449 , the functions @xmath455 and @xmath456 can be written as follows : @xmath457 it follows from that @xmath353 can be written as @xmath458,\ ] ] and because of @xmath459 on the imaginary axis , and the fact that @xmath451 , the two terms on the right hand side are complex conjugates , so @xmath460.\ ] ] using the formula @xmath461 in terms of hankel functions , see @xcite and we observe that @xmath462 hence , on the positive imaginary axis , @xmath463 = -\frac{\sqrt{2n}\pi}{2}(j_{\nu}(n\pi y)\cos\nu\pi - y_{\nu}(n\pi y)\sin \nu\pi).\ ] ] using and , this proves the first formula . similarly , for @xmath449 , @xmath464 . \ ] ] in this case , we use @xmath465 see @xcite , and to obtain @xmath466 so @xmath467 = \frac{\sqrt{2n } \pi}{2}j_{\nu}(n\pi y).\ ] ] we use and , and this completes the proof . next , we will obtain estimates of the previous functions @xmath353 and @xmath354 for large @xmath15 . [ lem : asymptotic_j1_j2 ] for @xmath47 there exist constants @xmath468 such that for all @xmath13 we have @xmath469 for the proof , we consider the following expansions : as @xmath470 , @xmath471 and for @xmath472 we have @xmath473 as @xmath474 , we have @xmath475 where @xmath476 . see for instance ( * ? ? ? * formulas 10.7.34 , 10.17.34 ) . from this , it follows that @xmath477 from , we claim that there exist two constants @xmath478 such that @xmath479 using a similar argument , we have @xmath480 and also @xmath481 and putting all the estimates together we get the bounds in the lemma . as a consequence of lemma [ lem : j1_j2_bessel ] and lemma [ lem : asymptotic_j1_j2 ] we obtain the following bounds for @xmath353 and @xmath354 for @xmath449 : @xmath482 next , we need an estimate for @xmath133 ( see formula ) , with @xmath483 , @xmath484 $ ] where we recall that @xmath485 is the intersection of the lens with the imaginary axis . [ lem : lemmad1 ] for @xmath47 , there exists a constant @xmath486 such that for all sufficiently large @xmath15 , latexmath:[\[\label{boundd1 } we write first @xmath448 with @xmath449 in and use the parity of the function @xmath225 to get the following expression : @xmath488 using the asymptotic expansions , and , we claim that there exist two constants @xmath489 and @xmath490 , depending on @xmath3 , such that @xmath491 satisfies @xmath492 and @xmath493 since @xmath494 , both bounds hold uniformly for @xmath495 . since the integrand in is a real function , we can bound @xmath496 from above by another szeg function : @xmath497 this last szeg function is explicit , since for a general exponent @xmath498 we have @xmath499 as a consequence , substituting @xmath448 with @xmath500 $ ] , and @xmath501 , @xmath502 and by the same argument with @xmath503 , @xmath504 the bound in the lemma follows for @xmath449 from these two estimates , for some constant @xmath505 . finally , from the definition of @xmath330 , see , we have that if @xmath506 , then @xmath507 , so the modulus is equal and the bound holds also in this case . now we write together all the estimates computed before to obtain bounds for the functions @xmath443 and @xmath444 defined in . [ lem : bounds - eta1-eta2 ] for @xmath47 , there exist constants @xmath508 such that for @xmath15 large enough and @xmath509 $ ] , we have the bounds @xmath510 we collect the results on @xmath330 ( see formula ) , @xmath280 ( we use the fact that this function does not depend on @xmath15 and formula ) , @xmath353 and @xmath354 ( formula ) . then for some constant @xmath511 we simplify the bound to @xmath512 also , @xmath513 and the result follows . in order to estimate the norms of @xmath406 and @xmath407 we need the @xmath514 norm of @xmath443 and @xmath444 , see formula . for this we use the estimate in lemma [ lem : bounds - eta1-eta2 ] and the following bound on @xmath216 : for every @xmath515 we have @xmath516 we consider @xmath517 with @xmath518 . the other cases follow by symmetry . let @xmath519 . then by and , @xmath520 and so @xmath521 . by analytic continuation we find @xmath522 then @xmath523 since @xmath524 is purely imaginary , we obtain by taking the real part and letting @xmath525 , @xmath526 where we used for @xmath86 . the integral can be evaluated explicitly and it gives . without loss of generality we assume in what follows that @xmath527 is small enough so that @xmath528 for @xmath529 . in order to estimate integrals involving the functions @xmath258 , we use , together with the following technical lemma . [ lem : bound - integral ] for any @xmath498 , there exists a constant @xmath530 such that for @xmath15 large enough @xmath531 we split the integral into two parts and we estimate @xmath532 where for the second integral we used that @xmath533 is decreasing on @xmath534 $ ] and so @xmath535 for @xmath536 $ ] . the first integral of is estimated by extending the integral to @xmath537 and the result is that it is @xmath538 as @xmath4 . the second integral in is @xmath539 as @xmath4 . this gives the result . combining the estimates in , , and we obtain , whenever @xmath540 , @xmath541 and @xmath542 as @xmath4 . to obtain one has to consider the three different integrals coming from square of the factor @xmath543 in , and retain the largest one . hence , using and - we have the bounds @xmath544 thus @xmath406 and @xmath407 are bounded operators between the hilbert spaces @xmath545)$ ] and @xmath546)$ ] . in addition from , we get @xmath547 and similarly @xmath548 it follows from and that the operators @xmath413 and @xmath414 are invertible for @xmath15 large enough , and then we can solve the equations and . thus we define the entries of the matrix @xmath390 as follows : @xmath549 in and we use @xmath550 to denote the identically - one function in @xmath551)$ ] and @xmath546)$ ] , respectively . then and hold true , which means that the equations in hold . this then also means that the jump condition in the rh problem [ rhforphat ] is satisfied . the equations allow us to give estimates on @xmath391 . first of all we obtain from - , , and that @xmath552 ) } = \mathcal{o}(1 ) , \qquad \| \widehat p_{22}\|_{l^2([-2i\varepsilon,0 ] ) } = \mathcal{o}(1),\ ] ] and then by @xmath553 ) } & \leq \| k_1 \| \ , \| \widehat p_{11}\|_{l^2([0,2i\varepsilon ] ) } = \mathcal{o}(n^{-\nu}(\log n)^{-\nu}),\\ \| \widehat p_{21 } \|_{l^2([0 , 2i \varepsilon ] ) } & \leq \| k_2 \| \ , \| \widehat p_{22}\|_{l^2([-2i\varepsilon,0 ] ) } = \mathcal{o}(n^{\nu}(\log n)^{-\nu } ) . \end{aligned}\ ] ] for pointwise estimates we use the distances @xmath554 ) , \qquad d_-(z ) = \operatorname{dist}(z,[-2i\varepsilon,0]).\ ] ] then by the first equation in , we get for @xmath555 $ ] , @xmath556 where we used the cauchy - schwarz inequality , and @xmath557 is the @xmath558 norm on @xmath559 $ ] . thus by and , @xmath560 as @xmath4 , uniformly for @xmath555 $ ] . using similar arguments , we obtain @xmath561 as @xmath4 , and the @xmath290 terms are uniform in @xmath93 . observe that all @xmath290 terms tend to @xmath138 as @xmath4 , since @xmath7 . it follows from that @xmath562 as @xmath145 and therefore @xmath415 satisfies the rh problem [ rhforphat ] . for @xmath389 we have @xmath563 . from we then immediately find that the estimates in lemma [ lem : phat ] hold , and the lemma is proved . this also completes the proof of proposition [ propo8 ] . having @xmath370 as in proposition [ propo8 ] we define the final transformation @xmath564 as @xmath565 recall that @xmath344 is the solution of the rh problem [ rhforq ] . then @xmath566 has jumps on a contour @xmath567 that consists of @xmath568 together with the circle of radius @xmath569 around @xmath138 , see figure [ figr ] . note that the jumps of @xmath370 and @xmath344 coincide on @xmath570 , so that @xmath566 has an analytic continuation across that interval . ] from rh problem [ rhforq ] and the definition it follows that @xmath566 satisfies the following rh problem . [ rhforr ] * @xmath571 is analytic . * @xmath566 satisfies the jump condition @xmath572 on @xmath567 where @xmath573 * as @xmath156 , @xmath574 in order to solve this rh problem asymptotically for large @xmath15 , we need to show that the jump matrices for @xmath416 are close to the identity matrix uniformly for @xmath575 , see figure [ figr ] . the jump matrix @xmath576 in the rh problem for @xmath566 satisfies for some constant @xmath272 , @xmath577 as @xmath4 , where the @xmath290 terms are uniform . for @xmath575 with @xmath578 , we have @xmath579 . on the boundary of the disks around the endpoints we have @xmath580 , see and on the rest of @xmath567 except @xmath241 we have @xmath581 for some @xmath582 , see . on the circle @xmath583 , the jump is @xmath584 . we use and the fact that @xmath585 , uniformly for @xmath583 , to find that @xmath586 as given in . for @xmath587 we get from and @xmath588 from and , we obtain for @xmath589 $ ] , @xmath590 follows . the case @xmath591 can be handled in a similar way . what is left are the intervals @xmath592 and @xmath593 . for @xmath594 we find from and that @xmath595 using - we rewrite this as @xmath596 here we note that @xmath597 , which follows by standard arguments from the rh problem [ rhforphat ] , and therefore @xmath598 . then a little calculation shows that reduces to @xmath599 where @xmath600 the functions @xmath601 and @xmath602 are analytic on @xmath603 and so we do not have to take the @xmath604-boundary value . then it follows from and the estimates in and that all entries in @xmath605 are uniformly bounded as @xmath4 . then by and we find for @xmath606 . a similar argument shows that @xmath607 is exponentially close to the identity matrix for @xmath608 as well , and the lemma follows . as a consequence of , the biggest estimates for @xmath609 are on the circle @xmath583 . for @xmath47 , the jump matrix satisfies ( recall @xmath91 is given by ) @xmath610 uniformly for @xmath611 where @xmath567 is the union of contours depicted in figure [ figr ] . note that @xmath612 as @xmath4 , but the rate of convergence is remarkably slow . following standard arguments , we now find that for @xmath15 sufficiently large , the rh problem [ rhforr ] for @xmath566 is solvable , and @xmath613 uniformly for @xmath614 . the convergence rate in may not be optimal , since some of the bounds in the analysis may not be as sharp as possible . note that for @xmath78 we only have @xmath615 , which is a very slow convergence . since all of the transformations @xmath616 are invertible , we then also find that the rh problem for @xmath123 is solvable for @xmath15 large enough . in particular we find that the polynomial @xmath617 exists for @xmath15 large enough . following the transformations of the deift zhou steepest descent analysis and using formula , we obtain asymptotic information about @xmath618 in the complex plane , see and . consider the region in figure [ figr ] which is outside the lens and outside of the disks around @xmath619 . in this case @xmath620 , and by , , , , @xmath621 which means that @xmath622 using . here @xmath91 is given again by . we observe that @xmath623 , from , and using , , and we get @xmath624 as @xmath46 . similarly , we also see that @xmath625 as @xmath4 and follows . since the lens can be taken arbitrarily close to the interval @xmath37 $ ] and the disks can be taken arbitrarily small , the asymptotics is valid uniformly on any compact subset of @xmath286 $ ] . this proves theorem [ th1 ] . inside the lens , but away from the endpoints and the origin , we use the relation between the functions @xmath227 and @xmath246 . let @xmath93 be in the lens with @xmath105 . then we have @xmath626 for @xmath627 , and therefore @xmath628.\ ] ] since @xmath629 away from the endpoints , and @xmath630 away from the origin ( if @xmath578 ) , see and , we obtain @xmath631.\ ] ] for @xmath99 , and @xmath627 . we are going to simplify the expression and we do it for @xmath105 , @xmath183 . first we use , , and in to get @xmath632.\ ] ] from we have @xmath633 , @xmath634 and so @xmath635.\end{gathered}\ ] ] next we use to write @xmath636 where @xmath637 denotes the branch that is real and positive for @xmath638 and @xmath639 is given by . thus @xmath640 . \end{gathered}\ ] ] the two terms in parenthesis are inverse of each other . we write all contributing factors in exponential form . we have by , and @xmath641 for @xmath105 , @xmath183 , and we note that by and @xmath642 as @xmath4 . finally , we write @xmath643 and inserting into we find , where we also use , to simplify the first factor . a similar calculation leads to the same formula for @xmath113 with @xmath105 and @xmath275 . it follows from and that the leading factor in the outer asymptotics of @xmath174 does not vanish for @xmath284 $ ] . let @xmath644 be the monic polynomial . then we find from that @xmath645 uniformly for @xmath93 in compact subsets of @xmath286 $ ] . this implies that for any given compact subset @xmath646 $ ] , the polynomial @xmath69 does not have any zeros in @xmath647 for @xmath15 large enough . in other words , all zeros of @xmath69 tend to the interval @xmath37 $ ] as @xmath4 . in addition we find from that the zeros of @xmath69 have @xmath136 as limiting density . this follows from standard arguments in potential theory , see e.g. @xcite . this proves theorem [ th0 ] . let @xmath109 be the neighborhood of @xmath98 as in theorem [ th3 ] . theorem [ th2 ] will follow from the asymptotic approximation that is valid uniformly for @xmath93 in @xmath648 with @xmath99 . [ lem : allzeros ] there is a constant @xmath50 such that for large @xmath15 all zeros in @xmath649 satisfy @xmath650 it is enough to consider @xmath99 . let @xmath651 then by we have that zeros of @xmath69 in @xmath649 with @xmath105 are in the region where @xmath652 this leads to @xmath653 and so there is a constant @xmath50 such that all zeros in @xmath649 satisfy @xmath654 if @xmath15 is large enough . note that @xmath655 thus if is not satisfied then either @xmath656 or @xmath657 . in both cases it follows that @xmath658 because of this can not happen for zeros of @xmath69 in @xmath649 if @xmath15 is large enough , and the lemma follows . the lemma is the main ingredient to prove theorem [ th2 ] . in the proof we use @xmath659 , to denote positive constants that do not depend on @xmath15 or @xmath93 . the constants will depend on @xmath57 . it is easy to see from the definition that @xmath660 for @xmath661 this implies that for some constant @xmath296 @xmath662 there are also constants @xmath663 such that @xmath664 see . thus if @xmath665 then by and @xmath666 and thus there are no zeros in @xmath649 with @xmath665 by lemma [ lem : allzeros ] if @xmath15 is large enough . for @xmath667 we have by and @xmath668 it follows from this and lemma [ lem : allzeros ] that for large @xmath15 , there are no zeros with @xmath669 if @xmath670 . now assume @xmath671 with @xmath672 and @xmath105 . . then by taylor expansion @xmath674 and , see also , @xmath675 and @xmath290 terms are uniform for @xmath93 in the considered region . then since @xmath136 and @xmath676 are real , we have @xmath677 thus if @xmath678 then by the above and @xmath679 and from lemma [ lem : allzeros ] it follows that @xmath680 is not a zero if @xmath681 is large enough . thus for large @xmath15 all zeros @xmath682 of @xmath69 in @xmath649 satisfy @xmath683 then @xmath684 is a zero of @xmath0 , see , and the real part of this zero is @xmath685 which differs from @xmath34 by an amount less than @xmath686 . this proves theorem [ th2 ] . we thank daan huybrechs for suggesting the problem and for stimulating conversations . a. deao gratefully acknowledges financial support from projects fwo g.0617.10 and fwo g.0641.11 , funded by fwo ( fonds wetenschappelijk onderzoek , research fund flanders , belgium ) , and projects mtm201234787 and mtm2012 - 36732c0301 , from ministerio de economa y competitividad de espaa ( spanish ministry of economy and competitivity ) . kuijlaars is supported by ku leuven research grant ot/12/073 , the belgian interuniversity attraction pole p07/18 , fwo flanders projects g.0641.11 and g.0934.13 , and by grant no . mtm2011 - 28952-c02 of the spanish ministry of science and innovation . m. j. atia , a. martnez - finkelshtein , p. martnez - gonzlez , and f. thabet , quadratic differentials and asymptotics of laguerre polynomials with varying complex parameters , _ j. math ( 2014 ) , 5280 . _ p. m. bleher and t. bothner , exact solution of the six vertex model with domain wall boundary conditions . critical line between disordered and antiferroelectric phases . _ random matrices : theory appl . 01 , 1250012 ( 2012 ) . _ p. deift , t. kriecherbauer , k.t - r mclaughlin , s. venakides , and x. zhou , uniform asymptotics for polynomials orthogonal with respect to varying exponential weights and applications to universality questions in random matrix theory , comm . pure appl . math . 52 ( 1999 ) , 13351425 . a. foulqui moreno , a. martnez - finkelshtein , and v. l. sousa , on a conjecture of a. magnus concerning the asymptotic behavior of the recurrence coefficients of the generalized jacobi polynomials . _ j. approx . theory 162 ( 2010 ) , 807831 .

_ we present an overview of observational progress in the study of extragalactic globular cluster systems . globular clusters turn out to be excellent tracers not only for the star - formation histories in galaxies , but also for kinematics at large galactocentric radii . their properties can be used to efficiently constrain galaxy formation and evolution . after a brief introduction of the current methods and futures perspectives , we summarize the knowledge gained in various areas of galaxy research through the study of globular clusters . in particular , we address the star - formation histories of early - type galaxies ; globular cluster population in late - type galaxies and their link to early - type galaxies ; star and cluster formation during mergers and violent interactions ; and the kinematics at large radii in early - type galaxies . the different points are reviewed within the context of galaxy formation and evolution . _ finally , we revisit the globular cluster luminosity function as a distance indicator . despite its low popularity in the literature , we demonstrate that it ranks among one of the most precise distance indicators to early - type galaxies , provided that it is applied properly . * extragalactic globular cluster systems * + * a new perspective on galaxy formation and evolution * + markus kissler - patig + european southern observatory + karl - schwarzschild - str . 2 , 85748 garching , germany + [email protected] , http://www.eso.org/mkissler

astro-ph0002070

this brief review on extragalactic globular cluster systems is derived from a lecture given for the award of the ludwig - bierman - preis of the astronomische gesellschaft in gttingen during september 1999 . the oral version aimed at introducing , mostly from an observer s point of view , this field of research and at emphasizing its tight links to galaxy formation and evolution . the scope of this written follow - up is _ not _ to give a complete review on globular cluster systems but to present recent discoveries , including examples , and to set them into the context of galaxy formation and evolution . the choice of examples and the emphasis of certain ideas will necessarily be subjective , and we apologize at this point for any missing references . excellent recent reviews can be found in the form of two books : `` globular cluster systems '' by ashman & zepf ( 1998 ) , as well as `` globular cluster systems '' by harris ( 2000 ) . these include a full description of the globular clusters in the local group ( not discussed here ) , as well as an extensive list of references , including to older reviews . the plan of the article is the following . in section 2 , we give an introduction and the motivation for studying globular cluster systems with the aim of understanding galaxies . section 3 presents current and future methods of observations , and the rational behind them . this section reviews recent progress in optical and near - infrared photometry and multi - object , low - resolution spectroscopy . it can be skipped by readers interested in results rather than methods . in section 4 we present in turn the status of our knowledge on globular cluster systems in ellipticals , spirals and mergers . what are the properties of the systems ? how are the galaxy types linked ? and do mergers produce real ` globular ' clusters ? in section 5 , we discuss sub - populations of globular clusters and their possible origin . the most popular scenarios to explain the presence of globular cluster sub - populations around galaxies are listed . the pros and cons , as well as the expectations of each scenario are discussed . we present , in section 6 , some results from the study of globular cluster system kinematics . finally , in section 7 , we revisit the globular cluster luminosity function as a distance indicator . under which conditions can it be used , and how should it be applied to minimize any systematic errors ? it is compared to other distance indicators and shown to do very well . some conclusions and an outlook are given in section 8 . as a reminder , globular clusters are tyically composed of @xmath1 to @xmath2 stars clustered within a few parsecs . they are old , although young globular - cluster - like objects are seen in mergers , and their metallicity can vary between [ fe / h]@xmath3 dex and [ fe / h]@xmath4 dex . we refer to a globular cluster system as the totallity of globular clusters surrounding a galaxy . the two fundamental questions in galaxy formation and evolution are : _ 1 ) _ when and how did the galaxies assemble ? and _ 2 ) _ when and how did the galaxies form their stars ? a third question could be whether , and to what extend , the two first points are linked . generally speaking , in order to answer these questions from an observational point of view , one can follow two paths . the first would be to observe the galaxies at high redshift , right at the epoch of their assembly and/or star formation . we will consider this to be the `` hard way '' . these observations are extremely challenging for many reasons ( shifted restframe wavelength , faint magnitudes , small angular scales , etc ... ) . nevertheless , they are pursued by a number of groups through the observations of absorption line systems along the line of sights of quasars , or the detection of high - redshift ( e.g. lyman break ) galaxies , etc ... ( see e.g. combes , mamon , & charmandaris 1999 , bunker & van breugel 1999 , mazure , le fevre , & le brun 1999 for recent proceedings on the rapidly evolving subject ) . the second path is to wait until a galaxy reaches very low redshift and try to extract information about its past . this would be the `` lazy way '' . this is partly done by the study of the diffuse stellar populations at 0 redshift and the comparison of its properties at low redshift . such studies on fundamental relations ( fundamental plane , mg-@xmath5 , dn-@xmath5 ) tend to be consistent with the stellar populations evolving purely passively and having formed at high redshifts ( @xmath6 ) . alternatively , one can study merging events among galaxies at low- to intermediate - redshift ( e.g. van dokkum et al . 1999 ) in order to understand the assembly of galaxies . how do globular clusters fit into these picture ? globular cluster studies could be classified as the `` very lazy way '' , since they reach out to at most redshifts of @xmath7 . however , globular clusters are among the oldest objects in the universe , i.e. they witnessed most , if not all , of the history of their host galaxy . the goal of the globular cluster system studies is therefore to extract the memory of the system . photometry and spectroscopy are used to derive their ages and chemical abundances which are used to understand the epoch(s ) of star formation in the galaxy . kinematic information obtained from the globular clusters ( especially at large galactocentric radii ) can be used to help understanding the assembly mechanism of the galaxies . figure 1 illustrates extragalactic globular cluster studies . we show the galaxy ngc 1399 surrounded by a number of point sources . if these point sources could be resolved , they would look like one of the galactic globular clusters ( here shown as m 15 ) . however , even with diffraction limited imaging from space , we can not resolve clusters at distances of 10 to 100 mpc into single stars , and have to study their integrated properties . the study of a globular cluster system is therefore equivalent to studying the integrated properties of a large number of globular clusters surrounding a galaxy , in order to derive their individual properties and compare them , as well as the properties of the system as a whole , with the properties of the host galaxy . the purely practical advantages of observing objects at @xmath8 is the possibility to study the objects in great detail . very low @xmath9 observations are justified if the gain in details outbalances the fact that at high @xmath9 one is seeing events closer to the time at which they actually happened . one example that demonstrates that the gain is significant in the case of extragalactic globular clusters , is the discovery of several sub - populations of clusters around a large number of early - type galaxies . the presence of two or more distinct star - formation epochs / mechanisms in at least a large number if not all giant galaxies was not discovered by any other type of observations . the old age of globular clusters is often advanced as argument for their study , since they witnessed the entire past of the galaxy including the earliest epochs . if this would be the whole truth , globular clusters would not be suited to study the recent star formation epochs . nor would they present a real advantage over stars , which can be old too . what are the advantages of observing globular clusters as tracers of the star - formation / stellar populations instead of studying directly the diffuse stellar population of the host galaxies ? _ globular clusters trace star formation _ a number of arguments support the fact that globular clusters indeed trace the star formation in galaxies . however , we know that some star formation can occur without forming globular clusters . one example is the large magellanic cloud which , at some epochs , produced stars but no clusters ( e.g. geha et al . 1998 ) . on the other hand , we know that major star formation episodes induce the formation of a large number of star clusters . for example , the violent star formation in interacting galaxies is accompanied by the formation of massive young star clusters ( e.g. schweizer 1997 ) . also , the final number of globular clusters in a galaxy is roughly proportional to the galaxy luminosity , i.e. number of stars ( see harris 1991 ) . this hints at a close link between star and cluster formation . additional support for such a link comes from the close relation between the number of young star clusters in spirals and their current star formation rate ( larsen & richtler 1999 ) . in summary , globular clusters are not perfect tracers for star formation , as they will not form during every single little ( i.e. low rate ) star formation event . but they will trace the major ( violent ) epochs of star formation , which is our goal . _ from a practical point of view _ globular clusters exist around all luminous ( @xmath10 ) galaxies observed to date . their number , that scales with the mass of the galaxy , typically lies around a few hundreds to a few thousands . furthermore , globular clusters can be observed out to @xmath0 100 mpc . this is not as far as the diffuse light can be observed , but far enough to include many thousands of galaxies of all types and in all varieties of environments . the study of globular clusters is therefore not restricted to a specific type or environment of galaxies : unbiased samples can be constructed . from this point of view , diffuse stellar light and globular clusters are equally appropriate . _ the advantages of globular clusters over the diffuse stellar population _ globular clusters present a significant advantage when trying to determine the star formation history of a galaxy : they are far simpler structures . a globular cluster can be characterized by a single age and single metallicity , while the diffuse stellar population of a galaxy needs to be modeled by an unknown mix of ages and of metallicities . studying a globular cluster system returns a large number of discrete age / metallicity data points . these can be grouped to determine the mean ages and chemical abundances of the main sub - populations present in the galaxy . along the same line , and as shown above , globular clusters form proportionally to the number of stars . that is , the number of globular clusters in a given population reflects the importance of the star formation episode at its origin . counting globular cluster in different sub - populations indicates right away the relative importance of the different star formation events . in contrast , the different populations in the diffuse stellar light appear luminosity weighted : a small ( in terms of mass ) but recent star formation event can outshine a much more important but older event that has faded . _ the bonus _ as for stellar populations , kinematical informations can be derived from the spectra originally aimed at determining ages and metallicities . globular clusters have the advantage that they can be traced out to galactocentric distances unreachable with the diffuse stellar light . the dynamical information of the clusters can be used to study the assembly of the host galaxy . in section 6 , we will come back to this point . the bottom line is that globular clusters are good tracers for the star formation history of their host galaxies , and eventually allow some insight into their assembly too . they present a number of advantages over the study of stellar populations , and complement observations at high redshift . their study allows new insight into galaxy formation and evolution . this section is intended to give a feeling for the observational methods used to study globular cluster systems . it addresses the problems still encountered in imaging and spectroscopy , as well as the improvements to be expected with future instruments . to set the stage : we are trying to analyze the light of objects with typical half - light radii of 1 to 5 pc , at distances of 10 to 100 mpc ( i.e. half - light radii of 0.01 to 0.1 ) , and absolute magnitudes ranging from @xmath11 to @xmath12 ( i.e. @xmath13 ) . the galaxies in the nearest galaxy clusters ( virgo , fornax ) have globular cluster luminosity functions that peak in magnitude around @xmath14 , and globular clusters have typical half - light radii of @xmath0 0.05 . we intend to study both the properties of the individual clusters , as well as the ones of the whole cluster system . for the individual globular clusters , our goal is to derive their ages , chemical abundances , sizes and eventually masses . this requires spectral information ( the crudest being just a color ) and high angular resolution . for the whole system , our goal is to determine the total numbers , the globular cluster luminosity function , the spatial distributions ( extent or density profile , ellipticity ) , and any radial dependencies of the cluster properties ( e.g. metallicity gradients ) . these properties should also be measured for individual sub - populations , if they are present . the requirements for the systems are therefore deep , wide - field imaging , and the ability to distinguish potential sub - populations from each other . globular clusters outside the local group were , for a long time , exclusively studied with optical photometry . optical , ground - based photometry ( reaching @xmath15 in a field @xmath16 5@xmath17 5 ) turns out to be sufficient to determine most morphological properties of the systems ( see sect . 4 ) . the depth allows to reasonably sample the luminosity function , and a field of several arcminutes a side usually covers the vast majority of the system . problems with optical photometry arise when trying to determine ages and metallicities . it is well known that broad - band optical colors are degenerate in age and metallicity ( e.g. worthey 1994 ) . a younger age is compensated by a higher metallicity in most broad - band , optical colors . to some extent the problem is solved by the fact that most globular clusters are older than several gyr , and colors do not depend significantly on age in that range . this , of course , means that deriving ages from optical colors is hopeless , except for young clusters as seen e.g. in mergers . for old clusters , the goal is to find a color that is as sensitive to metallicity as possible . the widely used @xmath18 color is the least sensitive color to metallicity . @xmath19 and @xmath20 do better in the johnson - cousins system ( e.g. couture et al . 1991 for one of the first comparisons ) , but the mini break - through came with the use of washington filters ( e.g. geisler & forte 1990 ) . these allowed the discovery of the first multi - modal globular cluster color distributions around galaxies ( zepf & ashman 1993 ) . however , the common use of the johnson system , combined with large errors in the photometry at faint magnitudes , do still not allow a clean separation of individual globular clusters into sub - populations around most galaxies . another problem with ground - based imaging is that its resolution is by far insufficient to resolve globular clusters . this prohibits the unambiguous identification of globular clusters from foreground stars and background galaxies . in most cases the over - density of globular clusters around the host galaxy is sufficient to derive the general properties of the system . control fields should be used ( although often left out because of the lack of observing time ) , and statistical background subtraction performed . the individual identification of globular clusters became possible with wfpc2 on hst . globular clusters appear barely extended in wfpc2 images , which allows on the one hand to reliably separate them from foreground stars and background galaxies , and on the other hand to systematically study for the first time globular cluster sizes outside the local group ( kundu & whitmore 1998 , puzia et al . 1999 , kundu et al . 1999 ) . the disadvantage of wfpc2 observations is that the vast majority was carried out in @xmath18 , the least performant system in terms of metallicity sensitivity . furthermore , the wfpc2 has a small field of view which biases all the analysis towards the center of the galaxies , making it very hard to derive the global properties of a system without large extrapolations or multiple pointings . in summary , ground - based photometry returns the general properties of the systems , and eventually of the sub - systems when high quality photometry is obtained . it suffers from confusion when identifying individual clusters , and is limited in age / metallicity determinations . space photometry is currently as bad in deriving ages / metallicities , but allows to determine sizes of individual clusters . the current small fields , however , restrict the studies of whole systems . future progress is expected with the many wide - field imagers coming online , provided that deep enough photometry is obtained ( errors @xmath21 mag at @xmath22 ) . these will provide a large number of targets for spectroscopic follow - up . in space , the acs to be mounted on hst will superseed the wfpc2 . the field of view remains modest , but the slightly higher resolution will support further size determinations , and the higher throughput will allow a more clever choice of filters , including @xmath23 and @xmath24 . since the introduction of 1024@xmath171024 pixel arrays in the near infrared a couple of years ago , imaging at wavelength from 1.2@xmath25 m to 2.5@xmath25 m became competitive in terms of depths and field size with optical imaging ( see figure 2 ) . historically , the first near - infrared measurements of extragalactic globular clusters were carried out in m31 ( frogel , persson & cohen 1980 ) and the large magellanic cloud ( persson et al . why extend the wavelength range to the near infrared ? for old globular clusters , @xmath26 is a measure of the temperature of the red giant branch that is directly dependent on metallicity but hardly on age . @xmath26 is even more sensitive to metallicity than the washington @xmath27 index . the combination of optical and near - infrared colors is therefore superior to optical imaging alone , both for deriving metallicities , and for a clean separation of cluster sub - populations ( see figure 3 ) . it is also used to detect potential sub - populations were optical colors failed to reveal any . in young populations , @xmath26 is most sensitive to the asymptotic giant branch which dominates the light of populations that are 0.2 to 1 gyr old . the combination of optical and near infrared colors can be used to derive ages ( and metallicities ) of these populations ( e.g. maraston , kissler - patig , & brodie 2000 ) . the disadvantages of complementing optical with near - infrared colors is the need for a second instrument ( usually a second observing run ) in addition to the optical one . near - infrared observations will continue fighting against the high sky background in addition to the background light of the galaxy which requires blank sky observations . overall , obtaining near - infrared data is still very time consuming when compared to optical studies . for example , a deep @xmath28 image of a galaxy will require a full night of observations . currently both depth and field size do not allow the near infrared to fully replace optical colors for the study of morphological properties or the globular cluster luminosity function . but this might happen in the future whth the ngst . the immediate future looks bright , with a number of `` wide - field '' imagers being available , such as isaac on the vlt , sofi on the 3.5 m ntt , the omega systems on the 3.5 m calar alto , etc .. and 2k @xmath17 2k infrared arrays coming soon . the ideal future instrument would have a dichroic which would allow to observe simultaneously in the near - infrared and the optical . spectroscopy is the only way to unambiguously associate a globular cluster with its host galaxy by matching their radial velocities . also , it is the most precise way to determine the metallicity of single objects , and the only way to determine individual ages . obviously , it is also the only way to get radial velocities . ideally , one would like a spectrum of each globular cluster identified from imaging . in practice , good spectra are still hard to obtain . early attempts with 4m - class telescopes succeeded in obtaining radial velocities , but mostly failed to determine reliable chemical abundances ( see sect . 6 ) . with the arrival of 10m - class telescopes , it became feasible to obtain spectra with high enough signal - to - noise to derive chemical abundances ( kissler - patig et al . 1998a , cohen , blakeslee & ryzhov 1998 ) . such studies are still limited to relatively bright objects ( @xmath29 ) and remain time consuming ( @xmath0 3h integration time for low - resolution spectroscopy ) . figure 4 shows a few examples of globular clusters in ngc 1399 . high - resolution spectroscopy in order to measure internal velocity dispersions of individual clusters is still out of reach for old clusters , and was only carried out for two nearby super star clusters ( ho & fillipenko 1996a , b ) , in addition to several clusters within the local group . even low - resolution spectroscopy is currently still limited to follow - ups on photometric studies , targeting a number of selected , representative clusters , rather than building up own spectroscopic samples . current problems are the low signal - to - noise , even with 10m - telescopes , that prohibit very accurate age or metallicity determinations for individual clusters . the multiplexity of the existing instruments ( fors1 & 2 on the vlt , lris on keck ) is low and only allows to spectroscopy a limited number of selected targets . finally , the absorption indices that are being measured on the spectra in order to determine the various element abundances are not optimally defined . these indices lie in the region 3800 to 6000 and were designed for spectra with 8to 9 resolution . they often include a number of absorption lines in the bandpass ( or pseudo - continuum ) other than the element to be measured . this introduces an additional dependence e.g. of the balmer indices on metallicity , etc ... using a slightly higher resolution might help defining better indices . the immediate future of spectroscopy are instruments such as vimos on the vlt or deimos on keck that will allow a multiplexity of 100 to 150 . these will allow to increase the exposure times and slightly the spectral resolution to solve a number of the problems mentioned above . these will also allow to obtain several hundred radial velocities of globular clusters around a given galaxy in a single night , improving significantly the potential of kinematical studies ( see sect . in this section , we present some properties of globular cluster systems and of young clusters in ellipticals , spirals and mergers . in the last section we mentioned what are the properties measured in globular cluster systems : the metallicity distribution can be obtained from photometry ( colors ) or spectroscopy ( absorption line indices ) . the luminosity function of the clusters is computed from the measured magnitudes folded with any incompleteness or contamination function . the total number of clusters ( and eventually number of metal - poor and metal - rich clusters ) is obtained by extrapolating the observed counts over the luminosity function and eventually applying any geometrical completeness for the regions that are not covered . for the latter , one uses also knowledge about the spatial distribution ( position angle , ellipticity ) and radial density profile of the globular cluster system . for young star clusters , the color distribution no longer reflects the metallicity distribution , but a mix of ages and metallicities . more complex comparisons with population synthesis models and/or spectroscopy are needed to disentangle the two quantities . of interest for young clusters are also the mass distribution ( derived from the luminosity function ) that helps to understand how many of the newly formed clusters will indeed evolve into massive `` globular '' clusters . early - type galaxies have the best studied globular cluster systems . spirals have the two systems studied in most details ( i.e. the milky way and m31 ) due to our biased location in space , but a far larger sample now exists for early - type galaxies . despite looking remarkably similar in many respects ( e.g. globular cluster luminosity function ) , globular cluster systems in early - type galaxies also show a large scatter in a number of properties . for example , the number of globular clusters normalized to the galaxy light ( specific frequency , see harris & van den bergh 1981 ) appears to scatter by a factor of several , mainly driven by the very high specific frequencies of central giant ellipticals ( and recently also observed in faint dwarf galaxies , see durrel et al . 1996 , miller et al . 1998 ) . furthermore , the radial density profiles are very extended for large galaxies , while following the galaxy light in the case of intermediate ellipticals ( e.g. kissler - patig 1997a ) . in the early 90 s , zepf & ashman ( 1993 ) discovered the presence of globular cluster sub - populations in several early - type galaxies . we will come back to the origin of the sub - populations in sect . , we will discuss the implications of sub - populations on our understanding of the globular cluster system properties . until the early 90 s , properties were derived for the whole globular cluster system . since then , it became clear that many properties need to take into account the existence of ( at least two ) different sub - populations , in order to be explained . probably the first work to show this most clearly was the presentation of the properties of blue and red clusters in ngc 4472 by geisler et al . taking into account the existence and different spatial distribution of blue and red clusters , they explained two properties of whole systems at once . first , the color gradient observed in several systems could be explained by a varying ratio of blue to red clusters with radius ( without any gradient in the individual sub - populations ) . second , the mean color of the systems was previously thought to be systematically bluer than the diffuse galaxy light . it turns out that the color of the red sub - population matches the color of the galaxy , while it is the presence of the blue `` halo '' population that makes the color of the whole globular cluster system appear bluish . it has not yet been demonstrated that the scatter in the specific frequency and in the slopes of the radial density profiles also originate from different mixes of blue to red sub - populations , but this could be the case . the few studies that investigated separately the morphological properties of blue and red clusters ( geisler et al . 1996 , kissler - patig et al . 1997 , lee et al . 1998 , kundu & whitmore 1998 ) found the metal - poor ( blue ) population to be more spherically distributed and extended than the metal - rich population that has a steeper density profile , tends to be more flattened and appears to follow the diffuse stellar light of the galaxy in ellipticity and position angle ( cf . fig . 5 ) . thus , a larger fraction of blue clusters in a galaxy would mimic a flatter density profile of the whole globular cluster system . furthermore , the specific frequency of the blue clusters ( when related to the blue light ) appears to be very high ( @xmath30 see harris 2000 ) . this , by the way , could be explained if the latter came from small fragments similar to the dwarf ellipticals observed today , that also show high specific frequency values ( although not as high , but in the range 10 to 20 ) . thus , an overabundance of blue clusters would imply a high specific frequency . incidentally , the shallow density profiles are found in the galaxies with the highest specific frequencies ( see kissler - patig 1997a ) . we can therefore speculate that the properties of the entire globular cluster systems of these massive ( often central ) giant ellipticals can be explained by a large overabundance of metal - poor globular clusters originating from small fragments . the scatter in the globular cluster system properties among ellipticals could then ( at least partly ) be explained by a varying fraction of metal - poor `` halo '' and metal - rich `` bulge '' globular clusters . observationally , this could be verified by determining the number ratios of metal - rich and metal - poor globular clusters in a sample of galaxies showing different globular cluster radial density profiles and specific frequencies . the number of studies investigating the properties of metal - poor and metal - rich populations needs to increase in order to confirm the general properties of these two groups . we end with a word of caution : the existence of such sub - populations has been observed in only @xmath31% of all early - type galaxies studied ( e.g. gebhardt & kissler - patig 1999 ) , and still remains to be demonstrated in all cases . furthermore , the exact formation process of these sub - populations is still unclear ( see sect . 5 ) . the study of globular cluster systems of late - type spirals started with the work of shapley ( 1918 ) on the milky way system . despite a head - start of nearly 40 years compared to studies in early - type galaxies , the number of studied systems in spirals lags far behind the one in ellipticals . this is mainly due to the observational difficulties : globular clusters in spirals are difficult to identify on the inhomogeneous background of disks . furthermore , internal extinction in the spiral galaxies make detection and completeness estimations difficult , and photometry further suffers from confusion by reddened hii regions , open clusters or star forming regions . the best studied cases ( milky way and m31 ) show sub - populations ( e.g. morgan 1959 , kinman 1959 , zinn 1985 ; ashman & bird 1993 , barmby et al . 1999 ) associated in our galaxy with the halo and the bulge / thick disk ( minniti 1995 , ct 1999 ) . beyond the local group , spectroscopy is needed to separate potential sub - populations . both abundances and kinematics are needed , while colors suffer too much from reddening to serve as useful metallicity tracers . spectroscopic studies have been rare in the past , but are now becoming feasible ( see sect . 3.3 and 4.1 ) . for example , a recent study of m81 allowed to identify a potential thick disk population beside halo and bulge populations ( see schroder et al . 2000 and references therein ) . some of the open questions are whether all spirals host halo and bulge clusters , and whether one or both populations are related to the metal - poor and metal - rich populations in early - type galaxies . the number of globular clusters as traced by the specific frequency appears roughly constant in spirals of all types independently of the presence of a bulge and/or thick disk ( e.g. kissler - patig et al . this would mean that spirals are dominated by metal - poor populations , with their globular cluster systems only marginally affected by the presence of a bulge / thick disk . if metal - poor globular clusters indeed formed in pre - galactic fragments , then one would expect the metal - poor populations in spirals and ellipticals to be the same . we know that the globular cluster luminosity functions are extremely similar , but the metallicity distributions and other properties remain to be derived and compared ( see burgarella et al . 2000 for a first attempt ) . finally , a good understanding of the globular cluster systems in spirals will also help predicting the resulting globular cluster system of a spiral spiral merger . predictions can then be compared to the properties of systems of elliptical galaxies in order to constrain this mode of galaxy formation . after some speculations and predictions that massive star clusters could / should form in mergers ( harris 1981 , schweizer 1987 ) , these were finally discovered in the early 90 s ( lutz 1991 , holtzman et al . since then a number of studies focussed on the detection and properties of these massive young star clusters ( see schweizer 1997 , and reviews cited in sect . 1 for an overview ) . the most intense debate around these young clusters focussed on whether or not their properties were compatible with a formation of early - type galaxies through spiral spiral mergers . it was noticed early on ( harris 1981 , van den bergh 1982 ) that ellipticals appeared to host more clusters than spirals , and thus that mergers would have to produce a large number of globular clusters . moreover , the specific frequency of ellipticals appeared higher than in spirals , i.e. mergers were supposed to form globular clusters extremely efficiently . in a second stage , a number of studies investigated whether or not these newly formed clusters would resemble globular clusters , and/or would survive as bound clusters at all . the above questions are still open , except maybe for the last one . the young clusters studied to date show luminosities , sizes , and masses ( when they can be measured ) that are compatible with them being bound stellar clusters and able to survive the next several gyr ( see schweizer 1997 for a summary of the studies and extensive references ) . whether they will have the exact same properties as old globular clusters in our milky way is still controversial . first spectroscopic measurements found the young clusters in ngc 7252 compatible with a normal initial mass function ( imf ) ( schweizer & seitzer 1998 ) , while in ngc 1275 the young clusters show anomalies and potentially have a flatter imf ( brodie et al . 1998 ) which would compromise their evolution into old globular clusters , as we know them from the galaxy . the mass distribution of these young cluster was first found to be a power - law ( e.g. meurer 1995 ) , as opposed to a log - normal distribution for old globular clusters . this result is likely to suffer from uncertainties in the conversion of luminosities into masses , when neglecting the significant age spread among the young clusters ( see fritze - von alvensleben 1999 ) . however , deeper data seem to rule out the possibility that the initial mass distribution has already the same shape as the one observed for old clusters ( see whitmore et al . 1999 , zepf et al . 1999 ) . but the slope of the mass distributions could be affected during the evolution of the system by dynamical destruction at the low - mass end . finally , whitmore et al . ( 1999 ) recently found a break in the mass function of the young clusters of the antennae galaxies , similar to the characteristic mass of the old clusters further supporting similar mass functions for young and old cluster populations ( see also sect . overall , the young clusters might or might not resemble old galactic globular clusters , but some will survive as massive star clusters and could mimic a population of metal - rich globular clusters . the most interesting point remains the number of clusters produced in mergers . obviously , this will depend on the gas content ( ` fuel ' ) that is provided by the merger ( e.g. kissler - patig , et al . most gas - rich mergers form a large number of star clusters , but few of the latter have masses that would actually allow them to evolve into massive globular clusters as we observe them in distant ellipticals . harris ( 2000 ) reviews comprehensively this issue and other problems related with a scenario in which all metal - rich globular clusters of ellipticals would have formed in mergers . the main problem with such a scenario is that the high specific frequency of ellipticals should be due to metal - rich clusters , which is usually not the case . potential other problems , depending on the exact enrichment history , are that large ellipticals would be build up by a series of mergers that should probably produce an even broader metallicity distribution than observed ; and that radial metallicity gradients might be expected to be steeper in high specific frequency ellipticals . in summary , mergers are the best laboratories to study younger stellar populations and the formation of young stellar cluster , but how important they are in the building of globular cluster systems ( and galaxies ) remains uncertain . however , a good understanding of these clusters is crucial for the understanding of globular cluster systems in early - type galaxies , since merger events must have played a role at some stage . in this section we come back to the presence of multiple sub - population of globular clusters around a number of giant galaxies . we will briefly review the different scenarios present in the literature that could explain the properties of such composite systems and discuss their pros and cons . sub - populations of globular clusters were first identified in the milky way ( morgan 1959 , kinman 1959 , zinn 1985 ) , and associated with the halo ( in the case of the metal - poor population ) and the `` disk '' ( in the case of the metal - rich population . the `` disk '' clusters are now better associated with the bulge ( e.g. minniti 1995 , ct 1999 ) . the presence of multiple component populations in other giant galaxies was first detected by zepf & ashman ( 1993 ) . obviously the multiple sub - populations get associated with several distinct epochs or mechanisms of star / cluster formation . the simple scenario of a disk disk merger explaining the presence of two populations of globular clusters ( ashman & zepf 1992 ) found a strong support in the community for 5 - 6 years , partly because of a lack of alternatives . it was backed up by the discovery of newly formed , young star clusters in interacting galaxies ( lutz et al . 1991 , holtzman et al . 1992 ) . only recently , other scenarios explaining the presence of at least two distinct populations were presented and discussed . we will make a ( somewhat artificial ) separation in four scenarios and briefly outline them and their predictions . the fact that mergers could produce new globular clusters was mentioned in the literature early after toomre ( 1977 ) proposed that ellipticals could form out of the merging of two spirals ( see harris 1981 and schweizer 1987 ) . but the first crude predictions of the spiral - spiral merger scenario go back to ashman & zepf ( 1992 ) . they predicted two populations of globular clusters in the resulting galaxy : one old , metal - poor population from the progenitor spirals and one newly formed , young , metal - rich population . the metal - poor population would be more extended and would have been transfered some of the orbital angular momentum by the merger . the metal - rich globular clusters would be more concentrated towards the center and probably on more radial orbits . in situ scenarios see all globular clusters forming within the entity that will become the final galaxy . in this scenario , globular clusters form in the collapse of the galaxy , which happens in two distinct phases ( see forbes et al . 1997 , harris et al . 1998 , harris et al . 1999 ) . the first burst produces metal - poor globular clusters and stars ( similar to searle & zinn 1978 ) and provokes its own end e.g. by ionizing the gas or expelling it ( e.g. harris et al . the second collapse happens shortly later ( 1 - 2 gyr ) and is at the origin of the metal - rich component . both populations are linked with the initial galaxy . accretion scenarios were reconsidered in detail to explain the presence of the large populations of metal - poor globular clusters around early - type galaxies . in these scenarios , the metal - rich clusters belong to the seed galaxy , while the metal - poor clusters are accreted from or with dwarf galaxies ( e.g. richtler 1994 ) . ct et al . ( 1998 ) showed in extensive simulations that the color distributions could be reproduced . hilker ( 1998 ) and hilker et al . ( 1999 ) proposed the accretion of stellar as well as gas - rich dwarfs that would form new globular when accreted . in such scenarios , the metal - poor clusters would not be related to the final galaxies but rather have properties compatible with that of globular clusters in dwarf galaxies . furthermore , this scenario is the only one that could easily explain metal - poor cluster that are younger than metal - rich ones . in a slightly differently scenario , kissler - patig et al . ( 1999b ) mentioned the possibility that central giant ellipticals could have accreted both metal - poor and metal - rich clusters from surrounding medium - sized galaxies . pre - galactic scenarios were proposed long ago by peebles & dicke ( 1968 ) , when the jeans mass in the early universe was similar to globular cluster masses . meanwhile , it was reconsidered in the frame of globular cluster systems ( kissler - patig 1997b , kissler - patig et al . 1998b , burgarella et al . the metal - poor globular clusters would have formed in fragments before the assembly of the galaxy , later - on building up the galaxy halos and feeding with gas the formation of the bulge . in that scenario too , the metal - poor globular clusters do not have properties dependent from the final galaxy , while the metal - rich clusters do . also , metal - poor clusters are older than metal - rich clusters . overall , the scenarios are discussed in the literature as different but do not differ by much . the first scenario explains the presence of the metal - rich population , as opposed to the last two that deal with the metal - poor population . these three scenarios are mutually not exclusive . only in situ models connect the metal - rich and metal - poor components . for the metal - rich clusters , the question resumes to whether they formed during the collapse of the bulge / spheroid , or whether they formed in a violent interaction . although an early , gas - rich merger event at the origin of the bulge / spheroid would qualify for both scenarios . in the case of metal - poor clusters , the difference between the last three scenarios is mostly semantics . they differ slightly on when the clusters formed , and models two and four might expect differences in whether or not the properties of the clusters are related to the final galaxy . but the bottom line is that the boarder - line between the scenarios is not very clear . explaining the building up of globular cluster systems is probably a matter of finding the right mix of the above mechanisms , and this for every given galaxy . the predictions of the different scenarios are fairly fuzzy , and no scenario makes clear , unique predictions . nevertheless , we can present the pros and cons to outline potential problems with any of them . pros : we know that new star cluster form in mergers ( e.g. above mentioned reviews , and see schweizer 1997 ) , and will populate the metal - rich sub - population of the resulting galaxy . note also , that the merger scenario is the only one that predicted bimodal color distributions rather then explaining them after fact . cons : we do not know _ i ) _ if all early - type galaxies formed in mergers , _ ii ) _ if the star clusters formed in mergers will indeed evolve into globular clusters ( e.g. brodie et al . 1998 ) , _ iii ) _ if all mergers produce a large number of clusters ( which depends on the gas content ) . furthermore , we would then expect the metal - rich populations to be significantly younger in many galaxies ( according e.g. to the merger histories predicted by hierarchical clustering models ) . there are still problems in explaining the specific frequencies and the right mix of blue and red clusters in early - type galaxies in the frame of the merger scenario ( e.g. forbes et al . 1997 ) . pros : searle & zinn ( 1978 ) list the evidences for our milky way halo globular clusters to have formed in fragments building up the halo . the massive stars in such a population would quickly create a hold of the star / cluster formation for a gyr or two . cons : if a correlation between metal - poor clusters and galaxy is expected , the scenario would be ruled out . a clear age sequence from metal - poor to metal - rich clusters is predicted but not yet verified . this scenarios is not in line with hierarchical clustering models for the formation of galaxies ( kauffmann et al . 1993 , cole et al . 1994 ) , should the latter turn out to be the right model for galaxy formation . pros : dwarf galaxies are seen in great numbers around giant galaxies , and hierarchical clustering scenarios predict even more at early epochs . dwarf galaxies do get accreted ( e.g. sagittarius in our galaxy ) . we observe `` free - floating '' populations around central cluster galaxies ( e.g. hilker et al . 1999 ) and the color distributions of globular cluster systems can be reproduced ( ct et al . 1998 ) . cons : we are missing detailed dynamical simulations of galaxy groups and clusters to test whether the predicted large number of dwarf galaxies gets indeed accreted ( and when ) . we do not know whether the ( dwarf ) galaxy luminosity function was indeed as steep as required at early times to explain the large accretion rates needed . also , the model does not provide a physical explanation for the metal - rich populations . pros : similar to the above , we observe a `` free - floating '' , spatially extended populations of globular clusters around central galaxies . the properties of the metal - poor populations do not seem to correlate with the properties of their host galaxies ( burgarella et al . 2000 ) . the metal - poor globular cluster are observed to be very old ( e.g. ortolani et al . 1995 for our galaxy ; kissler - patig et al . 1998a , cohen et al . 1998 , puzia et al . 1999 for analogies in extragalactic systems ) . cons : galaxies and galaxy halos might not have formed by the agglomeration of independent fragments . no physical model exists , except a broad compatibility with hierarchical clustering models ( see also burgarella et al . 2000 ) . some pros and cons are listed only under one scenario but apply obviously to others . it should be noted that these pros and cons apply to `` normal '' globular cluster systems . it has been noted that several galaxies host very curious mixes of metal - poor and metal - rich clusters ( gebhardt & kissler - patig 1999 , harris et al . 2000 ) that pose challenges to all scenarios . fine difference will require a much more detailed abundance analysis of the individual clusters in sub - populations , as well as their dynamical properties and ( at least relative ) ages for the different globular cluster populations . these might allow to identify a unique prediction supporting the one or the other formation mode , or constrain the importance of each formation mechanism . in this section , we briefly discuss recent results from kinematical studies of extragalactic globular clusters . the required measurements were discussed in sect . 3 . kinematics can be used both to understand the formation of the globular cluster systems , as well as to derive dynamics of galaxies at large radii . globular cluster system kinematics are used since a long time to constrain their formation . in the milky way , kinematics support the association of the various clusters with the halo and the bulge ( see harris 2000 and references therein ) . in m31 , similar results were derived ( huchra et al . 1991 , barmby et al . 1999 ) . in m81 , the situation appears very similar again ( schroder et al . 2000 ) . beyond the local group , radial velocities for globular clusters are somewhat harder to obtain . nevertheless , studies of globular cluster kinematics in elliptical galaxies started over a decade ago ( mould et al . 1987 , 1990 , huchra & brodie 1987 , harris 1988 , grillmair et al . 1994 ) . figure 6 illustrates one example where the kinematics of globular clusters allowed to gain some insight into the globular cluster system formation ( from kissler - patig et al . the figure shows the velocity dispersion as a function of radius around ngc 1399 , the central giant elliptical in fornax . the velocity dispersion of the globular clusters increases with radius , rising from a value not unlike that for the outermost stellar measurements at 100 , to values almost twice as high at @xmath0 300 . 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 . thus , a large fraction of the globular clusters which we associate with ngc 1399 could rather be attributed to the whole of the fornax cluster . by association , this would be true for the stars in the cd envelope too . this picture strongly favors the accretion or pre - galactic scenarios for the formation of the metal - poor clusters in this galaxy . as another example , fig . 7 shows the velocity dispersion and rotational velocity for the metal - poor and metal - rich globular clusters around m87 , the central giant elliptical in virgo . there is some evidence that the rotation is confined to the metal poor globular clusters . if , as assumed , the last merger was mainly dissipationless ( and did not form a significant amount of metal - rich clusters ) , this kinematic difference between the two sub populations could reflect the situation in the _ progenitor _ galaxies of m87 . these would then be compatible ( see hernquist & bolte 1992 ) with a formation in a gas - rich merger event ( see ashman & zepf 1992 ) . generally , the data seem to support the view that the metal - poor globular clusters form a hot system with some rotation , or tangentially biased orbits . the metal - rich globular clusters have a lower velocity dispersion in comparison , and exhibit only weak rotation , if at all ( cohen & ryzhov 1997 , kissler - patig et al . 1999b , sharples et al . 1999 , kissler - patig & gebhardt 1999 , cohen 2000 ) . the interpretation of these results in the frame of the different formation scenarios presented in sect . 5 is unclear , since no scenario makes clear and unique predictions for the kinematics of the clusters . furthermore , some events unrelated to the formation of the globular clusters can alter the dynamics : e.g. a late dissipationless mergers of two ellipticals could convey angular momentum to both metal - rich and metal - poor clusters , bluring kinematical signatures present in the past . detailed dynamical simulations of globular cluster accretion and galaxy mergers are necessary in order to compare the data with scenario predictions . but clearly , kinematics can help understanding differences in the metal - poor and metal - rich components , exploring intra - cluster globular clusters , and studying the formation of globular cluster systems as a whole . kinematical studies of globular clusters can also be used to study galaxy dynamics . the globular clusters do only represent discrete probes in the gravitational potential of the galaxy , as opposed to the diffuse stellar light that can be used as a continuous probe with radius , but globular clusters have the advantage ( such as planetary nebulae ) to extend further out . globular clusters can be measured out to several effective radii , probing the dark halo and dynamics at large radii . the velocity dispersion around ngc 1399 , presented above , is one example . another example was presented by cohen & ryzhov ( 1997 ) who derived from the velocity dispersion of the globular clusters in m87 a mass of @xmath32 at 44kpc and a mass - to - light ratio @xmath30 , strongly supporting the presence of a massive dark halo around this galaxy . with the same data , kissler - patig & gebhardt ( 1998 ) derived a spin for m87 of @xmath33 , at the very high end of what is predicted by cosmological n - body simulations . the authors suggested as most likely explanation for the data a major ( dissipationless ) merger as the last major event in the building of m87 . these examples illustrate what can be learned about the galaxy formation history from kinematical studies of globular cluster system . in the future , instruments such as vimos and deimos will allow to get many hundreds velocities in a single night for a given galaxy . these data will allow to constrain even more strongly galaxy dynamics at large radii . in this section , we will review the globular cluster luminosity function ( gclf ) as a distance indicator . the method is currently `` unfashionable '' in the literature mainly because some previous results seem to be in contradiction with other distance indicators ( e.g. ferrarese et al . we will try to shade some light on the discrepancies , and show that , if the proper corrections are applied , the gclf competes well with other extragalactic distance indicators . a nice overview of the method is given in harris ( 2000 ) , including some historical remarks and a detailed description of the method . a further review on the gclf method was written by whitmore ( 1997 ) , who addressed in particular the errors accompanying the method . we will only briefly summarize the method here . the basics of the method are to measure in a given filter ( most often @xmath34 ) the apparent magnitudes of a large number of globular clusters in the system . the so constructed magnitude distribution , or luminosity function , peaks at a characteristic ( turn - over ) magnitude . the absolute value for this characteristic magnitude is derived from local or secondary distance calibrators , allowing to derive a distance modulus from the observed turn - over magnitude . figure 8 shows a typical globular cluster luminosity function with its clear peak ( taken from della valle et al . 1998 ) . the justification of the method is mainly empirical . apparent turn - overs for galaxies at the same distance ( e.g. in the same galaxy cluster ) can be compared and a scatter around 0.15 mag is then obtained , without correcting for any external error . similarly , a number of well observed apparent turn - over magnitudes can be transformed into absolute ones using distances from other distance indicators ( cepheids where possible , or a mean of cepheids , surface brightness fluctuations , planetary luminosity function , ... ) and a similar small scatter is found ( see harris 2000 for a recent compilation ) . taking into account the errors in the photometry , the fitting of the gclf , the assumed distances , etc ... this hints at an internal dispersion of the turn - over magnitude of @xmath35 , making it a good standard candle . from a theoretical point of view , this constancy of the turn - over magnitude translates into a `` universal '' characteristic mass in the globular cluster mass distributions in all galaxies . whether this is a relict of a characteristic mass in the mass function of the molecular clouds at the origin of the globular clusters , or whether it was implemented during the formation process of the globular clusters is still unclear . the absolute turn - over magnitude lies around @xmath36 , and the determination of the visual turn - over is only accurate if the peak of the gclf is reached by the observations . from an observational point of view , this means that the data must reach e.g. @xmath37 to determine distances in the fornax or virgo galaxy clusters ( d@xmath38 mpc ) , and that with hst or 10m - class telescopes reaching typically @xmath39 , the method could be applied as far out as 120 mpc ( including the coma galaxy cluster ) . the observational advantages of this method over others are that globular clusters are brighter than other standard candles ( except for supernovae ) , and do not vary , i.e. no repeated observations are necessary . further , they are usually measured at large radii or in the halo of ( mostly elliptical ) galaxies where reddening is not a concern . a large number of distance determinations from the gclf were only by - products of globular cluster system studies , and often suffered from purely practical problems of data taken for different purposes . first , a good estimation of the background contamination is necessary to clean the globular cluster luminosity function from the luminosity function of background galaxies which tends to mimic a fainter turn - over magnitude . next , the finding incompleteness for the globular clusters needs to be determined , in particular as a function of radius since the photon noise is changing dramatically with galactocentric radius . proper reddening corrections need to be applied and might differ whether one uses the `` classical '' reddening maps of burstein & heiles ( 1984 ) or the newer maps from schlegel et al . ( 1998 ) . when necessary , proper aperture correction for slightly extended clusters on wfpc2/hst images has to be made ( e.g. puzia et al . finally , several different ways of fitting the gclf are used : from fitting a histogram , over more sophisticated maximum - likelihood fits taking into account background contamination and incompleteness . the functions fitted vary from gaussians to student ( t@xmath40 ) functions , with or without their dispersion as a free parameter in addition to the peak value . in addition to these , errors in the absolute calibration will be added ( see below ) . furthermore , dependences on galaxy type and environment were claimed , although the former is probably due to the mean metallicity of globular clusters differing in early- and late - type galaxies , while the latter was never demonstrated with a reliable set of data . all the above details can introduce errors in the analysis that might sum up to several tenths of a magnitude . the fact that distance determinations using the gclf are often a by - product of studies aiming at understanding globular cluster or galaxy formation and evolution , did not help in constructing a very homogeneous sample in the past . the result is a very inhomogeneous database ( e.g. ferrarese et al . 1999 ) dominated by large random scatter introduced in the analysis , as well as systematic errors introduced by the choice of calibration and the complex nature of globular cluster systems ( see below ) . nevertheless , most of these problems were recognized and are overcome by better methods and data in the recent gclf distance determinations . harris ( 2000 , see also kavelaars et al . 2000 ) outline what we will call the classical way of measuring distances with the gclf . this method implies that the gclf is measured from all globular clusters in a system . in addition , it uses the gclf as a `` secondary '' distance indicator , basing its calibration on distances derived by cepheids an other distance indicators . the method compares the peak of the observed gclf with the peak of a compilation of gclfs from mostly virgo and fornax ellipticals , adopting from the literature a distance to these calibrators . this allowed , among others , harris group to determine a distance to coma ellipticals and to construct the first hubble diagram from gclfs in order to derive a value for h@xmath41 ( harris 2000 , kavelaars et al . 2000 ) . in practice , an accurate gclf turn - over is determined ( see above ) and calibrated without any further corrections using m@xmath42 ( harris 2000 ) or m@xmath43 using virgo alone ( kavelaars et al . 2000 ) . the advantages of this approach are the following . using all globular clusters ( instead of a limited sub - population ) often avoids problems with small number statistics . this is also the idea behind using virgo gclfs instead of the spars milky way gclf as calibrators . the virgo gclfs , derived from giant elliptical galaxies rich in globular clusters , are well sampled and do not suffer from small number statistics . further , since most newly derived gclfs come from cluster ellipticals , one might be more confident to calibrate these using virgo ( i.e. cluster ) ellipticals , in order to avoid any potential dependence on galaxy type and/or environment . however , the method has a number of caveats . the main one is that giant ellipticals are known to have globular cluster sub - populations with different ages and metallicities . this automatically implies that the different sub - populations around a given galaxy will have different turn - over magnitudes . by using the whole globular cluster systems , one is using _ a mix _ of turn - over magnitudes . one could in principal try to correct e.g. for a mean metallicity ( as proposed by ashman , conti & zepf 1995 ) , but this correction depends on the population synthesis model adopted ( see puzia et al . 1999 ) and implies that the mix of metal - poor to metal - rich globular clusters is known . this mix does not only vary from galaxy to galaxy ( e.g. gebhardt & kissler - patig 1999 ) , but also varies with galactocentric radius ( e.g. geisler et al . 1996 , kissler - patig et al . it results in a displacement of the turn - over peak and the broadening of the observed gclf of the whole system . the virgo ellipticals are therefore only valid calibrators for other giant ellipticals with a similar ratio of metal - poor to metal - rich globular clusters _ and _ for which the observations cover similar radii . this is potentially a problem when comparing ground - based ( wide - field ) studies with hst studies focusing on the inner regions of a galaxy . or when comparing nearby galaxies where the center is well sampled to very distant galaxies for which mostly halo globulars are observed . in the worse case , ignoring the presence of different sub - populations and comparing very different galaxies in this respect , can introduce errors a several tenths of magnitudes . another caveat of the classical way , is that relative distances to virgo can be derived , but absolute magnitudes ( and e.g. values of h@xmath41 ) will still dependent on other methods such as cepheids , surface brightness fluctuations ( sbf ) , planetary nebulae luminosity functions ( pnlf ) , and tip of red - giant branches ( trgb ) , i.e. the method will never overcome these other methods in accuracy and carry along any of their potential systematic errors . as an alternative to the classical way , one can focus on the metal - poor clusters only . the idea is to isolate the metal - poor globular clusters of a system and to determine their gclf . as a calibrator , one can use the gclf of the metal - poor globular clusters in the milky way , which avoids any assumption on the distance of the lmc and will be independent of any other extragalactic distance indicator . for the milky way gclf , the idea is to re - derive an absolute distance to each individual cluster , resulting in individual absolute magnitudes and allowing to derive an absolute luminosity function . individual distances to the clusters are derived using the known apparent magnitudes of their horizontal branches and a relation between the absolute magnitude of the horizontal branch and the metallicity ( e.g. gratton et al . the latter is based on hipparcos distances to sub - dwarfs fitted to the lower main sequence of chosen clusters . this methods follows a completely different path than methods based at some stage on cepheids . in particular , the method is completely independently from the distance to the lmc . in practice , an accurate gclf turn - over ( see above ) for the metal - poor clusters in the target galaxy is derived and calibrated , without any further corrections , using m@xmath44 derived from the metal - poor clusters of the milky way ( see della valle et al . 1998 , drenkhahn & richtler 1999 ; note that the error is statistical only and does not include any potential systematic error associated with the distance to galatic globular clusters , currently under debate ) . the advantages of this method are the following . this method takes into account the known sub - structures of globular cluster systems . using the metal - poor globular clusters is motivated by several facts . first , they appear to have a true universal origin ( see burgarella et al . 2000 ) , and their properties seem to be relatively independent of galaxy type , environment , size and metallicity . thus , to first order they can be used in all galaxies without applying any corrections . in addition , the milky way is justified as calibrator even for gclfs derived from elliptical galaxies . further , they appear to be `` halo '' objects , i.e. little affected by destruction processes that might have shaped the gclf in the inner few kpc of large galaxies , or that affect objects on radial orbits . they will certainly form a much more homogeneous populations than the total globular cluster system ( see previous sections ) . using the milky way as calibrator allows this method to be completely independent on other distance indicators and to check independently derived distances and value of h@xmath41 . the method is not free from disadvantages . first , selecting metal - poor globular clusters requires better data than are currently used in most gclf studies , implying more complicated and time - consuming observations . second , even with excellent data a perfect separation of metal - poor and metal - rich clusters will not be possible and the sample will be somewhat contaminated by metal - rich clusters . worse , the sample size will be roughly halved ( for a typical ratio of blue to red clusters around one ) . this might mean that in some galaxies less than hundred clusters will be available to construct the luminosity function , inducing error @xmath45 on the peak determination due to sample size alone . finally , the same concerns applies as for the whole sample : how universal is the gclf peak of metal - poor globular clusters ? this remains to be checked , but since variations of the order of @xmath35 seem to be the rule for whole samples , there is no reason to expect a much larger scatter for metal - poor clusters alone . two examples of distance determinations from metal - poor clusters were given in della valle et al . ( 1998 ) , and puzia et al . ( 1999 ) . the first study derived a distance modulus for ngc 1380 in the fornax cluster of @xmath46 ( not including a potential systematic error of up to 0.2 ) . in this case , the gclf of the metal - poor and the metal - rich clusters peaked at the same value , i.e. the higher metallicity was compensated by a younger age ( few gyr ) of the red globular cluster population , so that it would not make a difference whether one uses the metal - poor clusters alone or the whole system . as a comparison , values derived from cepheids and a mean of cepheids / sbf / pnlf to fornax are @xmath47 ( ferrarese et al . 1999 ) and @xmath48 ( from kavelaars et al . 2000 ) . in the case of ngc 4472 in the virgo galaxy cluster , puzia et al . ( 1999 ) derived turn - overs from the metal - poor and metal - rich clusters of @xmath49 and @xmath50 respectively . using the metal - poor clusters alone , their derived distance is then @xmath51 . this compares with the cepheid distance to virgo from 6 galaxies of @xmath52 and to the mean of cepheids / sbf / trgb / pnlf of @xmath53 ( from kavelaars et al . both cases show clearly the excellent agreement of the gclf method with other popular methods , _ despite the completely different and independent calibrators used_. the accuracy of the gclf method will always be limited by the sample size and lies around @xmath54 . a nice example of the `` classical way '' is the recent determination of the distance to coma . at the distance of @xmath55 mpc the separation of metal - poor and metal - rich globular clusters is barely feasible anymore , and using the full globular cluster systems is necessary . kavelaars et al . ( 2000 ) derived turn - over values of m@xmath56 and m@xmath57 for the two galaxies ngc 4874 and ic 4051 in coma , respectively . using virgo ellipticals as calibrators and assuming a distance to virgo of @xmath53 , they derive a distance to coma of @xmath58 mpc . adding several turn - over values for distant galaxies ( taken from lauer et al . 1998 ) , they construct a hubble diagram for the gclf technique and derive a hubble constant of h@xmath59 km s@xmath60 mpc@xmath60 . this example demonstrates the reach in distance of the method . in summary , we think that the method is mature now and that most errors in the analysis can be avoided , as well as good choices for the calibration made . in the future , with hst and 10m - class telescope data , a number of determinations in the 100 mpc range will emerge , and eventually , using metal - poor globular clusters only , this will give us a grasp on the distance scale completely independent from distances based at some stage on the lmc distance or cepheids . all the previous section should have made clear that globular clusters can be used for a very wide variety of studies . they can constrain the star formation history of galaxies , in particular on the two or more distinct epochs of star formation in early - type galaxies . they can help explaining the building up of spiral galaxies , and the star formation in violent interactions . they can be useful to study galaxy dynamics at large galactocentric radii . and finally , they provide an accurate distance indicator , independent of cepheids and the distance to the lmc . this makes the study of globular cluster systems one of the most versatile fields in astronomy . extragalactic globular cluster research experienced a boom in the early 90s with the first generation of reliable ccds , and the first imaging from space . we can expect a continuation of the improvement of optical imaging , but more important , the field will benefit from the advancement in near - infrared imaging , and most of all , of the upcoming multi - object spectrographs on 10 m telescopes . the next little revolution in this subject will come with the determination of hundreds of globular cluster abundances around a large number of galaxies . the next 5 years will be an exiting time . first of all , i would like to thank the astronomische gesellschaft for awarding me the ludwig - bierman price . i feel extremely honored and proud . for his constant support , i would like to thank tom richtler , who introduced me to the fascinating subject of globular clusters . for the most recent years , i would like to thank jean brodie for her collaboration and for giving me the first opportunity to use a 10 m telescope to satisfy my curiosity . i am grateful to my current collaborators thomas puzia , claudia maraston , daniel thomas , denis burgarella , veronique buat , sandra chapelon , michael hilker , dante minniti , paul goudfrooij , linda schroder and many others , for helping me to keep up the flame . as usual , i would be lost without karl gebhardt s codes and sharp ideas . i am grateful to klaas de boer and simona zaggia for comments on various points . and last but not least , i am very thankful to steve zepf for a critical reading of the manuscript .

given a closed , oriented surface , the algebraic intersection of closed curves induces a symplectic form on the first homology group of . if is equipped with a riemannian metric , the first homology group of inherits a norm , called the stable norm . we study the norm of the bilinear form , with respect to the stable norm .

1302.4692

let @xmath0 be a closed ( i.e. compact , without boundary ) manifold of dimension two , different from the 2-sphere , equipped with an orientation 2-form @xmath3 . if @xmath4 and @xmath5 are two @xmath6 closed curves on @xmath0 which intersect transversally , we call algebraic intersection of @xmath7 and @xmath5 the number @xmath8 where * @xmath9 denotes the tangent vector to @xmath4 at @xmath10 * the sum is taken over all pairs of parameter values @xmath11 such that @xmath12 . it is classical that this number only depends on the homology classes of @xmath4 and @xmath5 . we denote it by @xmath13,\left[\beta\right])$ ] . the map @xmath1 extends by linearity to a symplectic ( i.e bilinear , antisymmetric , nondegenerate ) form on the first homology @xmath14 of @xmath0 . the central question in this paper is _ when @xmath0 is endowed with a riemannian metric @xmath2 , how much can two curves of a given length intersect ? _ this amounts to evaluating the norm of the bilinear form @xmath1 with respect to a certain norm on @xmath14 , called the stable norm . informally speaking the stable norm measures the size , relative to the metric @xmath2 , of a homology or cohomology class . various equivalent definitions exist , see @xcite . we shall use that of @xcite : for @xmath15 and a vector @xmath16 , we denote by @xmath17 its riemannian norm . the _ comass _ of a differential one - form on @xmath0 is given by @xmath18 equation ( [ comass ] ) defines a norm on the space @xmath19 of smooth 1-forms on @xmath0 . we get a norm on the first cohomology of @xmath0 by taking the infimum of the comass over all smooth closed 1-forms in a given cohomology class : @xmath20=c \}.\ ] ] the norm @xmath21 is called the stable norm on @xmath22 . we denote in the same way the dual norm on @xmath14 . we say a homology class @xmath23 is integer if @xmath24 is the image in @xmath25 of an element of @xmath26 . when @xmath0 is an orientable surface of genus @xmath27 , and the homology class @xmath24 is integer , the stable norm has a nice expression , see @xcite : @xmath28 is the minimum of all sums @xmath29 , where * the index @xmath30 ranges over @xmath31 * @xmath32 denotes the length with respect to @xmath2 * the @xmath33 are integer numbers * the @xmath34 are pairwise disjoint simple closed geodesics * @xmath35 $ ] . the norm of the bilinear form @xmath36 with respect to the stable norm on @xmath37 is then defined as : @xmath38 observe that in the above expression , the supremum is actually a maximum , since the function @xmath39 is zero - homogeneous , so it is actually defined on the projectivized of @xmath14 , which is compact . when there is no ambiguity on @xmath0 and @xmath2 , we shall sometimes abbreviate the notation @xmath40 to @xmath41 . while , from a geometrical standpoint , the stable norm is the most natural norm on @xmath14 , from the complex analysis viewpoint , the most natural norm is the @xmath42-norm . for any differential one - form @xmath43 and for @xmath44 we denote by @xmath45 the norm , with respect to the metric @xmath2 , of the corresponding linear form on @xmath46 . then we define the @xmath42-norm of @xmath43 by the formula @xmath47 where @xmath48 denotes the volume element of the metric @xmath2 . we define the @xmath42-norm of a cohomology class @xmath49 as @xmath50 , over all 1-forms @xmath51 . it is a remarkable fact ( see @xcite ) that this infimum is actually a minimum , and is achieved by the unique harmonic 1-form in the cohomology class @xmath49 . the norm on @xmath37 dual to the @xmath42-norm on @xmath22 will also be called @xmath42-norm , and will be denoted by the same symbol . the original motivation for this article was to compare the stable norm and the @xmath42-norm . this is done in section [ comparison ] , and our result is : [ compa1 ] let * @xmath52 be a closed , oriented surface equipped with a riemannian metric * @xmath53 be the total volume of @xmath52 . then for all @xmath54 , we have @xmath55 this theorem was originally proved as equation ( 4.8 ) of @xcite , see also @xcite . in section [ section compa ] we give a short and simple proof . the first inequality , which is a straightforward consequence of the cauchy - schwarz inequality , has been extended to higher dimensions in @xcite . it is also used in @xcite . now that we ve been introduced to the number @xmath40 , we want to know more about it . a trivial , but nice observation , is that theorem [ compa1 ] entails @xmath56 the first question that comes to mind is [ question 1 ] is the lower bound of equation ( [ poincare - vol - eq ] ) optimal ? if not , what is the best possible lower bound ? is it realized by some surfaces , and if so , how to characterize such surfaces ? such as it is , question [ question 1 ] is readily answered by @xcite , and the answer is that @xmath57 if and only if @xmath0 is the two - torus and the metric @xmath2 is flat . the `` if '' part may be checked by elementary calculations and we leave it as an exercise . the `` only if '' holds because , by @xcite , if the stable norm and the @xmath42-norm are proportional , then each harmonic 1-form has constant norm . then proposition 6.2 of @xcite implies that @xmath52 is a flat torus . this answer to question [ question 1 ] prompts new questions : [ question 2 ] if we fix a genus @xmath58 for @xmath0 , what is the optimal lower bound ? is it realized by some riemannian metrics ? if so , are those metrics of constant curvature ? another obvious question is [ question 3 ] does @xmath40 have an upper bound involving known geometric quantities such as the length of a homological systole ( the length of a shortest , non - separating closed geodesic ) ? the best we can do about questions [ question 2 ] , [ question 3 ] is summed up in corollary [ bounds in arbitrary curvature ] which we restate here for the commodity of the reader : let * @xmath0 be a closed , oriented surface of genus @xmath59 * @xmath2 be a riemannian metric on @xmath0 * @xmath60 be the diameter of @xmath52 * @xmath61 be the length of a homological systole of @xmath52 . then we have @xmath62 in section [ section constant curvature ] we specialize to metrics of constant negative curvature , and we obtain the [ encadrement courbure -1 ] let * @xmath0 be a closed , oriented surface of genus @xmath58 * @xmath2 be a riemannian metric of constant curvature @xmath63 on @xmath0 * @xmath64 be the length of a homological systole of @xmath52 . then there exist positive numbers @xmath65 and @xmath66 , which depend only on the genus @xmath27 of @xmath0 , such that when @xmath64 is small enough , @xmath67 it would be interesting to know if there is a more precise asymptotic estimate for the behaviour of @xmath40 when @xmath2 tends to infinity in the moduli space @xmath68 of surfaces of genus @xmath27 and curvature @xmath63 . at least we know that @xmath40 does not have a maximum in @xmath68 , but the following question remains : does @xmath40 have a minimum when @xmath52 ranges over the moduli space @xmath68 of surfaces of genus @xmath27 ? if so , which surfaces realize the minimum ? there is still an obvious question that we havent addressed : given a surface @xmath52 , by which homology classes is @xmath40 realized , as the maximum in equation ( [ def k ] ) ? when is it realized by ( the homology classes of ) simple closed geodesics ? in the case of flat tori , it can be checked by elementary calculations that for almost every flat torus ( with respect to lebesgue measure on the moduli space of flat tori ) , @xmath40 is not realized by the homology classes of simple closed geodesics . in the case of surfaces of constant negative curvature , we propose the following conjecture , inspired by theorem 10.7 of @xcite : for any @xmath69 , for almost every @xmath52 in @xmath68 , @xmath40 is realized by the homology classes of simple closed geodesics . first let us recall some basic facts . let @xmath43 and @xmath70 be two closed @xmath71-forms on @xmath0 , and let @xmath49 and @xmath72 be their respective cohomology classes . the wedge product @xmath73 is a @xmath74form on @xmath0 , so there exists @xmath75 in @xmath76 such that @xmath77 = \lambda \left[\omega\right],\ ] ] where @xmath78 is the volume form of @xmath0 . the number @xmath75 only depends on the cohomology classes @xmath49 and @xmath72 , we denote it @xmath79 for the sake of brevity . recall that the poincar duality @xmath80 is the map from @xmath81 to @xmath37 induced by the wedge product of @xmath71-forms : for any @xmath82 , we have @xmath83 where @xmath84 denotes the usual duality bracket beween @xmath81 and @xmath37 . by @xcite , , the poincar duality maps the wedge product in @xmath81 to the intersection pairing in @xmath37 , that is , for any @xmath82 , we have @xmath85 we shall need the following three lemmas : [ poincare l2 ] the poincar duality map @xmath86 is an isometry with respect to the @xmath42-norm . let @xmath49 be any cohomology class in @xmath22 . recall that the @xmath42-norm of @xmath49 is @xmath87 , where @xmath88 is hodge s star operator . then by the definition of the dual @xmath42-norm on @xmath14 , we have @xmath89 and by the schwarz inequality , the supremum above is achieved for @xmath90 , so we have @xmath91 [ poincare - eq ] for any @xmath92 and any @xmath23 we have @xmath93 since the wedge product and the intersection pairing are poincar dual , we have @xmath94 , and by equation ( [ poincare bracket ] ) we have @xmath95 [ poincare norme k ] the norm of the inverse poincar duality map from @xmath37 to @xmath81 , with respect to the stable norm , is @xmath41 . the norm of the inverse poincar duality map from @xmath37 to @xmath81 is @xmath96 by lemma [ poincare - eq ] @xmath97 let @xmath43 be a differential 1-form on @xmath0 . we have @xmath98 thus , @xmath99 taking the infimum of either member of equation ( [ l2comass ] ) over all closed 1-forms in the cohomology class @xmath100 $ ] , we get @xmath101 and since @xmath102 we get @xmath103 which is the first inequality of theorem [ compa1 ] . now let us prove the second inequality . equation ( [ thm 1 , eq 1 ] ) and lemma [ poincare norme k ] tell us that for any homology class @xmath24 , we have @xmath104 now lemma [ poincare l2 ] says that @xmath105 , which completes the proof . we get a better understanding of @xmath41 by noticing that in its definition we may restrict to simple closed geodesics : [ k simple closed geodesic ] we have @xmath106 first let us point out that for any simple closed geodesics @xmath4 and @xmath107 , we have @xmath108 \|_s $ ] and @xmath109 \|_s $ ] , so @xmath110 , \left[\beta\right])|}{\| \left[\alpha\right ] \|_s \| \left[\beta\right ] \|_s } \leq k,\ ] ] whence @xmath111 to establish the reverse inequality , first observe that in the definition of @xmath41 we may restrict to integer homology classes : @xmath112 because rational homology classes are dense in @xmath14 . now take * two integer homology classes @xmath113 and @xmath114 * simple closed geodesics @xmath115 and @xmath116 * integers @xmath117 and @xmath118 such that * @xmath119 + \ldots a_k \left[\alpha_k\right]$ ] * @xmath120 * @xmath121 + \ldots b_p \left[\beta_p\right]$ ] * @xmath122 . without loss of generality we may assume that @xmath123 then @xmath124 whence @xmath125 which proves that @xmath126 and the lemma . + let @xmath127 be a positive real number and let @xmath4 and @xmath5 be two simple closed geodesics , such that @xmath128 set @xmath129 . replacing , if necessary , @xmath4 and @xmath5 by shorter curves whose algebraic intersection is @xmath130 , we may assume that @xmath4 and @xmath5 minimize the product @xmath131 among all pairs of curves whose algebraic intersection is @xmath130 . then the following lemma tells us that all intersections of @xmath4 and @xmath5 have the same sign , that is , the algebraic intersection of @xmath4 and @xmath5 coincides with the number of their intersection points . therefore in the definition of @xmath41 we may restrict to simple closed geodesics , where @xmath132 . let @xmath130 be a positive integer , and let @xmath4 , @xmath5 be simple closed geodesics such that @xmath133 and @xmath134 then @xmath135 [ thm : samesign ] by contradiction : assume that their exists a geodesic arc @xmath136 of @xmath4 with endpoints @xmath137 and @xmath138 on @xmath5 , such that the sign of the intersection at @xmath137 is different from the sign at @xmath138 . let @xmath139 be the geodesic arc on @xmath5 connecting @xmath137 and @xmath138 traversing @xmath5 in the positive sense and let @xmath140 the remaining part of @xmath5 ( see fig . [ fig : onesign ] ) . (.38*.30 ) @xmath5 + (.38*.51 ) @xmath4 + (.48*.81 ) @xmath139 + (.50*.51 ) @xmath136 + (.43*.44 ) @xmath137 + (.55*.44 ) @xmath138 + now assume without loss of generality that @xmath141 . we construct a new curve @xmath142 by connecting @xmath140 with @xmath136 . by homotoping @xmath142 away from @xmath4 with a small deformation , the intersection points @xmath137 and @xmath138 disappear . now the closed geodesic @xmath143 in the free homotopy class of @xmath142 is strictly smaller than @xmath5 but @xmath144 . this contradicts the maximality of @xmath145 . let @xmath52 be a closed , oriented riemannian surface . let @xmath64 be the length of a homological systole @xmath146 and let @xmath147 be the diameter of @xmath52 . then @xmath148 let @xmath149 be a shortest closed geodesic such that @xmath150 , and let @xmath151 be its length . then we have @xmath152 . we shall prove , by contradiction , that @xmath153 , which entails the proposition . assume @xmath154 . then there exist two points @xmath137 and @xmath138 on @xmath149 whose distance is not realized by a geodesic arc on @xmath149 . let @xmath155 be a geodesic arc , which is not an arc of @xmath149 , and realizes the distance between @xmath137 and @xmath138 . the points @xmath137 and @xmath138 divide @xmath149 into two arcs , @xmath156 and @xmath157 . denote by @xmath158 and @xmath159 the curves we obtain by connecting the arcs with the same name . both of these curves are strictly shorter than @xmath149 . now we distinguish two cases . _ case 1 : _ @xmath155 and @xmath146 do not have two consecutive ( along @xmath155 ) intersections with the same sign ( this includes the case when @xmath155 and @xmath146 have one , or zero , intersection point ) . then the algebraic intersection between @xmath155 and @xmath146 has absolute value zero or one . therefore one of the curves @xmath158 and @xmath159 has algebraic intersection @xmath160 with @xmath146 . since it is shorter than @xmath149 , this contradicts the minimality of @xmath149 . (.68*.40 ) @xmath155 + (.33*.62 ) @xmath146 + (.34*.37 ) @xmath161 + (.61*.43 ) @xmath162 + (.45*.63 ) @xmath163 + _ case 2 : _ @xmath155 and @xmath146 have two intersection points @xmath161 and @xmath162 , consecutive along @xmath155 , with the same sign . let @xmath163 be the closed curve obtained by joining the arc of @xmath155 between @xmath161 and @xmath162 , with an arc of @xmath146 , of length @xmath164 ( see fig . [ fig : l1d ] ) . then @xmath165 on the other hand , @xmath163 is homotopic to a closed curve which intersects @xmath146 exactly once . therefore it is homotopic to a closed geodesic @xmath49 , such that @xmath166 ( see @xcite , proof of theorem 17.3.1 ) , which is shorter than @xmath149 , a contradiction . + in the following we will obtain an upper bound on @xmath41 . to this end we prove the following proposition . [ arbitrcurvature ] let * @xmath52 be a closed , oriented riemannian surface * @xmath64 be the length of a homological systole of @xmath52 * @xmath167 be simple closed geodesics in @xmath52 . then @xmath168 take a real number @xmath169 . we cut @xmath4 and @xmath5 into segments of length @xmath170 and at most one segment of smaller length . let @xmath171 and @xmath172 be the respective numbers of segments obtained . let @xmath173 and @xmath174 be a pair of these segments in @xmath4 and @xmath5 respectively . we shall prove , by contradiction , that the algebraic intersection of @xmath173 and @xmath174 is at most one . assume to the contrary . then there exist two intersection points @xmath175 and @xmath176 of @xmath173 and @xmath174 , consecutive along @xmath173 , such that the intersections of @xmath173 and @xmath174 at @xmath175 and @xmath176 have the same sign . let @xmath177 be the closed curve formed by subsegments of @xmath173 and @xmath174 glued at @xmath175 and @xmath176 . then @xmath177 is homotopic to a curve @xmath178 which intersects @xmath4 exactly once ( see fig . [ fig : intsign ] ) . (.36*.51 ) @xmath4 + (.28*.60 ) @xmath5 + (.51*.67 ) @xmath173 + (.44*.90 ) @xmath174 + (.44*.66 ) @xmath175 + (.55*.66 ) @xmath176 + (.46*.82 ) @xmath178 + (.54*.84 ) @xmath177 + in particular @xmath178 is non - separating . on the other hand , the length of @xmath178 is @xmath179 , which contradicts the definition of @xmath64 . we have proven the inequality @xmath180 by construction @xmath181 @xmath182 so @xmath183 substituting this into equation ( [ eq : upper1 ] ) , and since @xmath170 is arbitrarily close to @xmath184 , we obtain the claim . summarizing equations ( [ poincare - vol - eq ] , [ k_low_l1 ] and [ arbitrcurvature - eq ] ) , we obtain the following bounds . [ bounds in arbitrary curvature ] let * @xmath52 be a closed , oriented riemannian surface * @xmath60 be its diameter * @xmath64 be the length of a homological systole of @xmath52 * @xmath185 be the total volume of @xmath52 * @xmath186 . then @xmath187 [ all_bounds ] from now on we assume that the genus of @xmath0 is @xmath188 and the metric @xmath2 has curvature @xmath63 everywhere . recall that we denote by @xmath68 the moduli space of surfaces of genus @xmath27 , that is , the set of metrics of curvature @xmath63 on @xmath0 , modulo isometries . + + let @xmath189 be a simple closed geodesic on @xmath0 . let @xmath190 be the supremum of all @xmath191 , such that the geodesic arcs of length @xmath191 emanating perpendicularly from @xmath189 are pairwise disjoint . a _ collar _ around @xmath189 or _ cylinder _ of width @xmath192 , @xmath193 , is defined by @xmath194 by @xcite , theorem 4.3.2 we have : let @xmath0 be a closed , oriented surface of genus @xmath195 , endowed with a metric @xmath2 of curvature @xmath63 . let @xmath189 be a simple closed geodesic in @xmath52 . then @xmath196 if @xmath197 is another simple closed geodesic that does not intersect @xmath189 , then @xmath198 and @xmath199 are disjoint . [ thm : col_lem ] the main result of this section is : let * @xmath0 be a closed , oriented surface @xmath0 of genus @xmath195 * @xmath2 be a metric of curvature @xmath63 on @xmath0 * @xmath146 be a homological systole of @xmath52 * @xmath200 be its length . we have @xmath201 [ thm : inequ_k ] since @xmath202 and @xmath203 are equivalent to @xmath204 , when @xmath64 goes to zero , theorem [ thm : inequ_k ] implies theorem [ encadrement courbure -1 ] . this means that @xmath40 tends to infinity if and only if @xmath64 , the length of a homological systole of @xmath52 goes to zero . in particular @xmath40 is unbounded . by @xcite there exist simple closed geodesics @xmath205 and @xmath206 such that * @xmath207 , \ldots , \left[\beta_{2s}\right]$ ] is a basis of @xmath208 as a vector space * @xmath209 , \left[\beta_{2i}\right ] ) = 1 $ ] for all @xmath210 * @xmath211 in particular @xmath212 as @xmath213 , \left[\beta_{2}\right ] ) = 1 $ ] , we get : @xmath214 let @xmath215 be the set of non - separating simple closed geodesics , such that @xmath216 it follows from @xcite , theorem 4.1.1 that @xmath217 . for @xmath218 we set @xmath219 , @xmath220 let furthermore @xmath221 and @xmath222 be the connected boundary components of @xmath223 and let @xmath224 be a geodesic arc realizing the distance between these two boundaries . we will gather some useful facts about the geometry of these collars in the following : (.77*.50 ) @xmath225 + (.58*.65 ) @xmath223 + (.28*.78 ) @xmath226 + (.30*.19 ) @xmath221 + (.64*.19 ) @xmath222 + (.49*.33 ) @xmath227 + (.42*.46 ) @xmath224 + * by the collar theorem the @xmath228 are pairwise disjoint . * for @xmath229 we have : @xmath230 is a monotonically decreasing function and @xmath231 . * @xmath232 * as @xmath233 it follows from this formula and the definition of @xmath234 that @xmath235 (.82*.02 ) @xmath236 + let @xmath127 be a positive real number and let @xmath177 and @xmath197 be two simple closed geodesics , such that @xmath237,[\delta])|}{l_g(\gamma ) \cdot l_g(\delta ) } + \epsilon .\ ] ] set @xmath238,[\delta])|$ ] . replacing , if necessary , @xmath177 and @xmath197 by shorter curves whose algebraic intersection is @xmath130 , we may assume that @xmath177 and @xmath197 minimize the product @xmath239 among all pairs of curves whose algebraic intersection is @xmath130 . then lemma [ thm : samesign ] tells us that all intersections of @xmath177 and @xmath197 have the same sign , that is , the algebraic intersection of @xmath177 and @xmath197 coincides with the number of their intersection points . we will obtain our result by distinguishing two cases : either @xmath177 or @xmath197 is one of the @xmath215 , or neither @xmath177 nor @xmath197 is one of the @xmath215 . set @xmath240 for @xmath241 , let @xmath242 be the number of intersection points of @xmath177 and @xmath197 in @xmath243 . we have : @xmath244 where @xmath245 and @xmath246 are defined by @xmath247 here we use the fact that the algebraic intersection of @xmath177 and @xmath197 coincides with the number of their intersection points . we will establish two independent upper bounds on @xmath245 and @xmath246 to prove our theorem . + to find an upper bound on @xmath245 , we establish bounds on each @xmath223 . denote by * @xmath248 the number of intersection points of @xmath177 and @xmath197 in @xmath223 * @xmath249 . . we have : @xmath250 now fix an index @xmath30 and let us work in the cylinder @xmath223 . let @xmath251 be the disjoint union of geodesic arcs of @xmath177 and @xmath197 , respectively , which traverse @xmath223 . we have : @xmath252 now we may assume without loss of generality that @xmath253 it follows that @xmath254 we now determine an upper bound on the intersection number and a lower bound on the length of @xmath255 and @xmath156 . to this end we will define the winding number of an arc traversing @xmath223 and prove two lemmas concerning the intersection of two such geodesic arcs . then we will provide a lower bound on the length of an arc . + the next three lemmata will also be used in the proof of the upper bound on @xmath246 , where we shall need to apply them to pairs of geodesic arcs , which are not necessarily formed by an arc of @xmath177 and an arc of @xmath197 traversing a cylinder @xmath223 . let @xmath49 be a geodesic arc traversing a cylinder @xmath223 . let @xmath256 be another geodesic arc traversing @xmath223 or let @xmath257 be the simple closed geodesic @xmath227 . then @xmath49 and @xmath257 intersect under the same sign at any intersection point . [ thm : sign_cyl ] by contradiction : assume that there exists a geodesic arc @xmath258 of @xmath49 with endpoints @xmath137 and @xmath138 on @xmath257 , consecutive along @xmath257 , such that the sign of the intersection at @xmath137 is different from the sign at @xmath138 . let @xmath259 be the geodesic arc on @xmath257 connecting @xmath137 and @xmath138 in @xmath223 . consider a lift of @xmath258 to the hyperbolic plane . by abuse of notation we denote the lift of @xmath258 by the same symbol . we also denote the lifts of @xmath137 and @xmath138 on @xmath258 by the same symbols . let @xmath260 be the lift of @xmath259 intersecting @xmath261 at @xmath137 in the hyperbolic plane . now , due to the topology of the cylinder , @xmath138 lies also on @xmath260 . hence in the hyperbolic plane @xmath137 and @xmath138 are connected by the two different geodesic arcs @xmath258 and @xmath260 . these arcs belong to two different geodesics passing through @xmath137 and @xmath138 . but in the hyperbolic plane there can be only one geodesic through any two distinct points ( see @xcite , theorem 1.1.4 ) , a contradiction . + let @xmath49 be a geodesic arc traversing @xmath223 . with respect to its fixed endpoints on @xmath262 @xmath49 is in the homotopy class @xmath263= [ b ' \cdot a \cdot b''].\ ] ] here @xmath264 and @xmath265 are directed geodesic arcs that meet @xmath227 perpendicularly on opposite sides of @xmath227 , and @xmath266 is a directed arc on @xmath227 . we define the _ orientation _ @xmath267 of @xmath49 with respect to @xmath227 as @xmath268 and the _ winding number _ of @xmath49 as @xmath269 so we have @xmath270 , where @xmath271 denotes the sign function . + note that due to lemma [ thm : sign_cyl ] an arc @xmath49 has only one intersection point with @xmath227 . denote by the floor function the mapping @xmath272 , defined by @xmath273 for all @xmath274 . the number of intersection points of two geodesic arcs traversing a cylinder @xmath223 and the sign of their intersections are related to the winding numbers of these arcs in the following way . let @xmath49 and @xmath257 be two geodesic arcs traversing a cylinder @xmath223 . let @xmath275 and @xmath276 be the winding numbers of the arcs @xmath49 and @xmath257 , respectively . if @xmath49 and @xmath257 intersect @xmath227 under the same sign , then @xmath277 furthermore @xmath278 determines the sign of the intersection of @xmath49 and @xmath257 at any intersection point . + if @xmath49 and @xmath257 intersect @xmath227 under different signs , then @xmath279 furthermore @xmath280 determines the sign of the intersection of @xmath49 and @xmath257 at any intersection point . + [ thm : up_intgd1 ] we first consider the case , where @xmath49 and @xmath257 intersect @xmath227 from the same side . + + _ case 1 : @xmath281 _ + + we first treat the case where @xmath282 . + by lemma [ thm : sign_cyl ] @xmath49 and @xmath257 always intersect under the same sign at any intersection point . hence we have @xmath283 if @xmath282 , then @xmath49 intersects @xmath227 perpendicularly . now @xmath257 winds at least @xmath284 times around @xmath223 and intersects @xmath49 at least @xmath284 times . there might be an additional intersection point , but not more than one , that is , @xmath285 furthermore , the sign of the intersection of @xmath49 and @xmath257 at any intersection point is determined by the orientation @xmath286 of @xmath257 , that is , @xmath287 therefore our lemma is true if @xmath282 . + the intersection number of two curves only depends on the homotopy class with fixed endpoints of the curves . let @xmath189 and @xmath288 be two curves with fixed endpoints and with homotopy classes @xmath289= [ c]$ ] and @xmath290=[d]$ ] . if @xmath282 we obtain the following result : @xmath291 if @xmath292 , we apply a dehn twist to the cylinder , which we define in the following via fermi coordinates . + we recall that @xmath293 . the _ fermi coordinates _ with base point @xmath294 are an injective parametrization @xmath295 such that * @xmath296 and @xmath297 , for all @xmath298 * @xmath299 is an arc - length parametrization of an oriented geodesic arc @xmath300 that intersects @xmath227 perpendicularly in @xmath301 * @xmath302 . let @xmath303 be a real number . a dehn twist of order @xmath304 , @xmath305 is defined in fermi coordinates by @xmath306 . + let @xmath307 be the dehn twist of order @xmath308 . then the winding number @xmath309 of the geodesic arc @xmath72 in the homotopy class ( with fixed extremities ) of @xmath310 is @xmath311 . the winding number @xmath312 of the geodesic arc @xmath313 in the homotopy class of @xmath314 is @xmath315 . since @xmath316 is isotopic to the identity , we have @xmath317 . by ( [ eq : wind01 ] ) we have @xmath318 that is , @xmath319 it follows furthermore from ( [ eq : wind02 ] ) that @xmath320 determines the sign of the intersection of @xmath257 and @xmath49 at any intersection point . this completes the proof in the case , where here @xmath49 and @xmath257 intersect @xmath227 from the same side . + we now consider the case , where @xmath49 and @xmath257 intersect @xmath227 from different sides . + + _ case 2 : @xmath321 _ + + in this case let @xmath322 be the geodesic that coincides pointwise with @xmath49 , but which traverses @xmath223 in the opposite sense . let @xmath323 be the winding number of the arc @xmath322 . we have that @xmath324 as @xmath325 we have that @xmath326 as @xmath327 , we can apply the result from _ case 1_. therefore the statement for _ case 2 _ follows from _ case 1_. this completes the proof of lemma [ thm : up_intgd1 ] . + we now give two lower bounds for the length of a geodesic arc traversing a cylinder @xmath223 . let @xmath49 be a geodesic arc traversing a cylinder @xmath223 and let @xmath275 be its winding number . we have : @xmath328 [ thm : low_length ] we lift @xmath49 to @xmath329 in the hyperbolic plane ( see fig . [ fig : col_lift ] ) . let @xmath330 be the lift of @xmath227 and let @xmath331 be a lift of @xmath224 ( see fig . [ fig : collar ] ) . let @xmath332 be a fundamental domain of @xmath223 , whose boundary is @xmath331 . (.30*.62 ) @xmath333 + (.40*.63 ) @xmath329 + (.33*.51 ) @xmath334 + (.55*.67 ) @xmath332 + (.51*.30 ) @xmath331 + (.46*.46 ) @xmath176 + (.77*.49 ) @xmath330 + denote by @xmath334 the arc which we obtain from the orthogonal projection of @xmath329 onto @xmath330 in the hyperbolic plane . let @xmath176 be the midpoint of @xmath329 . here , due to the symmetry of the situation , the midpoint of @xmath329 lies on @xmath330 and is also the midpoint of @xmath334 . let @xmath333 be a triangle with vertices @xmath176 , an endpoint of @xmath329 and an endpoint of @xmath334 ( see fig . [ fig : col_lift ] ) . + it follows from the geometry of the hyperbolic right - angled triangle @xmath333 ( see @xcite , p. 454 ) that @xmath335 as @xmath336 and @xmath337 we obtain from the above equation that @xmath338 as @xmath339 is a strictly increasing function on @xmath340 and as @xmath341 , it follows from the above equation that @xmath342 from these two inequalities we obtain again by the monotonicity of the @xmath339 function on @xmath340 that @xmath343 here the first inequality is the first inequality of the lemma . it follows furthermore from the definition of the winding number that @xmath344 which proves the second inequality in lemma [ thm : low_length ] . + we will denote in the following the winding number of an arc @xmath345 by @xmath346 and the winding number of an arc @xmath347 by @xmath348 . now let @xmath349 be the natural numbers such that @xmath350 we may assume without loss of generality that @xmath351 . it follows from lemma [ thm : up_intgd1 ] that @xmath352 it follows from the definition of @xmath353 and lemma [ thm : low_length ] that @xmath354 here the inequality @xmath355 in ( [ eq : low_gd1 ] ) follows by combining the two inequalities in ( [ eq : cla ] ) . using the above inequalities we obtain from inequality ( [ eq : ki_gd ] ) for @xmath356 : @xmath357 if @xmath358 , we use the second inequality in ( [ eq : low_gd1 ] ) to derive the same upper bound . + + combining the estimates for the @xmath359 we obtain that @xmath360 to prove the last inequality we have to show that @xmath361 we obtain this result by combining the two inequalities @xmath362 and @xmath363 in ( [ eq : cla ] ) . + as the function @xmath364 is monotonously decreasing in the interval @xmath365 $ ] ( see fig . [ fig : acla ] ) , we obtain from the above inequality for @xmath245 that @xmath366 here the last inequality follows from the fact that we have at most @xmath367 cylinders @xmath223 . (.82*.02 ) @xmath10 + we recall that @xmath368 to obtain an upper bound on @xmath246 we now construct a comparison surface @xmath369 from @xmath0 such that @xmath370 where @xmath371 is the shortest non - separating simple closed geodesic in @xmath369 . then we construct two comparison curves @xmath372 and @xmath373 , such that @xmath374,[\delta'])|.\ ] ] now for every @xmath375 , we can approximate our non - smooth surface @xmath376 with a smooth surface @xmath377 such that the distance function of @xmath377 is @xmath378-close to that of @xmath376 . it follows from this remark and by applying proposition [ arbitrcurvature ] to @xmath377 that @xmath379,[\delta'])|}{l_g(\gamma')\cdot l_g(\delta ' ) } \leq k(m',g ' ) \leq \frac{9}{l_g(\alpha'_1)^2 } \leq 144 . \label{eq : k2}\ ] ] we construct a surface @xmath376 with a singular riemannian metric in the following way . we cut out all collars @xmath223 from @xmath0 and then reconnect the open ends . here we identify the sides in the following way . for all @xmath380 , let @xmath381 be an isotopy @xmath382 \rightarrow b_i , \text { \ \ such that \ \ } j_i(\cdot,0)=id \text { \ \ and \ \ } j_i(b_i,1)=\partial_2 b_i.\ ] ] here @xmath381 should satisfy the following condition . for all @xmath383 $ ] and all @xmath384 @xmath385 @xmath386 is a geodesic arc in @xmath223 with endpoint @xmath137 and that intersects @xmath227 perpendicularly . we define @xmath387 we call @xmath262 the image of @xmath221 in @xmath369 . + now we have to show that the length of a non - separating simple closed curve @xmath189 in @xmath369 is bigger than @xmath388 . we distinguish two cases : either @xmath189 intersects @xmath389 , or not . + consider first the case , where @xmath189 does not intersect a @xmath262 in @xmath369 . now the @xmath215 are the non - separating simple closed geodesics in @xmath0 , such that @xmath233 . in @xmath0 all non - separating simple closed curves of length smaller than @xmath388 are contained in the union @xmath390 of cylinders @xmath223 . as @xmath391 does not intersect this set , we have that @xmath392 any simple closed curve @xmath189 in @xmath369 that intersects a @xmath262 either intersects a boundary of @xmath226 or is contained in @xmath226 . in the first case @xmath189 is longer than the distance between a boundary of @xmath225 and @xmath262 . in the second case @xmath189 is either contractible , or is freely homotopic to @xmath262 , which is the shortest curve in its free homotopy class in @xmath369 . as the @xmath393 are chosen in a way such that for all @xmath380 @xmath394 we have in any case that @xmath395 summarizing these cases we obtain that the length of any non - separating simple closed curve in @xmath369 is bigger than @xmath388 . therefore inequality ( [ eq : lprime ] ) holds . this proves our upper bound on @xmath246 in ( [ eq : k2 ] ) . first we construct two comparison curves @xmath372 and @xmath373 , such that @xmath397,[\delta'])|.\ ] ] to this end we will replace all arcs of @xmath177 and @xmath197 traversing a cylinder @xmath225 with shorter arcs in @xmath226 . proceeding this way with all @xmath398 , we obtain the comparison curves @xmath178 and @xmath396 from @xmath177 and @xmath197 , respectively . + let in the following @xmath225 be a fixed cylinder . before we present the construction , we will first gather some information about the way @xmath177 and @xmath197 intersect in @xmath399 . we recall that @xmath251 are the arcs of @xmath177 and @xmath197 traversing @xmath223 . the following lemma shows that all arcs of @xmath177 and all arcs of @xmath197 intersect @xmath227 under the same sign . more precisely : let * @xmath400 and @xmath345 be two distinct arcs of @xmath177 traversing @xmath223 * @xmath401 and @xmath347 be two distinct arcs of @xmath197 traversing @xmath223 . then @xmath402 [ thm : sameint ] (.58*.65 ) @xmath223 + (.48*.33 ) @xmath227 + (.40*.51 ) @xmath255 + (.58*.42 ) @xmath403 + (.45*.43 ) @xmath189 + (.34*.44 ) @xmath404 + (.67*.44 ) @xmath405 + we will prove the statement by contradiction . recall that we assume that @xmath177 is the shortest simple closed geodesic , such that @xmath406,[\delta])| = n$ ] . consider the arcs @xmath407 of @xmath177 traversing @xmath223 . we assume that there exist two such arcs that intersect @xmath227 under a different sign . let without loss of generality @xmath255 and @xmath403 be these two arcs . as @xmath177 is a simple closed geodesic , it has no self - intersection and these two arcs can not intersect . we can therefore connect the endpoints of @xmath255 and @xmath403 on @xmath221 with an arc @xmath404 on @xmath221 and the endpoints of these two arcs on @xmath222 with an arc @xmath405 on @xmath222 , such that together with the arcs @xmath255 and @xmath403 the arcs form a simple closed curve @xmath189 ( see fig . [ fig : col_g1g2 ] ) . as @xmath189 is contractible to a point , @xmath289 = 0 $ ] . hence replacing @xmath255 and @xmath403 of @xmath177 with directed arcs corresponding to @xmath404 and @xmath405 , but with inverse direction , we can create a new curve @xmath408 , such that @xmath409 = [ \tilde{\gamma}]$ ] . but it follows from the first inequality in lemma [ thm : low_length ] and ( [ eq : cla ] ) that @xmath410 hence @xmath411 . a contradiction to the minimality of @xmath412 . as the same proof applies to @xmath197 , we obtain our lemma . let @xmath346 be the winding number of the arc @xmath345 and @xmath413 be the winding number of the arc @xmath414 . then @xmath415 [ thm : samewind ] we will prove the lemma for two arcs @xmath345 and @xmath416 of @xmath177 . if @xmath417 , then @xmath418 . it follows from lemma [ thm : up_intgd1 ] that @xmath419 and @xmath345 intersects @xmath416 . a contradiction to the fact that @xmath177 is a simple closed geodesic and therefore has no self - intersection . + let @xmath420 and @xmath421 be two arcs of @xmath177 and @xmath197 , respectively , with minimal absolute value of the winding number . let @xmath422 and @xmath423 be the winding numbers of these arcs . we have @xmath424 we set furthermore @xmath425 let @xmath49 be a geodesic arc traversing @xmath223 . with respect to its fixed endpoints on @xmath262 @xmath49 is in the homotopy class @xmath426= [ b ' \cdot a \cdot b'']$ ] . here @xmath264 and @xmath265 are directed geodesic arcs that meet @xmath227 perpendicularly on opposite sides of @xmath227 , and @xmath266 is a directed arc on @xmath227 . we recall that @xmath267 , the orientation of @xmath49 , is defined by @xmath427 if @xmath275 is the winding number of the arc @xmath49 , then @xmath428 . we have : let @xmath346 be the winding number of the arc @xmath345 and @xmath413 be the winding number of the arc @xmath414 . + if @xmath429 then @xmath430 for all @xmath431 . + if @xmath432 then @xmath433 for all @xmath434 . + furthermore @xmath435 [ thm : less2 ] we will prove the lemma for an arc @xmath345 of @xmath177 . the statement about the orientation @xmath436 of @xmath345 follows from lemma [ thm : samewind ] . it follows from the triangle inequality that @xmath437 here the last inequality follows from lemma [ thm : samewind ] . now if @xmath438 then @xmath439 and the inequality is true . if @xmath429 then @xmath430 and the inequality follows from the definition of @xmath440 and @xmath441 . + we now define the comparison curves @xmath178 and @xmath396 of @xmath369 . let @xmath442 be a geodesic arc of @xmath177 or @xmath197 that traverses @xmath223 with endpoints @xmath384 and @xmath443 and let @xmath444 be its winding number . we first replace @xmath442 with the geodesic arc @xmath445 with the same endpoints @xmath137 and @xmath138 on @xmath262 , such that its winding number @xmath446 has the following value : + if @xmath447 then @xmath448 and we set @xmath449 if @xmath450 then @xmath451 and we set @xmath452 denote by @xmath453 and @xmath454 the curves which we obtain this way from @xmath177 and @xmath197 . we call @xmath455 the arcs of @xmath453 and @xmath454 traversing @xmath223 and denote by @xmath456 and @xmath457 the winding number of the arc @xmath458 and @xmath459 , respectively . + now let @xmath460 be the arc on @xmath461 , such that @xmath462 replacing all arcs @xmath442 of @xmath177 and @xmath197 in all @xmath393 with corresponding arcs @xmath460 in @xmath369 we obtain @xmath178 and @xmath396 . [ claim ] @xmath463 , @xmath464 and @xmath465,[\delta'])|.$ ] [ thm : claim ] before launching into the proof of claim [ claim ] , let us explain the idea a little bit . on the one hand , we need the comparison curves @xmath178 and @xmath396 to be shorter than the original curves @xmath177 and @xmath197 , respectively . so we ensure that the winding numbers of the arcs of @xmath453 and the arcs of @xmath454 in @xmath223 are no greater than the winding numbers of the corresponding arcs of @xmath177 and @xmath197 . this is the reason for the @xmath466 and @xmath467 in the definition . on the other hand , to prove the statement about the intersection number : @xmath465,[\delta'])|$ ] , what we need to do is to make sure that the intersections between the comparison curves @xmath453 and @xmath454 have the same sign as those of the original curves . recall that by lemma [ thm : samesign ] @xmath177 and @xmath197 intersect always under the same sign and have @xmath468 intersection points outside the union of the cylinders @xmath393 . as @xmath177 and @xmath453 and @xmath197 and @xmath454 coincide in @xmath469 , @xmath453 and @xmath454 have at least @xmath470 intersection points and their sign of intersection at any intersection point in @xmath469 is the same . we will show that due to the @xmath471 in our definition there are no two consecutive intersections of @xmath453 and @xmath454 with different sign . it follows that @xmath472 , whence @xmath473 . as furthermore @xmath178 and @xmath396 are the image of @xmath453 and @xmath454 , respectively , under a continuous deformation of the surface @xmath0 , it follows that @xmath474 in total we obtain : @xmath475 now let us prove claim [ claim ] . to simplify our proof we may assume without loss of generality that @xmath476 we will first show : + + _ @xmath463 and @xmath464 . _ + + to this end we first show that @xmath477 and @xmath478 . + consider an arc @xmath458 of @xmath453 . it follows from equation ( [ eq : changewind1 ] ) that @xmath479 from which follows by the triangle inequality that @xmath480 here the last inequality follows from lemma [ thm : less2 ] and the fact that @xmath481 for all @xmath482 . + now let @xmath483 be an arc of @xmath454 . it follows from equation ( [ eq : changewind1 ] ) that @xmath484 we distinguish two cases , @xmath485 and @xmath486 . + a. @xmath487 + we obtain from the triangle inequality and lemma [ thm : less2 ] that @xmath488 applying again the triangle inequality we have that @xmath489 here the last inequality follows from the hypothesis and the fact that the function @xmath481 for all @xmath482 . hence @xmath490 . + b. @xmath491 + it follows from equation ( [ eq : changewind1 ] ) that in this case @xmath492 in this case we apply the triangle inequality twice to @xmath493 and obtain @xmath494 here the last inequality follows from lemma [ thm : less2 ] and the fact that the function @xmath481 for all @xmath482 . hence in any case @xmath477 and @xmath478 . + let @xmath442 be an arc of @xmath177 or @xmath197 traversing @xmath223 . let @xmath445 be the replacement arc of @xmath442 according to equation ( [ eq : changewind1 ] ) . the deformation @xmath381 collapses the cylinder @xmath223 onto @xmath262 . here the arc @xmath445 of @xmath453 or @xmath454 is deformed into an arc @xmath460 of @xmath178 or @xmath396 in @xmath369 . if @xmath446 is the winding number of an arc @xmath445 , then @xmath460 winds @xmath495 times around @xmath262 . hence , as @xmath496 , we have that @xmath497 as @xmath442 traverses @xmath223 we conclude with lemma [ thm : low_length ] that @xmath498 here the second inequality follows from the first part of inequality ( [ eq : cla ] ) . hence our replacement arc @xmath460 of @xmath178 or @xmath396 is always shorter than the arc @xmath442 from @xmath177 or @xmath197 . this proves the first part of claim [ thm : claim ] . now we prove the second part , that is , @xmath499,[\delta'])|.\ ] ] this amounts to showing that there are no two consecutive intersections of @xmath453 and @xmath454 with different sign . for this it suffices to show that the sign of the intersection between @xmath453 and @xmath454 at any point @xmath175 inside a cylinder @xmath223 is always equal to @xmath500 . we will therefore show that all comparison arcs @xmath458 and @xmath459 intersect under the same sign as @xmath345 and @xmath347 in a cylinder @xmath223 . + + therefore we have to treat two cases . either both @xmath177 and @xmath197 intersect @xmath227 from the same side or @xmath177 and @xmath197 intersect @xmath227 from different sides . here we will deduce the result of the second case from the result of the first case . + + _ case i : @xmath501 _ + + by lemma [ thm : up_intgd1 ] it is sufficient to show that the sign of the difference of the winding numbers does not change , that is , @xmath502 recall that we assume that @xmath447 . in this case we have @xmath503 we will prove our statement depending on whether @xmath346 and @xmath348 have different sign or are equal to zero or whether they have the same sign or are equal to zero . + a. _ ( @xmath504 and @xmath505 ) or ( @xmath506 and @xmath507 ) _ + we assume without loss of generality that @xmath508 and @xmath505 . we have that @xmath509 as @xmath508 and @xmath505 it follows that @xmath510 if and only if @xmath511 . this case implies that @xmath512 it follows from equation ( [ eq : changewind1 ] ) that @xmath513 and @xmath514 . hence @xmath515 and ( [ cond_sign ] ) holds . + conversely if @xmath511 does not hold then either @xmath348 or @xmath346 is different from zero . we may assume that @xmath516 and we have that @xmath517 it follows from equation ( [ eq : changewind1 ] ) and from inequality ( [ eq : max_gd ] ) that @xmath518 here the last inequality follows from the fact that @xmath508 and @xmath519 . furthermore @xmath520 we now show that @xmath521 . as @xmath522 it follows that @xmath523 and condition ( [ cond_sign ] ) follows with ( [ eq : sign_a ] ) . + to show that @xmath524 we distinguish the two subcases @xmath485 and @xmath486 . + a. @xmath525 + then as @xmath526 @xmath527 here the last inequality follows by distinguishing the cases @xmath528 and @xmath529 . in the latter case we use the fact that @xmath530 . + b. @xmath531 + then it follows with inequality ( [ eq : max_gd ] ) : @xmath532 b. _ ( @xmath504 and @xmath507 ) or ( @xmath506 and @xmath505 ) _ + we assume without loss of generality that @xmath533 and @xmath505 . we will further subdivide this case into the subcases @xmath534 and @xmath535 . + a. @xmath534 + + in this case it follows from equation ( [ eq : changewind1 ] ) that @xmath536 hence @xmath537 and the condition ( [ cond_sign ] ) is fulfilled . + b. @xmath538 + we have due to the definition of @xmath539 that @xmath540 we note that due to lemma [ thm : samewind ] @xmath541 . furthermore due to the definition of @xmath542 and @xmath441 we have that @xmath543 . hence @xmath544 here the last inequality follows from our hypothesis . it follows from equation ( [ eq : changewind1 ] ) that @xmath545 hence @xmath546 here by the definition of @xmath539 , @xmath547 and by lemma [ thm : less2 ] @xmath548 . hence @xmath549 . therefore @xmath550 and the condition ( [ cond_sign ] ) is fulfilled . this proves that in any case condition ( [ cond_sign ] ) is fulfilled . this settles the claim in the case , where @xmath177 and @xmath197 intersect @xmath227 from the same side . + + _ case ii : @xmath551 _ + + if @xmath177 and @xmath197 intersect @xmath227 from different sides , it is sufficient to show that @xmath552 this follows from lemma [ thm : up_intgd1 ] . + in this case let @xmath553 be the oriented geodesic that coincides pointwise with @xmath177 , but which has opposite orientation . let @xmath554 be the arcs of @xmath553 traversing @xmath223 . let @xmath555 be the winding number of the arc @xmath556 . we have that @xmath557 . + applying equation ( [ eq : changewind1 ] ) to the arcs of @xmath197 and @xmath553 , we obtain that @xmath558 now @xmath553 and @xmath197 intersect @xmath227 under the same sign . hence condition ( [ cond_sign ] ) from _ case i _ is fulfilled , that is , @xmath559 combining the two previous equations we obtain : @xmath560 and therefore condition ( [ cond_sign2 ] ) is fulfilled . this settles the claim in the case where @xmath177 and @xmath197 intersect @xmath227 from different sides . hence the remaining second part of claim [ thm : claim ] is true . + this finishes the construction of the curves @xmath178 and @xmath396 . from this we get the upper bound on @xmath246 in equation ( [ eq : k2 ] ) . summarizing _ case 1 _ , we obtain from the inequalities ( [ eq : k1 ] ) and ( [ eq : k2 ] ) in ( [ eq : k12 ] ) that @xmath561 as @xmath378 is arbitrarily small , we obtain the upper bound stated in theorem [ thm : inequ_k ] from this inequality . we may suppose that @xmath562 . in this case we have to verify the upper bound in theorem [ thm : inequ_k ] for @xmath563,[\delta])|}{l_g(\alpha_i ) \cdot l_g(\delta)}= \frac{n}{l_g(\alpha_i ) \cdot l_g(\delta)}.\ ] ] now @xmath197 intersects @xmath227 @xmath130 times . to this end it has to traverse @xmath130 times the collar @xmath225 of @xmath227 . analogous to lemma [ thm : low_length ] we obtain from the length of the @xmath130 arcs @xmath564 of @xmath197 traversing @xmath225 : @xmath565 now from the monotonicity of the function @xmath566 in the interval @xmath365 $ ] ( see fig . [ fig : acla ] ) it follows that @xmath567 again we obtain our upper bound in theorem [ thm : inequ_k ] as @xmath378 is arbitrarily small . while working on this article the second author has been supported by the alexander von humboldt foundation . + daniel massart and bjoern muetzel + department of mathematics , universit montpellier 2 + place eugne bataillon , 34095 montpellier cedex 5 , france + e - mail : [email protected]_ and [email protected]_

the collaboration is performing a search for neutrinoless double beta decay ofge with the eponymous detector . the experiment has been installed and commissioned at the laboratori nazionali del gran sasso and has started operation in november 2011 . the design , construction and first operational results are described , along with detailed information from the r&d phase . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1212.4067

the gerda experiment ( germanium detector array @xcite ) is a search for the neutrinoless double beta ( @xmath1 ) decay of @xmath0ge . the observation of such a decay would prove that lepton number is not conserved , and that the neutrino has a majorana component @xcite . a discovery of @xmath1 decay would have significant implications on particle physics and other fields , including cosmology @xcite . the importance of the topic has stimulated the development of several experimental approaches to the search for @xmath1 decay on a number of isotopes which undergo double beta decay . for recent reviews on the state of knowledge concerning double beta decay and on running or planned experiments , see refs . @xcite . the experimental signature for @xmath1 decay is a line in the summed electron energy spectrum appearing at the @xmath2-value for the reaction , @xmath3 . the experimental result is a measurement of , or a limit on , the half life , @xmath4 , for the process . within the three neutrino model and assuming the existence of a significant majorana component a positive observation of @xmath1 decay would possibly give access to the neutrino mass hierarchy as well as information on the absolute values of the neutrino masses . the latter is only possible with knowledge of the nuclear matrix elements , @xmath5 , as discussed in ref . the statements on the mass also require an understanding of whether the @xmath1 process is solely due to the majorana nature of the neutrino , or whether additional new physics processes beyond the standard model contribute . a recent review of the particle physics implications of a discovery of @xmath1 decay was given in ref . @xcite . nuclides that are potentially @xmath1 emitters will decay via the standard model allowed @xmath6 decay . both reactions are second order weak interactions , and therefore have extremely long half lives . values have been directly measured for @xmath6 decay in about ten cases and these are in the range 10@xmath710@xmath8 yr @xcite . the half lives for @xmath1 decay , assuming the process exists , are expected to be substantially longer . consequently , @xmath1 decay experiments must be sensitive to just a few events per year for a source with a mass of tens to hundreds of kilograms . backgrounds must typically be reduced to the level of one event per year in the region of interest ( roi ) , an energy interval of the order of the energy resolution around @xmath3 . experiments looking for @xmath1 decay of @xmath0ge operate germanium diodes normally made from enriched material , i.e. the number of @xmath0ge nuclei , the isotopic fraction @xmath9 , is enlarged from 7.8 % to 86 % or higher . in these type of experiments , the source is equal to the detector which yields high detection efficiency . additional advantages of this technique are the superior energy resolution of 0.2 % at @xmath3=2039 kev compared to other searches with different isotopes and the high radiopurity of the crystal growing procedure . disadvantages are the relatively low @xmath3 value since backgrounds typically fall with energy and the relative difficulty to scale to larger mass compared to e.g. experiments using liquids and gases . there is a considerable history to the use of @xmath0ge for the search for @xmath1 decay . after initial experiments @xcite , the heidelberg - moscow ( hdm ) collaboration @xcite and igex @xcite were the driving forces in this field setting the most stringent limits . in 2004 a subgroup of the hdm collaboration @xcite claimed a 4@xmath10 significance for the observation of @xmath1 decay with a best value of @xmath4=1.19@xmath11 yr ; the quoted 3@xmath10 range is @xmath12 yr . to scrutinize this result , and to push the sensitivity to much higher levels , two new @xmath0ge experiments have been initiated : majorana @xcite and gerda @xcite . the latter has been built in the infn laboratori nazionali del gran sasso ( lngs ) at a depth of 3500mw.e . ( water equivalent ) . whereas majorana further refines the background reduction techniques in the traditional approach of operating germanium detectors in vacuum , gerda submerses bare high - purity germanium detectors enriched in @xmath0ge into liquid argon ( lar ) following a suggestion by ref . @xcite ; lar serves simultaneously as a shield against external radioactivity and as cooling medium . phase i of the experiment is currently taking data and will continue until a sensitivity is reached which corresponds to an exposure of 15 kg@xmath13yr with a background index ( bi ) of 10@xmath14 cts/(kev@xmath13kg@xmath13yr ) @xcite ) . this will be sufficient to make a strong statement on the existence of @xmath1 decay in @xmath0ge for the best value given in ref . @xcite . phase ii of gerda is planned to acquire an exposure of 100 kg@xmath13yr at a bi of 10@xmath15 cts/(kev@xmath13kg@xmath13yr ) . for pure majorana exchange and the case that no signal is seen , this will constrain the effective neutrino mass @xmath16 to less than about @xmath17 mev with the precise value depending on the choice of matrix elements @xcite . the gerda experiment is described in detail in the following sections . an overview of experimental constraints and the design is presented first . this is followed by a description of the ge detectors . then , the experimental setup , electronic readout , data acquisition ( daq ) and data processing are described . as gerda phase i has been fully commissioned and has started data production , the main characteristics of its performance are given in the final section . the experimental challenge is to have nearly background free conditions in the roi around @xmath3 . typically , background levels are quoted in units of counts per kev per kilogram per year , cts/(kev@xmath13kg@xmath13yr ) , since the number of background events roughly scales with the detector mass , energy resolution and running time . defining @xmath18 as the width of the roi where a signal is searched for , the expected background is the bi multiplied by @xmath18 in kev and the exposure in kg@xmath13yr . gerda has set the goal to keep the expected background below 1 event . for @xmath19 kev and exposures mentioned above , this implies a bi of 0.01 and 0.001 cts/(kev@xmath13kg@xmath13yr ) , respectively , for the two phases of gerda . the main feature of the gerda design is to operate bare ge detectors made out of material enriched in @xmath0ge ( @xmath20ge ) in lar . this design concept evolved from a proposal to operate ge detectors in liquid nitrogen ( ln@xmath21 ) @xcite . it allows for a significant reduction in the cladding material around the diodes and the accompanying radiation sources as compared to traditional ge experiments . furthermore , the background produced by interactions of cosmic rays is lower than for the traditional concepts of hdm , igex or majorana due to the lower z of the shielding material . other background sources include neutrons and gammas from the decays in the rock of the underground laboratory , radioactivity in support materials , radioactive elements in the cryogenic liquid ( intrinsic , such as @xmath22ar and @xmath23ar , as well as externally introduced , such as radon ) as well as internal backgrounds in the ge diodes . these backgrounds were considered in the design and construction phase of gerda and resulted in specific design choices , selection of materials used and also in how detectors were handled . natural ge ( @xmath24ge ) contains about 7.8% @xmath0ge , and could in principle be used directly for a @xmath1 decay experiment . indeed , the first searches for @xmath1 decay used natural ge detectors @xcite . enriched detectors allow for a better signal - to - background ratio and also yield reduced costs for a fixed mass of @xmath0ge in the experiment . the improvement in signal - to - background ratio originates from two sources : ( @xmath25 ) many background sources , such as backgrounds from external gamma rays , are expected to scale with the total mass of the detector ; and ( @xmath26 ) the cosmogenic production of @xmath27ge and @xmath28co in the ge diodes occurs at a higher rate for @xmath24ge than for @xmath20ge . the lower overall cost is due to the fact that the high cost of enrichment is more than offset by the cost of producing the extra crystals and diodes required for @xmath24ge detectors . [ fig : overview ] shows a model of the realized design : the core of the experiment is an array of germanium diodes suspended in strings into a cryostat filled with lar . the lar serves both as cooling medium and shield . the cryostat is a steel vessel with a copper lining used primarily to reduce the gamma radiation from the steel vessel . the cryostat is placed in a large water tank , that fulfills the functions of shielding the inner volumes from radiation sources within the hall , such as neutrons , as well as providing a sensitive medium for a muon veto system . a similar experimental setup has been proposed previously in ref . the detectors are lowered into the lar volume using a lock system located in a clean room on top of the water tank . a further muon veto system is placed on top of the clean room in order to shield the neck region of the cryostat . these installations are supported by a steel superstructure . all components are described in the subsequent sections . artists view ( ge array not to scale ) of the gerda experiment as described in detail in the following sections : the germanium detector array ( 1 ) , the lar cryostat ( 2 ) with its internal copper shield ( 3 ) and the surrounding water tank ( 4 ) housing the cherenkov muon veto , the gerda building with the superstructure supporting the clean room ( 5 ) and the lock ( 6 , design modified ) . various laboratories behind the staircase include the water plant and a radon monitor , control rooms , cryogenic infrastructure and the electronics for the muon veto . ] the gerda detector laboratory ( gdl ) , located underground at lngs , has been used for r&d for gerda as well as for auxiliary tests . it is a grey room equipped with a clean bench , a glove box and wet chemistry for etching . ge diodes submerged in ln@xmath21 or lar can be characterized in a clean environment without any exposure to air . the detector handling described in sec . [ sec : det ] and now adopted for gerda was developed in gdl . the liquid argon germanium ( large ) apparatus was installed inside gdl to investigate properties of lar , such as the scintillation light output . it is used for studies of background suppression in germanium detectors by observing the coincident scintillation light of the liquid argon @xcite and to exploit the lar scintillation light pulse shape properties to recognize the interacting particle @xcite . large is a 1 m@xmath29 low - background copper cryostat with a shield consisting of ( from outside to inside ) 20 cm polyethylene , 23 cm steel , 10 cm lead and 15 cm copper . the inner walls are covered with a reflector foil with a wavelength shifter coating . the shifted light is detected by nine 8 etl 9357 photomultiplier tubes ( pmts ) from electron tubes limited ( etl ) @xcite . calibration sources ( @xmath30th , @xmath31ra , @xmath28co , @xmath32cs ) have been placed in- and outside of the cryostat and the event rejection by pulse shape discrimination and scintillation light detection were studied @xcite . as a consequence of these measurements gerda decided to implement a lar scintillation light veto for phase ii . large has also been used to understand the background coming from the decay of @xmath23ar . in addition to gdl , screening facilities at lngs , in particular gempi @xcite and gator @xcite , have been used extensively . additional screening facilities have been used at different locations , including heidelberg , geel , and baksan . finally , many of the institutes in the gerda collaboration have laboratories which have been extensively used in r&d and testing related to the experiment . a full monte carlo simulation of the gerda experiment and of many of the related r&d facilities has been setup in the form of a general and flexible framework based on geant4 @xcite , which is called mage @xcite . mage has been widely used for gerda - related simulations and background studies . conversely , most of the experimental test stands provided experimental data that were used to validate and benchmark mage . a detailed simulation of the large setup is also available within mage . a few specific gerda - related simulations were run using other codes than mage . in particular , a dedicated simulation code was developed to estimate the residual background in the detector array due to external @xmath33-rays , produced either in the surrounding rocks or in the cryostat volume @xcite . the simulation code shield @xcite was used to optimize the shielding required for the transportation of geo@xmath34 enriched in @xmath0ge from the enrichment plant to the underground storage site @xcite . neutron spectra and fluxes produced by @xmath35s from the @xmath30th calibration sources via the ( @xmath35,n ) reactions were calculated through the sources-4a code @xcite . this section describes the germanium detectors that represent the core of the gerda experiment . for phase i all eight detectors from the former hdm and igex experiments @xcite were refurbished and redeployed . for phase ii new material amounting to 50 kg @xmath20geo@xmath21 and 34 kg of @xmath36geo@xmath21 was purchased . the @xmath36ge , material depleted in @xmath0ge below 0.6 % , was used to check the supply chain and methods of phase ii diode production @xcite . the production and characterization of the new detectors is ongoing . phase i detectors are based on standard p - type hpge detector technology from canberra semiconductor nv , olen @xcite . standard closed - end coaxial detectors have a `` wrap around '' n@xmath37 conductive lithium layer ( @xmath381 mm ) that is separated from the boron implanted p@xmath37 contact by a groove ; the groove region is usually passivated . the detector geometry for one of the enriched detectors is shown schematically in fig . [ fig : detschem ] . schematic drawing of a @xmath20ge diode currently operated in gerda . the ranges of dimensions for the eight detectors are given in units of mm . the masses range from 0.98 to 2.9 kg . ] in normal dc coupled readout , the p@xmath37 surface ( @xmath381 @xmath39 m ) is connected to a charge sensitive amplifier and the n@xmath37 surface is biased with up to + 4600 v. in the alternative readout scheme with ac coupling , the n@xmath37 contact is grounded and the p@xmath37 contact biased with negative high voltage ( hv ) . the analog signal is still read out from the p@xmath37 contact but coupled with a hv capacitor to the amplifier . operation of bare hpge detectors in cryogenic liquids is a non - standard technique . the success of gerda depends strongly on the long - term stability of the ge detectors operated in lar . before deploying the enriched detectors in lar , bare @xmath24ge detectors built with the same technology as the phase i detectors were used for tests in gdl . a long - term study of the leakage current ( lc ) of bare detectors operated in ln@xmath21 and lar under varying @xmath33-irradiation conditions was performed . irradiation of a first prototype detector in lar with @xmath33 s resulted in a continuous increase of the lc ( see fig . [ fig : lc ] , left ) . the ionizing radiation created the expected bulk current in the detector ( @xmath3840 pa ) , observed as a step at the start of the @xmath33-irradiation at @xmath402 d. this was then followed by a continuous increase of the lc . after about one day of irradiation , at @xmath403.5 d , the source was removed and the lc stabilized at a higher value than prior to the irradiation ( @xmath18lc @xmath4130 pa ) . no increase of the lc was observed with the same detector assembly in ln@xmath21 after one week of irradiation . the process is reversible as the lc was partly restored by irradiation with the same source but without applying hv ; the lc was completely restored to its initial value by warming up the detector in methanol baths . these measurements are the first observation of @xmath33-radiation induced leakage current increase for detectors of this design operated in this way . the @xmath33-radiation induced lc was measured for different hv bias values , source - detector configurations and hv polarities @xcite . measurements with three prototype detectors using different sizes of groove passivation ( large area , reduced and none ) were performed . it was found that reducing the size of the passivation layer strongly suppresses the @xmath33-radiation induced lc ( see fig . [ fig : lc ] , right ) . the most likely explanation is that the lc increase is induced by the collection and trapping of charges produced by the ionization of lar on the passivated surface of the detector . no @xmath33-radiation induced lc increase was observed with the prototype without passivation layer . for all stability measurements @xcite , the detectors were biased above their nominal operation voltage . the lc , continuously monitored with high accuracy , was at a few tens of pa for each detector , similar to the values measured at the detector manufacturer . detectors with no passivation layer showed the best performance in lar . consequently , all gerda phase i detectors were reprocessed without the evaporation of a passivation layer . our positive results on the long - term stability of ge detectors in lar and ln@xmath21 contradict the statements made in ref . @xcite . the enriched phase i detectors ang 1 - 5 from the hdm and rg 1 - 3 from the igex collaborations were originally produced by ortec . in addition , six detectors made of @xmath24ge are available from the genius - tf experiment @xcite . they have been stored underground and therefore their intrinsic activity is low . thus , they have been used in the commissioning phase of gerda . details of the characterization of the enriched detectors before they were dismounted from vacuum cryostats in 2006 are reported in ref . @xcite . the phase i detectors , @xmath20ge and @xmath24ge , were modified at canberra , olen @xcite , in the period from 2006 to 2008 . the detector ang 1 had a previous reprocessing at the same manufacturer in 1991 . the work was performed according to the standard manufacturer technology , however the passivation layer on the groove was omitted . leakage current and capacitance of each detector were measured in ln@xmath21 at the manufacturer site after the reprocessing @xcite . c | c c c c c r detector & serial nr . & diam . & length & total & operat . & abundance + name & ortec & ( mm ) & ( mm ) & mass ( g)&bias ( v ) & @xmath9 + ang 1 & @xmath42 ) & 58.5 & 68 & 958 & 3200 & 0.859 ( 13 ) + ang 2 & p40239a & 80 & 107 & 2833 & 3500 & 0.866 ( 25 ) + ang 3 & p40270a & 78 & 93 & 2391 & 3200 & 0.883 ( 26 ) + ang 4 & p40368a & 75 & 100 & 2372 & 3200 & 0.863 ( 13 ) + ang 5 & p40496a & 78.5 & 105 & 2746 & 1800 & 0.856 ( 13 ) + rg 1 @xmath43 ) & 28005-s & 77.5 & 84 & 2110 & 4600 & 0.8551 ( 10 ) + rg 2 @xmath43 ) & 28006-s & 77.5 & 84 & 2166 & 4500 & 0.8551 ( 10 ) + rg 3 @xmath43 ) & 28007-s & 79 & 81 & 2087 & 3300 & 0.8551 ( 10 ) + gtf 32 & p41032a & 89 & 71 & 2321 & 3500 & 0.078 ( 1 ) + gtf 42 & p41042a & 85 & 82.5 & 2467 & 3000 & 0.078 ( 1 ) + gtf 44 & p41044a & 84 & 84 & 2465 & 3500 & 0.078 ( 1 ) + gtf 45 & p41045a & 87 & 75 & 2312 & 4000 & 0.078 ( 1 ) + gtf 110 & p41110a & 84 & 105 & 3046 & 3000 & 0.078 ( 1 ) + gtf 112 & p41112a & 85 & 100 & 2965 & 3000 & 0.078 ( 1 ) + @xmath42 ) produced by canberra , serial nr . b 89002 . + @xmath43 ) as different types of measurements vary , an uncertainty of 2 % is taken in evaluations . the detector dimensions after the reprocessing , the operating bias determined in the lar test bench of gdl and with the abundance of @xmath0ge measured earlier are reported in table [ tab : det ] . a total of @xmath38300 g was removed from the detectors during reprocessing resulting in 17.7 kg enriched diodes for phase i. the active masses of the detectors were assessed at typically @xmath3887 % by comparing @xmath33-ray detection efficiencies to monte carlo simulations of the diodes with dead layer thicknesses varied @xcite . this assessment will be refined with in - situ gerda data . cosmogenically produced isotopes @xmath27ge and @xmath28co can lead to an internal contamination that represents a background in the region of interest . the detectors are always stored at an underground facility to avoid exposure to cosmic rays . this applies also for the reprocessing steps , where the detectors were stored underground at the hades facility @xcite , located at a depth of about 500mw.e . at a distance of 15 km from the detector manufacturer . the total exposure above ground was minimized to @xmath385days @xcite . at the start of phase i in november 2011 , the estimated bi contribution from the cosmogenically produced @xmath28co is on average about @xmath44 cts/(kev@xmath13kg@xmath13yr ) . the bulk of the @xmath28co activity comes from the production before the underground installation of the detectors for the hdm and igex experiments . the contribution from @xmath27ge is negligible since it decayed away . drawing of a phase i detector assembly . the signal contact is realized by a conical copper piece ( `` chinese hat '' ) that is pushed by a silicon spring onto the p@xmath37 contact ( inset left top ) . high voltage is applied to the n@xmath37 contact by a copper strip ( not shown ) pressed by a copper disc which in turn is electrically insulated by a ptfe cylinder ( inset bottom left ) . the force to achieve good electrical contact is actuated through a copper screw . masses and dimensions of the assembly are given for the rg3 detector . ] the mounting scheme of the detectors has competing requirements . it must have a low mass to minimize sources of radiation near to the detectors . however , the construction must be sufficiently sturdy to provide safe suspension . it must support the cables for detector bias and readout . furthermore , the diodes must remain electrically isolated from all other materials . the chosen support design is depicted in fig . [ fig : detsupp ] where the contacting scheme is shown as well . in order to reach the background goals of gerda , the amount of material is minimized . only selected high radiopurity materials were used : copper ( @xmath3880 g ) , ptfe ( @xmath3810 g ) , and silicon ( @xmath381 g ) . the results of the @xmath33 ray spectroscopy measurements ( see sec . [ ssec : screening ] ) , combined with monte carlo simulations give an upper limit on the bi contribution from the detector support of @xmath45 cts/(kev@xmath13kg@xmath13yr ) . one of the prototype detectors was mounted in a support of the phase i design to test the electrical and mechanical performance . this confirmed the mounting procedure , the mechanical stability , the signal and hv contact quality , and the spectroscopic performance of this design . during this test , the energy resolution was the same as was achieved previously when the same detector was mounted in a standard vacuum cryostat , i.e. @xmath382.2 kev full width at half maximum ( fwhm ) at the 1332 kev spectral line of @xmath28co . [ fig : detcan ] shows one of the phase i detectors before and after mounting in its custom made support structure . the phase i detectors were mounted in their final low - mass supports in 2008 and their performance parameters ( leakage current , counting efficiency , energy resolution ) were measured in lar as a function of bias voltage @xcite . the detector handling was performed in gdl entirely within an environment of n@xmath21 gas . the lc of the majority of the detectors was at the same level as measured at the detector manufacturer after reprocessing . the detectors ang 1 , ang 3 and rg 3 showed high lcs even after successive thermal cycling and required additional reprocessing to reach an acceptable performance . spectroscopic measurements were performed , as described in ref . @xcite , with the preamplifier mounted in a gaseous ar environment in the neck of the lar cryostat . the energy resolutions of the phase i detectors was between 2.5 and 5.1 kev ( fwhm ) for the 1332 kev spectral line of @xmath28co . an improvement of the energy resolution of the detectors was observed after polishing the diode surface in the location of the hv contact . since november 2011 all the enriched phase i detectors have been inserted into the gerda cryostat . left : a phase i detector after reprocessing at canberra , olen . the conductive lithium layer ( n@xmath37 contact ) and the boron implanted bore hole ( p@xmath37 contact ) are separated by a groove . right : the detector is mounted upside down in a low - mass holder ( groove no longer visible ) . ] [ cols="^ , > , < , < , < , < , < " , ] finally , radioactivity measurements of the gerda front - end electronics have been performed ( see sec . [ ssec : frontend ] ) . particular efforts have been made to produce a low radioactivity version of the circuit . some of the key points to achieve these results are : manufacturing of the printed circuit board on a specifically selected low - radioactivity substrate ( cuflon ) , minimization of the number of tantalum decoupling capacitors , integration of low value capacitors as stray capacitance between traces directly on board , and careful selection of passive physical components and soldering paste . to reach a bi @xmath46 @xmath47 cts/(kev@xmath13kg@xmath13yr ) , the monte carlo predicts a maximum allowed activity for the front end electronics of 2 mbq for @xmath31ra and 500 @xmath39bq for @xmath30th with a separation of 30 cm between the electronics and the top detectors . the average measured activity of a set of three preamplifiers is ( 286@xmath4828 ) @xmath39bq and ( 150@xmath4824 ) @xmath39bq in @xmath31ra and @xmath30th , respectively , including the pins . thus , the radiopurity limits are met for the 5 pcbs presently in use . the construction of the apparatus was completed in june 2010 . the commissioning phase started with the operation of refurbished @xmath24ge diodes from the genius - tf experiment @xcite , in order to minimize the potential risks for the @xmath20ge detectors . a larger background than expected at @xmath3 and an intense line at 1525 kev was discovered . the origin and mitigation was studied in the following months ( see sec . [ ssec : backgrounds ] ) . in june 2011 a string of @xmath20ge diodes was deployed for further preliminary tests including various operational configurations of the detectors and the electric stray fields . the commissioning phase was completed on november 1 , 2011 . all components had met their design specifications and an adequate background index was reached ; thus , physics data taking of phase i was started on november 9 , 2011 . a blinding window of 40 kev width around @xmath3 is in place since january 11 , 2012 . the raw data are written to disk , however events with energies from 2019 to 2059 kev are not exported to the tier1 data and are therefore not available for analysis . the blinding window will be opened when a sufficient exposure is acquired and the calibrations and selection cuts are finalized . while it is not the scope of this paper to discuss the physics analysis and results , the principal performance of the apparatus is summarized here . results are shown demonstrating that a low background has actually been reached via thorough material selection and screening . the stability of the performance of the complete gerda setup at lngs is inferred from the energy calibrations and the first spectra . the performance particulars are obtained on the basis of physics runs between november 2011 and may 2012 , which resulted in an exposure of 6.10 kg@xmath13yr for the enriched detectors and 3.17 kg@xmath13yr for the natural detectors . these data are collected with an overall live time ( calibration runs subtracted ) of about 90% . the pmts of the muon veto have been checked for pulse height stability for more than one year . a satisfying individual stability is reflected in the constant average light output per muon event per day ( fig . [ fig : mvrate ] , squares and right scale ) . this constancy is mandatory for a reliable determination of the muon rate that is shown by the crosses in fig . [ fig : mvrate ] ( left scale ) . except for short term fluctuation the rate is consistent with a 2 % sinusoidal variation with a period of about one year . this is a well - known phenomenon @xcite that will be verified when a longer period of data is available . the observed muon rate in gerda results in a preliminary value of ( [email protected])@xmath13 cts/(m@xmath49s ) which compares very well with the recent result of ( [email protected])@xmath13 cts/(m@xmath49s ) @xcite . the average light output per event and day ( squares , right scale ) of the cherenkov muon veto . the daily rates ( crosses , left scale ) are rather constant ] fig . [ fig : multiplicity ] shows the multiplicity @xmath50 , the number of cherenkov pmts fired . the spectrum is taken with trigger signals from both muon veto systems with a threshold of 1 photoelectron ( p.e . ) . the expected light yield is roughly 200 to 300 photons for every centimeter traversed by a muon . since almost all surfaces of the water tank and cryostat are covered with the vm2000 foil , one would expect that most of the muon events will cause a high multiplicity of triggered pmts . the low coverage of 0.5% of the surface by pmts is compensated by the reflectivity and wavelength shifting properties of the vm2000 foil . there is , indeed , a rise towards high multiplicities as predicted by the mc simulations , but there is also a prominent enhancement observed in the low multiplicity region below @xmath50=20 which measured multiplicities of all mounted cherenkov pmts without cut on the number of detected photoelectrons ( full line ) and with a cut of @xmath5130 p.e . ( dashed line ) . ] is not present in mc . the low multiplicity bump around @xmath50=10 vanishes for events triggered by the plastic panels only . therefore , it is unlikely that it is caused directly by muons . the hypothesis of local radioactivity creating scintillation light in the vm2000 foil is still investigated . triggers from the water cherenkov can not contribute to @xmath50@xmath465 . the increase close to @xmath50=0 originates from triggers of the triple layered plastic scintillator when the muon hits the plastic but misses the water . increasing the trigger threshold to 30 p.e . ( dashed line ) removes the intensity at low @xmath50 . the lower limit of the muon detection efficiency ( mde ) is estimated for a threshold of 30 p.e . amounting to @xmath52 = ( [email protected])% @xcite . mc estimates @xcite give a value of @xmath52 = ( [email protected])% . the latter , however , includes an energy deposition of the muon within the full detector array . this selects automatically longer tracks within the water , which in turn produce more detectable light . a more refined determination employing coincidences between the plastic and cherenkov detectors is under way . alternatively , one can estimate the efficiency that a muon detected in the ge detector array is accompanied by a muon trigger . two different event types were taken into account : ( i ) all events with @xmath538.5 mev deposited in germanium , but only one germanium detector fired ; and ( ii ) all events with @xmath534 mev deposited in germanium , but at least two germanium detectors fired ( @xmath35 emitters from the u / th decay chain have energy @xmath534 mev , but they would release their energy within one detector ) . in the commissioning runs , a total of 79 events were selected by these cuts , while 78 of them were also found in the muon veto . the muon `` rejection efficiency '' is calculated as @xmath54 % ( median with 68 % central interval ) , which is in a good agreement with the simulations . with the measured efficiency the background index due to un - identified prompt muons is estimated as @xmath55 cts/(kev@xmath13kg@xmath13yr ) , which is well below the specifications needed for phase i and ii @xcite . initially , eight detectors from igex and hdm have been in operation in the gerda cryostat . two of them , ang 1 and rg 3 , developed high leakage currents at the beginning of phase i. these detectors have been removed from the analysis of phase i data . for some time however , they have been used as veto to suppress multi - site events . the remaining total mass for analysis is 14.6 kg with an average enrichment of 86% in @xmath0ge corresponding to 165 moles . the energy resolution of the germanium detectors is plotted for several energy calibrations with the @xmath30th source . ] energy calibrations are performed on a ( bi)weekly basis with the @xmath30th sources . spectra of the six active enriched detectors are shown in fig . [ fig : calibdets ] , including scaled subplots for the 583.2 and the 2614.5 kev lines . the high count rates cause pile up that would manifest itself in tails on the low energy side of the peaks . proper pile up rejection algorithms and further data quality cuts have been applied before the fitting @xcite . l c c c c experiment & diodes & @xmath18e & exposure & background index + diode environment & & ( kev ) & ( kg@xmath13yr ) & 10@xmath14cts/(kev@xmath13kg@xmath13yr ) + igex @xcite + vacuum , cu enclosed & enr & 2000 - 2500 & 4.7 & 26 + hdm @xcite + vacuum , cu enclosed & enr & 2000 - 2100 & 56.7 & 16 + + lar & nat & 1839 - 2239 & 0.6 & 18@xmath483 + lar , cu mini - shroud & nat & 1839 - 2239 & 2.6 & [email protected] + ditto & enr & 1839 - 2239 & 0.7 & 4.3@xmath56 + gerda phase i + lar , cu mini - shroud & nat & 1839 - 2239@xmath42 & 1.2 & 3.5@xmath57 + lar ( diodes ac - coupled ) & nat & 1839 - 2239@xmath42 & 1.9 & 6.0@xmath57 + lar , cu mini - shroud & enr & 1939 - 2139@xmath42 & 6.1 & 2.0@xmath58 + @xmath59 ) excluding the blinded region of @xmath3@xmath4820 kev the peaks are fitted well by a gaussian and an error function representing the background . the results are shown by the red lines and the fwhm of the gaussian is given in kev . values between 4.2 to 5.3 kev ( fwhm ) at 2614 kev have been obtained . these can be translated to a mass weighted average of 4.5 kev ( fwhm ) at @xmath3=2039.01(5 ) kev @xcite . the resolution of the 2614.5 kev line for all detectors during the first months of data taking is shown in fig . [ fig : calibtime ] . no significant variation or trend is visible for this period . the same is also true for the gain , which normally changed only after some power cycling or temperature drifts . the 2614.5 kev @xmath33-line positions in the calibration spectra are stable in time as shown in fig . [ fig : calstab ] as they fall into a range of @xmath481.3 kev . the shifts observed between two calibrations can be scaled linearly to the interesting energy at @xmath3 . the two lines at @xmath481.3 kev shown in fig . [ fig : calstab ] correspond to @xmath481 kev at @xmath3 . the gain shifts within the roi thus are typically less than 1 kev . this value is small compared to the average fwhm of 4.5 kev and shows that the data from all periods can be added in the search for the peak of the @xmath1 decay . variations of the 2614.5 kev @xmath33 line between successive calibrations . the green lines indicate @xmath481 kev variations at @xmath3 if scaled linearly in energy . ] the commissioning of gerda started with a string of three bare low background @xmath24ge diodes , and yielded a surprisingly large bi on the order of the hdm and igex experiments ( @xmath60 cts/(kev@xmath13kg@xmath13yr ) , see table [ tab : bi ] ) . as another surprise , the line at 1525 kev from @xmath23k , the daughter of @xmath23ar , appeared in the spectra with an intensity much higher than expected on the basis of the upper limit for the ratio @xmath23ar/@xmath61ar determined by v.d . et al . _ @xcite . the published limit of @xmath464.3@xmath6210@xmath63 g / g at 90% confidence level converts to an upper limit of 41 @xmath39bq / kg for @xmath23ar . these observations led to the working hypothesis that charged ions , and in particular the progeny @xmath23k , are drifting in the electric field of the bare ge diodes that are biased with 3 to 4 kv via the large n@xmath64 surface ( see sec . [ sec : det ] and fig . [ fig : detschem ] ) . the concentration of radioactive impurities near the ge diodes can increase . further studies with different bias schemes confirmed this hypothesis . a major improvement of the bi was achieved by enclosing the string of detectors with a cylinder , made of 60 @xmath39 m thick cu foil , called `` mini - shroud '' ( bi @xmath65 cts/(kev@xmath13kg@xmath13yr ) ) . the volume of lar from which the ions can be collected onto the surface of the detectors is reduced and bulk convection of the lar near the detectors is prevented . in fact , operating the ge diodes in ac - coupled mode ( @xmath66 surface grounded and @xmath67 contact biased ) without mini - shroud but with adequate shielding of the @xmath67 contact , i.e. without external electrical stray field , yielded a similar bi of @xmath68 cts/(kev@xmath13kg@xmath13yr ) ( see next to last line in table [ tab : bi ] ) . for the phase i physics run , the hermeticity of the mini - shroud , as well as the shielding of the hv cables , was further improved in order to avoid any leakage of electric field lines into the lar volume . the improvement with respect to the precursor experiments is evident . the stability of the bi must be proven for a longer period of time . c c |r@ / lc| r@ / lc| r@ @xmath48 l & & & & + isotope & energy & & rate & & rate & + & & & & & & + ' '' '' @xmath69k & 1460.8 & 85 & 15 & @xmath70 & 125 & 42 & @xmath71 & 181 & 2 + ' '' '' @xmath28co & 1173.2 & 43 & 38 & @xmath72 & 182 & 152 & @xmath73 & 55 & 1 + & 1332.3 & 31 & 33 & @xmath74 & 93 & 101 & @xmath75 & 51 & 1 + @xmath32cs & 661.6 & 46 & 62 & @xmath76 & 335 & 348 & @xmath77 & 282 & 2 + @xmath30ac & 910.8 & 54 & 38 & @xmath78 & 294 & 303 & @xmath72 & 29.8 & 1.6 + ' '' '' & 968.9 & 64 & 42 & @xmath79 & 247 & 230 & @xmath80 & 17.6 & 1.1 + @xmath81tl & 583.2 & 56 & 51 & @xmath82 & 333 & 327 & @xmath83 & 36 & 3 + & 2614.5 & 9 & 2 & @xmath84 & 10 & 0 & @xmath85 & 16.5 & 0.5 + ' '' '' @xmath86pb & 352 & 740 & 630 & @xmath87 & 1770 & 1688 & @xmath88 & 138.7 & 4.8 + ' '' '' @xmath86bi & 609.3 & 99 & 51 & @xmath89 & 351 & 311 & @xmath90 & 105 & 1 + ' '' '' & 1120.3 & 71 & 44 & @xmath91 & 194 & 186 & @xmath92 & 26.9 & 1.2 + ' '' '' & 1764.5 & 23 & 5 & @xmath93 & 24 & 1 & @xmath94 & 30.7 & 0.7 + ' '' '' & 2204.2 & 5 & 2 & @xmath95 & 6 & 3 & @xmath96 & 8.1 & 0.5 + an analysis of the intensity of the 1525 kev line gives a concentration for @xmath23ar that is about twice the literature limit . this estimate is based on the assumption of a homogeneous distribution of this isotope outside the mini - shroud . the intensity of @xmath33 lines was investigated in order to identify sources of backgrounds . the results are compiled in table [ tab : backgroundlines ] for the natural and the enriched detectors in comparison to numbers from hdm @xcite . the rate estimates are based on a bayesian approach starting with a flat prior probability distribution function . the general observation is an achieved reduction by about a factor of 10 with respect to the hdm experiment . the composition of the background in relation to the screening results will be discussed in a future publication . additional contributions to the bi will result from radioactive surface contaminations such as @xmath97pb as well as from cosmogenically produced radioisotopes within the diodes . these contributions are expected to be small and will require large data sets to evaluate . finally , it is worth to mention that auxiliary experiments were performed to study the cross sections of cosmogenic activation of steel and other constructional materials @xcite , the inelastic neutron scattering @xcite , the neutron activation cross sections , and the @xmath33 decay spectra @xcite . the deduced contributions to the bi are in the order of few 10@xmath98 cts/(kev@xmath13kg@xmath13yr ) . first energy spectra for enriched and natural diodes are shown in fig . [ fig : spectra - ge ] . spectra taken with enriched ( red ) and non - enriched ( blue ) detectors during the same time period . the @xmath24ge spectrum has been normalized to match the exposure of @xmath20ge . the blinding window of @xmath3@xmath4820 kev is indicated as green bar . identified @xmath33 lines are indicated . ] notice , that the spectrum from the natural detectors has been renormalized to match the exposure of the enriched diodes . the low energy part is dominated by the @xmath99 decay of @xmath22ar which has an endpoint energy of 565 kev . the well known activity of @xmath100ar)= [ [email protected](stat)@xmath480.08(syst ) ] bq / kg ( ref . @xcite ) describes the observed intensity . the enhancement of @xmath6 events in the range from 600 to 1400 kev for the enriched detectors is clearly visible . the bi of @xmath101 @xmath102 cts/(kev@xmath13kg@xmath13yr ) for the enriched detectors is evaluated in the energy region @xmath3@xmath48100 kev with the 2019 to 2059 kev window excluded ( green bar in fig . [ fig : spectra - ge ] ) . this value is an order of magnitude lower than the one for the very same detectors in their previous shielding in the hdm and igex experiments ( see table [ tab : bi ] ) . gerda searches for @xmath1 decay of @xmath0ge using a new experimental concept . bare germanium diodes are operated successfully in a 4 m diameter cryostat filled with lar , requiring only a small amount of radiopure materials as mechanical and electrical support . shielding against external background is achieved by lar and an additional shell of 3 m of water . the experiment started commissioning in may 2010 and in november 2011 with physics data taking ( phase i ) . the experience gained so far shows that all components work well . 1 . the operation of the cryostat inside the water tank is stable and safe . 2 . bare germanium diodes are operated reliably in liquid argon over a long time and the implemented handling procedure ensures that many operational cycles do not deteriorate the performance . the readout electronics is balancing the partially conflicting requirements of good energy resolution , low radioactivity , and operation at lar temperature . the water tank instrumentation ensures a high veto efficiency of muon events and only a tolerable loss of 3 out of 66 pmts have stopped functioning during a 2 year period . data acquisition and monitoring of the ambient parameters operate reliably . the implemented software allows for a fast reconstruction of the data together with a good monitoring of data quality . the experience from the ( bi-)weekly calibrations shows that the gain drifts of the entire readout chain are typically smaller than 1 kev at @xmath3 . this is small enough to ensure that adding all data will not result in relevant shifts of peak positions or deteriorations of resolutions . the surprisingly large background from @xmath23k , the @xmath23ar progeny , experienced during the commissioning can be mitigated by two methods : encapsulation of each detector string by a closed thin - walled copper cylinder or ac coupling of the detectors . in both cases the electric field outside of the encapsulation is minimized . the phase i background is determined currently to @xmath101@xmath102 cts/(kev@xmath13kg@xmath13yr ) . this value and the intensities of gamma lines show an order of magnitude improvement compared to the previous hdm and igex experiments . in the absence of a signal and given the current bi , expects to set 90 % probability lower limits of @xmath103 yr for an exposure of 20 kg@xmath13yr . the gerda experiment is supported financially by the german federal ministry for education and research ( bmbf ) , the german research foundation ( dfg ) via the excellence cluster universe , the italian istituto nazionale di fisica nucleare ( infn ) , the max planck society ( mpg ) , the polish national science centre ( ncn ) , the russian foundation for basic research ( rfbr ) , and the swiss national science foundation ( snf ) . the institutions acknowledge also internal financial support . preparing and setting up the infrastructure of the gerda experiment was made possible only through the indispensable help of r. adinolfi falcone , t. apfel , p. aprili , j. baumgart , g. bucciarelli , m. castagna , f. costa , n. dambrosio , s. flicker , d. franciotti , h. fuchs , h. hess , v. mallinger , p. martella , b. mrk , d. orlandi , m. reissfelder , t. schwab , r. sedlmeyr , s. stallio , r. tartaglia , e. tatananni , m. tobia , d. wamsler , g. winkelmller , and t. weber . 000 i. abt _ et al . _ , ( gerda collaboration ) , _ gerda : the germanium detector array for the search of neutrinoless @xmath104 decays of @xmath0ge at lngs _ , letter of intent to lngs , see also : hep - ex/04040390 , proposal to lngs 38/04 ( 2004 ) , http://www.mpi-hd.mpg.de/gerda/ ; phd theses : http://www.mpi-hd.mpg.de/gerda/public/phd.html . j. schechter and j.w.f . valle , phys . rev . * d 25 * , 2951 ( 1982 ) . s. davidson , e. nardi and y. nir , phys . rept . * 466 * , 105 ( 2008 ) . f. avignone iii , s. r. elliot , j. engel , rev . phys . * 80 * , 481 ( 2008 ) . barabash , phys . nucl . * 74 * , 603 ( 2011 ) . gomez - cadenas _ et al . _ , rivista nuovo cimento * 35 * , 29 ( 2012 ) . s. r. elliot , mod . . lett . * a 27 * , 1230009 ( 2012 ) . b. schwingenheuer , procs . 12@xmath105 international conference on topics in astroparticle and underground physics , munich , sep 5 - 9 , 2011 . m. doi , t. kotani and e. takasugi , prog . . suppl . * 83 * , 1 ( 1985 ) . j. suhonen and o. civitarese , nucl . phys . * a847 * , 207 ( 2010 ) , and refs . therein ; j. suhonen , j. phys . , conf . ser . * 375 * , 042026 ( 2012 ) . j. menendez , a. poves , e. caurier , and f. nowacki , nucl . a818 * , 139 ( 2009 ) , and refs . therein . v. a. rodin _ _ , nucl . phys . * a766 * , 107 ( 2006 ) [ erratum - ibid . * a793 * , 213 ( 2007 ) ] ; f. simkovic _ et al . _ , phys . rev . * c 79 * , 055501 ( 2009 ) ; v. a. rodin , j. phys . , conf . ser . * 375 * , 042025 ( 2012 ) . j. barea and f. iachello , phys * c 79 * , 044301 ( 2009 ) . w. rodejohann , int . j. mod e20 * , 1833 ( 2011 ) . e. fiorini _ et al . _ , phys . lett . * b 25 * , 602 ( 1967 ) ; + f.t . avignone iii _ lett . * 50 * , 721 ( 1983 ) ; + j. j. simpson _ et al . _ , rev . lett.*53 * , 141 ( 1984 ) ; + e. bellotti _ _ , phys . lett . * b 146 * , 450 ( 1984 ) ; + h. ejiri _ et al . _ , phys . * a448 * , 271 ( 1986 ) ; + d. caldwell , j. phys . * g 17 * , s137 ( 1991 ) ; + p. fisher _ et al . _ , phys . lett . * b218 * , 257 ( 1989 ) ; + a. morales _ et al . _ , j. phys . * g 17 * , s211 ( 1991 ) ; + j. busto _ et al . _ , nucl . phys . * a513 * , 291 ( 1990 ) ; + a.a . a5 * , 1299 ( 1990 ) ; + i.v . kirpichnikov , preprint itep91 - 91 ( 1991 ) . + hdm - collaboration : h. v. klapdor - kleingrothaus _ et al . j. * a12 * , 147 ( 2001 ) . igex - collaboration : c. e. aalseth _ et al . d * 65 * , 092007 ( 2002 ) ; + c. e. aalseth _ et al . _ , c * 59 * , 2108 ( 1999 ) ; + d. gonzalez _ et al . _ , nucl . phys . b procs . * 87 * , 278 ( 2000 ) ; + d. gonzalez _ et al . methods * a515 * , 634 ( 2003 ) h.v . klapdor - kleingrothaus _ et al . _ , b586 * , 198 ( 2004 ) . majorana - collaboration : r. henning , cipanp 2009 conference proceedings , aip conf . proc . * 1182 * , 88 ( 2009 ) ; j. wilkerson _ et al . _ , j. phys . , conf * 375 * , 042010 ( 2012 ) . g. heusser , ann . nucl . part . * 45 * , 543 ( 1995 ) .. a. smolnikov and p. grabmayr , phys . rev . * c81 * , 028502 ( 2010 ) . g. zdesenko _ et al . _ , j. phys . g : nucl . part . * 27 * , 2129 ( 2001 ) . m. di marco , p. peiffer and s. schnert , nucl . . suppl . * 172 * , 45 ( 2007 ) . p. peiffer _ et al . _ , jinst * 3 * , p08007 ( 2008 ) . 8 pmt , type 9357 , 9350 kb & kb9354 kb . et enterprises limited , uxbridge , uk . m. agostini _ _ , j. phys . , conf ser . * 375 * , 042009 ( 2012 ) . h. neder , g. heusser , m. laubenstein , appl . * 53 * , 191 ( 2000 ) . l. baudis _ _ , jinst * 6 * p08010 , 2011 geant4 collaboration , s. agostinelli _ et al . _ , instrum . methods a * 506 * , 250 ( 2003 ) . geant4 collaboration , j. allison _ et al . _ , ieee trans . nucl . sci . * 53 * , 270 ( 2006 ) . m. boswell _ et al . _ , ieee trans . sci . * 58 * 1212 , ( 2011 ) , arxiv:1011.3827 [ nucl - ex ] . i. barabanov _ et al . methods * a606 * , 790 ( 2009 ) . dementyev and n.m . sobolevsky , radiat . measure . * 30 * , 553 ( 1999 ) . i. barabanov _ et al . methods * b251 * , 115 ( 2006 ) . w.b . wilson _ _ , sources-4a , technical report la-441 13539-ms los alamos ( 1999 ) . m. agostini _ et al . _ , physics b proceedings supplement * 229 - 232c * , 489 ( 2012 ) . , lammerdries 25 , b-2250 , olen , belgium . m. barnab heider _ et al . _ , nss08 , ieee conf . publication 3513 ( 2008 ) . m. barnab heider , phd thesis , u. heidelberg , may 2009 . klapdor - kleingrothaus , nucl . methods * a566 * , 472 ( 2006 ) . j. hellmig , h.v . kleingrothaus , z. phys . * a359 * , 351 ( 1997 ) ; l. baudis _ et al . _ , methods * a481 * , 149 ( 2002 ) . o. chkvorets , phd thesis , u. heidelberg , 2008 . , boeretang 200 , be-2400 mol , belgium . klapdor - kleingrothaus _ et al . methods * a522 * , 371 ( 2004 ) . v. khournoukov , priv . communication ; measurements performed by the botchvar insitute , moscow , russia . j. r. de laeter _ , pure appl . chem . * 75 * , 683 ( 2003 ) . m. barnab heider _ et al . _ , jinst * 5 * , p10007 ( 2010 ) . currently known as joint stock company `` production association electrochemical plant '' ( jsc `` pa electrochemical plant '' ) , uranium enrichment enterprise of the state atomic energy corporation `` rosatom '' . ppm pure metals , gmbh , in langelsheim , germany . canberra oak ridge , usa . e. haller , advances in physics * 30 * , 93 ( 1981 ) . canberra broad energy ge ( bege ) detector , www.canberra.com/products/485.asp . d. budj _ et al . _ , jinst * 4 * , p10007 ( 2009 ) . m. agostini _ et al . _ , jinst * 6 * , p03005 ( 2011 ) . k. zuber , phd thesis , u. heidelberg , 1992 ; analysis of ang 1 by j. echternach . avignone , priv . comm . , 2011 . w. maneschg _ et al . _ , nucl . methods * a593 * , 448 ( 2008 ) . torlon type 4503 , torlon is a trademark of solvay advanced polymers . g. heusser _ et al . isot . * 52 * , 691 ( 2000 ) . synthetic charcoal by carbo - act international ( nl ) . makrolon is a registered trade mark of basf . haberstroh , advances in cryogenic engineering * 53 * , 1201 ( 2008 ) . ld301 hart pressure transmitter , smar research corporation , ny , usa . m. balata , _ et al . methods * a 370 * , 605 ( 1996 ) ; m.g . et al . _ , ultrapure water * 13 * , 59 ( 1996 ) . institute of environmental science and technology , arlington heights ( il ) , usa . habia cable , upplands vsby , sweden . sami s.p.a , liscate ( mi ) italy . teledyne reynolds uk , newbury , berkshire , uk . f. froborg , phd thesis , u. zurich , 2012 . national instruments corporation , austin ( tx ) , usa . m. knapp , phd thesis , kepler center for astro and particle physics , u. tbingen , 2009 . daylighting film df2000ma ( vm2000 ) , 3 m , st . paul , mn , usa . rg213c / u by jowo systemtechnik , delmenhorst , germany . caen s.p.a . , viareggio , italy . f. ritter _ et al . methods * a617 * ( 2010 ) 420 . f. ritter , phd thesis , kepler center for astro and particle physics , u. tbingen , 2012 . scintillator ups-923a is based on polystyrol with additive of ptp ( 2% ) and popop ( 0.03% ) built by the kharkov crystal institute , ukraine . hm6780 - 2 ( @xmath106 8 mm ) hamamatsu photonics k.k . , hamamatsu city , japan pmt-85 ( @xmath106 30 mm ) by kvadrotech , moscow , ru . s - type y11(250)m , kuraray , japan . ( the middle of the fiber is made of polystyrol with additive 0.02% of wavelength shifter y11 . the internal part of fiber consists of pmma , external from the fluorinated pmma ) . s. riboldi _ et al . _ , ieee nuclear science symposium and int . workshop on room temperature semiconductor detectors , oct 30 - nov 6 , p. 1386 polyflon company , one willard road , norwalk , ct 06851 , usa . stycast , emerson & cuming , westerlo , belgium . r. isocrate _ _ , _ md2s - digital sampling electronics for the mars detector _ , lnl annual report 2004 , 226 . c.a . et al . _ , lnl annual report 2004 , 228 . j. stein , f. scheuer , w. gast and a. georgiev , nucl . instrum . methods * b 113 * , 141 ( 1996 ) . qt cross platform application frame work , qt project hosting , oslo , norway . struck innovative systeme gmbh , 22399 hamburg , germany . pas9817/ao dac , precision analog systems co. , plantation ( fl ) usa . m. agostini _ et al . _ , j. phys . , conf * 375 * , 042027 ( 2012 ) . rene brun and fons rademakers , root nucl . methods * a 389 * , 81 ( 1997 ) . m. agostini , l. pandola , p. zavarise and o. volynets , jinst * 6 * , p08013 ( 2011 ) . m. agostini , l. pandola and p. zavarise , j. phys . : conf . ser . * 368 * , 012047 ( 2012 ) . _ et al . _ , j. phys . ser * 375 * , 042028 ( 2012 ) . hewlett packard procurve switch 2810 - 48 g . dell power edge r300 , quad core xeon x3323 , 2.5 ghz . r. brugnera _ et al . _ , jinst * 7 * , p100712 ( 2012 ) . postgresql database management system . ajax ( asynchronous javascript and xml ) , j.j . ajax : a new approach to web applications _ , adaptive path 2005 . h. neder , _ et al . isot . * 53 * , 191 ( 2000 ) . dupont performance elastomers , wilmington ( de ) , usa . , or ethylen - propylen - dien - monomer ( epdm ) . pfeiffer vakuum , asslar , germany . bd sensors , thierstein , germany . b. bellini _ et al . _ , submitted to jcap , arxiv:1202.6403 g. douysset _ et al . _ , rev . lett.*86 * , 4259 ( 2001 ) . v.d . ashitkov _ et al . _ , instruments and experimental techniques , * 46 * , 153 - 160 ( 2003 ) . m. laubenstein and g. heusser , appl . rad . and isotopes * 67 * , 750 ( 2009 ) . a. domula , phd thesis in preparation , tu dresden . g. meierhofer _ et al . _ , phys . rev . c * 81 * , 027603 ( 2010 ) ; g. meierhofer _ et al . _ , j. * a40 * , 61 ( 2009 ) . j. marganiec _ et al . _ , c * 79 * , 065802 ( 2009 ) ; j. marganiec _ et al . _ , j. phys . * g35 * , 014022 ( 2008 ) . g. meierhofer _ et al . _ , j. * a48 * , 20 ( 2012 ) . p. benetti _ et al . _ , methods * a574 * , 83 ( 2007 ) .

using a combination of analytical theory , monte carlo simulations , and three dimensional self - consistent field calculations , we study the equilibrium properties and the switching behavior of adsorption - active polymer chains included in a homopolymer brush . the switching transition is driven by a conformational change of a small fraction of minority chains , which are attracted by the substrate . depending on the strength of the attractive interaction , the minority chains assume one of two states : an exposed state characterized by a stem - crown - like conformation , and an adsorbed state characterized by a flat two - dimensional structure . comparing the monte carlo simulations , which use an edwards - type hamiltonian with density dependent interactions , with the predictions from self - consistent - field theory based on the same hamiltonian , we find that thermal density fluctuations affect the system in two different ways . first , they renormalize the excluded volume interaction parameter inside the brush . the properties of the brushes can be reproduced by self - consistent field theory if one replaces by an effective parameter , where the ratio of second virial coefficients depends on the range of monomer interactions , but not on the grafting density , the chain length , and . second , density fluctuations affect the conformations of chains at the brush surface and have a favorable effect on the characteristics of the switching transition : in the interesting regime where the transition is sharp , they reduce the free energy barrier between the two states significantly . the scaling behavior of various quantities is also analyzed and compared with analytical predictions .

1505.01614

polymer brushes are organic surface layers formed by polymer chains that are grafted at one end to a substrate @xcite . since the thickness of brush layers is typically in the nanometer range , they are interesting for the design of functional surfaces @xcite with applications in a wide variety of areas ranging from colloidal stabilization @xcite , lubrication @xcite , controlled friction and adhesion @xcite , anti - fouling @xcite , biocompatibility@xcite , drug delivery @xcite , and smart stimuli - responsive materials @xcite . in this respect , multicomponent polymer brushes are particularly promising @xcite . if the brush chains are covalently bound to the substrate , they can not phase separate on a global scale , but they can still develop structure on the nanoscale . the resulting brush morphologies are controlled both by the intrinsic properties of the brush , e.g. , chemical properties of the chains ( compatibility or incompatibility ) , the grafting densities , the chain lengths , and by the environment - related parameters , such as solvent selectivity , substrate preference , temperature , and the ph . due to the ability of polymer brushes to selectively respond to environmental stimuli , they can be used to design materials that can reversibly switch / tune their surface properties , e.g. , with respect to wettability @xcite , permeability @xcite , friction @xcite , and optical properties @xcite . stimuli - induced phase separation provides the basic mechanism for the change in the brush surface properties depending on which of the two microphases forms the outer part of the brush . a typical example for such a morphology - related switchable surface is a mixed polymer brush with equal amounts of hydrophobic and hydrophilic polymers grafted on a substrate . by treatment of different solvents , the surface composition can change and then the wettability of this material switches @xcite . morphology change necessarily involves slow highly cooperative chain dynamics and therefore a typical response time turns out to be in the range of several minutes or larger @xcite . in a recent letter , we have proposed a new class of brush - based switches @xcite , which rely on a radical conformational change of individual adsorption - active minority chains in an otherwise inert brush . the selective adsorption driving the transition may arise , e.g. , from electrostatic interactions or hydrogen bonding between active groups in the minority chains and the substrate . these active groups can serve as responsive sensors that trigger the switching transition . the most radical change is associated with the end group of the minority chain . in the adsorbed state it resides in close proximity of the substrate deeply buried within the brush , while in the other state it is exposed to the environment . thus the transition could potentially promote a specific immune - like response . chemically or biologically active groups can be attached to the free end of the minority chain to serve as practically useful sensors . based on theoretical arguments and simple one - dimensional mean - field calculations , we demonstrated that even a small chain length increment of about 10 % produces a sharp transition from the exposed state to the adsorbed state . the transition time is expected to be very short since the free energy barrier between these two thermodynamically stable states is about several @xmath3 . a further increase in the minority chain length will lead to sharper transitions , but at the same time to longer transition times due to higher barriers . the strong response of the chain conformation of a minority chain @xcite to small variations of its length is related to the recently reported `` surface instabilities '' in polymer brushes @xcite , which can be used to sense solvent quality @xcite . in our work , we proposed to use single chains as switches triggered by a change in substrate - polymer interaction . one major benefit of this switch is that it does not involve cooperative rearrangements of many chains , therefore the switching transition is fast as compared to the existing examples in mixed brushes . the arguments presented in ref . @xcite rely on several assumptions . first , the presence of the minority chains was taken to have no effect on the surrounding polymer brush . second , thermal fluctuations of the majority brush component were disregarded . in the current paper , we present extensive monte carlo ( mc ) simulations of a coarse - grained model for polymer brushes with a single immersed adsorption - active minority chain . to assess the influence of fluctuations separately , we have also performed three - dimensional self - consistent field ( scf ) calculations for comparison . we investigate in detail both lateral and longitudinal characteristics of chain conformations and the various characteristics of the transition . the remainder of the paper is organized as follows : sec . [ sec : mc_model ] outlines the mc model and the simulation scheme . [ sec : brush ] presents the mc results for a homogeneous monodisperse brush and comparison with the established analytical scf theory , and analyzes the dependence of the renormalized virial coefficient on the effective interaction range . sec.[sec : theory ] presents a more detailed derivation of the theory for the adsorption - active minority chain sketched in ref . @xcite . in sec . [ sec : results ] the main results of the mc simulations are presented and discussed . this section also compares mc and 3-d scf calculations with the emphasis on fluctuation effects . a general discussion is given in the final section [ sec : conclusions ] . in the simulation community , a variety of numerical methods have been developed and used to investigate polymer brush systems , ranging from molecular dynamics @xcite , dissipative particle dynamics @xcite , and monte carlo ( mc ) simulations @xcite to numerical self - consistent field ( scf ) calculations @xcite . the present study mainly relies on mc simulations , while for comparison we also present results obtained from 3-d scf theory . the mc simulation model and scheme are described as follows , and the detailed description of scf method is shown in the appendix . in the mc simulations , we adopt a coarse - grained off - lattice model first proposed by laradji et al @xcite . in this approach , a particle - based representation of the polymers is combined with an edwards type hamiltonian @xcite , which defines non - bonded interactions in terms of local monomer densities . compared to the more commonly used coarse - grained polymer models with hard - core monomer interactions , the laradji model has two advantages : first , it can be simulated very efficiently , since the explicit evaluation of pair interactions is often the most time consuming part in a simulation . second , it uses soft potentials , hence equilibration times are comparatively short . the model does not account for packing effects and for topological constraints ( which may restrict the conformational phase space for strongly adsorbed , quasi - two dimensional polymers ) . however , these are not in the focus of the present study . specifically , the model system is a monodisperse brush in a good solvent containing a single minority chain in a volume @xmath4 . we use periodic boundaries along @xmath5 and @xmath6 directions , while impenetrable boundary walls are placed at @xmath7 and @xmath8 . polymer chains are modelled by the discrete gaussian bead - spring model with spring constant @xmath9 , where @xmath10 is the statistical bond length , @xmath11 is the boltzmann constant , and @xmath12 the temperature . we will use @xmath10 as the basic length unit , and @xmath3 as the energy unit . all chains are attached to a substrate placed at @xmath13 at one end , where @xmath13 is chosen smaller than @xmath10 . the impenetrable wall at @xmath7 exerts an attractive potential to all minority chain beads with strength @xmath14 in a range of @xmath15 . there is no explicit solvent in the model and all the relevant interactions in the good solvent case are represented through an effective excluded volume potential between monomers ( beads ) . the edwards type hamiltonian of the system reads @xmath16 where @xmath17 , @xmath18 denotes the position of the @xmath19-th bead in the @xmath20-th chain ( the minority chain has the index @xmath21 ) , @xmath22 is the total number of brush chains , @xmath23 denotes the local density of minority chain monomers , @xmath24 , and @xmath25 the total density of monomers with @xmath26 being the local density of brush monomers . ( here and in the following , the subscript @xmath27 will be used to denote the brush chains , and @xmath28 will denote the minority chain . ) the first two terms in the hamiltonian describe the gaussian stretching energy ( bonded interactions ) between two neighboring monomers in the same chain . the third term represents the non - bonded effective interactions between polymer beads . for a good solvent the excluded volume parameter @xmath29 is larger than zero . the fourth term describes the adsorption between the substrate and the minority beads with a step - like adsorption potential , @xmath30 in the mc simulations , local densities are extracted from the position of the beads by using the particle - to - mesh technique @xcite , which provides a way for the smoothing of density operators . in the present work we use the zeroth order scheme . the system is divided into cubic cells of size @xmath31 whose centers define the grid points , and the local densities evaluated by counting the total number of beads in the corresponding cells divided by the cell volume , are defined at these grid points . the smoothed monomer densities for the minority chain and brush chains are denoted as @xmath32 and @xmath33 , respectively . this implies that two beads interact only when they are in the same cell , and thus the size of the cell is indirectly linked to the interaction strength of non - bonded interactions . higher order schemes are conceivable , but computationally more expensive . after the density operators are smoothed over the cell volume , the hamiltonian acquires the familiar form used in the scf theory @xcite , hence scf calculations and mc simulations can be compared directly . in the present study , the brush chain length is fixed at @xmath34 . in most mc of calculations we choose the excluded volume parameter @xmath35 which would correspond to a flory - huggins parameter @xmath36 in an explicit solvent solution ( athermal solvent condition ) , and the size of the averaging cell is taken @xmath37 . we model relatively dense monodisperse polymer brushes in good solvent with the surface grafting density @xmath38 , with the corresponding overlap parameters @xmath39 , where @xmath40 is the mean radius of gyration for an ideal brush polymer . the grafting points of the chains on the substrate were fixed on a regular square lattice.the system size was chosen @xmath41 , @xmath42 , and @xmath43 in most cases . to assess the influence of finite - size effects , we have also carried out simulations of larger systems with size @xmath44 , @xmath45 , and @xmath43 for selected parameter values . the results were identical within the error . mc simulations @xcite were carried out according to the standard metropolis criterion . at every mc update , we try to move the position of one chosen monomer to a new position with a distance in space comparable to the bond length . this trial move results in an energy change including the bonded energy and non - bonded energy , and it is accepted or rejected according to the boltzmann probability in the usual way . in all the simulations @xmath46 mc steps per monomer were performed to equilibrate the system , and another @xmath46 mc steps to extract statistical averages . in order to get good statistics sampling for the single minority chain , an additional @xmath47 mc steps per monomer updating the minority chain monomers were performed during the process of evaluating statistical averages . final statistical quantities were obtained by averaging the results from 48 separate independent mc runs . we first analyze the simulation results for homogenous monodisperse brushes which are very well understood . in contrast to common mc polymer models our model contains an additional free parameter , the cell size @xmath27 , defining the averaging volume for the local density operator , and the excluded volume parameter @xmath29 . [ fig : pure_brush ] shows the density profiles of homogeneous brushes with two different grafting densities @xmath48 and @xmath49 , and various values of the averaging cell size @xmath27 . it is clear that the choice of cell size has a pronounced effect on the brush density profile . to rationalize this effect we turn to the analytical scf theory developed over 20 years ago @xcite and subsequently verified by simulations and experiments . within the second virial approximation for good solvent conditions and assuming gaussian elasticity for the chains , the density profile @xmath50 has a parabolic form @xmath51 where the brush height is @xmath52 , and @xmath53 an effective interaction parameter . gaussian elasticity is an important ingredient in the analytical theory and the fact that the mc simulation model is also based on a hamiltonian with a gaussian term describing chain connectivity allows a more direct comparison . it is known that corrections due to finite chain extensibility become important for dense grafting , but this is outside the scope of the present paper . on the other hand , modifications of the chain elasticity due to the excluded volume effects are naturally taken into account by our mc simulations . the mean - field potential associated with the density , profile , @xmath54 is also parabolic @xmath55 where @xmath56 is the potential at the grafting surface . according to the theory ( see for example @xcite ) , the brush density profiles evaluated for the same model at different values of @xmath57 and @xmath58 should collapse in the rescaled coordinates @xmath59 vs @xmath60 with one adjustable parameter related to the interaction parameter @xmath61 @xmath62 \label{eq : collapse}\ ] ] the rescaled mc density profiles with @xmath38 and @xmath63 are presented in fig . [ fig : density_scaled ] , for a specific choice of the averaging cell size , @xmath37 and the `` bare '' excluded volume parameter @xmath35 . the quality of the collapse is good , and one can extract the value of the effective interaction parameter @xmath64 , which differs from @xmath29 . the same procedure was repeated for several other values of the @xmath27 parameter , and different bare excluded volume parameters @xmath29 . to evaluate the difference between @xmath65 and @xmath66 , we calculate the corresponding exact second exact virial coefficient in the simulation model , taking into account the finite grid size @xmath27 . for given interaction parameter @xmath67 , it is given by @xmath68 . the ratio @xmath69 for the best fit values of @xmath65 are plotted as a function of @xmath70 in fig . [ fig : v_b ] . the data for different values of @xmath66 collapse nicely . we conclude that the renormalization effect can be presented in terms of a dimensionless ratio @xmath71 , which is a universal function of a single dimensionless parameter @xmath70 ( or , equivalently , @xmath72 ) . universality in this context does not mean a broad class of different models but rather that other model parameters , such as @xmath57 and @xmath58 do not enter explicitly . it is clear from fig . [ fig : v_b ] that at cell size @xmath73 or larger , the curve saturates and @xmath53 approaches the bare value . hence the system approaches mean - field behavior if the interaction range @xmath27 of the monomers becomes large , as one would expect . at smaller @xmath74 , local monomer correlations become important and renormalize the effective interaction parameter . renormalization of monomer interaction parameters has also been observed in edwards - type simulations of polymer melts @xcite . the theory was briefly sketched in ref . @xcite . in this section , we present it in more detail . we consider a relatively dense monodisperse polymer brush in a good solvent containing a single surface - active minority chain . the length of the minority chain is denoted as @xmath75 , while the length of the brush chain is denoted as @xmath76 . the minority chain and the brush chains are chemically different and have different interactions with the substrate , i.e. , the substrate adsorbs minority chain monomers with strength @xmath14 , while it is neutral to the brush chain monomers . both the minority and the majority chains are flexible with the same statistical kuhn length @xmath10 which is taken as the unit length ; the excluded volume parameter is also the same for both chain types and is positive corresponding to good solvent conditions . this is fully consistent with the mc model . we treat the inter- and intrachain interactions in the mean - field approximation and neglect the back effect of the change in the minority chain conformation on the surrounding brush . hence , the conformation of a minority chain is affected by a fixed mean - field potential profile consisting of the repulsive contribution determined by the brush density and a short - ranged attraction due to the solid substrate . the minority chain itself is described by an ideal continuum model , since intrachain excluded volume effects are screened out considerably within the brush thickness . the analytical description of the minority polymer chain is based on the continuum approach . the green s function @xmath77 , i.e. the total statistical weight of the minority chain grafted at the substrate ( @xmath78 ) as a function of the free end position , @xmath79 , is a solution of the edwards equation @xmath80 taken at @xmath81 , with the initial condition @xmath82 . the total potential @xmath83 , where @xmath84 is the adsorption attraction potential , and @xmath85 is the mean - field brush potential given by eq . ( [ eq : brush_potential ] ) . the adsorption potential @xmath84 is described in the analytical scf approach as an attractive pseudopotential @xmath86 where @xmath87 is the adsorption interaction parameter in the continual model . this parameter was introduced by de gennes @xcite to replace the real adsorption potential through the boundary condition @xmath88^{-1}\frac{\partial g(z , s)}{\partial z}\big|_{z=0}=-c$ ] . close to the substrate , @xmath89 , the brush potential changes very little , @xmath90 where @xmath91 is a shorthand notation for the numerical coefficient the adsorbed state is thus approximately described by the standard green s function modified by the potential @xmath92 @xcite @xmath93 integration over @xmath79 gives the partition function @xmath94 in order to connect this expression to the mc model we note that @xmath95 has the meaning of the negative free energy of adsorption per segment ( the chemical potential ) in an asymptotically long chain , i.e. , @xmath96 . when comparing the actual brush density profile with the theoretical parabolic shape , one concludes that the pseudopotential must account for the combined effect of the actual adsorption potential ( a step of unit width with energy @xmath97 ) and the short - range depression in the brush density near the wall . it is known that for weak adsorption , the free energy per monomer is determined by the crossover exponent @xmath98 . for an ideal chain , @xmath99 while the value for a chain with excluded volume was a subject of extensive investigations and prolonged debates @xcite . for any practical purposes , the adsorption of relatively short chains is very accurately described by the ideal value @xmath99 . here , for weak adsorption , we adopt the expression @xmath100 where @xmath20 and @xmath101 are model - dependent constants to be extracted from the mc data . it is also known that at stronger adsorption , the free energy deviates from eq . ( [ eq : mu_epsilon ] ) , and for very large @xmath14 , it can simply be estimated as @xmath102 , where the constant shift is given by the limiting entropy difference per monomer between the coil and the fully adsorbed state . the state with the free end exposed at the brush edge has no contacts with the surface . hence the green s function is given by the known solution of the edwards equation for a purely parabolic potential and a neutral solid surface @xcite @xmath103^{3/2}ze^{-\frac{3\pi}{4n_b}\cot\big(\frac{\pi n}{2n_b}\big)z^2}. \ ] ] for longer minority chains , @xmath104 , the green s function of eq . ( [ eq : green_exposed ] ) shows unlimited monotonic increase with @xmath79 since the solution refers to a potential extending to infinity and to an infinitely extendable gaussin chain . in a realistic situation the minority chain does not make excursions well beyond @xmath105 . to account for this fact we truncate @xmath106 at @xmath105 . if minority chains are just slightly longer @xmath107 , ( which is the situation of interest in this paper ) , the green s function simplifies to @xmath108 this gives the partition function @xmath109 assuming @xmath110 . eqs ( [ eq : partition_ads ] ) and ( [ eq : partition_ex ] ) clearly show that in the limit of @xmath111 , @xmath112 , @xmath113 the free energies of the two states are extensive and switching become a classical first - order phase transition . based on this analytical approach , the transition properties between the adsorbed state and exposed state can be investigated . for example , one could define the transition point as the condition @xmath114 . omitting logarithmic correlations this gives @xmath115 . under the condition @xmath116 one can use the asymptotic expression eq . ( [ eq : mu_epsilon ] ) to obtain the adsorption energy at the transition @xmath117 for larger grafting densities and chain length differences one expects deviation from the scaling form of of eq . ( [ eq : transition_point ] ) , since the quadratic dependence of @xmath118 may have a limited range . in the extreme case of a very long minority chain , @xmath119 , the transition point satisfies the condition @xmath120 . since the mean - field potential @xmath92 is in the range @xmath121 , the limiting value of @xmath122 remains of order 1 . the sharpness of the transition is characterized by the width of the adsorption parameter change , @xmath123 , sufficient to produce a reliable switching from the adsorbed state with the average distance of the free end of the minority chain @xmath124 to the exposed state with @xmath125 : @xmath126 where the derivative is evaluated at the transition point itself . a standard recipe for a two - state model @xcite gives @xmath127 which combined with eq . ( [ eq : partition_ads ] ) results in @xmath128 finally , the switching time is essentially determined by the barrier separating the two relevant states . hence we estimate the barrier height at the transition point , @xmath129 , in terms of the change in the green s function of the exposed state , @xmath130 . it follows from eq . ( [ eq : green_ex ] ) that up to a numerical coefficient close to unity @xmath131 [ fig : snapshots ] shows typical simulation snapshots of a brush containing a minority chain @xmath132 ( monomers ) in an exposed and an adsorbed state . there is a substrate surface below and a virtual surface above showing the brush height @xmath133 . as can be seen a small fraction of monomer units are located above @xmath133 . these monomers form a tenuous brush exterior . the minority chain is shown in green color . it is longer than the majority brush chains with @xmath34 . we characterize the length of the minority chain by a parameter @xmath134 ( in the snapshot , @xmath135 is given by @xmath136 ) . the minority chain in the exposed state has a stem - crown - like conformation . the main features of the snapshot can be also seen in fig . [ fig : density_z ] where the longitudinal monomer density profiles are shown for minority chains with different @xmath135 in the exposed state at @xmath137 ( panel a ) , and in the adsorbed state at @xmath138(panel b ) . the density profile of a homogeneous brush is also shown , and for comparison both profiles are normalized in the same way . it is clear that at @xmath137 and @xmath139 the minority chain is identical to brush chains . for longer minority chains , @xmath140 , the density profile deviates from a parabolic , monotonically decreasing form ( apart from a narrow depletion layer near the substrate ) shape . it is well understood that the standard parabolic shape is due to a specific very broad end - monomer distribution . `` partial '' density profiles produced by chains with a given end position are initially increasing with @xmath79 with a subsequent sharp drop . only after averaging over end positions one obtains the actual density profile . the most pronounced effect is observed for the longest minority chain with @xmath136 . the profile can be understood if we take chain conformations with the end - monomer positioned on average close to @xmath105 ( shown by the vertical dashed line ) with a typical fluctuation of a few monomer lengths ( @xmath141 for shorter minority chains the average end - hight is below @xmath133 and the fluctuations are relatively larger , which makes the difference to the majority chains less dramatic . in the strongly adsorbed state the density profile exponentially decreases with @xmath79 and is insensitive to @xmath135 , see fig . [ fig : density_n_e0.7 ] . even though the minority chains are highly dilute , they perturb the brush locally . furthermore , the minority chain itself of course has lateral structure . fig . [ fig : lateral_probability ] shows the lateral distribution of the end monomer @xmath142 as a function of the in - plane radial distance @xmath143 from the grafting point for different adsorption energies @xmath14 at grafting density @xmath144 and chain length @xmath132 . here @xmath142 was obtained by counting how often the minority chain end was located in a cylindrical shell ranging from @xmath145 to @xmath146 and dividing this by the volume of the shell and the total sampling number . the thickness of the shell was chosen as @xmath147 . the results for @xmath137 in fig . [ fig : lateral_probability ] show that @xmath142 exhibits gaussian behavior for nonadsorbing chains . with increasing adsorption strength , the distribution becomes broader , and a depletion region develops close to @xmath148 : chain ends are pushed away from the area close to the grafting point , due to the fact that this area is already covered by adsorbed monomers . this is also shown by the radial distribution density @xmath149 , which gives the probability to find the free end of the minority chain located at distances ranging from @xmath145 to @xmath146 with @xmath147 . at strong enough adsorption the maximum of @xmath150 moves towards larger values of @xmath145 since the flattened adsorbed state dominates ( fig.[fig : lateral_probability ] , inset ) . for @xmath144 , @xmath132 , and different adsorption strengths as indicated . the inset shows the lateral probability given by @xmath151 , where @xmath152=1 . the data are obtained by mc simulations ] defined as @xmath153 vs the adsorption strength for @xmath154 and different minority chain lengths . the data are obtained by mc simulations ] to measure more quantitatively how the minority chain extends in the lateral direction , we calculate the parallel root - mean - squared radius of gyration defined as @xmath155 , where @xmath156 and @xmath157 are the mean - square radius of gyration of the minority chain in the @xmath5 and @xmath6 directions , separately . fig . [ fig : lateral_size ] shows @xmath158 as a function of the adsorption strength at @xmath48 for different minority chain lengths . from these curves it can be seen that the lateral size of the minority chain remains almost unchanged up to @xmath159 , and then starts to increase , which is another signature of the conformational transition from the desorbed to the adsorbed state . [ fig : rxyg ] shows the lateral size of the minority chain in the exposed state with @xmath137 and in a strongly adsorbed state with @xmath160 , as a function of the minority chain length for several grafting densities . in both the exposed state and the adsorbed state , @xmath161 is consistent with a power law . the best fit for the exposed state ( @xmath137 ) gives @xmath162 , with @xmath163 , which is larger than the flory exponent expected for ideal chains ( @xmath164 ) , and smaller than that expected for free chains in a good solvent ( @xmath165 ) . this can be understood since part of the chain well within the brush experiences semi - dilute conditions with screened excluded volume interactions while the part near the tenuous brush exterior is swollen . as one increases the grafting density at fixed length , the lateral size of the exposed chain decreases . this has two reasons : first , the brush thickens with increasing grafting density , and consequently the fraction of swollen minority chain part outside the brush ( the crown ) decreases . second , the monomer density inside the ( semi - dilute ) brush increases , which leads to a stronger screening of excluded volume interactions and a decrease of lateral swelling of the inner part of the minority chain ( the stem ) . according to fig . [ fig : lateral_size ] , the shrinking does not depend noticeably on the chain length @xmath166 between minority chain and brush chains , i.e. , on the size of the crown . hence we conclude that the second effect is probably dominant . for strongly adsorbed chains ( @xmath160 ) , we find a relation @xmath167 , with @xmath168 , which is close to the flory exponent for two dimensional self - avoiding chains ( @xmath169 ) . in the strongly adsorbed state the lateral extension is almost insensitive to the grafting density , see fig . [ fig : rxyg ] , upper line . from the scaling point of view the adsorption blob size is quite small and thus screening of the excluded volume interactions by other chains is ineffective . ) and in a strongly adsorbed state @xmath160 . the symbols denote the results from mc simulations , lines are obtained by a linear fitting after taking the logarithm of the original data ] fig . [ fig : zend_mc ] presents the average distance @xmath170 from the free end of the adsorption - active minority chains to the solid substrate as a function of the adsorption strength @xmath171 , for minority chains of several lengths in a brush with @xmath34 at @xmath144 . the two lower curves represent the behavior of shorter minority chains with @xmath172 and @xmath173 . in the desorbed state , short minority chains exist in a coil - like conformation and hence the adsorption transition is expected to be continuous . the curves are further smeared out due to the finite length of the chain . in contrast , even a small positive increment @xmath134 in the length of minority chain results in a well - pronounced sharp transition from an exposed state in which the free end of the minority chain is localized at the outer space of the brush to an adsorbed state where the chain is localized close to the substrate as illustrated in fig . [ fig : density_z ] fig . [ fig : zend_mc_scf ] compares the mc results with the prediction of the three - dimensional scf , using the interaction parameter @xmath53 extracted from the mc simulations of pure brushes ( fig . [ fig : v_b ] ) . the scf results differ from the mc date in two important aspects . first , they do no capture the attractive contact interaction of the minority chain with the surface accurately . this is not surprising , given that the interaction vanishes on the scale of one grid size , and it leads to a shift of the adsorption curves by a constant value @xmath174 . second , the density fluctuations at the brush surface ( see fig . [ fig : density_scaled ] ) effectively reduce the ( spatial ) range of the repulsive potential , hence the minority chain is less stretched in the exposed state . this effect is even observed for minority chains that have the same length than the brush chains , @xmath175 . we will see below that this leads to a reduction of the free energy barrier between the exposed and the adsorbed state . after shifting the scf curves by @xmath176 and rescaling them with a factor @xmath177 which was obtained empirically , they are in very good agreement with the mc results . curves such as those shown in fig . [ fig : zend ] , which give the average distance of the minority chain free end from the surface as a function of the adsorption strength , can be used to localize the crossover between the adsorbed and the desorbed state . the transition point @xmath178 is then defined as the point where the slope of the curve is maximal ( for @xmath179 ) . the value of the slope at this point is used to extract the width of the transition , @xmath123 . this is done as follows . first we find the point where @xmath170 vs @xmath171 has the maximum slope , and calculate this slope as @xmath180 . then we draw a line through this maximum slope point with the slope @xmath180 and find its intersection with ( i ) the abscissa and ( ii ) with a line parallel to the abscissa that corresponds to the value of @xmath170 at @xmath137 . the absolute value of the difference of the adsorption strength at these two intersection points ( i.e. , difference of their abscissa values ) defines the transition width @xmath181 in order to study the transition mechanism we analyze the distribution of minority chain ends , which we denote as @xmath182 . the product @xmath183 represents the probability to find the free end in a layer with its @xmath79 coordinate ranging from @xmath79 to @xmath184 . in the mc scheme , @xmath182 is evaluated as follows : we split the simulation box along the @xmath79 direction into @xmath185 layers , where each layer has the volume @xmath186 and an index @xmath187 ranging from @xmath188 to @xmath189 . then @xmath190 is obtained by counting how often the free end was located in the @xmath187th layer , divided by @xmath191 and by the total size of the sample . [ fig : pz ] shows examples of distributions @xmath192 for a minority chain of length @xmath193 in a brush with grafting density @xmath49 . one can clearly see that two populations of different conformations ( adsorbed and exposed ) are involved in the transition . in the transition region these populations coexist which is reflected by a bimodal distribution . specifically one can see that at low adsorption strength ( e.g. , @xmath194 ) , the chain end is mostly located outside of the brush , whereas at high adsorption strength ( e.g. , @xmath195 ) , the chain end is localized close to the surface . at intermediate adsorption strengths ( @xmath196 ) , the adsorbed and the exposed state coexist and the minority chain switches back and forth from one to the other . a conventional way to analyze a phase transition is to construct a landau free energy " from the logarithm of the statistical distribution of the order parameter @xcite . we will use the position @xmath79 of the free end of the minority chain as the order parameter . in this case the landau free energy is @xmath197 . selected free energy curves are shown in fig . [ fig : freeenergy ] . the free energy as a function of chain end position has one minimum close to @xmath7 for large @xmath14 ( adsorbed state ) , and one minimum close to @xmath198 for small @xmath14 ( exposed state ) . for intermediate @xmath14 , these two minima compete with each other . an alternative definition of the transition point @xmath199 is the point where both minima have equal depth . however , we wish to stress that at finite chain length , there is no classical phase transition , but rather a crossover point between two regimes . the height of the free energy barrier between the exposed state and the adsorbed state , @xmath200 , is extracted from the landau function curve at this transition point . the equal depth condition for the transition point makes the barrier height definition unambiguous . in the following we compare the transition characteristics as obtained by simulations to the scaling prediction of the analytical theory laid out in section [ sec : theory ] . [ fig : mc_transition_point ] displays the transition point , i.e. the adsorption energy @xmath178 as a function of the scaling parameter @xmath201 combining the grafting density , and the ratio of the minority - to - majority chain lengths . two definitions of @xmath178 have been proposed above , and data points corresponding to both definitions are presented . each set collapses into a single master curve close to the straight line suggested by eq.([eq : transition_point ] ) but due to the effect of finite chain length the lines are separate . we expect the difference to vanish in the appropriate thermodynamic limit @xmath202 , where the transition becomes truly sharp . the general trend is that the transition point shifts to larger adsorption strength @xmath203 with increasing minority chain length and grafting density . this is to be expected , since both factors stabilize the exposed state against the competing adsorbed state . in fig . [ fig : mc_transition_point ] , the intercept and the slope are linked to the model - specific dependence of the adsorbed chain s free energy on @xmath14 , which were therefore treated as free parameters by the theory . one has to be cautious , however , not to overestimate the agreement between theory and simulations . the scaling predictions rely on two assumptions : ( i ) logarithmic corrections are neglected , implying that the relevant free energies per chain are much larger than 1 ( in @xmath204 units ) , @xmath205 ; ( ii ) the adsorption is assumed to be weak , such that the expansion of @xmath118 in eq . ( [ eq : mu_epsilon ] ) can be applied , @xmath206 . at brush chain lengths @xmath207 , it is difficult to satisfy both conditions simultaneously . in systems with small values of @xmath135 and/or small @xmath57 , the first condition becomes questionable , whereas in systems with larger @xmath135 and/or larger @xmath57 , the second condition may be violated . therefore , the observed reasonably good collapse onto a straight line might be the result of cancellation effects . the two most important characteristics of the transition , as far as potential applications are concerned , are the transition sharpness ( or effective width ) and the waiting time characterizing the transition kinetics . the present work does not include dynamic simulations , so we use the free energy barrier at the transition as an indirect measure of the waiting time . the transition width , denoted as @xmath123 and characterized by the inverse slope of @xmath208 , is plotted in fig . [ fig : sharpness_raw ] as a function of the minority chain length increment @xmath135 , while the barrier height ( in units of @xmath204 ) is displayed in the fig . [ fig : barrier_raw ] . we present both the mc results ( symbols ) and the scf results ( lines ) . the transition becomes sharper for longer minority chains and larger grafting densities . the difference between the mc and the scf results for @xmath209 ( fig . [ fig : sharpness_raw ] ) is noticeable but not dramatic . as discussed in sec . [ sec : theory ] ( eq . ( [ eq : sharpness scaling ] ) ) , the transition sharpness is essentially determined by the response of the free energy of the adsorbed state to variations of the adsorption strength @xmath14 , which does not seem to be affected by fluctuations very much . in contrast , the effect of fluctuations on the free energy barrier is quite strong . the free energy barrier grows approximately linearly with @xmath135 and also increases with the grafting density . the absolute values of the barrier height obtained by mc simulations are smaller than those from scf method by roughly a factor of 2 ( fig . [ fig : barrier_raw ] ) . this is clearly a pronounced effect of density fluctuations implying strong consequences on the transition kinetics . the same data are replotted in fig . ( [ fig : transition_scaled1 ] ) with scaling coordinates as suggested by the analytical theory , see eqs ( [ eq : sharpness scaling]),([eq : barrier scaling ] ) . the curve for @xmath123 vs. @xmath210 is expected to pass through the origin since in the thermodynamic limit , one has @xmath211 , and the transition becomes jump - like with @xmath212 . at finite brush chain length @xmath76 and for larger values of the scaling parameter @xmath210 finite chain length effects and logarithmic corrections become important . this presumably explains why the data do not collapse on a master curve , and why the scf data do nt even seem to converge towards the expected behavior @xmath213 at @xmath211 . the mc data are better compatible with a limiting straight line passing through the origin , and the slope has the correct order of magnitude . according to the theoretical expectation ( eqs ( [ eq : transition_point ] ) and ( [ eq : sharpness scaling ] ) ) the slope in fig . [ fig : sharpness_scaled1 ] should exceed that in fig . [ fig : mc_transition_point ] by a factor of @xmath214 . the actual slopes differ by a factor of 2.3 . [ fig : barrier_scaled1 ] displays the energy barriers at the transition as a function of @xmath215 . the scf data are quite close to collapsing onto a single straight line , as predicted by eq.([eq : barrier scaling ] ) while the mc data are not . although the direct proportionality to the chain length increment @xmath135 is still satisfied reasonably well , the effect of increasing the grafting density is stronger than predicted by the theory . this is probably due to the fact that the values of the barrier height @xmath200 obtained from mc simulations lie in the range between 0.5 and 5 @xmath216 , whereas the theory assumes that the two states are well separated , implying @xmath217 . thus the asymptotic behavior assumed by the theory is not yet reached , which may explain the poor data collapse . it is worth noting that the data do seem to obey an apparent scaling @xmath218 as illustrated by fig.[fig : barrier_scaled2 ] . we are not aware of a theoretical reason for such a scaling . the observed apparent scaling might be the purely fortuitous result of various corrections to the analytical theory , which relies on several somewhat questionable assumptions as discussed above . to summarize , we have investigated in detail conformational properties and phase transition behavior of a type of stimuli - responsive polymer materials using particle - based monte carlo simulations and continuum - based 3-dimensional self - consistent field theory . this type of stimuli - responsive materials is designed based on a polymer brush with a small fraction of adsorption - active minority chains introduced in the brush , and the minority chain can work as a responsive sensor . the basic mechanism for this responsive sensor relies on the conformational switch of the minority chain . the adsorption between the substrate and the minority chain serves as the trigger that switches the chain conformations . in practice , the trigger could also be a temperature or solvent composition change . upon varying the adsorption strength , two states were observed , one being an exposed state , in which the free end of the minority chain is located at the outside surface of the brush and possesses a stem - crown - like configuration , and the other being an adsorbed state , in which the minority chain is located at the adsorbing substrate in a nearly 2-dimensional confinement . , the barrier height is significantly smaller in the mc simulations . ] one important purpose of the present study was to highlight the effect of fluctuations by comparing two numerical methods applied to one and the same model , namely mc simulations and 3d scf calculations . the comparison shows that the scf theory roughly captures the sharpness of the transition ( fig . [ fig : sharpness_scaled1 ] , but overestimates the barrier height ( fig . [ fig : barrier_scaled1 ] ) . as discussed earlier , density fluctuations at the outer brush surface ( see fig . [ fig : density_scaled ] ) effectively reduce the range of the repulsive barrier created by the brush and hence the degree of chain stretching in the exposed state ( see fig.[fig : zend_mc_scf ] ) . as a consequence , the free energy barrier between the adsorbed and the exposed state is also reduced . the practical importance of this effect can be seen when plotting the barrier height against the sharpness of the transition ( fig . [ fig : barrier_vs_sharpness ] ) . the transition sharpness is a measure for the sensitivity of the switch to small changes in the environment . it can be tuned by adjusting the minority chain length or the grafting density . assuming that it has been tuned to a certain value , fig . [ fig : barrier_vs_sharpness ] demonstrates that the corresponding barrier height between coexisting states at the transition is greatly reduced in the mc simulations compared to the scf prediction . since the barrier height determines the time required for switching between states , this can drastically reduce the response times of the switch . for example , for @xmath219 , the reduction can be as high as @xmath220 , and according to a simple arrhenius - type estimate this would lead to a speedup of the switching time by three orders of magnitude . the switching time is one major issue for the quality of sensors or other responsive materials . in our previous work @xcite , we performed overdamped brownian dynamics simulations of a single switch chain in a static brush potential to estimate the switching time . by tracking the position of the free end bead of the minority chain , we found that this bead could jump between two well separated positions ( which correspond to the exposed and adsorbed state , respectively ) during a time about @xmath221 , where @xmath222 is the characteristic relaxation time of a single monomer , which is typically on the order of @xmath223 seconds for flexible chains . we thus estimated the switching time to be on the order of milliseconds . however , in that calculation , the brush was represented in a simplified manner by a parabolic profile and the coupling between the minority chain and the brush chains was neglected , which presumably leads to a large deviation from the true switch time . molecular dynamic simulations that model a brush - minority chain system in a dynamic manner are clearly desirable . in the present study , we restrict ourselves to systems with good solvent conditions , and assume that two - body interactions dominate , thus we only consider excluded volume interactions . however , for dense brushes , higher order contributions will become important . higher order terms in the free energy virial expansion provide corrections that lead to some change in the brush height and in the overall free energy of the competing states . hence one would expect corrections to the properties of the switching transition , in particular the barrier height at transition . in the case of very dense grafting these corrections may be significant but this falls outside the scope of the present paper . besides fluctuations , another factor that must be taken into account in real systems is polydispersity . in the present study , the brush polymers were taken to be monodisperse . real polymer brushes , however , are always polydisperse , and it has been demonstrated that polydispersity significantly alters both the equilibrium properties @xcite and dynamical properties @xcite of materials . using a one - dimensional scf theory , we have performed a preliminary study of a system with an adsorption - active chain embedded in a polydisperse brush with a continuous schulz - zimm length distribution . we found that for @xmath224 , the switch sharpness is similar to that of the monodisperse brush obtained by 1-dimensional scf theory ; however , the switch barrier is reduced . this suggests that the switching mechanism should be at least as effective in polydisperse brushes than in monodisperse brushes , the switching might even be faster . further and more detailed studies are currently under way . an important question pertaining to the mc scheme used in our work is the choice of the size of the density averaging cell , @xmath27 , which has the meaning of an interaction range . in section [ sec : brush ] we have shown that the effective excluded volume parameter appears as the result of a renormalization of the `` bare '' parameter due to density fluctuations . for a given set of parameters @xmath58 and @xmath225 , where @xmath10 is the statistical segment length unit , the brush profile is uniquely defined by @xmath53 or , more precisely , the corresponding second virial coefficient in the mc model , @xmath226 , while the renormalization factor @xmath71 was shown to be a function of a single dimensionless variable @xmath70 . the renormalization of monomer interaction parameters has been studied for single flexible swollen chains @xcite , in the context of dense polymer systems @xcite , and in the field - theoretical context @xcite . we would like to emphasize that polymer brushes simulated by mc in the laradji version @xcite represent a class of model systems where renormalization effects are quite strong and can be extracted easily . the basic tenet is that large - scale properties should depend only on the renormalized excluded volume parameter while its bare value may become relevant on small length scales comparable to @xmath10 or @xmath27 . one consequence in our system is that the brush profiles very close to the outer surface can not be described by a single renormalized @xmath53 ( see fig . [ fig : density_scaled ] ) , and that the surface interaction parameter @xmath14 has to be renormalized independently from @xmath29 ( as has been done in fig . [ fig : zend_mc_scf ] ) . however , even after taking into account the renormalization of interaction parameters , we find that the properties of chains in close vicinity and above the brush surface are still strongly influenced by density fluctuations . the density fluctuations at the brush surface significantly reduce the free energy barrier between the exposed and the adsorbed state in fig.[fig : barrier_vs_sharpness ] . thus , the minority chain serves as a delicate probe of both local and large - scale fluctuation effects . * acknowledgments * financial support by the deutsche forschungsgemeinschaft ( grants no . schm 985/13 - 1 , and no . 13 - 03 - 91331-nnio - a ) is gratefully acknowledged . s. q acknowledges support from the german science foundation ( dfg ) within project c1 in sfb trr 146 . simulations have been carried out on the computer cluster mogon at jgu mainz . the scf method used in the present work is formulated in the continuous 3d space , starting from the same hamiltonian with that in the mc method mentioned in the text . in the scf approach @xcite , the partition function @xmath227 is rewritten as a functional integral , i.e. , @xmath228 , with the free energy functional expressed as @xmath229-\ln q_m[\omega+u_\mathrm{ads}],\end{aligned}\ ] ] where @xmath230 is the total density distribution , @xmath231 is the corresponding conjugate auxiliary potential , and @xmath232 , @xmath233 are the single chain partition functions . in the calculation we set the excluded volume parameter @xmath234 ( where @xmath53 is obtained from the mc simulation ) . extremizing the free energy functional with respect to the fluctuating fields , i.e. , the density @xmath230 and the potential @xmath231 , one obtains the scf equations @xmath235 where @xmath236 and @xmath237 ( @xmath238 ) satisfy the modified diffusion equation @xmath239 with @xmath240 for @xmath241 , @xmath242 for @xmath243 , dirichlet boundary conditions along the @xmath79 direction , and periodic boundary conditions along the @xmath5 and @xmath6 directions . the functions @xmath237 and @xmath236 ( @xmath238 ) correspond to spatial integrals of the single - chain propagator between monomer @xmath244 and the free or grafted end , respectively , over all possible end positions . thus the initial condition for @xmath237 ( free end ) is @xmath245 , and the initial condition for @xmath236 ( grafted end ) is chosen self - consistently such that the density of graft points is reproduced correctly , i.e. , @xmath246 for the brush chains , and @xmath247 for the minority chain , where @xmath248 is positioned at the center of the substrate . the transition properties are extracted from the corresponding minority chain propagator . for example , one can calculate the distribution of free end via @xmath249 . from these distributions the transition point and transition barrier can be obtained in a similar way as that used in the mc simulations . the scf equations are closed and iterative methods are used to find their solutions . the auxiliary potential @xmath231 is used as the iterative variable , and a new potential for the next iteration is generated @xmath250 , where @xmath251 is a small number controlling the stability of the iteration and the speed of converging . in most of the cases we use @xmath252 . the iteration process stops if the iteration error is smaller than @xmath253 . there are several ways to solve the modified diffusion equation numerically , for example , real space finite difference schemes @xcite , pseudo - spectral schemes @xcite , and so forth . usually , pseudo - spectral schemes are reliable and efficient . however , in our system , we encountered numerical problems when chains are strongly stretched ( e.g. at large grafting densities , or when chains are long ) . in such cases , the propagators sometimes assumed negative values , and the scf equations did not converge . similar problems were reported in refs . therefore , we adopt the doublas - brian scheme @xcite , a real - space finite - difference method which did not suffer from this problem . the volume of the simulation system is chosen equal to that of the mc method . it is divided uniformly into @xmath254 grid cells , and continuous quantities are discretized on the vortices of these cells . the chain contour including the minority chain is uniformly discretized into 400 steps . about 300 hundred iteration steps are necessary for the iteration to reach convergence . 99 halperin , a. ; tirrell , m. ; lodge t. p. _ adv . sci _ * 1992 * , 100 , 31 - 71 . milner , s. t. _ science _ * 1991 * , 251 , 905 - 914 . russell , t. p. _ science _ * 2002 * , 297 , 964 - 967 . cohen stuart m. ; huck w. ; genzer j. ; mller m. ; ober c. ; stamm m. ; sukhorukov g. b. ; szleifer i. ; tsukruk v. v. ; urban m. ; winnik f. ; zauscher s. ; luzinov i. ; minko s. ; _ nature materials _ * 2010 * , 9 , 101 - 113 . chen t. ; ferris r. ; zhang j. ; ducker r. ; zauscher s. _ prog . sci . _ * 2010 * , 35 , 94 - 112 . jaquet , b. ; wei , d. ; reck , b. ; reinhold , f. ; zhang , x. ; wu , h. ; morbidelli , m. _ colloid polym . sci . _ * 2013 * , 291 , 1659 - 1667 . klein j. _ annu . _ * 1996 * , 26 , 581 - 612 lger l. ; raphal e. ; hervet h. _ adv . sci . _ * 1999 * , 138 , 185 - 225 . chen y .- w . ; chang y. ; lee r .- h . ; li w .- t . ; chinnathambi a. ; alharbi s. a. ; hsiue g .- h . _ langmuir _ * 2014 * , 30 , 9139 - 9146 . papaphilippou , p. ; loizou , l. ; popa , n. c. ; han , a. ; vekas , l. ; odysseos , a. ; krasia - christoforou , t. _ biomacromolecules _ * 2009 * , 10 , 2662 - 2671 . mura s. ; nicolas j. ; couvreur p. _ nature materials _ * 2013 * , 12 , 991 - 1003 . draper j. ; zuzinov i. ; minko s. ; tokarev i. ; stamm m. ; _ langmuir _ * 2004 * , 20 , 4064 - 4075 . motornov m. ; minko s. ; eichhorn k .- j . ; nitschke m. ; simon f. ; stamm m. _ langmuir _ * 2003 * , 19 , 8077 - 8085 . ma y. ; dong w .- f . ; hempenius m. a. ; mhwald h. ; vancso j. _ nature material _ * 2006 * , 5 , 724 - 729 de beer s. _ langmuir _ * 2014 * , 30 , 8085 - 8090 . lynch j. g. ; jaycox g. d. _ polymer _ * 2014 * , 55 , 3564 - 3572 skvortsov a. m. ; klushin l. i. ; gorbunov a. a. _ macromolecules _ * 1997 * , 30 , 1818 - 1827 merlitz h. ; he g .- wu c .- x . ; sommer j .- u . _ macromolecules _ * 2008 * , 41 , 5070 - 5072 romeis d. ; merlitz h. ; sommer j .- u . _ j. chem . phys . _ * 2012 * , 136 , 044903 . merlitz h. ; he g .- wu c .- x . ; sommer j .- u . lett _ * 2009 * , 102 , 115702 romeis d. ; sommer j .- u . _ j. chem . phys . _ * 2013 * , 139 , 044910 . dimitrov d. i. ; milchev a. ; binder k. _ j. chem . phys . _ * 2007 * , 127 , 084905 . pal s. ; seidel c. _ macromol . theory simul . _ * 2006 * , 15 , 668 - 673 . lai p .- y . ; binder k. _ j. chem . phys . _ * 1991 * , 95 , 9288 - 9299 . laradji m. ; guo h. ; zuckermann m. j. _ phys . e _ * 1994 * , 49 , 3199 - 3206 wang j. ; mller m. _ j. phys . b _ * 2009 * , 113 , 11384 - 11402 . van lehn r. c. ; alexander - katz a. _ j. chem . phys . _ * 2011 * , 135 , 141106 sommer j .- u . ; klos j. s. ; mironova o. n. _ j. chem . phys . _ * 2013 * , 139 , 244903 scheutjens , j. m. h. ; fleer g. j. _ j. phys . chem _ * 1979 * 83 , 1619 - 1635 matsen m. w. ; gardiner j. m. ; _ j. chem . phys . _ * 2001 * , 115 , 2794 - 2804 . mller m. _ phys . * 2002 * 65 , 030802 . meng d. ; wang q. _ j. chem . phys . _ * 2009 * , 130 , 134904 . chantawansri t. l. ; hur s .- m . ; garca - cervera c. j. ; ceniceros h. d. ; fredrickson g. h. _ j. chem . phys . _ * 2011 * , 134 , 244905 suo t. ; yan d. _ j. chem . phys . _ * 2011 * , 134 , 054901 . helfand e. _ j. chem . phys . _ * 1975 * , 62 , 999 - 1005 besold g. ; guo h. ; zuckermann m. j. _ j. polym . part b : polym . _ * 2000 * , 38 1053 - 1068 detcheverry f. a. ; kang h. ; ch . daoulas k. ; mller m. ; nealey p. f. ; de pablo j. j. _ macromolecules _ * 2008 * , 41 , 4989 - 5001 qi s. ; behringer h. ; schmid f. _ new j. phys . _ * 2013 * , 15 , 125009 . frenkel d and smit b * 2002 * , _ understanding molecular simulation : from algorithms to applications _ ( new york : academic press ) rubin r. j. _ j. chem . phys . _ * 1965 * , 43 , 2392 - 2407 . alexander s. _ j. phys . france _ * 1977 * , 38 , 977 - 981 . de gennes p. g. _ macromolecules _ * 1980 * , 13 , 1069 - 1075 . zhulina e. ; borisov o. ; priamitsyn v. , _ j. coll . * 137 * , 495 ( 1990 ) . milner s. t. ; witten t. a. ; cates m. e. _ macromolecules _ * 1988 * , 21 , 2610 - 2619 . de gennes p. g. _ rep . * 1969 * , 32 , 187 - 205 skvortsov a. m. ; gorbunov a. a. ; leermakers f. a. m. ; fleer g. j. _ macromolecules _ * 1999 * , 32 , 2004 - 2015 . grassberger p. _ j. phys . a : math . gen . _ * 2005 * , 38 , 323 - 331 . klushin l. i. ; polotsky a. a. ; hsu h .- p . ; markelov d. a. ; binder k. ; skvortsov a. m. _ phys . rev . e _ * 2013 * , 87 , 022604 . challa m. ; landau d. ; binder k. _ phase transit . _ * 1990 * , 24 - 26 , 343 - 369 . klushin l. i. ; skvortsov a. m. _ j. phys . a : math . * 2011 * , 44 , 473001 fredrickson g. h. ; sides s. w. _ macromolecules _ * 2003 * , 36 , 5415 - 5423 de vos w. m. ; leermakers f. a. m. _ polymer _ * 2009 * , 50 , 305 - 316 . qi s. ; yan d. _ j. chem . phys . _ * 2008 * , 129 , 204902 qi s. ; zhang x. ; yan d. _ j. chem . phys . _ * 2010 * , 132 , 064903 pandav g. ; ganesan v. _ phys . _ * 2013 * , 139 , 214905 grosberg a. yu . ; erukhimovich i. ya . ; khokhlov a. r. _ phys . lett _ a * 1980 * , 78 , 163 . wang z .- g . ; _ j. chem . phys . _ * 2002 * , 117 , 481 . kudlay a. ; stepanow s. _ j. chem . phys . _ * 2003 * , 118 , 4272 . alexander - katz a. ; moreira a. g. ; sides s. w. ; fredrickson g. h. _ j. chem . _ * 2005 * , 122 , 014904 . grzywacz p. ; qin j. ; morse d. c. _ phys . e _ * 2007 * , 76 , 061802 . schmid f. _ j. phys . : condens . matter _ * 1998 * , 10 , 8105 - 8136 fredrickson g. h. ; ganesan v. ; drolet f. _ macromolecules _ * 2002 * , 35 , 16 - 39 fredrickson g. h. * 2006 * _ the equilibrium theory of inhomogeneous polymers _ ( oxford : oxford university press ) drolet f. ; fredrickson g. h. _ phys . lett . _ * 1999 * , 83 , 4317 - 4320 tzeremes g. ; rasmussen k. . ; lookman t. ; saxena a. _ phys . rev . e _ * 2002 * 65 , 041806 mller m. ; schmid f. _ adv . sci . _ * 2005 * , 185 , 1 - 58 . lapidus l. ; and pinder g. f. , * 1997 * _ numerical solution of partial differential equations in science and engineering _ , ( john wiley and sons , inc . , new york )

we have performed light - scattering measurements in dilute and semidilute polymer solutions of polystyrene in toluene when subjected to stationary temperature gradients . five solutions with concentrations below and one solution with a concentration above the overlap concentration were investigated . the experiments confirm the presence of long - range nonequilibrium concentration fluctuations which are proportional to , where is the applied temperature gradient and is the wave number of the fluctuations . in addition , we demonstrate that the strength of the nonequilibrium concentration fluctuations , observed in the dilute and semidilute solution regime , agrees with theoretical values calculated from fluctuating hydrodynamics . further theoretical and experimental work will be needed to understand nonequilibrium fluctuations in polymer solutions at higher concentrations . 2

cond-mat0002379

density and concentration fluctuations in fluids and fluid mixtures can be investigated experimentally by light - scattering techniques . the nature of these fluctuations when the system is in equilibrium is a subject well understood@xcite . here we shall consider fluctuations in nonequilibrium steady states ( ness ) , when an external and constant temperature gradient is applied , while the system remains in a hydrodynamically quiescent state . that is , we shall deal with fluctuations that are intrinsically present in thermal nonequilibrium states in the absence of any convective instabilities . such fluctuations have received considerable attention during the past decade@xcite . it was originally believed that , because of the existence of local equilibrium in ness , the time correlation function of the scattered - light intensity would be the same as in equilibrium , but in terms of spatially varying thermodynamic and transport properties corresponding to the local value of temperature . however , it has been demonstrated that qualitative differences do appear . the first complete expression for the spectrum of the nonequilibrium fluctuations in a one - component fluid subjected to a stationary gradient was obtained by kirkpatrick et al.@xcite by using mode - coupling theory . they showed that the central rayleigh line of the spectrum would be substantially modified as a result of the presence of a temperature gradient . because of a coupling between the temperature fluctuations and the transverse - velocity fluctuations through the temperature gradient , spatially long - range nonequilibrium temperature and viscous fluctuations appear , modifying the rayleigh spectrum of the scattered - light intensity . their results were subsequently confirmed on the basis of fluctuating hydrodynamics@xcite . the effect is largest for the transverse - velocity fluctuations in the direction of the temperature gradient which corresponds to the situation that the scattering wave vector , @xmath3 , is perpendicular to the temperature gradient @xmath1 , which configuration will be assumed throughout the present paper . in that case , the strengths of the nonequilibrium temperature and viscous fluctuations are predicted to be proportional to @xmath0 . the dependence on @xmath4 implies that , in real space , the nonequilibrium correlation functions become long ranged@xcite . the spatially long - range nature of the correlation functions in ness is nowadays understood as a general phenomenon arising from the violation of the principle of detailed balance@xcite . experimentally , the long - range nature of the nonequilibrium fluctuations can be probed by light - scattering measurements at small wave numbers @xmath2 , i.e. at small scattering angles @xmath5 . such experiments have been performed in one - component liquids and excellent agreement between theory and experiments has been obtained@xcite . in binary systems , the situation is a little more complicated . in liquid mixtures or in polymer solutions a temperature gradient will induce a concentration gradient through the soret effect . this induced concentration gradient is parallel to the temperature gradient and has the same or opposite direction depending on the sign of the soret coefficient , @xmath6 . in this case , nonequilibrium fluctuations appear , not only because of a coupling between the temperature fluctuations and the transverse - velocity fluctuations through the temperature gradient , but also because of a coupling between the concentration fluctuations and the transverse - velocity fluctuations through the induced concentration gradient . the nonequilibrium rayleigh - scattering spectrum has been calculated for binary liquid mixtures , both on the basis of mode - coupling theory@xcite and on the basis of fluctuating hydrodynamics@xcite , with identical results . in liquid mixtures in thermal nonequilibrium states not only nonequilibrium temperature and nonequilibrium viscous fluctuations , but also nonequilibrium concentration fluctuations exist . the strengths of all three types of nonequilibrium fluctuations are again predicted to be proportional to @xmath0 . the theory was extended by segr et al.@xcite to include the effects of gravity and by vailati and giglio@xcite to include time - dependent nonequilibrium states . the nature of the nonequilibrium fluctuations in the vicinity of a convective instability has also been investigated@xcite , in which case the nonequilibrium modes become propagative . the influence of boundary conditions , which may become important when @xmath3 is parallel to @xmath1 , was considered by pagonabarraga et al.@xcite . experiments have been performed to study nonequilibrium fluctuations in liquid mixtures@xcite . the three types of nonequilibrium fluctuations have been observed in liquid mixtures of toluene and n - hexane@xcite and the strength of all three types of nonequilibrium fluctuations were indeed proportional to @xmath0 , as expected theoretically . initially , it seemed that also the prefactors of the amplitudes of these nonequilibrium fluctuations were in agreement with the theoretical predictions@xcite . however , a definitive assessment was hampered by a lack of reliable experimental information on the soret coefficient@xcite . to our surprise , subsequent accurate measurements of the soret coefficient of liquid mixtures of toluene and n - hexane obtained both by khler and mller@xcite and by zhang et al.@xcite yielded values for the soret coefficient that were about 25% lower than the values needed to explain the quantitative magnitude of the amplitudes of the observed nonequilibrium fluctuations@xcite . subsequent measurements of the nonequilibrium concentration fluctuations in a mixture of aniline and cyclohexane , obtained by vailati and giglio@xcite did not have sufficient accuracy to resolve this issue . as an alternative approach , we decided to investigate nonequilibrium concentration fluctuations in a polymer solution . in a polymer solution the same nonequilibrium enhancement effects are expected to exist , but the mass - diffusion coefficient , @xmath7 , in this case is several orders of magnitude lower than in ordinary liquid mixtures . in addition , the soret coefficient is two orders of magnitude larger that the soret coefficient of ordinary liquid mixtures@xcite . these two facts , as discussed below , simplify the theory because , as also happens in an equilibrium polymer solution , the concentration fluctuations become dominant and they are readily observed by light scattering . both the data acquisition and the data analysis become much easier in this case . thus a polymer solution would seem to be an ideal system to further investigate nonequilibrium concentration fluctuations . for this purpose we have selected solutions of polystyrene in toluene , for which reliable information on the thermophysical properties is available . we have performed small - angle rayleigh - scattering experiments in polystyrene - toluene solutions subjected to various externally applied temperature gradients . a summary of our results has been presented in a physical review letter@xcite . in the present paper we provide a full account of the experiment and of the analysis of the experimental data . a theory specifically developed for the fluctuations in polymer solutions in thermal nonequilibrium states is not yet available in the literature . however , for polymer solutions in the dilute and semidilute solution regime , as long as entanglement effects can be neglected , it should be possible to use the same fluctuating hydrodynamics equations as those for ordinary liquid mixtures . as mentioned in the introduction , the complete expression for the rayleigh spectrum of a binary liquid in the presence of a stationary temperature gradient was evaluated by law and nieuwoudt@xcite . the dynamic structure factor contains three diffusive modes . the decay rate @xmath8 of one of these modes is determined by the kinematic viscosity @xmath9 ; this mode disappears in thermal equilibrium , i.e. when @xmath10 . the other two modes are also present in the rayleigh spectrum of a liquid mixture in equilibrium and have the decay rates @xmath11 given by : @xmath12^{1/2 } , \label{eq : lamda}\ ] ] where @xmath13 is the thermal diffusivity , @xmath7 the binary mass - diffusion coefficient , in the case of a polymer solution to be referred to as collective diffusion coefficient@xcite , and where @xmath14 with @xmath15 ^ 2 \over c_{p , c } } ~ \left ( { \partial \mu \over \partial w } \right)_{p , t}. \label{eq : epsilon}\ ] ] in eq . ( [ eq : epsilon ] ) @xmath16 is the concentration expressed as mass fraction of the polymer , @xmath17 the isobaric specific heat capacity of the solution at constant concentration , and @xmath18 is the difference between the chemical potentials per unit mass of the solvent and the solute . as usual , the soret coefficient , @xmath6 , specifies the ratio between temperature and concentration gradients and it is defined through the steady - state phenomenological equation : @xmath19 the two modes with decay rates @xmath11 incorporate a coupling between temperature and concentration fluctuations . this coupling becomes important in compressible fluid mixtures near a critical locus@xcite . however , when @xmath20 , the dynamic structure factor of a binary system in ness consists of just three exponentials with decay rates given by : @xmath21 physically , it means that the temperature gradient @xmath1 induces nonequilibrium viscous fluctuations appearing as a new term in the rayleigh spectrum and it leads to nonequilibrium enhancements of the temperature and concentration fluctuations . the condition @xmath20 is always fulfilled in the low - concentration limit@xcite . but for some liquid mixtures @xmath22 is also small at all concentrations . for instance , in an equimolar mixture of toluene and n - hexane segr et al . found @xmath23 . for the dilute polymer solutions considered in the present paper , @xmath24 . hence , the approximations implying the presence of three diffusive modes with the simple decay rates given by eq . ( [ eq : decay ] ) are even more justified for a polymer solution than for the liquid mixtures studied in previous papers@xcite . furthermore , in ordinary polymer solutions @xmath25 ; such a small value simplifies the analysis of the concentration fluctuations , because the decay in the time correlation function coming from the concentration mode ( decay rate @xmath26 ) will be well separated from the decays of the modes with decays rates @xmath27 and @xmath28 . actually , for the small angles employed in our experiments , the decay time of the @xmath26 mode is typically 0.5 - 1.5 s , whereas the decay times of the other two modes are around @xmath29 s. in addition , for a dilute polymer solution , the rayleigh factor ratio , that determines the ratio of the scattering intensities of the concentration fluctuations and the temperature fluctuations@xcite , is much larger than unity so that the contributions of temperature fluctuations are negligibly small in practice . moreover , when the experimental correlograms are analyzed beginning at @xmath30 s , viscous and temperature fluctuations have already decayed . this kind of approximation is usually assumed in the theory of light scattering from polymer solutions in equilibrium states@xcite . in conclusion , for heterodyne light - scattering experiments in sufficiently dilute polymer solutions , the expression for the time - dependent correlation function of the scattered light becomes : @xmath31 e^{-d k^{2 } t } , \label{eq : corr}\ ] ] where @xmath32 is the signal to local - oscillator background ratio representing the amplitude of the correlation function in thermal equilibrium . in eq . ( [ eq : corr ] ) the term @xmath33 represents the nonequilibrium enhancement of the concentration fluctuations . this term is anisotropic and depends on the scattering angle . when @xmath34 , @xmath35 reaches a maximum given by : @xmath36 where the strength of the enhancement , @xmath37 , is given by@xcite : @xmath38^{2 } { s_t}^2 \over \nu d } ~ \left ( { \partial \mu \over \partial w } \right)_{p , t } ~ \left [ 1 + 2 { d \over d_t } ( 1 + \zeta ) \right ] , \label{eq : enhan11}\ ] ] where @xmath39 is a dimensionless correction term related to the ratio of @xmath40 and @xmath41 . as mentioned earlier for the polymer solutions to be considered , @xmath42 is of the order of @xmath43 , so that the correction term inside the square brackets can be neglected in practice . hence , the expression for @xmath37 reduces to : @xmath38^{2 } { s_t}^2 \over \nu d } ~ \left ( { \partial \mu \over \partial w } \right)_{p , t}. \label{eq : enhan2}\ ] ] the dependence of the nonequilibrium enhancement @xmath35 of the concentration fluctuations on @xmath4 indicates that the correlations in real space are long ranged . actually , a @xmath4 dependence in fourier space corresponds to a linear increase of the correlation function in real space @xcite . the rapid increase of the strength of the nonequilibrium fluctuations with decreasing values of the wavenumber will saturate for sufficiently small @xmath2 due to the presence of gravity . the gravity effect was predicted by segr et al.@xcite and has been confirmed by some beautiful experiments of vailati and giglio@xcite . our set of working equations , eqs . ( [ eq : corr ] ) , ( [ eq : enhan1 ] ) and ( [ eq : enhan2 ] ) , can be also deduced from the theory of vailati and giglio for nonequilibrium fluctuations in time - dependent diffusion processes@xcite . in that paper only concentration and velocity fluctuations are considered around a nonequilibrium time - dependent state . by applying an inverse fourier transform to eq . ( 25 ) of vailati and gilio@xcite , we obtain for the autocorrelation function of the concentration fluctuations for @xmath44 : @xmath45 , \label{eq : giglio1}\ ] ] where @xmath46 is the rayleigh - number ratio , which accounts for the effects of gravity and is defined in eq . ( 22 ) of vailati and giglio@xcite . @xmath47 is the static structure factor , defined in eq . ( 26 ) of the same paper . it should be noticed that , for consistency , in our eq . ( [ eq : giglio1 ] ) we have used the symbol @xmath16 instead of the @xmath48 used by vailati and giglio . both symbols have the same meaning , as it is stated after eq . ( 2 ) of vailati and giglio that : @xmath48 is the weight fraction of the denser component " . ( 26 ) of vailati and giglio for the static structure factor @xmath47 can be written as : @xmath49 . \label{eq : giglio2}\ ] ] 2 the derivation of vailati and giglio@xcite was originally developed for time - dependent isothermal diffusion processes in the presence of gravity , where only concentration gradients are present . the validity of this theory for stationary nonequilibrium states is obvious , and we can consider @xmath50 to be constant , neglecting the weak dependence on space and time considered by vailati and giglio@xcite . neglecting gravity effects is equivalent to assuming that the rayleigh - number ratio is almost zero , @xmath51 , as can be easily shown from eq . ( 22 ) in the paper of vailati and giglio@xcite . introducing eq . ( [ eq : soret ] ) into eq . ( [ eq : giglio2 ] ) and neglecting @xmath46 , one readily verifies that eqs . ( [ eq : giglio1 ] ) and ( [ eq : giglio2 ] ) are equivalent to eqs . ( [ eq : enhan1 ] ) and ( [ eq : enhan2 ] ) . we note that eqs . ( [ eq : corr ] ) , ( [ eq : enhan1 ] ) and ( [ eq : enhan2 ] ) can also be obtained as a limiting case of the equations derived by schmitz for nonequilibrium concentration fluctuations in a colloidal suspension @xcite . schmitz considered a colloidal suspension , in the presence of a constant gradient , @xmath52 , in the volume fraction , @xmath53 , of the colloidal particles , maintained against diffusion by continuous pumping of solvent between two semipermeable and parallel walls . the relevant expression is eq . ( 7.7 ) in the article of schmitz@xcite , which gives the dynamic structure factor , @xmath54 , of the nonequilibrium concentration fluctuations as a function of the wavenumber @xmath2 and the frequency @xmath55 . the expression obtained by schmitz includes non - local and memory effects due to the large particle sizes in comparison with the scattering wavelength . the non - local and memory effects cause the transport coefficients to depend on the frequency and the wave number . in our case these effects can be neglected , because the polymer molecules are much smaller that the colloidal particles considered by schmitz and the condition @xmath56 , where @xmath57 is the radius of gyration , is fulfilled . with this simplification the original expression obtained by schmitz reduces to : @xmath58 ~ { 2 d k^2 \over \omega^2 + ( d k^2)^2},\ ] ] where @xmath35 is given by : @xmath59 in deriving eq . ( [ eq : schmitz2 ] ) we have used the relationship between pumping rate and volume fraction gradient , as given by eqs . ( 7.6 ) and ( 7.11 ) in the paper of schmitz@xcite . we have also converted the various concentration units used by schmitz : @xmath60 which is the number of particles per unit volume ( @xmath61 ) and @xmath48 which is the number of particles per unit mass ( @xmath62 ) , @xmath63 being the density of the suspension and @xmath64 the mass of one particle . furthermore , we are using the difference between solvent and solute chemical potentials per unit mass , whereas in the paper of schmitz @xmath18 is the difference in chemical potentials per particle . although the theory of schmitz for the nonequilibrium concentration fluctuations was derived for the case that the concentration gradient is produced by a solvent flow , the results will be applicable to ness under a constant temperature gradient . since the solvent flow does not appear explicitly in eq . ( [ eq : schmitz2 ] ) , we may also apply the equation to a system subjected to a stationary volume - fraction gradient caused by other driving forces , such as a temperature gradient . for our polymer solutions @xmath65 , the difference being less than 0.5% . with this approximation and by using eq . ( [ eq : soret ] ) , it is straightforward to show the equivalence of the theory for nonequilibrium concentration fluctuations in a colloidal suspension and in a polymer solution , independent of whether the concentration gradient is established by solvent pumping or induced by a temperature gradient . the polystyrene used to prepare the polymer solutions was purchased from the tosoh corporation ( japan ) ; it has a mass - averaged molecular weight @xmath66 and a polydispersivity of 1.01 , as specified by the manufacturer@xcite . the solvent toluene was fisher certified reagent purchased from baker chemical co. with a stated purity of better than 99.8% . by using the correlation proposed by noda et al.@xcite , @xmath67 , the radius of gyration , @xmath57 , and the overlap concentration , @xmath68 , for this polymer in toluene solutions may be estimated as @xmath69 nm and @xmath70 , respectively . therefore , for the viscoelastic and entanglement effects to be negligible , the concentrations employed in this work should not be much higher than @xmath71 . five polystyrene / toluene solutions below @xmath68 and one slightly above @xmath68 were prepared gravimetrically as described by zhang et al.@xcite . to assure the homogeneity of the mixture , the solutions were agitated ( magnetic stirrer or shaking by hand)for at least one hour before further use . the light - scattering experiments were performed with an apparatus specifically designed for small - angle rayleigh scattering in an horizontal fluid layer subjected to a vertical temperature gradient@xcite . a diagram of the optical cell is shown in fig . [ f1 ] . the following paragraphs will describe the apparatus in detail , and references to the symbols in fig.[f1 ] will be used . the polymer solution in the actual light - scattering cell ( e ) is confined by a circular quartz ring ( d ) that is sandwiched between top and bottom copper plates ( a and f , respectively ) . the cell is filled through two stainless steel capillary tubes ( g ) which had been soldered into holes in the top and bottom plates . the inner and the outer diameters of the quartz ring ( d ) are 1.52 cm and 2.05 cm , respectively . two identical optical windows ( b ) are used for letting a laser beam pass through the liquid solution . both windows are cylindrical and made from sapphire because of the relatively high thermal conductivity of this material . the windows are epoxied into the centers of the top and the bottom copper plates by a procedure similar to the one described by law et al.@xcite . once the optical windows are installed , the top and bottom copper plates are sealed against the quartz ring by indium o - rings . to avoid heat conduction from the upper to the lower plate other than through the liquid , the two copper plates are held together by teflon screws ( not shown in the figure ) . the tension of these teflon screws was adjusted to set the flat surfaces of the optical windows parallel to within 10 seconds of arch . this was accomplished by passing a laser beam through the cell and monitoring the interference pattern of the reflected beams from the inner surfaces of the windows while the screws were fastened . once the cell was mounted , the distance @xmath72 between the windows was accurately determined by measuring the angular variation of these interference fringes . the result was @xmath73 cm . this value was confirmed by also measuring the separation with a cathetometer . since the nonequilibrium enhancement of the fluctuations given by eq . ( [ eq : enhan1 ] ) depends on @xmath4 , small scattering angles are required to obtain measurable values of @xmath35 ; in practice scattering angles @xmath5 between 0.4@xmath74 and 0.9@xmath74 were used . a major difficulty with such small - angle experiments is the presence of strong static scattering from the optical surfaces . to reduce the background scattering , we needed very clean windows . the optical surfaces of the windows and the other optical components of the experimental arrangement were cleaned by applying first a windex solution and then acetone with cotton swabs . moreover , we employ thick cell windows to remove the air - window surfaces from the field of view of the detector . in addition , the outer surface of the windows and the other optical surfaces were broad - band ( 488 - 623 nm ) anti - reflection coated ( cvi laser corporation ) , to reduce the back reflections and forward scattering intensity . as a result of these efforts , we obtained high enough signal to background ratios to allow accurate measurements of the scattered - light intensities . to observe the intrinsic nonequilibrium fluctuations , it is essential to avoid any convection in the liquid layer . this is accomplished by heating the horizontal fluid layer from above . furthermore , bending and defocusing of the light beam due to the refractive - index gradient induced by the temperature gradient could cause serious limitations in the resolution of the angles . this difficulty is avoided by employing a vertical incident light beam , parallel to the temperature gradient . to eliminate stray deflections caused by air currents near the cell , the whole assembly is covered by a plexiglass box with a small hole at the top to let the laser beam pass through . the temperature at the top plate was maintained by a computer controlled resistive heater winding ( c ) . the temperature at the bottom plate is controlled by circulating constant - temperature fluid from a forma model 2096 bath through channels in the base of the cell ( i ) , which is in good thermal contact with the bottom plate . with these devices , the temperatures of both the hot and cold plates can be held constant to within @xmath7520 mk . these temperatures remain fixed over the collection time of any experimental run . temperatures of both plates are monitored by thermistors inserted in holes ( h ) next to the sample windows and deep enough to be located very close to the liquid layer . due to the relatively high thermal conductivity of the sapphire windows and the symmetry of the arrangement , the lateral temperature gradients are small . numerical modeling of the thermal conduction process yields negligible differences between the measured temperature gradient and the actual temperature gradient between the windows . a coherent beam ( @xmath76 nm ) from a 6 mw stabilized cw he - ne laser is focused with a 20 cm focal length lens onto the polymer solution in the scattering cell . the scattering angle , @xmath5 is selected with a small pinhole ( 500 @xmath18 m ) located 197 mm after the cell . the collecting pinhole is placed in a plane orthogonal to the transmitted beam . the distance between the point where the beam hits the collection plane and the pinhole was carefully measured with a vernier micrometer scale . in the original light - scattering experiments of law et al.@xcite in ness of a one - component liquid , the scattering wave number , @xmath2 , was determined from equilibrium light - scattering measurements and the known value of the thermal diffusivity , @xmath13 . in the present experiments we have determined @xmath2 by directly measuring the scattering angle @xmath5 . since the intensities of the nonequilibrium fluctuations depend on @xmath4 , the measurement of @xmath2 has to be done with considerable accuracy . by working out the ray tracing problem , taking into account refraction at the two window surfaces , we can deduce the scattering angle , @xmath5 , from the location of the pinhole relative to that of the transmitted beam for this calculation the thickness of the bottom window , the distance from the window to the collecting plane , the refractive index of the window and the refractive index , @xmath60 , of the solution are needed . it was assumed that the scattering volume is in the middle of the cell , but since the cell is very thin the possible corrections to this assumption are negligible . the distances were measured with an accuracy of @xmath77 cm . the value 1.7660 of the refractive index of the window at the wavelength @xmath78 of the incident light was obtained from the manufacturer . the refractive index , @xmath60 , of the polystyrene solutions as a function of the polymer concentration at 25@xmath74c was measured with a thermostated abb refractometer as described by zhang et al.@xcite . for the polystyrene solutions considered in the present paper the refractive index can be represented by@xcite : @xmath79 the scattering wave number @xmath2 is related to the scattering angle @xmath5 through the bragg condition @xmath80 , where @xmath78 is the wavelength of the incident light . taken into account the experimental errors in the different magnitudes relevant to the calculation , we were able to determine @xmath2 with an accuracy of 1.5% . as discussed in section ii , we want to investigate the concentration fluctuations , which yield the dominant contribution to the rayleigh spectrum . since typical decay times for this mode at the small angles employed are always larger than 0.5 s , we are interested in the experimental correlograms for times starting around 10@xmath81 s. since this value is well above the region where photomultiplier ( pm ) afterpulsing effects are important , unlike in previous work@xcite , cross - correlation was not necessary for our measurements . this simplifies the experimental arrangement . the light exiting through the pinhole is focused , through corresponding optics , onto the field selecting pinhole in front of a single pm . the signal from the pm and the corresponding discriminator is analyzed with an alv-5000 multiple tau correlator . rayleigh - scattering measurements were obtained for six different solutions of polystyrene in toluene , with concentrations @xmath16 in weight fraction ranging from 0.50% to 4.00% . for each solution , the measurements were performed at three to five scattering angles , ranging from 0.4@xmath74 to 0.9@xmath74 , which corresponds to scattering wave numbers @xmath2 ranging from 900 @xmath82 to 2000 @xmath82 . these small angles cause the stray light from window - surface scattering to be dominant , assuring that the measurements are in the heterodyne regime . hence , the light scattered from the inner surfaces of the windows plays the role of a local oscillator and provides the background with which our signal , the light scattered from the polymer solution , is mixed . before starting the experiments , the light - scattering cell was carefully cleaned by flushing the cell with pure toluene for at least one hour . next , a gentle stream of nitrogen gas was continuously directed through the cell to dry the inner walls and to remove dust particles . after this cleaning procedure the polymer solution was introduced into the cell through a 0.5 @xmath18 m millipore millex hplc teflon filter . we have been careful to remove bubbles from the light - scattering cell while we were filling it with the polymer solution . this cleaning and filling procedure was repeated each time the concentration of the solution inside the cell was changed . it should be noted that a portion of the same polymer solution was introduced into an optical beam - bending cell for measuring the diffusion coefficient , @xmath7 , and the soret coefficient , @xmath6 , of the solutions as described by zhang et al.@xcite . these independent measurements of the diffusion and soret coefficients will be used to compare the results of our light - scattering measurements with the theoretical predictions . in the case of polymer solutions it is especially important to have independent measurements of these quantities for the same polymer / solvent system because , as commented below , there is a sizable dispersion in the literature data , mainly caused by the dependence of these coefficients on parameters difficult to control , such as the polydispersity of the sample . once the cell was filled with a polystyrene solution with known polymer concentration , we started the experiments by setting the temperature of the top and the bottom plates of the cell at 25@xmath74c . a light - scattering angle was then selected with the collection pinhole , and the distance in the collection plane between the pinhole and the center of the transmitted forward - beam spot was measured accurately . as already mentioned , this procedure , with the corresponding calculations , yielded scattering wave vectors @xmath2 with an accuracy of 1.5% . once a scattering angle was selected , the optics was arranged to focus the light exiting through the collecting pinhole into the field - stop pinhole at the pm . when the temperature of the polymer solution had stabilized , we used the alv-5000 to collect at least ten equilibrium light - scattering correlation functions , with the polymer solution in thermal equilibrium at 25@xmath74c . the photon count rate of these measurements ranged from 0.3 to 2.5 mhz , depending on the run . each correlation lasted from 30 to 60 minutes , with signal to background ratios from @xmath83 to @xmath84 . these small signal to background ratios confirm that our measurements are in the heterodyne regime . after having completed the measurements with the polymer solution in thermal equilibrium , we applied various values of the temperature gradient to the polymer solution by increasing and decreasing the temperatures of the top and bottom plates symmetrically , so that all experimental results correspond to the same average temperature of 25@xmath74c . the maximum temperature difference employed was 4.1@xmath74c , which corresponds to a maximum temperature gradient of 34.6 k @xmath82 . the variation in the thermophysical properties of the solutions is negligible in this small temperature interval , and the property values corresponding to 25@xmath74c have been used for the calculations . after changing the temperatures , we waited at least two hours , to be sure that the concentration gradient was fully developed . then , for each value of the temperature gradient , about six correlograms were taken . in fig.[fnon2 ] , typical experimental light - scattering correlograms , obtained with the alv-5000 correlator at @xmath85 @xmath82 are shown for the solution with polymer mass fraction @xmath86 , as a function of the temperature gradient @xmath1 . the figure shows the well sampled amplitude of the correlation at short times for the correlograms . a simple glance at fig . [ fnon2 ] confirms that the amplitude of the correlation function increases with increasing values of the temperature gradient @xmath1 . the correlograms , in the range from 10@xmath81 s to 1 s , depending on the run , were fitted to a single exponential , in accordance with eq . ( [ eq : corr ] ) . in all cases , both in equilibrium and in nonequilibrium , very good fits were obtained . actually , the intensity of the scattered light is observed over a range of wave numbers corresponding to the non - zero aperture of the collecting pinhole . this effect can be accounted for by representing the experimental correlation function in terms of a gaussian convolution , as explained in previous publications@xcite . however , we found that the resulting corrections to the parameter values obtained from fits to a single exponential amounted to less than 1% for the present experiments and could be neglected . hence , experimental values for the decay rate , @xmath87 , and for the prefactor @xmath88 $ ] were obtained by directly fitting the time - correlation function data of the scattered light to a single exponential , as given by eq . ( [ eq : corr ] ) . the experimental values obtained for the decay rates , @xmath87 , for the polymer solutions at various values of the scattering wave number @xmath2 and the temperature gradient @xmath1 are presented in table [ t1 ] . each value displayed in table [ t1 ] is the average from several ( at least five ) experimental correlograms obtained at the same conditions . the prefactor of the exponential obtained from the fitting procedure was multiplied by the average count rate during the run to get the average intensity ( in 10@xmath89 counts / s ) in each run . the dimensionless enhancement , @xmath35 , of the concentration fluctuations at a given nonzero value of @xmath1 was obtained from the ratio of the nonequilibrium to the equilibrium intensities , measured at the same @xmath2 , by means of eq . ( [ eq : corr ] ) . the experimental values obtained for @xmath35 are presented in table [ t2 ] . again , each value displayed in table[t2 ] is an average of several correlograms measured at the same conditions . the uncertainties quoted in tables[t1 ] and [ t2 ] represent standard deviations of the values obtained from the sets of experimental correlograms . dividing the decay rates displayed in table [ t1 ] by the square of the known wave numbers , a mass - diffusion coefficient , @xmath7 , is obtained for each @xmath16 , @xmath2 and @xmath1 . as an example , we show in fig . [ f68 ] the values of @xmath7 thus obtained for the polymer solution with @xmath90 obtained from the light - scattering measurements at various values of the wave number @xmath2 , as a function of the applied temperature gradient @xmath1 . the horizontal line in fig . [ f68 ] represents the weighted average value of @xmath7 . it is readily seen that the experimental @xmath7 is independent of @xmath1 , which implies that the applied temperature gradient does not affect the translational diffusion dynamics of the polymer molecules and that we indeed observed concentration fluctuations for all @xmath1 . figure [ f68 ] also demonstrates that @xmath7 is independent of the wave number @xmath2 . thus the decay rates in table [ t1 ] are indeed proportional to @xmath91 , which confirms the correctness of our measurements of @xmath2 . for each concentration we determined a simple value of @xmath7 as an average of the experimental data obtained at various @xmath2 and @xmath1 . in the averaging process the individual accuracies were taken into account . the resulting values of @xmath7 for the different polymer solutions are presented in table [ td ] and displayed in fig . [ fd ] as a function of the polymer concentration . for this purpose we prefer to use the concentration @xmath48 in g @xmath92 , because this unit is more widely employed in the literature for polymer solutions . to change the concentration units we used the relationship : @xmath93 as reported by scholte@xcite . the error bars in fig.[fd ] have been calculated by adding 3% of the value of @xmath7 to the standard deviations given in table [ td ] , so as to account for the uncertainty in the wave number @xmath2 . in fig . [ fd ] we have also plotted the values obtained by zhang et al . for the collective diffusion coefficient of the same polymer solutions with an optical beam - bending technique@xcite . this figure shows that the values obtained from the two methods for the diffusion coefficient agree within the experimental accuracy . the straight line displayed in fig . [ fd ] represents the linear relationship : @xmath94 with the values [ eq : dvalues ] @xmath95 and @xmath96 for the diffusion coefficient at infinite dilution , @xmath97 , and the hydrodynamic interaction parameter , @xmath98 , as determined by zhang et al.@xcite for the polymer solution with the same molecular mass @xmath66 . it may be observed that the @xmath97 and @xmath98 values proposed by zhang et al.@xcite yield a satisfactory description of the dependence of our @xmath7 values on the concentration . an extensive survey of literature values for @xmath97 and @xmath98 of polystyrene in toluene solutions was performed , and ten different references were examined @xcite . most authors have studied the molecular weight dependence of these parameters and propose relationships which usually have the form of power laws . the results of our survey are presented in table[t3 ] . in some cases where the scaling equations are not directly given by the authors , we have performed the corresponding fits to obtain the power - law dependence of @xmath97 and @xmath98 on @xmath99 . all data in table [ t3 ] correspond to polystyrene in toluene at the temperature of 25@xmath74c . while @xmath98 should be nearly independent of temperature because toluene is a good solvent@xcite , @xmath97 depends on temperature through a stokes - einstein relation . neglecting the dependence on temperature of the hydrodynamic radius of polystyrene in toluene@xcite , we represent the dependence of @xmath97 on the temperature , @xmath100 , by : @xmath101 where @xmath102 is the viscosity of the pure solvent ( toluene ) as a function of temperature which can be found in the literature@xcite . equation ( [ eq : dependt ] ) was used for making temperature corrections in cases where the values of @xmath97 reported in the literature had been measured at temperatures other than 25@xmath74c . furthermore , we did not consider literature values measured at temperatures more than @xmath7510 k from 25@xmath74c . the sixth column of table [ t3 ] contains the values extrapolated to @xmath103 , from the @xmath104 and @xmath105 relationships shown in the second column of the same table . from the information in table [ t3 ] we conclude that @xmath106 @xmath107 s@xmath108 and @xmath109 @xmath110 g@xmath108 , to be compared with the values @xmath111 @xmath107 s@xmath108 and @xmath112 @xmath110 g@xmath108 quoted in eq . ( [ eq : dvalues ] ) and adopted by us . note that standard deviations of the extrapolated values for @xmath97 and @xmath98 from the literature are @xmath113 and @xmath114 , respectively . the corresponding spread of the literature values of @xmath7 is indicated by the two dashed lines in fig . the upper line was calculated by taking for @xmath97 and @xmath98 the literature averages plus their standard deviations ; the lower line was calculated by taking for @xmath97 and @xmath98 the literature averages minus their standard deviations . having confirmed the validity of our experimental results for the decay rates , we now consider the nonequilibrium enhancements , @xmath115 , of the concentration fluctuations reported in table [ t2 ] . as can be readily seen , the nonequilibrium enhancements show a dramatic increase with increasing temperature gradients . the experimental values obtained for the enhancement are plotted in fig . [ f2 ] as a function of @xmath116 , for the six polymer solutions investigated . the results of a least - squares fit of the experimental points to a straight line going through the origin are also displayed . the information presented in fig . [ f2 ] confirms that , in the range of scattering wave vectors investigated , the nonequilibrium enhancement of the concentration fluctuations is indeed proportional to @xmath117 in accordance with eq . ( [ eq : enhan1 ] ) . the slope of the lines in fig . [ f2 ] yields experimental values for the strength of the enhancement , @xmath118 , listed in table [ ta ] . to compare the experimental results with the theoretical prediction , eq . ( [ eq : enhan2 ] ) , we need several thermophysical properties of the solutions , namely : the concentration derivative of the difference in the chemical potentials per unit mass , @xmath119 , the zero - shear viscosity , @xmath120 , the mass - diffusion coefficient , @xmath7 , and the soret coefficient , @xmath6 . a ) _ the derivative of the difference in chemical potentials was calculated from its relationship with the osmotic pressure@xcite , @xmath121 , which , for the small concentrations used can be simplified to : @xmath122 values for the concentration derivative of the osmotic pressure of the solutions , were obtained from the extensive work of noda et al.@xcite . specifically , the universal function : @xmath123\ ] ] represents the behavior of this property from the very dilute to the concentrated regime . the overlap concentration , @xmath124 g @xmath92 , was calculated , as in section iii , from the correlations proposed by the same authors@xcite . _ b ) _ to calculate the zero - shear viscosity , we employed the huggins relationship : @xmath125 c + k_{\rm h } [ \eta]^2 c^2 ) , \label{eq : nu}\ ] ] where @xmath126 is the viscosity of the pure solvent , for which the value 552.7 mpa@xmath127s was taken , @xmath128 $ ] is the intrinsic viscosity of polystyrene in toluene and @xmath129 the huggins coefficient for the same system . the intrinsic viscosity was obtained from the correlation@xcite : @xmath128 = 9.06~10^{-3 } m_w^{0.74}~({\rm cm^3~g^{-1}})$ ] . for the huggins coefficient , the usual value in a good solvent@xcite , @xmath130 , was adopted ; no molecular weight dependence has been reported in the literature for @xmath129 . values for the collective mass - diffusion coefficient , @xmath7 , of our solutions were presented in the previous section , when analyzing the decay rates of the correlograms . they are displayed in table [ td ] . for a continuous representation of @xmath7 as a function of concentration we use eq . ( [ eq : dif ] ) , with the parameters given by eq . ( [ eq : dvalues ] ) . _ d ) _ the soret coefficient , @xmath6 , was measured by zhang et al.@xcite for the same solutions used in our light - scattering experiments . it is worth noting that the soret coefficients measured by zhang et al . agree , within experimental error , with other recent @xmath6 values for polystyrene in toluene reported in the literature@xcite . since the strength of the enhancement depends on the square of the soret coefficient , we need a good continuous representation of these data to make a theoretical prediction of @xmath131 as a function of the concentration . we assume that @xmath6 scales as the inverse of @xmath7 , as rationalized by brochard and de gennes@xcite , and use a relationship proposed by nystrom and roots@xciteand used successfully in zhang , et al.@xcite for the diffusion coefficient and soret coefficient . the equation@xcite for @xmath6 is @xmath132 where @xmath133 and @xmath134 are exponents evaluated with @xmath135 , @xmath136 is a scaling variable , and @xmath137 . the virial constants @xmath138 and @xmath139 are defined by making a series expansion of eq . ( [ eq : nystrom ] ) around @xmath140 to give @xmath141 , where @xmath139 is the first virial coefficient of @xmath6 . we used the values of @xmath142 @xmath110 g@xmath108 and @xmath143 k@xmath108 found by zhang et al.@xcite for this molecular weight and then did a weighted , least - squares fit to the concentration dependent @xmath6 data of zhang et al . to find a value of @xmath144 . we find @xmath145 . equation ( [ eq : nystrom ] ) gives an excellent representation of the experimental soret coefficient data , as shown in fig . [ fsoret ] . in fig . [ f3 ] we present a comparison between the experimental values for the nonequilibrium - enhancement strength , @xmath131 , and the values calculated form eq . ( [ eq : enhan2 ] ) with the information for the various thermophysical properties as specified above . the error bars associated with the experimental data displayed in fig . [ f3 ] have been calculated by adding 6% to the statistical errors quoted in table [ ta ] so as to account for a 1.5% uncertainty in the values of the wave number @xmath2 . the theoretical values have an estimated uncertainty of at least 5% . taking into account these uncertainties , we conclude that the observed strength of the nonequilibrium concentration fluctuations is in agreement with the values predicted on the basis of fluctuating hydrodynamics in the concentration range investigated . the existence of long - range concentration fluctuations in polymer solutions subjected to stationary temperature gradients has been verified experimentally . as in the case of liquid mixtures@xcite , the nonequilibrium enhancement of the concentration fluctuations has been found to be proportional to @xmath0 . unlike the case of liquid mixtures@xcite , good agreement between the experimental and theoretical values for the strength of the enhancement of the concentration fluctuations , @xmath131 , has been found here . this indicates the validity of fluctuating hydrodynamics to describe nonequilibrium concentration fluctuations in dilute and semidilute polymer solutions . further theoretical and experimental work will be needed to understand nonequilibrium concentration fluctuations in polymer solutions at higher concentrations . our present results complement the considerable theoretical and experimental progress recently made in understanding the dynamics of concentration fluctuations of polymer solutions under shear flow . it has been demonstrated that concentration fluctuations in semidilute polymer solutions subjected to shear flow are also enhanced dramatically@xcite . this enhancement of the intensity of the concentration fluctuations in shear - induced ness , is similar to the enhancement reported here , when the ness is achieved by the application of a stationary temperature gradient . on the other hand , the long - range nature of the concentration fluctuations when a concentration gradient is present , has also recently been observed in a liquid mixture by shadowgraph techniques@xcite , yielding additional evidence of the interesting nature of this topic . we are indebted to j.f . douglas for valuable discussions and to s.c . greer for helpful advice concerning the characterization of the polymer sample . j.v.s . acknowledges the hospitality of the institute for theoretical physics of the utrech university , where part of the manuscript was prepared . the research at the university of maryland is supported by the u.s . national science foundation under grant che-9805260 . j.m.o.z . was funded by the spanish department of education during his postdoctoral stage at maryland , when part of the work was done . 99 j. berne and s. pecora , _ dynamic light scattering _ ( mcgraw - hill , new york , 1980 ) . j.p . boon and s. yip , _ molecular hydrodynamics _ ( mcgraw - hill , new york , 1980 ) . dorfman , t.r . kirkpatrick , and j.v . sengers , annu . phys . chem . * 45 * , 213 ( 1994 ) . kirkpatrick , e.g.d . cohen , and j.r . dorfman , phys . * 42 * , 862 ( 1979 ) , * 44 * , 472 ( 1980 ) , phys . a * 26 * , 950 , 972 , 995 ( 1982 ) . d. ronis and i. procaccia , phys . a * 26 * , 1812 ( 1982 ) b.m . law and j.v . sengers , j. stat . phys . * 57 * , 531 ( 1989 ) . r. schmitz and e.g.d . cohen , j. stat . phys . * 39 * , 285 ( 1985 ) ; * 40 * , 431 ( 1985 ) . g. grinstein , d .- h . lee , and s. sachdev , phys . * 64 * , 1927 ( 1990 ) . segr , r.w . gammon , j.v . sengers , and b.m . law , phys . a * 45 * , 714 ( 1992 ) . li , p.n . segr , r.w . gammon , and j.v . sengers , physica a * 204 * , 399 ( 1994 ) . j.c . nieuwoudt and b.m . law , phys . a * 42 * , 2003 ( 1990 ) . b.m . law and j.c . nieuwoudt , phys . a * 40 * , 3880 ( 1989 ) . segr , r. schmitz , and j.v . sengers , physica a * 195 * , 31 ( 1993 ) . segr and j.v . sengers , physica a * 198 * , 96 ( 1993 ) . a. vailati and m. giglio , phys . e * 58 * , 4361 ( 1998 ) . hohenberg and j.b . swift , phys . a * 46 * , 4773 ( 1992 ) h. van beijeren and e.g.d . cohen , j. stat . phys . * 53 * , 77 ( 1988 ) . i. pagonabarraga , j.m . rub , and l. torner , physica a * 173 * , 111 ( 1991 ) . segr , r.w . gammon , and j.v . sengers , phys . e * 47 * , 1026 ( 1993 ) . li , p.n . segr , j.v . sengers , and r.w . gammon , j. phys . condens . matter * 6 * , a119 ( 1994 ) . a. vailati and m. giglio , phys . * 77 * , 1484 ( 1996 ) ; prog . sci . * 104 * , 76 ( 1997 ) . w. khler and b. mller , j. chem . phys . * 103 * , 4367 ( 1995 ) . k.j . zhang , m.e . briggs , r.w . gammon , and j.v . sengers , j. chem . phys . * 104 * , 6881 ( 1996 ) . [ lbkai2 ] k.j . zhang , m.e . briggs , j.v . sengers , r.w . gammon , and j.f . douglas , j. chem . phys . * 111 * , 2270 ( 1999 ) . li , k.j . zhang , j.v . sengers , r.w . gammon , and j.m . ortiz de zrate , phys . lett . * 81 * , 5580 ( 1998 ) . anisimov , v.a . agayan , a.a . povodyrev , j.v . sengers , and e.e . gorodetskii , phys . e * 57 * , 1946 ( 1998 ) . r. schmitz , physica a * 206 * , 25 ( 1995 ) . i. noda , y. higo , n. ueno , and t. fujimoto , macromolecules * 17 * , 1055 ( 1984 ) . y. higo , n. ueno , and i. noda , polym . j. * 15 * , 367 ( 1983 ) . b.m . law , p.n . segr , r.w . gammon , and j.v . sengers , phys . rev . a * 41 * , 816 ( 1990 ) . scholte , j. polym . part a-2 * 8 * , 841 ( 1970 ) . pusey , j.m . vaughan , and g. williams , j. chem . fadaray trans . * 70 * , 1696 ( 1970 ) . b. appelt and g. meyerhoff , macromolecules * 13 * , 657 ( 1980 ) . v. petrus , b. porsch , b. nystrm , and l.- o. sundelf , makromol . chem . * 183 * , 1279 ( 1982 ) . e. ozdemir and r.w . richards , polymer * 24 * , 1097 ( 1983 ) . m. ramanathan and m.e . mcdonnell , macromolecules * 17 * , 2093 ( 1984 ) . varma , y. fujita , m. takakashi , and t. nose , j. polym . * 22 * , 1781 ( 1984 ) . k. huber , s. bantle , p. lutz and w. burchard , macromolecules * 18 * , 1461 ( 1985 ) . freeman , d.s . soane , and m.m . denn , macromolecules * 23 * , 245 ( 1990 ) . w. khler , c. rosenauer , and p. rossmanith , int . j. thermophys . * 16 * , 11 ( 1995 ) . p. vidakovic and f. rondelez , macromolecules * 18 * , 700 ( 1985 ) . h. yamakawa , _ modern theory of polymer solutions , _ ( harper&row , new york , 1971 ) . y. takakashi , y. isono , i. noda , and m. nagasawa , macromolecules * 18 * , 1002 ( 1985 ) . papazian , polymer * 10 * , 399 ( 1969 ) . f. brochard and p.g . de gennes , c.r . acad . sc . paris * 293-ii * , 1025 ( 1981 ) . b. nystrm , and j. roots , polymer * 33 * , 1548 ( 1992 ) . the expression for the exponent @xmath146 given in reference [ [ lbkai2 ] ] is given with the wrong sign . x.l . wu , d.j . pine , p.k . dixon , phys . lett . * 66 * , 2408 ( 1991 ) . milner , phys . lett . * 66 * , 1477 ( 1991 ) . a. onuki , phys . lett . * 62 * , 2472 ( 1989 ) . a. vailati and m. giglio , nature * 390 * , 262 ( 1997 ) . @xmath16&@xmath2&@xmath1=0.0&@xmath1=8.5&@xmath1=12.7&@xmath147=15.8&@xmath1=16.9 & @xmath1=21.8&@xmath1=29.1&@xmath147=34.6 + & ( @xmath82)&(k @xmath82)&(k @xmath82)&(k @xmath82)&(k @xmath82 ) & ( k @xmath82)&(k @xmath82)&(k @xmath82 ) + 0.50%&1318&[email protected]&&&[email protected]&&&[email protected]&[email protected] + & 1538&[email protected]&&&[email protected]&&&[email protected]&[email protected] + & 1757&[email protected]&&&[email protected]&&&[email protected]&[email protected] + & 1977&[email protected]&&&&&&&[email protected] + 1.00%&1318&[email protected]&&[email protected]&&&&[email protected]&[email protected] + & 1538&1.4 @xmath750.3 & & [email protected]&&&[email protected] + & 1757&[email protected]&&&[email protected]&&[email protected]&[email protected] + & 1977&[email protected]&&[email protected]&&&[email protected]&[email protected] & [email protected] + 1.50%&1318&[email protected]&&&[email protected]&&[email protected]&[email protected] + & 1538&[email protected]&&&[email protected]&&[email protected]&[email protected] + & 1757&[email protected]&&&[email protected]&&[email protected]&[email protected] + & 1977&[email protected]&&&[email protected]&&&[email protected]&[email protected] + 2.00%&872&[email protected]&[email protected]&[email protected]&&[email protected] + & 1392&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 1538&[email protected]&&&[email protected]&&[email protected]&[email protected] & [email protected] + & 1757&[email protected]&&&[email protected]&&[email protected]&[email protected] & [email protected] + & 1977&[email protected]&&&[email protected]&&[email protected]&[email protected] & [email protected] + 2.50%&1030&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 1350&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 1880&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + 4.00%&1353&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 1525&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 1854&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + & 2025&[email protected]&[email protected]&[email protected]&&[email protected] & [email protected] + @xmath16&@xmath2&@xmath1=8.5&@xmath1=12.7&@xmath1=15.8&@xmath147=16.9&@xmath1=21.8 & @xmath1=29.1&@xmath1=34.6 + & ( @xmath82)&(k @xmath82)&(k @xmath82)&(k @xmath82)&(k @xmath82 ) & ( k @xmath82)&(k @xmath82)&(k @xmath82 ) + 0.50%&1318&&&[email protected]&&&[email protected]&[email protected] + & 1538&&&[email protected]&&[email protected]&&[email protected] + & 1757&&&[email protected]&&[email protected]&&[email protected] + & 1977&&&&&&&[email protected] + 1.00%&1318&&[email protected]&&&[email protected]&&[email protected] + & 1538&&[email protected]&&&[email protected] + & 1757&&&[email protected]&&[email protected]&[email protected] + & 1977&&[email protected]&&&[email protected]&[email protected]&[email protected] + 1.50%&1318&&&[email protected]&&[email protected]&[email protected] + & 1538&&&[email protected]&&[email protected]&[email protected] + & 1757&&&[email protected]&&[email protected]&[email protected] + & 1977&&&[email protected]&&&[email protected]&[email protected] + 2.00%&872&[email protected]&[email protected]&&[email protected] + & 1392&[email protected]&[email protected]&&[email protected]&[email protected] + & 1538&&&[email protected]&&[email protected]&[email protected]&[email protected] + & 1757&&&[email protected]&&[email protected]&[email protected]&[email protected] + & 1977&&&[email protected]&&[email protected]&[email protected]&[email protected] + 2.50%&1030&[email protected]&[email protected]&&[email protected]&[email protected] + & 1350&[email protected]&[email protected]&&[email protected]&[email protected] + & 1880&[email protected]&[email protected]&&[email protected]&[email protected] + 4.00%&1353&[email protected]&[email protected]&&[email protected]&[email protected] + & 1525&[email protected]&[email protected]&&[email protected]&[email protected] + & 1854&[email protected]&[email protected]&&[email protected]&[email protected] + & 2025&[email protected]&[email protected]&&[email protected]&[email protected] +

we consider the totally asymmetric simple exclusion process on a ring with stationary initial conditions . the crossover between kpz dynamics and equilibrium dynamics occurs when time is proportional to the power of the ring size . we obtain the limit of height function along the direction of characteristic line in this time scale . the two - point covariance function in this scale is also discussed .

1610.04601

in this paper we consider the totally asymmetric simple exclusion process ( tasep ) on a ring of size @xmath1 which we denote by @xmath2 . the dynamics of tasep on the ring is the same as that of tasep on @xmath3 except the particle at the site @xmath4 , once it jumps , moves to the site @xmath5 if @xmath5 is empty , here the @xmath6 denotes the element @xmath7 in @xmath2 for @xmath8 . let @xmath9 the occupation variable of this model , @xmath10 . @xmath11 is @xmath12 if the site @xmath6 is occupied or @xmath13 if the site @xmath6 is empty . we extend the occupation variable to @xmath3 periodically by defining @xmath14 for all @xmath15 . define the following height function @xmath16 where @xmath17 counts the number of particles jumping through the bond from @xmath13 to @xmath12 on @xmath2 during the time interval @xmath18 $ ] . note that @xmath19 , where @xmath20 counts the number of particles jumping through the bond from @xmath21 to @xmath22 on @xmath2 during the time interval @xmath18 $ ] . although @xmath23 are both periodic in @xmath24 , @xmath25 is not periodic except when the system is half - filled . indeed , we have @xmath26 for all @xmath27 and @xmath28 , where @xmath29 is the number of particles on the ring . we are interested in the fluctuations of @xmath25 when @xmath30 and @xmath24 both increase of order @xmath31 , and @xmath32 go to infinity proportionally . the scale @xmath33 is called the relaxation time scale , which was first studied by gwa and spohn @xcite . at this relaxation time scale , one expects to see a crossover between the kpz dynamics and the gaussian dynamics and hence the fluctuations are of great interest to both math and physics communities . the crossover limiting distributions were obtained only recently by prolhac @xcite and baik and liu @xcite . in @xcite , prolhac obtained ( not rigorously ) the limit of the current fluctuations for step , flat and stationary initial conditions in the half particle system ( with the restriction @xmath34 ) . independently , baik and liu also obtained the limit in a more general setting of @xmath35 and @xmath1 for flat and step initial conditions in @xcite . the main goal of this paper is to extend the work of @xcite to the stationary initial condition and prove the rigorous limit theorem of @xmath25 in relaxation time scale . compared to @xcite , there are some other differences except for the rigorousness : we consider a more general setting of stationary initial conditions than @xcite , and more generally the height function @xmath25 , than the current in @xcite , which is equivalent to @xmath36 . hence the limiting distribution obtained in this paper , @xmath37 in theorem [ thm : limit_height_uniform ] , contains two parameters of time @xmath38 and location @xmath39 , in contrast to that of only time parameter in @xcite . due to the ring structure , the number of particles is invariant and hence it is natural to consider the following _ uniform _ initial condition of @xmath35 particles : initially all possible configurations of @xmath35 particles on the ring of size @xmath1 are of equal probability , i.e. , @xmath40 . this initial condition is stationary , and is the unique one for fixed number of particles @xmath35 and ring size @xmath1 @xcite . for this uniform initial condition , there is a _ characteristic _ line @xmath41 in the space - time plane , here @xmath42 is the density of the system . the main theorem of this paper is on the fluctuations of @xmath43 near the characteristic line in relaxation time scale . [ thm : limit_height_uniform ] let @xmath44 and @xmath45 be two fixed constants satisfying @xmath46 . suppose @xmath47 is a sequence of integers such that @xmath48 for all @xmath1 . we consider the tasep on a ring of size @xmath1 with @xmath47 particles . assume that they satisfy the uniform initial condition . denote @xmath49 . let @xmath38 and @xmath50 be two fixed constants satisfying @xmath51 and @xmath52 . suppose @xmath53 then along the line @xmath54 we have @xmath55 for each @xmath56 . here @xmath37 is a distribution function defined in for any @xmath51 and @xmath57 . it satisfies @xmath58 and @xmath59 . [ rmk:01 ] in @xcite , prolhac obtained when @xmath60 and @xmath61 ( and hence @xmath62 , @xmath63 ) with a different formula of the limiting distribution . his proof , as mentioned before , is not completely rigorous . note that if we write @xmath64 , then the line can be rewritten as @xmath65 this expression gives an intuitive reason why the limiting function @xmath37 is periodic on @xmath39 : it is the periodicity of the shifted height function @xmath66 . to better understand the parametrization in the above theorem , we compare it with the infinite tasep with stationary condition , i.e. , the stationary tasep on @xmath3 . suppose initially each site in @xmath3 is occupied independently with probability @xmath67 . then the height fluctuation converges along the line @xmath68 for any given constant @xmath52 , see @xcite , @xmath69 where @xmath70 is the baik - rains distribution defined in @xcite was denoted by @xmath71 . ] . theorem [ thm : limit_height_uniform ] of this paper shows that for the stationary tasep on a ring with uniform initial condition in relaxation time scale , similar limiting laws hold near the characteristic line . the difference is that for the ring tasep , the fluctuations have a periodicity on the parameter @xmath64 , which is not present in the infinite tasep model . the leading terms @xmath72 and @xmath73 in @xmath74 can be explained as following . the first term @xmath72 measures the change of height along the direction @xmath75 : for fixed @xmath76 , recall that @xmath77 grows as @xmath72 in the leading order and satisfies @xmath78 . the second term @xmath73 measures the time - integrated current at a fixed location : @xmath79 which grows as @xmath73 in the leading order . as an application of theorem [ thm : limit_height_uniform ] , we can express the limit of two - point covariance function in terms of @xmath37 . recall the occupation variable @xmath80 at the beginning of the paper . define the two - point covariance function @xmath81 where @xmath82 is the system density . it is known that for the stationary tasep , there is a relation between this two - point function @xmath83 and the height function @xmath84 : @xmath85 . this relation was proved for the infinite tasep in @xcite but the proof is also valid for tasep on a ring after minor modifications . using this identity and the tail estimate which is provided in the appendix [ sec : estimates ] , we obtain the following result . the proof is almost the same as that for the stationary tasep on @xmath3 , see @xcite , and hence we omit it . [ cor : space_time_covariance_uniform ] suppose @xmath86 and @xmath75 are defined as in theorem [ thm : limit_height_uniform ] with the same constants @xmath51 and @xmath57 . then we have @xmath87 if integrated over smooth functions in @xmath39 with compact support , where @xmath88 another application is that one can obtain the height fluctuations for other stationary tasep on a ring . note that the uniform initial conditions with @xmath89 form a complete basis for all stationary initial conditions . hence we may apply theorem [ thm : limit_height_uniform ] for other stationary initial conditions . one example is the bernoulli condition . suppose initially each site of the ring is occupied independently with probability @xmath67 , where @xmath67 is a constant satisfies @xmath90 . then we have the following result . [ cor : height_fluctuations_bernoulli ] suppose @xmath91 is a fixed constant . we consider the tasep on the ring of size @xmath1 with bernoulli initial condition of parameter @xmath67 . suppose @xmath52 , @xmath51 and @xmath56 are fixed constants . denote @xmath92 then @xmath93 here @xmath94 is a distribution function for arbitrary @xmath51 and @xmath57 , given by @xmath95 a formal proof is as following . assume there are @xmath96 particles initially . by applying theorem [ thm : limit_height_uniform ] , we obtain that @xmath97 converges to @xmath98 as @xmath99 , where @xmath64 . together with the central limit theorem , we obtain @xmath100 by a simple change of variables we arrive at . this argument can be made rigorous by a simple tail estimate on the number of particles and then by the dominated convergence theorem . since the argument is standard , we omit the details . recall that @xmath37 is symmetric on @xmath39 . hence by using the formula we have @xmath101 . however , different from @xmath37 , we do not expect @xmath102 . it is because by definition @xmath103 where @xmath104 is a standard gaussian random variable . hence formally @xmath105 here the two random variables on the right hand side of are not necessarily independent . this relation still strongly indicates that @xmath106 is not the same as @xmath107 . the organization of this paper is as following . in section [ sec : limiting_distribution ] we give the explicit formula and some properties of @xmath37 . the proof of theorem [ thm : limit_height_uniform ] is given in section [ sec : formulas ] and [ sec : asymptotics ] : the finite time distribution formula is provided in section [ sec : formulas ] and then the asymptotics in section [ sec : asymptotics ] . finally in the appendix [ sec : estimates ] we give some tail bounds related to the distribution function @xmath37 . the author would like to thank jinho baik , ivan corwin and peter nejjar for useful discussions . the limiting distribution @xmath37 is defined as following @xmath109 where the integral is over an arbitrary simple closed contour within the disk @xmath110 and with @xmath13 inside . the terms @xmath111 are given by @xmath112 and @xmath113 is given by @xmath114 here @xmath115 is the polylogarithm function defined by @xmath116 for all @xmath110 , and then to be analytical continuous to the complex plane except the half line @xmath117 . and @xmath118 is the usual logarithm function with branch cut @xmath119 . the operator @xmath120 is defined on the set @xmath121 with kernel @xmath122 where @xmath123 the terms @xmath124 and @xmath120 are defined in @xcite . they appeared in the limiting distributions of tasep on a ring with step initial condition in relaxation time scale , more explicitly in the distribution function @xmath125 as shown below @xmath126 see ( 4.10 ) of @xcite . it is known that they are well defined and bounded uniformly on the choice of @xmath127 ( but the bound may depend on the contour ) . furthermore , the fredholm determinant @xmath128 is periodic and symmetric on @xmath39 , which implies @xmath129 and @xmath130 . to ensure @xmath37 in is well defined , we still need to check that the derivative in the integrand exists and is uniformly bounded . the only non - trivial part is to check @xmath131 . this can be proved by directly using the super - exponential decaying property of the kernel . the argument is standard and we do not provide details . alternately , our analysis in section [ sec : asymptotics_delta_determinant ] also implies @xmath131 is a limit of a uniformly bounded sequence hence it is also uniformly bounded . see lemma [ lm : moments_convergent ] and [ lm : uniform_bound_difference_determinant ] . as we discussed in remark [ rmk:01 ] , the limiting distribution when @xmath63 was obtained in @xcite . numeric plots of our formula @xmath132 match the limiting distribution obtained in @xcite well , see figure [ fig : fu_br ] in this paper and fig.5.b in @xcite . however , a rigorous proof of the equivalence on @xmath132 and their formula ( see ( 10 ) of @xcite ) is still missing . with @xmath133 , @xmath134 , and @xmath12 respectively . and the solid line is the density function of the standard gaussian distribution . ] with @xmath133 , @xmath134 , and @xmath12 respectively . and the solid line is the density function of the standard gaussian distribution . ] for any fixed @xmath51 and @xmath135 , the function @xmath37 is a distribution function . the proof is not trivial and we prove it in the appendix [ sec : estimates ] . similar to @xmath125 , the function @xmath37 has the properties @xmath136 and @xmath59 . also by using the following simple identity ( see ( 11 ) of @xcite ) @xmath137 which can also be checked directly from the definition , we formally have @xmath138 the rigorous proof of this identity is similar to corollary [ cor : space_time_covariance_uniform ] and we do not provide details here . besides , we expect the following small @xmath38 and large @xmath38 limits of @xmath37 : 1 . for any fixed @xmath139 , we have ( see figure [ fig : fu_br ] for an illustration ) @xmath140 2 . for any fixed @xmath141 , we have ( see figure [ fig : fu_gaussian ] for an illustration ) @xmath142 in this section , we prove an exact formula for the height function with uniform initial condition . this formula turns out to be suitable for later asymptotic analysis . before stating the results , we need to introduce some notations . most of these notations are the same as in @xcite , and hence we just go through it quickly without further discussions . see section 7 of @xcite for more details . we fix @xmath1 and @xmath35 in this section , and denote @xmath143 the density of the system . for any @xmath144 , define a polynomial @xmath145 and its root set @xmath146 when @xmath147 , @xmath148 is a degenerated set of two points @xmath13 and @xmath12 with multiplicities @xmath35 and @xmath149 respectively . on the other hand , when @xmath150 , @xmath148 is asymptotically a set of @xmath1 equidistant points on a circle @xmath151 . for our purpose , we focus on the case when @xmath152 for such @xmath153 , @xmath154 contains @xmath149 points in the half plane @xmath155 and @xmath35 points in the second half plane @xmath156 . we denote @xmath157 and @xmath158 the sets of these @xmath149 and @xmath35 points respectively . then we define @xmath159 which are two monic polynomials with root sets @xmath157 and @xmath158 respectively . these two functions satisfy the following equation @xmath160 for all @xmath161 . for @xmath144 satisfying and arbitrary @xmath162 , we define a kernel @xmath163 acting on @xmath164 as following @xmath165 where the function @xmath166 is defined by @xmath167 we also define a function @xmath168 @xmath163 and @xmath169 are the same as @xmath170 and @xmath171 in @xcite ( with @xmath24 and @xmath172 replaced by @xmath173 and @xmath174 respectively ) but we emphasize the parameters @xmath172 and @xmath24 for our purpose . finally we denote @xmath175 the difference operator @xmath176 for arbitrary function @xmath177 . for an example , @xmath178 . now we state the formula for the distribution function of @xmath25 . [ thm : current_fluctuation ] suppose @xmath24 and @xmath179 are both integers satisfying @xmath180 . for the @xmath35-particle tasep on the ring of size @xmath1 with uniform initial condition , the distribution of the height function is given by @xmath181 where @xmath182 and the integral is over an arbitrary simple closed contour which contains @xmath13 inside and lies in an annulus @xmath183 . we consider an equivalent model : the tasep on @xmath184 . the configuration space @xmath184 is defined by @xmath185 the equivalence between tasep on @xmath184 and tasep with @xmath35 particles on the ring of size @xmath1 is as following : the ring tasep can be obtained by projecting the particles in tasep on @xmath184 to a ring of size @xmath1 , and on the other hand , in the tasep on a ring if we define @xmath186 to be the number of steps the @xmath172-th particle moved plus its initial location , then @xmath187 and we obtain the tasep on @xmath184 . see @xcite for more discussions on tasep on a ring and its equivalent models . it is easy to see that the uniform initial condition for the tasep of @xmath35 particles on a ring of size @xmath1 corresponds to the uniform initial condition in the following set @xmath188 in the system of tasep on @xmath184 . moreover , for any @xmath189 , we have the following relation between two models . in this case , @xmath190 . therefore @xmath191 if and only if @xmath192 , which is further equivalent to @xmath193 . the case when @xmath194 or @xmath195 follows immediately from the fact that @xmath196 . ] @xmath197 where @xmath198+n+1-\frac{b-{\ell}}{2},\\ a & = l\left[\frac{b-{\ell}-2}{2n}\right ] + { \ell}+1 , \end{split}\ ] ] and the notations @xmath199 , @xmath200 denote the probability of tasep on @xmath184 with initial configuration @xmath189 and the location of the @xmath201-th particle at time @xmath30 respectively . now we sum over all possible initial configurations @xmath189 each of which has probability @xmath202 , and obtain @xmath203 on the other hand , the one point distribution function for tasep on @xmath184 with arbitrary initial condition @xmath204 was obtained in @xcite ( see proposition 6.1 ) . more explicitly , we have @xmath205_{i , j=1}^n \frac{{{\mathrm d}}{\mathbf{z}}}{{\mathbf{z}}^{1-(k'-1)l } } , \end{split}\ ] ] where the integral is over any simple closed contour with @xmath13 inside . to proceed , we need the following two lemmas . [ lm : summation_initial_conf ] suppose @xmath206 , then we have @xmath207_{i , j=1}^n = \det\left [ w_i^{j-1}(w_i+1)^{-n+1}\right]_{i , j=1}^n-(-1)^{n-1}{\mathbf{z}}^{-l}\det\left [ w_i^{j}(w_i+1)^{-n}\right]_{i , j=1}^n.\ ] ] [ lm : fredholm_representation](theorem 7.2 in @xcite ) suppose @xmath153 is in the annulus @xmath183 as in . for any integer @xmath172 , we have the following identity . however , the proof is still valid if we drop the integral in both sides . ] @xmath208_{i , j=1}^n = { \mathcal{c}_n^{(2)}}({\mathbf{z}};k,{\ell}+1)\cdot \det\left(i+k_{{\mathbf{z}};k,{\ell}+1}^{(2)}\right).\ ] ] we first assume lemma [ lm : summation_initial_conf ] is true . by inserting to and then applying lemma [ lm : summation_initial_conf ] , we have @xmath209_{i , j=1}^n\frac{{{\mathrm d}}{\mathbf{z}}}{{\mathbf{z}}^{1-(k'-2)l}}.\ ] ] note that by using this expression is invariant under the following changes : @xmath210 and @xmath211 , therefore we can replace @xmath173 by @xmath212={\ell}+1 $ ] and @xmath201 by @xmath213=k+n$ ] . thus the above equation equals to @xmath214_{i , j=1}^n\frac{{{\mathrm d}}{\mathbf{z}}}{{\mathbf{z}}^{1-(k+n-2)l}}.\ ] ] by restricting @xmath153 in the annulus @xmath215 and applying lemma [ lm : fredholm_representation ] we immediately obtain . it remains to prove lemma [ lm : summation_initial_conf ] . we take the sum over @xmath189 in the following order : @xmath216 , @xmath217 . it is easy to check that the summation over @xmath189 is equivalent to that over @xmath218 : @xmath219 recurrently for @xmath220 and finally @xmath221 . note @xmath218 only appears in the @xmath222-th column in the determinant on the left hand side of . hence if we take the sum over all possible @xmath218 , all other columns in the determinant do not change except the @xmath222-th one . then for each @xmath223 , we have the following sum over @xmath218 on the @xmath222-th column @xmath224 where the second term is the same as the @xmath225 entry thus the determinant does not change if we remove this term . after taking the sum over @xmath226 , we obtain a new determinant whose first column is the same as before but the @xmath222-th column is @xmath227 for all @xmath223 . then we take the sum over @xmath228 . note that the bounds for @xmath228 are @xmath229 and @xmath230 . therefore we have @xmath231_{i , j=1}^n\\ = & \det\left [ w_i^{j-1}(w_i+1)^{-n+1}-\delta_1(j)w_i^{j-1}(w_i+1)^{-l}\right]_{i , j=1}^n\\ = & \det\left [ w_i^{j-1}(w_i+1)^{-n+1}\right]_{i , j=1}^n- \det\left [ w_i^{j-1}(w_i+1)^{-n+1-\delta_1(j)(l - n+1)}\right]_{i , j=1}^n , \end{split}\ ] ] where @xmath232 is the delta function and we used the linearity of the determinant on the first column in the second equation . comparing it with , it remains to show @xmath233_{i , j=1}^n=(-1)^{n-1}{\mathbf{z}}^{-l}\det\left [ w_i^{j}(w_i+1)^{-n}\right]_{i , j=1}^n.\]]by using the fact that @xmath234 and then exchanging the columns , the above equation is further reduced to @xmath235_{i , j=1}^n=\det\left [ w_i^{j}(w_i+1)^{-n}\right]_{i , j=1}^n,\ ] ] which follows from the simple identity @xmath236_{i , j=1}^n = \left [ w_i^{j}(w_i+1)^{-n}\right]_{i , j=1}^n \left[\delta_i(j)+\delta_i(j+1)\right]_{i , j=1}^n.\ ] ] in this section , we focus on the asymptotics of the formula and prove theorem [ thm : limit_height_uniform ] . we will follow the framework in @xcite , where they computed the asymptotics of two similar formulas , one of which contains exactly the same components @xmath169 , @xmath163 and @xmath237 as in this paper . however , there are the following two differences : 1 . [ item:01 ] in @xcite , the asymptotics of @xmath169 and @xmath237 was obtained with a special choice of parameters . more explicitly , the authors considered a case of discrete times @xmath30 and an order @xmath238 parameter @xmath172 . in this paper , we have a different setting of parameters , in which we let @xmath30 goes to infinity continuously and @xmath172 grows as @xmath239 . [ item:02 ] the formula in this paper contains a new feature . namely , we have difference operator @xmath175 in the formula , which was not present in @xcite . in the asymptotics , this @xmath175 , after appropriate scaling , converges to the differentiation with respect to @xmath240 . for [ item:01 ] , one can modify the calculations in @xcite to the new parameters . however , in this paper we instead consider a more general setting and find the conditions for parameters such that both @xmath169 and @xmath237 converge simultaneously . then we check our parameters as a special case . for [ item:02 ] , we need to find the asymptotics of @xmath241 and @xmath242 . the first one can be obtained straightforward , while the second one requires a bound estimate ( uniformly on @xmath1 and @xmath153 ) of each term in its expansion , which guarantees the convergence ( uniformly on @xmath153 ) of @xmath242 . in this subsection , we list the following general setting of the parameters . we suppose the density @xmath243 satisfies @xmath244 for some fixed positive constants @xmath245 . and we assume @xmath246 for some fixed constant @xmath51 . moreover , suppose @xmath247 and @xmath248 are two integer sequences which are bounded uniformly by @xmath31 and satisfy @xmath249 and @xmath250 where @xmath64 and @xmath240 are arbitrary fixed real constants , and the notation @xmath251 denotes the smallest distance between @xmath252 and all integers . recall the asymptotics is taken along the line @xmath253 , and the asymptotics along the line @xmath254 is the same . see and its discussions . the condition means that the points should be asymptotically on the @xmath255 lines . to understand the second condition , we need to view @xmath256 ( more precisely @xmath257 ) as the label of particle which is at the given location @xmath75 at time @xmath76 . first we extend the tasep on a ring to a periodic tasep on @xmath3 by making infinitely many identical copies of the particles on each interval of length @xmath1 , more precisely define @xmath258 for all @xmath172 and @xmath30 . with this setting , the labels of particles are in @xmath3 instead of @xmath259 . means the label of particle located at site @xmath75 at time @xmath76 is @xmath260 at the leading order ( more precisely @xmath261 due to our choice of initial labeling : the label is asymptotically @xmath35 at site @xmath13 initially ) , plus an @xmath262 fluctuation term . the term @xmath263 ( assuming @xmath264 , otherwise considering @xmath265 instead ) is asymptotically the number of particles initially in the interval @xmath266 $ ] , while @xmath267 is asymptotically the number of particles jumping through any given site during time @xmath268 $ ] . the above descriptions are in terms of stationary tasep on a ring with uniform initial condition . however , recalling the discussions at the beginning of section [ sec : asymptotics ] , the formula arising in step initial condition contains the same components @xmath169 and @xmath237 , whose asymptotics can be found within the same framework . thus the conditions and can also be interpreted similarly in terms of tasep on a ring with step initial condition . now we consider three different choices of parameters satisfying , and . the first choice is that we fix the label of particle @xmath256 and then let @xmath75 and @xmath76 go to infinity simultaneously . this choice corresponds to the case when an observer focuses on a tagged particle . now we rewrite the conditions and as @xmath269 and @xmath270 where @xmath271 is an integer sequence . these two equations imply that @xmath272 now we want @xmath76 grows as , and hence @xmath222 grows as @xmath273 $ ] , here the notation @xmath274 $ ] denotes the integer part of @xmath275 . for simplification , we ignore the @xmath276 in @xmath76 and obtain @xmath277+\frac{\gamma}{\rho_l } l -\frac{1}{\rho_l^2}k_l,\ ] ] which is a time scaling of tasep on a ring with step initial condition discussed in @xcite ( with their @xmath256 replaced by @xmath257 ) . see theorem 3.3 of @xcite . the second choice of parameters is that we fix the location @xmath75 and let @xmath256 and @xmath75 goes to infinity simultaneously . this choice corresponds to the case when an observer focuses on a fixed location . by a similar argument as above , we find that @xmath76 can be expressed as @xmath278-\frac{\gamma l}{1 - 2\rho_l}+\frac{{\ell}_l}{1 - 2\rho_l}\ ] ] when @xmath279 is of @xmath280 distance to @xmath281 . when @xmath61 , then the line @xmath282 is the characteristic line with a constant shift and this case is reduced to the next one we will discuss later . these scalings were discussed in @xcite , see theorem 3.4 of that paper . the third choice of parameters is that we fix the line @xmath283 . this is exactly the choice we pick in theorem [ thm : limit_height_uniform ] . it means that an observe moves along the direction of the characteristic line . in this case , the time parameter @xmath76 can grow continuously , and the location @xmath75 changes according to @xmath283 . finally the label of particle grows by the formula . note that in theorem [ thm : limit_height_uniform ] we have the height @xmath74 instead of the label of particles @xmath256 , hence to check one needs to use the relation @xmath284 in . for notation convenience , we will surpass the subscript @xmath1 in the asymptotic analysis from the next subsection to the end of section [ sec : asymptotics ] . in this subsection we follow the setting of @xcite ( see section 8) and give the explicit choice of integral contour . we also give the limit of @xmath157 and @xmath158 , and asymptocis of some parameter - independent components in @xmath169 . these results are all included in @xcite and hence we do not provide details . in , we set @xmath285 where @xmath127 is along any given simple closed contour within the unit disk @xmath110 and with @xmath13 inside . then becomes @xmath286 here @xmath287 is any number determined by . and it is easy to check the integrand above is invariant for @xmath288 , and hence the choice of @xmath153 , provided it satisfies , does not affect the integral . we first consider the limits of the nodes sets @xmath157 and @xmath158 with @xmath153 scaled as . it turns out that after rescaling these nodes sets converge to the sets @xmath289 and @xmath290 respectively . the explicit meaning of this convergence is described as below . [ lm : limit_nodes ] ( lemma 8.1 of @xcite ) let @xmath127 be a fixed number satisfying @xmath291 and let @xmath292 be a real constant satisfying @xmath293 . set @xmath294 where @xmath295 . define the map @xmath296 from @xmath297 to @xmath298 by @xmath299 then for large enough @xmath35 we have : a. @xmath300 is well - defined . b. @xmath300 is injective . c. the following relations hold : @xmath301 where @xmath302 , the image of the map @xmath300 , and @xmath303 for all @xmath304 . if we define the mapping @xmath305 in the same way but replace @xmath157 and @xmath298 by @xmath158 and @xmath306 respectively , the same results hold for @xmath305 . then we consider the limits of @xmath307 , @xmath308 and the following function @xmath309 the first two functions arise from the kernel @xmath310 , and the third function @xmath311 is part of @xmath169 . the limits of these three functions were obtained in @xcite as below . [ lm : limits_ind_ftns](lemma 8.2 of @xcite ) suppose @xmath312 and @xmath292 satisfy the conditions in lemma [ lm : limit_nodes ] . a. for complex number @xmath313 satisfying @xmath314 with some positive constant @xmath315 , set @xmath316 . then for sufficiently large @xmath35 @xmath317 if @xmath318 , where @xmath319 similarly for sufficiently large @xmath35 @xmath320 if @xmath321 , where @xmath322 b. for large enough @xmath35 we have @xmath323 where @xmath324 is defined in . finally , we need the expansions of two functions @xmath325 and @xmath326 along the line @xmath327 . these estimates were obtained in @xcite , see ( 9.36 ) and ( 9.37 ) of that paper . below we gave a quick summary of these estimates . write @xmath328 , where @xmath329 . it is direct to check that when @xmath330 @xmath331 here and below @xmath118 denotes the natural logarithm function with the branch cut @xmath119 . together with and , we have for @xmath330 @xmath332 when @xmath333 , it is also easy to check that @xmath334 for some positive constant @xmath315 . similarly , for @xmath330 , we have @xmath335 as we discussed before , the asymptotics of @xmath169 was obtained in @xcite with a specific choice of parameters . the idea is as following : write @xmath169 as @xmath336 , where @xmath311 is defined in and @xmath337 with the parameter setting in @xcite , they obtained ( see lemma 8.7 in @xcite ) @xmath338 where @xmath339 and @xmath340 are defined in . together with in lemma [ lm : limits_ind_ftns ] , one has @xmath341 the goal of this subsection is to check the proof of in @xcite also works under the more general setting , and . considering that the asymptotic analysis in @xcite was focusing on a different case which corresponds to the flat initial condition and appearing in the step case was only discussed briefly , and that some parts of the proof will be used in later discussions , we would like to go through the main steps of the proof of with the more general settings in this paper . however , we will not discuss many details on the calculations unless they are necessary . first we write the summation in @xmath342 as an integral . by using a residue computation , it is easy to see that @xmath343 where @xmath344 now we changes variables @xmath328 where @xmath329 . recall , it is sufficient to consider the integral over @xmath330 and the error is exponentially small @xmath345 . with this restriction and the assumptions that @xmath346 are bounded by @xmath31 , we have @xmath347 for notation simplification we denote the first three terms @xmath348 . by using the conditions - , it is easy to see that @xmath349 now by plugging , , and we obtain that equals to an exponentially small term @xmath345 plus @xmath350 by using the symmetry of the integral domain and integrating by parts , it is direct to check that the above quantity equals to @xmath351 where @xmath352 and @xmath353 are defined in . now we insert into the above equation , we obtain the right hand side equals to @xmath354 . combing with , we have . by definition , we have @xmath355 by applying and the following lemma , we obtain @xmath356 where @xmath292 is the same as in the previous subsection . [ lm : estimate_diff_consts ] for any fixed @xmath292 satisfying @xmath293 , we have @xmath357 by a residue computation similar to , we write the left hand side of as @xmath358 the rest of the proof is similar to but much easier . we omit the details . similarly to @xmath169 , the asymptotics of @xmath237 was also obtained in @xcite with a special setting of parameters . the argument can apply here for the general settings by a modification . below we only provide the main steps and omit the details . by using the property that @xmath359 for arbitrary @xmath360 , we can rewrite the determinant as @xmath361 with the kernel @xmath362 where @xmath363 with @xmath364 and @xmath365 here @xmath271 in is an integer sequence satisfying @xmath366 the existence of such @xmath222 is guaranteed by . moreover , since we assume @xmath30 and @xmath24 are both at most @xmath31 , we have @xmath367 . now we consider the asymptotics of @xmath368 . . then we have @xmath369 where @xmath370 is defined in . if we further assume @xmath330 , the asymptotics of @xmath371 can be obtained by using and @xmath372 where @xmath373 here in the second and third equations of we used the conditions and . thus we have @xmath374 together with lemma [ lm : limits_ind_ftns ] ( a ) , we immediately obtain the asymptotics of @xmath368 when @xmath375 . for the case when @xmath376 , one can show that @xmath368 decays on @xmath377 and grows on @xmath378 exponentially fast as @xmath379 . the proof is similar to the case discussed in @xcite and we do not provide details . the explicit asymptotics is described in the following lemma , which was proved for the special parameters in @xcite . [ lm : estimate_kernel ] ( lemma 8.8 of @xcite ) let @xmath292 be a fixed constant satisfying @xmath293 . a. when @xmath380 and @xmath381 , we have @xmath382 where @xmath383 and @xmath384 is defined by , and the error term @xmath385 in is independent of @xmath252 or @xmath386 . b. when @xmath387 and @xmath388 , we have @xmath389 where @xmath390 and @xmath391 is defined by , and the error term @xmath385 in is independent of @xmath392 or @xmath393 . c. when @xmath360 and @xmath394 , we have @xmath395 or @xmath396 here both error terms @xmath397 are independent of @xmath50 . the lemmas [ lm : limit_nodes ] and [ lm : estimate_kernel ] indicate the following result @xmath398 where @xmath120 is an operator on @xmath298 as defined in for @xmath399 . ] . a rigorous proof needs a uniform bound of the fredholm determinant on the left hand side and an error control when we change the space from @xmath148 to @xmath400 , both of which was considered in @xcite for their choice of parameters . however , the argument holds here as well and hence we omit the details . similar to the previous subsection , we write @xmath242 as @xmath402 . before presenting the result , we need the following two lemmas . [ lm : moments_convergent ] for any fixed positive integer @xmath403 , we have @xmath404_{i , j=1}^m \\ = & \sum_{\xi_1,\cdots,\xi_m\in{s}_{z,{{\rm left}}}}\left.\frac{{{\mathrm d}}}{{{\mathrm d}}y}\right|_{y=\tau^{1/3}x}\det\left[-{\mathcal{k}^{(2)}}_{z;y}(\xi_i,\xi_j)\right]_{i , j=1}^m . \end{split}\ ] ] [ lm : uniform_bound_difference_determinant ] there exists some constants @xmath405 and @xmath406 which do not depend on @xmath127 , such that for all positive integer @xmath403 we have @xmath407_{i , j=1}^m\right| \le 2mc^m\ ] ] for all @xmath408 . we first assume both lemmas hold . by using the dominated convergence theorem and the two lemmas above , we have @xmath409_{i , j=1}^m.\ ] ] moreover , the right hand side is uniformly bounded . this further implies @xmath410 is well defined and uniformly bounded . the above result thus can be written as @xmath411 uniformly on @xmath127 . now we prove lemmas [ lm : moments_convergent ] and [ lm : uniform_bound_difference_determinant ] . recall the definition of @xmath412 in . it is easy to check that @xmath413_{i , j=1}^m\\ = & \sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}\delta_k\det\left [ \frac{h_{2;k,{\ell}}(u_i)}{(u_i - v_i)(u_j - v_i)h_{2;k,{\ell}}(v_i)}\right]_{i , j=1}^m\\ = & \sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}\left(\prod_{i=1}^m\frac{(u_i+1)v_i}{(v_i+1)u_i}-1\right)\det\left [ \frac{h_{2;k,{\ell}}(u_i)}{(u_i - v_i)(u_j - v_i)h_{2;k,{\ell}}(v_i)}\right]_{i , j=1}^m . \end{split}\ ] ] here we emphasize the parameters in the function @xmath368 to avoid confusion . hence we have @xmath414_{i , j=1}^m\\ = & \sum_{\substack{u_1,\cdots , u_m\in{r}_{{\mathbf{z}},{{\rm left}}}\\ v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}}\sqrt{1-\rho}n^{1/2}\left(\prod_{i=1}^m\frac{(u_i+1)v_i}{(v_i+1)u_i}-1\right)\det\left [ \frac{h_{2;k,{\ell}}(u_i)}{(u_i - v_i)(u_j - v_i)h_{2;k,{\ell}}(v_i)}\right]_{i , j=1}^m . \end{split}\ ] ] note that there are only @xmath415 terms in the summation since @xmath416 , and when @xmath417 or @xmath418 for some @xmath419 the summand is exponentially small ( see lemma [ lm : estimate_kernel ] ) . therefore we can restrict the summation on all @xmath420 and @xmath421 s of at most @xmath422 distance to @xmath423 . we write @xmath424 and @xmath425 , where @xmath426 . then by applying lemma [ lm : estimate_kernel ] we have @xmath427_{i , j=1}^m\\ & + o(e^{-cn^{\epsilon/2 } } ) , \end{split}\ ] ] where the summation is over all possible @xmath428 and @xmath429 such that @xmath426 and @xmath430 and @xmath431 for all @xmath432 . and @xmath433 for @xmath386 and @xmath393 satisfying @xmath434 and @xmath435 . recall that the error terms in are all uniformly on @xmath428 and @xmath11 ( see lemma [ lm : estimate_kernel ] ) , and note that there are at most @xmath436 elements by lemma [ lm : limit_nodes ] part ( c ) . therefore equals to @xmath437_{i , j=1}^m + o(n^{(m+2)\epsilon-1/2}).\ ] ] now by using lemma [ lm : limit_nodes ] we know that these @xmath428 and @xmath429 s are chosen from a perturbation of @xmath438 and @xmath439 , the images of @xmath300 and @xmath305 respectively , see lemma [ lm : limit_nodes ] . the perturbation size is uniformly bounded by @xmath440 . similar to the reasoning from to , we can replace by @xmath441_{i , j=1}^m + o(n^{(m+2)\epsilon-1/2}).\ ] ] if we choose @xmath292 small such that @xmath442 , then the above quantity converges to @xmath443_{i , j=1}^m.\ ] ] finally we check that equals to the right hand side of . this follows from the facts that @xmath444 for all @xmath445 , and @xmath446 for all @xmath399 , and that @xmath447 . in this proof , we need the following result . [ claim:01 ] there exist a positive constant @xmath405 and @xmath406 uniformly on @xmath127 such that @xmath448 for all @xmath408 , where @xmath449 and @xmath450 note that @xmath451 is always positive and bounded by @xmath452 uniformly on @xmath148 . on the other hand , @xmath453 and @xmath454 are exponentially small when @xmath380 , @xmath387 are of distance @xmath455 , see lemma [ lm : estimate_kernel ] ( c ) . thus it is sufficient to prove the following inequality @xmath456 on the other hand , it is easy to check that @xmath457 uniformly for all @xmath330 , here @xmath458 is a constant independent of @xmath35 ( recall that @xmath459 depends on @xmath1 ) . we denote @xmath460 then is reduced to @xmath461 using lemmas [ lm : limit_nodes ] and [ lm : estimate_kernel ] , the left hand side of converges to @xmath462 as @xmath463 , where @xmath464 and @xmath465 . the rigorous proof of this convergence is similar to that of lemma [ lm : moments_convergent ] and hence we do not provide details . also it is easy to see that is finite . therefore holds for sufficiently large @xmath35 . now we prove lemma [ lm : uniform_bound_difference_determinant ] . this idea is to express the summand on the left hand side of as a sum of determinants @xmath466 $ ] where @xmath467 has similar structure of @xmath468 in claim [ claim:01 ] , and then apply the hadamard s inequality . the first step is to write @xmath469_{i , j=1}^m= \sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}\det\left[\frac{\sqrt{h_{2;k,{\ell}}(u_i)}\sqrt{h_{2;k,{\ell}}(u_j)}}{(u_i - v_i)(u_j - v_i)h_{2;k,{\ell}}(v_i)}\right]_{i , j=1}^m\ ] ] by using a conjugation , here @xmath470 is the square root function with any fixed branch cut . denote @xmath471 similarly to , we have @xmath472_{i , j=1}^m\\ = & n^{1/2}\sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}\left(\prod_{i=1}^m\frac{(u_i+1)v_i}{(v_i+1)u_i}-1\right)\det\left[h_{k,{\ell}}(u_i , u_j;v_i)\right]_{i , j=1}^m\\ = & \sum_{n=1}^m\sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}n^{1/2}\left(\frac{-\rho(u_n+1)}{(1-\rho)u_n}-1\right)\prod_{i=1}^{n-1}\frac{-\rho(u_i+1)}{(1-\rho)u_i}\det\left[\frac{(1-\rho)v_ih_{k,{\ell}}(u_i , u_j;v_i)}{-\rho(v_i+1)}\right]_{i , j=1}^m\\ & + \sum_{n=1}^m\sum_{v_1,\cdots , v_m\in{r}_{{\mathbf{z}},{{\rm right}}}}n^{1/2}\left(1-\frac{-\rho(v_n+1)}{(1-\rho)v_n}\right)\prod_{i=1}^{n-1}\frac{-\rho(v_i+1)}{(1-\rho)v_i}\det\left[\frac{(1-\rho)v_ih_{k,{\ell}}(u_i , u_j;v_i)}{-\rho(v_i+1)}\right]_{i , j=1}^m\\ = & \sum_{n=1}^m\det\left[a^{(n)}(u_i , u_j)\right]_{i , j=1}^m+\sum_{n=1}^m\det\left[\tilde a^{(n)}(u_i , u_j)\right]_{i , j=1}^m , \end{split}\ ] ] where @xmath473 and @xmath474 it is easy to check that @xmath475 and @xmath476 are bounded by @xmath477 defined in the claim [ claim:01 ] . by hadamard s inequality , we have @xmath478_{i , j=1}^m\right|\le \prod_{i=1}^m\sqrt{\sum_{1\le j\le m}|a^{(n)}(u_i , u_j)|^2}\le \prod_{i=1}^m\sqrt{\sum_{u'\in { r}_{{\mathbf{z}},{{\rm left}}}}|a(u_i , u')|^2}\ ] ] for all distinct @xmath479 . as a result , @xmath480_{i , j=1}^m\right|&\le \sum_{u_1,\cdots , u_m\in{r}_{{\mathbf{z}},{{\rm left}}}}\prod_{i=1}^m\sqrt{\sum_{u'\in { r}_{{\mathbf{z}},{{\rm left}}}}|a(u_i , u')|^2}\\ & = \left(\sum_{u\in{r}_{{\mathbf{z}},{{\rm left}}}}\sqrt{\sum_{u'\in { r}_{{\mathbf{z}},{{\rm left}}}}|a(u , u')|^2}\right)^m\le c^m \end{split}\ ] ] by the claim [ claim:01 ] . similarly we have @xmath481_{i , j=1}^m\right|\le c^m.\]]also note that @xmath482_{i , j=1}^m=0 $ ] if @xmath483 for some @xmath484 . by combing , and we obtain . now we prove theorem [ thm : limit_height_uniform](a ) . by using the estimates , , and , we obtain @xmath485 furthermore , by the discussions below and lemma [ lm : uniform_bound_difference_determinant ] , we know the left hand side of is uniformly bounded on @xmath127 . on the other hand , by using the sterling s formula and , we obtain @xmath486 theorem [ thm : limit_height_uniform ] follows immediately by inserting the above two estimates into . in this appendix , we give some tail bounds related to the function @xmath108 . these estimates are not optimal , however , they are sufficient to show that ( 1 ) @xmath37 is a distribution function , and ( 2 ) the @xmath487-th moments of @xmath488 converges to that of @xmath489 for any finite @xmath487 as @xmath99 , here @xmath490 and @xmath76 are defined in theorem [ thm : limit_height_uniform ] . the second statement follows in the same way as theorem 1 in @xcite . define @xmath492 and @xmath493 where the parameters and notations are the same as in theorem [ thm : limit_height_uniform ] , and we surpass the parameters @xmath491 and @xmath39 in the indices for simplification , and @xmath494 by using theorem [ thm : current_fluctuation ] , it is easy to check @xmath495 where @xmath496 is an point satisfying @xmath497 . ( right tail bound of @xmath505 ) [ prop : properties_gu ] there exist constants @xmath499 , @xmath500 and @xmath501 , such that @xmath506 for all @xmath507 and @xmath508 . , the term @xmath509 in should be replaced by @xmath510 . ] the idea of the proof is to map the periodic tasep to the periodic directed last passage percolation ( dlpp ) . the relation was discussed in @xcite and @xcite and we refer the readers to section 3.1 of @xcite for more details . here we just give a brief description . similarly to the mapping between the infinite tasep and usual dlpp , see @xcite , there is a mapping from periodic tasep to periodic dlpp described as following : let @xmath513 be the period vector , and @xmath514 be a lattice path with lower left corners @xmath515 for @xmath15 . it is easy to check that @xmath514 is invariant if translated by @xmath516 . let @xmath517 be random exponential variables with parameter @xmath12 for all lattice points @xmath518 which are on the upper right side of @xmath514 . we require @xmath519 for all @xmath518 . except for this restriction , all @xmath517 are independent . we then define @xmath520 where the maximum is over all the possible up / right lattice paths from @xmath521 to @xmath518 . we also define @xmath522 now we are ready to introduce the relation between particle location in periodic dlpp and last passage time in periodic dlpp , see ( 3.7 ) in @xcite , @xmath523 where we use the notation @xmath524 to denote the probability functions in periodic tasep and the equivalent periodic dlpp model . using and the relation between height function @xmath525 and the particle location @xmath526 , see , it is a direct to show the following @xmath527 where @xmath528 with @xmath529 the rest of this section is to show that there exist constants @xmath499 , @xmath304 , @xmath530 , and @xmath501 , such that @xmath531 for all @xmath532 and @xmath504 . then proposition [ prop : properties_fu ] follows immediately . the idea to prove is to compare the periodic dlpp with the usual dlpp . this idea was applied in @xcite for periodic tasep in sub - relaxation time scale . in the case we consider in this paper , we need a relaxation time analogous of the argument and hence a modification is needed . we first introduce some known results on dlpp model . the probability space for dlpp is that all the lattice points @xmath518 are associated with an i.i.d.exponential random variable @xmath517 , we use @xmath533 to denote the probability associated to this space . similarly to the periodic dlpp , we denote @xmath534 the last passage time from @xmath521 to @xmath518 , and @xmath535 the last passage time from the lattice path @xmath536 to @xmath518 . finally we define @xmath537 for arbitrary constants @xmath245 satisfying @xmath538 . from now on we fix these two constants @xmath44 and @xmath45 . it is known that @xcite @xmath539 where @xmath540 and @xmath541 . the following tail estimate is also needed , which is due to @xcite , @xmath542 for sufficiently large @xmath543 and @xmath544 satisfying @xmath545 . here @xmath546 , @xmath458 and @xmath547 are constants only depend on @xmath44 and @xmath45 . the last result in dlpp we need is about the transversal fluctuations . define @xmath548 to be the set of all lattice points @xmath549 satisfying @xmath550 we also define @xmath551 to be the maximal path from @xmath521 to @xmath518 in the usual dlpp . the following transversal fluctuation estimate is currently known : there exist constants @xmath552 , @xmath553 and @xmath554 such that @xmath555 for all @xmath556 and and @xmath544 satisfying @xmath557 . the analog of this estimate in poissonian version of dlpp was obtained in @xcite and their idea can be applied in the exponential case similarly . we hence do not provide a proof here , instead we refer the readers to a forthcoming paper @xcite by nejjar for more discussions . now we use and to prove . we pick @xmath558 equidistant points @xmath559 on the line @xmath560 such that @xmath561 here @xmath172 is some large parameter which will be decided later . note that @xmath562 , hence the above inequality is satisfied as long as @xmath172 is greater than certain constant . now note that @xmath563 since @xmath564 is at the upper right side of to the initial contour @xmath514 by definition , and @xmath565 , therefore @xmath566 on the other hand , by using we know that @xmath567 provided @xmath568 . finally , by inserting and then applying , we have @xmath569 provided @xmath570 and @xmath532 , where @xmath571 and @xmath405 are constants . by combing , and , we obtain @xmath572 finally we pick @xmath573 and follows immediately . now we assume @xmath240 is large and pick @xmath127 on the following circle @xmath575 we have @xmath576 provided @xmath577 . by tracking the error terms , the term @xmath578 is analytic in @xmath127 and can be expressed as @xmath579 with @xmath580 both bounded by @xmath578 . and similarly to section [ sec : asymptotics_determinant ] , we write @xmath581 as @xmath361 whose kernel is defined in . by a similar argument as lemma [ lm : estimate_kernel ] , one can show that the kernel decays exponentially @xmath582 for all @xmath583 and sufficiently large @xmath240 . here @xmath304 is a constant . the heuristic argument is as following : suppose @xmath584 with @xmath585 , then @xmath586 by . it is a direct to show that the leading term in the exponent of @xmath587 in lemma [ lm : estimate_kernel ] ( a ) ( after dropping the term @xmath588 , whose real part is independent of @xmath386 and hence cancels with the counterpart from @xmath589 ) is @xmath590 similar estimates for the @xmath589 in lemma [ lm : estimate_kernel ] ( b ) holds . therefore we have . finally , by using and , it is a direct to prove that @xmath591 for a different positive constant @xmath315 . since the above argument is similar to that in section [ sec : asymptotics_determinant ] , we do not provide details . m. prhofer and h. spohn . current fluctuations for the totally asymmetric simple exclusion process . in _ in and out of equilibrium ( mambucaba , 2000 ) _ , volume 51 of _ progr . _ , pages 185204 . birkhuser boston , boston , ma , 2002 .

in this paper we study the breakdown of normal hyperbolicity and its consequences for reaction dynamics ; in particular , the dividing surface , the flux through the dividing surface ( ds ) , and the gap time distribution . our approach is to study these questions using simple , two degree - of - freedom hamiltonian models where calculations for the different geometrical and dynamical quantities can be carried out exactly . for our examples , we show that resonances within the normally hyperbolic invariant manifold may , or may not , lead to a ` loss of normal hyperbolicity ' . moreover , we show that the onset of such resonances results in a change in topology of the dividing surface , but does not affect our ability to define a ds . the flux through the ds varies continuously with energy , even as the energy is varied in such a way that normal hyperbolicity is lost . for our examples the gap time distributions exhibit singularities at energies corresponding to the existence of homoclinic orbits in the ds , but these singularities are not associated with loss of normal hyperbolicity .

1304.3947

transition state theory has played , and continues to play , a fundamental role in how we think of chemical reactions . there are many excellent reviews of the subject ; see , for example , refs . the work of wigner @xcite highlighted the notion of the transition state as a dividing surface ( ds ) in phase space , having the property that trajectories crossed this surface ( only once ) in their evolution from reactants to products . this phase space point of view was relatively undeveloped for many years until the 1970 s when such a dynamical theory was fully realised for two degrees of freedom ( dof ) in the work of pechukas , pollak and child @xcite . in their work they realised the construction of such a ds having the ` no recrossing ' property , and a central role was played by a saddle - like ( in terms of stability ) periodic orbit ( po ) which serves as the ` anchor ' for the definition of the ds . this po was called a periodic orbit dividing surface ( pods ) . ( this choice of name was perhaps unfortunate as the po itself is not the ds [ in phase space ] but rather the _ boundary _ of the ds . ) the work of pechukas , pollak and child just cited was the first example of the use of a normally hyperbolic invariant manifold ( nhim ) in the context of transition state theory ( although see also ref . ) . for more than two dof a suitable generalisation of a saddle like ( or ` hyperbolic ' ) periodic orbit was required , and this was realised in the notion of a normally hyperbolic invariant manifold ( nhim ) . explicit discussion of nhims in the context of reaction theory was in refs ( see also refs ) . however , the usefulness of nhims for the definition of phase space dividing surfaces was fully realised only after the development of methods for computing nhims and their associated ds . in particular , nhims could be computed using the classical poincar - birkhoff normal form procedure , and its quantum analog @xcite . however , the poincar - birkhoff normal form procedure is local in nature since it involves a taylor expansion about an appropriate saddle point . practically , the expansion must be truncated at an appropriate order so that its ability to describe geometrical structures and dynamics at a desired accuracy must be assessed ( and this is , typically , problem dependent ) . very broadly speaking , the nhim , and its associated local dynamics , are accurately described by the truncated normal form for energies ` close ' to the energy of the saddle point . to date , our knowledge of nhims for specific problems relies heavily on the techniques used to compute them , in particular normal form theory . the local nature of these techniques makes nhims tools of limited utility in terms of our ability describe reaction dynamics at energies well above the reaction threshold . in particular , very little is known about how the nhim behaves as we increase the energy above that of the saddle . we expect the normal form to ` break down ' ( in some not very precisely defined sense ) . however , the breakdown of the normal form and the breakdown of the nhim are two separate issues . questions concerning the existence and properties of the nhim can in principle be considered independently of the methods used to compute the nhim , when normal form approximations may no longer apply . in this paper we study the behaviour of nhims as the energy above the saddle point is increased , and the dependence on energy of associated quantities related to reaction dynamics . our strategy will be to avoid the use of normal forms by considering simple models that allow us to explicitly compute the nhim , the ds , and the geometrical properties associated with reaction dynamics . we can view our approach as a decoupling of normal form issues from the dynamics of reaction . we then ask the question : `` what happens to a nhim as parameters ( the energy being a very important parameter ) are varied '' ? in other words , we study ` bifurcation of nhims ' . but , more specifically for reaction dynamics , we want to understand how such bifurcation of nhims affects the various quantities ( described below ) used to describe reaction dynamics . first , we need to describe some background regarding what we mean by the ` bifurcation of a nhim ' and then discuss our approach for studying these phenomena . roughly speaking ( for details , see section [ sec : nhimmodels ] ) a normally hyperbolic invariant manifold has the property that the _ linearized _ growth rates normal to the manifold dominate the _ linearized _ growth rates tangent to the manifold . the nature of the dynamics within the invariant manifold can be arbitrary , provided the growth rate conditions are satisfied . hence , there are several possibilities to consider : * bifurcation within the nhim , but with the normal hyperbolicity conditions not affected . * breakdown of the normal hyperbolicity conditions . * a combination of the above two phenomena bifurcation within the nhim and breakdown of the normal hyperbolicity conditions . how bifurcation within the nhim occurs and how the normal hyperbolicity conditions can break down will be discussed in detail when we consider explicit models in section [ sec : models ] . briefly , there are two ways in which such bifurcations can occur . one is when a parameter in the hamiltonian is varied , and the other is when the energy itself is varied . ( we will be considering only hamiltonian systems . ) of course , energy is a parameter , but its importance in understanding the dynamics is such that we highlight its effects explicitly . typically in bifurcation theory simple ` normal forms ' are developed that embody the essential features of a particular bifurcation . we will follow a similar approach here . our models will be two degree - of - freedom , separable , hamiltonian systems where we have explicit control and understanding of the dynamics in such a way that we can model the three possibilities noted above . modelling the three possibilities for separable , two dof hamiltonian systems is of course fairly straightforward , but that is the point . our goal is to determine the effect that the three scenarios outlined above have on quantities that are used to describe reaction dynamics . in particular , we want to consider the effect on : * the dividing surface ( ds ) separating reactants and products as a function of energy ( the ds is defined and explained in section [ subsec : ds ] ) . * the flux through the ds as a function of energy . * the gap time distribution as a function of energy ( these quantities are defined and explained in section [ sec : volumes ] ) . despite the fact that our examples are very simple ( completely integrable 2 dof hamiltonian systems ) , they allow for exact calculations and therefore provide benchmarks against which more complex examples can be compared . in particular , we note a previous study by li and coworkers @xcite which suggests that the breakdown of normal hyperbolicity of the nhim is responsible for the non definability of a relevant ds at high energy above the reaction threshold . this conclusion is contrary to that reached in the study of our model examples ( see below ) . a study of allahem and bartsch @xcite asserts that the transition state loses , and then regains normal hyperbolicity as energy is varied . a study of inarrea et al @xcite concludes that even though bifurcation of the nhim can occur , dividing surfaces can still be defined . these three studies are concerned with hamilltonian models that are much more complex than ours ( although still 2 dof ) . we also note an interesting study by yang @xcite , although the models studied there are _ not _ hamiltonian . the organisation of this paper is as follow . we review the concepts of nhims in section [ sec : nhimmodels ] where we discuss the definition of normal hyperbolicity in detail . in section [ sec : gen_models ] we discuss the general form for the two dof models that we study . these models allow us to explicitly compute nhims and quantify their geometrical and stability properties . in section [ sec : volumes ] we discuss concepts of phase space volumes , gap times , and reaction rates that are relevant for the present study . in section [ sec : models ] we describe the two models for which we will carry out explicit calculations . in section [ subsec : ds ] we discuss dividing surfaces and compute the flux through the dividing surfaces for our two examples , while section [ subsec : gapdistrib ] deals with the computation of the gap time distribution . in section [ subsecnormalhyperbol ] we discuss the loss of normal hyperbolicity and its consequences for the ds , flux , and gap time distribution in our examples . in section [ sec : concl ] we present our conclusions . the notion of a _ normally hyperbolic invariant manifold _ ( or nhim ) is by now a standard concept and tool in dynamical systems theory that is playing an increasingly important role in a variety of applications . the theoretical framework developed over the course of many years , beginning in the early part of the 20th century and reached a mature form in the works of fenichel @xcite and hirsch , pugh , and shub @xcite . in ref . a ( relatively ) elementary exposition of fenichel s approach to nhims is given with some discussion of applications . we begin by defining the notion of a normally hyperbolic invariant manifold ( nhim ) in a continuous time setting ( i.e. the setting of autonomous ordinary differential equations , rather than maps ) . there are numerous definitions that appear throughout the literature . we take the definition from ref . . from an ` applied ' point of view the definition may appear somewhat abstract , leading to difficulties understanding how the definition can be applied in concrete settings . however , this level of abstraction actually provides a great deal of flexibility for applications , as we shall see when we consider explicit models for illustrating different types of bifurcation phenomena associated with nhims and reaction dynamics . it is sufficient for our purpose to take phase space to be @xmath0 , which is a @xmath1 differentiable manifold . for inner products and norms on this space we will take the usual euclidean structures . phase space is taken to be even dimensional since we will be considering canonical hamiltonian systems , i.e. we consider equations of the following form : [ eq : ham ] @xmath2 for some function @xmath3 . we require the hamiltonian @xmath4 to be at least @xmath5 since we will assume the flow that it generates is at least @xmath6 . in practice , this is not a restriction since the explicit hamiltonians that we consider will be infinitely differentiable . however , in the theoretical description of nhims , it is important to explicitly specify the degree of differentiability of all geometrical objects for reasons that will become apparent as we develop the discussion of nhims further . note that in this work we will have no need to consider @xmath0 as a symplectic vector space , or to investigate the associated consequences of symplecticity ; this is likely to be a very fruitful avenue for future research . the reason is the following . roughly , in hamiltonian dynamical systems stability properties are divided into ` elliptic ' and ` hyperbolic ' , where bifurcation _ may _ occur when passing through the boundary between stability types . results describing ` elliptic stability ' ( e.g. the kolmogorov - arnold - moser ( kam ) and nekhoroshev theorems ) rely heavily on the hamiltonian structure for their proof . ` hyperbolic stability ' results , on the other hand , do not generally rely on the hamiltonian structure , and that is true for the development of nhims . it would nevertheless be interesting to know what restrictions hamiltonian structure places on the structure of nhims . for example , in ref . it is proven that invariant tori in canonical , time independent , hamiltonian systems _ can not _ be normally hyperbolic . we suppose that @xmath7 is a @xmath6 smooth flow defined on @xmath0 generated by . we will also need assume the flow to exist for all time , i.e. @xmath8 . as phase space is not compact , we will need to explicitly discuss the issue of existence of the flow for all time in our examples . we now state the definition of a nhim sufficient for our needs . an @xmath9 dimensional submanifold , @xmath10 of @xmath0 ( @xmath11 ) is said to be a normally hyperbolic invariant manifold for @xmath12 if @xmath10 is invariant under @xmath12 and there exists a splitting of the tangent bundle of @xmath13 into sub - bundles : @xmath14 having the following properties : invariance . : : the sub - bundles are invariant under @xmath15 , for all @xmath16 , @xmath8 , i.e. , + @xmath17 growth rates . : : there exists a constant @xmath18 and rates @xmath19 such that for all @xmath20 we have : + @xmath21 [ def : nhim ] we now provide some additional background related to the concepts in this definition . invariance . : : invariance of @xmath10 under @xmath12 means that for any @xmath20 , @xmath22 for all @xmath8 . we have _ not _ required @xmath10 to be boundaryless . making the assumption that @xmath10 has no boundary is usual when considering invariance of @xmath10 , since otherwise one would need to make some sort of assumption preventing trajectories from leaving @xmath10 by crossing the boundary of @xmath10 . we have _ not _ required @xmath10 to have no boundary since that is not natural for the examples we will consider . since we will be considering only autonomous hamiltonian systems the fact that trajectories are prevented from leaving by crossing the boundary of @xmath10 is ensured by constancy of the hamiltonian function , i.e. energy conservation . rates . : : note the rates @xmath23 . these , along with , , and encapsulate the idea of _ normal _ hyperbolicity , in the sense that the rate of growth or decay of tangent vectors _ normal _ to @xmath10 under the linearized dynamics dominate the rate of growth or decay of vectors tangent to @xmath10 under the linearized dynamics . note that vectors tangent to @xmath10 can still exhibit exponential growth or decay , but the _ linearized _ rate must be smaller than that normal to @xmath10 . energy conservation . : : the level set of the hamiltonian , @xmath24 is a @xmath25 dimensional surface ( except at possible singular points of the surface ) that is invariant under the flow generated by hamilton s equations . if it is compact and boundaryless , then the flow exists for all time . generically @xcite the @xmath26 dimensional manifold @xmath10 will intersect in a @xmath27 dimensional manifold , which we refer to as @xmath28 . if @xmath10 is normally hyperbolic , then @xmath29 is also normally hyperbolic . thus , for hamiltonian systems we will be considering both iso - energetic and non iso - energetic nhims . : : we will assume that : @xmath30 and it follows from the conditions of definition [ def : nhim ] that @xmath31 , @xmath32 and @xmath33 vary continuously with respect to @xmath34 , and therefore the dimensions of @xmath31 , @xmath32 and @xmath33 are constant with respect to @xmath34 . we also have : @xmath35 it follows from that : @xmath36 hence , @xmath9 is even . the need for bundles : linearization characteristics . : : the ` bundle ' concept and terminology , while perhaps unfamiliar , is necessary in our setting since we are considering the linearization about an invariant manifold , which is generally filled with different orbits ( as opposed to the more familiar situation of linearizing about a single orbit , such as an equilibrium point or periodic orbit ) . consider the unstable bundle , for example . invariance of this bundle under the linearized flow means that @xmath37 . this relation provides an unstable growth direction all along the orbit through @xmath34 , i.e. all along @xmath38 , and enables us to quantify the unstable growth along a given orbit . the ( disjoint ) union of all of the unstable subspaces over all points in @xmath10 gives the unstable bundle over @xmath10 . similar consideration hold for the stable bundle over @xmath10 and the tangent bundle of @xmath10 . more details can be found in the general references for nhims given earlier . [ [ existence - of - stable - and - unstable - manifolds . ] ] existence of stable and unstable manifolds . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as we have previously mentioned , the definition of nhim given in definition [ def : nhim ] specifies conditions on the _ linearized _ dynamics about @xmath10 . these provide _ sufficient _ conditions for proving results about the nonlinear dynamics . in particular , the sets of points having the following properties : @xmath39 can be shown to be @xmath40 manifolds ( for some constant @xmath41 ) , where @xmath42 , where , we recall , @xmath43 is the degree of differentiability of the flow . moreover , under the conditions of definition [ def : nhim ] , @xmath10 can be shown to be a @xmath44 invariant manifold . however , we emphasise again that the conditions of definition [ def : nhim ] provide _ sufficient _ conditions ( in terms of linearized dynamics ) for the smoothness properties of the manifolds . for example , in the models that we consider in section [ sec : models ] , @xmath10 is explicitly known and is infinitely differentiable , regardless of the value of the ratio @xmath45 . similarly , the flow will be infinitely differentiable in the models that we consider in section [ sec : models ] . [ [ persistence - under - perturbation . ] ] persistence under perturbation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + one of the most important properties of nhims is that they persist under @xmath46 perturbations . it is important to clearly define the term ` perturbation ' . for example , we are considering time independent hamiltonian systems . time - dependent erturbations would not satisfy the hypotheses of the perturbation theorem , unless the time - dependence could be recast in a way that the system became time independent , and the resulting system satisfies the hypotheses of the perturbation theorem . time dependence that is periodic or quasi periodic can be treated in this way , while more general time dependence results in a breakdown of compactness , leading to problems satisfying the hypotheses of the theorem . [ [ sufficient - conditions - and - breakdown - of - normal - hyperbolicity ] ] sufficient conditions and ` breakdown of normal hyperbolicity ' + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + definition [ def : nhim ] is a definition of a nhim . it _ can not _ be concluded that if the conditions of the definition are not satisfied , such as might occur for certain parameter values in a parameterized system , that the nhim undergoes a ` bifurcation ' . loss of hyperbolicity is a _ necessary , but not sufficient _ condition for bifurcation to occur , but the nature of a bifurcation depends on nonlinearity , which is outside the standard requirements of normal hyperbolicity as stated in definition [ def : nhim ] . we now consider the general form of a model that encompasses all of the specific examples that we will study . we consider a two dof hamiltonian system of the following form : @xmath47 with associated hamiltonian vector field : [ gen_model_2 ] @xmath48 note the particularly simple form of eqs . it has the form of two separable one degree - of - freedom hamiltonian systems ( hence it is completely integrable ) . the flow generated by has the general form : @xmath49 the equation for the ( typically ) three - dimensional energy surface is given by : @xmath50 we make the following assumption on the @xmath51 component of . at @xmath52 the system : [ gen_model_5 ] @xmath53 has a hyperbolic equilibrium point . the ( positive ) eigenvalue of the matrix associated with the linearization of about this equilibrium is @xmath54 . [ hyp_assump ] clearly , the set @xmath55 is a two dimensional invariant manifold in the four dimensional phase space . the intersection of the 3d energy surface with this 2d invariant manifold is given by : @xmath56 which is ( typically ) a one dimensional level set of @xmath57 ; an isoenergetic invariant manifold . now we will show that , under assumption [ hyp_assump ] , @xmath10 and @xmath28 are both normally hyperbolic invariant manifolds . we therefore need to show that definition [ def : nhim ] holds for @xmath10 . we begin by computing the linearization of the flow of about an arbitrary point on @xmath10 , which is denoted by @xmath58 . first , we transform @xmath59 to a set of coordinates that facilitates the computations . it follows from assumption [ hyp_assump ] that there exists a linear , symplectic transformation of coordinates , @xmath60 , such that the quadratic part of @xmath61 is diagonal in these coordinates , i.e. @xmath62 where @xmath63 is @xmath64 ( note : there are no linear terms in @xmath65 since @xmath66 is an equilibrium point ) . in these coordinates we can write the hamiltonian in the form : @xmath67 in these coordinates , the linearisation of the flow has the the following block diagonal form : @xmath68 where @xmath69 @xmath70 denotes the @xmath71 matrix of zeros , and we assume that we have transformed the @xmath51 coordinates of the lower right hand @xmath71 block so that the flow assumes the diagonal form , which follows from assumption [ hyp_assump ] . a general tangent vector in @xmath32 , where we use the shorthand notation @xmath72 , has the form @xmath73 then we have : @xmath74 which is a tangent vector in @xmath75 , where @xmath76 . since this argument holds for any @xmath20 and any @xmath77 it follows that the bundle @xmath78 is invariant ( i.e. holds ) . moreover , it follows from that the growth rate condition holds for all vectors in @xmath78 . a similar argument for the invariance of @xmath79 and the growth rate of vectors in @xmath79 follows . now we turn our attention to the tangent bundle of @xmath10 . a general vector in @xmath31 has the form : @xmath80 then using we have : @xmath81 which is in @xmath82 , where @xmath76 . here , @xmath83 and @xmath84 are functions of @xmath85 , but the explicit functional form is not important for this argument . since this argument holds for any @xmath20 , @xmath86 , it follows that @xmath87 is invariant . what we can _ not _ verify at this point is the growth rate condition . this will require explicit conditions on the flow on @xmath10 . the flow on @xmath10 can be very general , but normal hyperbolicity will require that is satisfied by this flow . this condition will be considered in the specific examples that we analyse in section [ sec : models ] . in this section we briefly review the concepts from classical reaction rate theory that are relevant to the present study . this section is adapted from the paper of collins et al@xcite where more background and details can be found . points in the @xmath88-dimensional system phase space @xmath89 are denoted @xmath90 . the system hamiltonian is denoted by @xmath91 , and the @xmath92 dimensional energy surface at energy @xmath93 , @xmath94 , is denoted @xmath95 . the corresponding microcanonical phase space density is @xmath96 , and the associated density of states for the complete energy surface at energy @xmath93 is @xmath97 the disjoint regions of phase space ( ` reactant ' , @xmath98 , and ` product ' , @xmath99 ) separated by the phase space dividing surface @xmath100 are denoted @xmath101 ; the region of phase space corresponding to @xmath102 will be denoted by @xmath103 , and that corresponding @xmath98 will be denoted by @xmath104 . the microcanonical density of states for points in region @xmath103 is @xmath105 with a corresponding expression for the density of states @xmath106 in @xmath107 . since the flow is everywhere transverse to @xmath108 , those phase points in the region @xmath103 that lie on trajectories that cross @xmath108 can be specified uniquely by coordinates @xmath109 , where @xmath110 is a point on @xmath111 , specified by @xmath112 coordinates @xmath113 , and @xmath114 is a time variable . the point @xmath115 is reached by propagating the initial condition @xmath110 forward for time @xmath114 @xcite . as all initial conditions on @xmath111 ( apart from a set of trajectories of measure zero lying on stable manifolds ) will leave the region @xmath103 in finite time by crossing @xmath116 , for each @xmath110 , we can define the _ gap time _ @xmath117 , which is the time it takes for the trajectory to traverse the region @xmath103 before entering the region @xmath104 . that is , @xmath118 . for the phase point @xmath115 , we therefore have @xmath119 . the coordinate transformation @xmath120 is canonical @xcite , so that the phase space volume element is @xmath121 with @xmath122 an element of @xmath112 dimensional area on ds(e ) . the magnitude @xmath123 of the flux through dividing surface @xmath111 at energy @xmath93 ( ` directional flux ' ) is given by @xmath124 where the element of area @xmath125 is precisely the restriction to ds(e ) of the appropriate flux @xmath112-form @xmath126 corresponding to the hamiltonian vector field associated with @xmath91 @xcite . the reactant phase space volume occupied by points initiated on the dividing surface with energies between @xmath93 and @xmath127 is therefore @xcite [ vol_1 ] @xmath128 where the _ mean gap time _ @xmath129 is defined as @xmath130 and is a function of energy @xmath93 . the _ gap time distribution _ , @xmath131 is of central interest in unimolecular kinetics @xcite : the probability that a phase point on @xmath111 at energy @xmath93 has a gap time between @xmath132 and @xmath133 is equal to @xmath134 . an important idealized gap distribution is the random , exponential distribution @xmath135 characterized by a single decay constant @xmath43 ( where @xmath43 depends on energy @xmath93 ) , with corresponding mean gap time @xmath136 . an exponential distribution of gap times is usually taken to be a necessary condition for ` statistical ' behavior in unimolecular reactions @xcite . the lifetime ( time to cross the dividing surface @xmath116 ) of phase point @xmath115 is @xmath137 , and the corresponding ( normalized ) reactant lifetime distribution function @xmath138 at energy @xmath93 is @xcite [ life_1 ] @xmath139 where the fraction of interesting ( reactive ) phase points having lifetimes between @xmath140 and @xmath141 is @xmath142 . it is often useful to work with the unnormalized lifetime distribution @xmath143 , where @xmath144 . equation gives the general relation between the lifetime distribution and the fraction of trajectories having lifetimes greater than a certain value for arbitrary ensembles @xcite . note that an exponential gap time distribution implies that the reactant lifetime distribution @xmath138 is also exponential @xcite ; both gap and lifetime distributions for realistic molecular potentials have been of great interest since the earliest days of trajectory simulations of unimolecular decay , and many examples of non - exponential lifetime distributions have been found @xcite . in this section we describe our two model systems . each model has nhims , and we describe the normal hyperbolicity properties of the nhims as a function of energy and other parameters . we then compute the flux through the associated ds as well as the gap time distribution , and we discuss the behaviour of these quantities in relation to the property of normal hyperbolicity . the two models to be studied are particular examples of the general class of models described in section [ sec : gen_models ] . our first example will consist of an integrable system whose hamiltonian is the sum of the hamiltonian of a simple pendulum and the hamiltonian of a symmetric double - well : @xmath145 with associated equations of motion : [ hamvf_ex1 ] @xmath146 the phase space structure for these two , uncoupled , one degree - of - freedom hamiltonian systems is illustrated in fig . [ fig : ex1_1 ] . since the trajectories of can be solved analytically , for any initial condition , it can be seen explicitly that the flow generated by exists for all time . following , the two dimensional non - isoenergetic surface : @xmath147 is an invariant manifold for . for @xmath148 , tangent vectors normal to @xmath10 experience exponential growth and decay under the linearised dynamics . however , we can not claim that it is a nhim according to definition [ def : nhim ] . whether or not this is in fact the case depends on the dynamics on @xmath10 . following , the intersection of the three dimensional energy surface with is given by : @xmath149 hence , the dynamics on @xmath10 is described by the dynamics of a ( symmetric ) two well potential , and an orbit of the pendulum with energy @xmath93 defines @xmath28 . we note that : * @xmath150 is an elliptic equilibrium point of the pendulum with energy @xmath151 . * surrounding the elliptic fixed point is a family of periodic orbits ( ` librations ' ) whose energies increase monotonically from @xmath151 to @xmath152 . * @xmath153 is a saddle point with energy @xmath154 that is connected by a pair of homoclinic orbits . * outside the pair of homoclinic orbits are two families of periodic orbits ( ` rotations ' ) whose actions increase monotonically with energy , from @xmath152 . the phase space structure associated with the simple pendulum on @xmath10 is shown in fig . [ fig : ex1_1 ] . we note that homoclinic orbits have zero frequency . hence , for one dof hamiltonian systems , they are examples of the phenomenon of _ resonance_. the presence of homoclinic orbits therefore allows us to examine the role of resonance in the breakdown of normal hyperbolicity explicitly . in our model we take ` reaction ' to correspond to a change of sign of the coordinate @xmath155 . hence , the energy of reacting trajectories in the @xmath156 coordinates must be greater than that of the saddle point at @xmath52 , i.e. , the energy must be greater than zero . hence , the total energy must be such that the energy in the @xmath51 subsystem is greater than zero . the second example consists of two uncoupled symmetric double - well hamiltonians : [ eq : ex2_1 ] @xmath157 with associated hamiltonian vector field : [ hamvf_ex2 ] @xmath158 again , the trajectories of can be solved analytically for any initial condition , and the flow generated by exists for all time . the discussion of the phase space structure is similar to that for example 1 above . following , the two dimensional non - isoenergetic surface : @xmath159 is an invariant manifold for . for @xmath148 , tangent vectors normal to @xmath10 experience exponential growth and decay under the linearised dynamics . again , whether or not @xmath10 is a nhim according to definition [ def : nhim ] depends on the dynamics on @xmath10 . following , the intersection of the three dimensional energy surface with is given by : @xmath160 hence , the dynamics on @xmath10 is described by the dynamics of a symmetric two well potential , and an orbit of the symmetric two well potential with energy @xmath93 defines @xmath28 . we note that : * @xmath150 is a saddle point of the symmetric two well potential with energy @xmath161 . * @xmath162 are elliptic equilibrium points with energy @xmath163 . * surrounding each elliptic equilibrium point is a family of periodic orbits whose energies increase monotonically away from the equilibrium point . at @xmath164 the periodic orbits merge into a pair of homoclinic orbits that connect the saddle point at the origin . * as the energy increases from zero there is a family of ( symmetric ) periodic orbits that surround both potential wells whose actions increase monotonically with energy . the phase space structure associated with the symmetric double well on @xmath10 is shown in fig . [ fig : ex2_1 ] . as in the previous model , ` reaction ' is associated with a change of sign of the @xmath155 coordinate . hence , the energy of reacting trajectories in the @xmath51 coordinates must be greater than the energy of the saddle point at @xmath52 , i.e. the energy must be greater than zero . the total energy must therefore be such that the energy in the @xmath51 component of the model is greater than zero . for each of our model hamiltonians we can construct a dividing surface in phase space having the ( local ) no - recrossing property . the codimension one non - isoenergetic surface defined by @xmath165 ( locally ) divides the phase space into two regions : one associated with the regions defined by @xmath166 and the other associated with the region defined by @xmath98 . the dividing surface restricted to a three dimensional fixed energy surface @xmath167 is given by : @xmath168 this dividing surface has two halves : @xmath169 and @xmath170 these two halves meet at an invariant manifold : @xmath171 the nature of this invariant manifold ( especially the normal hyperbolicity property ) depends on @xmath93 , the dynamics on @xmath28 , and a comparison of the linearised growth rates normal and tangent to @xmath93 , as we will see for our examples . in particular , we will examine the inter - relation between these characteristics . we now describe conditions under which @xmath172 and @xmath173 are surfaces having the no ( local ) re - crossing property ( and we will verify these conditions in our particular examples ) . these surfaces are defined by @xmath165 . therefore points on these surfaces leave if : @xmath174 and therefore it follows immediately from their definition that and have the ( local ) ` no - recrossing ' property . we denote the directional flux across these hemispheres by @xmath175 and @xmath176 , respectively , and note that @xmath177 . the magnitude of the flux is @xmath178 . the magnitude of the flux and related quantities are central to the theory of reaction rates , as we described in section [ sec : volumes ] . [ [ section ] ] in figure [ figdsex1 ] we plot the dividing surface ( i.e. ) for three different energies : @xmath179 ( energy below the separatrices of the pendulum ) , @xmath180 ( energy equal to the energy of the separatrices of the pendulum ) , and @xmath181 ( energy larger than the energy of the separatrices of the pendulum ) . we see that the ds undergoes a bifurcation as the energy passes through the energy of the separatrix . a central question is whether or not this bifurcation has any effect on quantities that are important for quantifying reaction dynamics , such as flux and gap times . + [ [ section-1 ] ] by stokes theorem , the directional flux across half the ds , at energy @xmath93 , is the area enclosed by the invariant manifold @xmath28 on the ( @xmath182 , @xmath183 ) plane . this area is just the integral @xmath184 and is related to the action variable of the simple pendulum by @xmath185 so that we have the relation @xmath186 . for the simple pendulum there are three different cases to consider for the calculation of the action integral : the librations , the separatrices , and the rotations . the computation of these integrals for the different cases is reviewed in the appendix [ appendixpendulum ] . here we just note the results : [ pendulumactions ] @xmath187 \label{librationaction } \\ j_s ( e ) & = \frac{4\alpha_1}{\pi } \label{separatrixaction } \\ j_r ( e ) & = \frac{4\alpha_1}{k_r \pi } \mathcal{e}_1 ( k_r ) \label{rotationaction}\end{aligned}\ ] ] where the subscripts @xmath188 , @xmath132 and @xmath189 denote libration , separatrix and rotation , respectively . @xmath190 and @xmath191 denote the complete elliptic integrals of the first and second kind , respectively and @xmath192 and @xmath193 are the moduli of the complete elliptic integrals , and are functions of the energy @xmath93 : [ pendulumactions_2 ] @xmath194 \label{librationmodulus } \\ k_r ( e ) & = \sqrt { \frac{2\alpha_1 ^ 2}{e + \alpha_1 ^ 2 } } \label{rotationmodulus}\end{aligned}\ ] ] with these expressions in hand we obtain expressions for the directional flux @xmath123 across half the ds : [ pendulumflux ] @xmath195 \label{librationflux } \\ \phi_s ( e ) & = 16\alpha_1 \label{separatrixflux } \\ \phi_r ( e ) & = \frac{16\alpha_1}{k_r } \mathcal{e}_1 ( k_r ) \label{rotationflux}\end{aligned}\ ] ] for the separatrix and the rotation cases an extra factor 2 appears because one must count the area enclosed by the curves having positive and negative momentum . the graph of the flux @xmath123 is shown in figure [ figfluxpendulum ] . [ [ section-2 ] ] in figure [ figdsex2 ] we plot the dividing surface ( i.e. , the surface ) for three different energies : @xmath196 ( energy below the separatrices of the two well potential ) , @xmath161 ( the energy of the separatrices of the two well potential ) , and @xmath181 ( energy larger than the energy of the separatrices of the two well potential ) . as for example 1 , the ds undergoes a bifurcation as the energy passes through the energy of the separatrix . + [ [ section-3 ] ] as for example 1 , the directional flux across half the ds is the area enclosed by the invariant manifold @xmath28 on the ( @xmath182 , @xmath183 ) plane . this area is the integral @xmath197 and is related to the action variable of the symmetric double - well by @xmath186 . for the symmetric double - well there are three different cases for the calculation of the action integral depending on the energy being less , greater or equal to the energy of the saddle point . the integrals for the different cases are ( cf . appendix [ appendixdoublewell ] ) : [ dwactions ] @xmath198 \label{dw1action } \\ j_2 ( e ) & = \frac{2\alpha_2^{3/2}}{3\pi } \label{dw2action } \\ j_3 ( e ) & = \frac{\sqrt{2\delta } b^2}{3k_3 ^ 2 \pi } \left [ ( 2k_3 ^ 2 - 1 ) \mathcal{e}_1 ( k_3 ) + ( 1-k_3 ^ 2 ) \mathcal{k}_1(k_3 ) \right ] \label{dw3action}\end{aligned}\ ] ] where the subscript 1 , 2 and 3 refers to the three different regimes of energy , that is , energy below , and above the energy of the saddle point , respectively . in theses expressions @xmath199 and @xmath200 are roots of the equation @xmath201 ( _ cf . _ appendix [ appendixdoublewell ] ) and @xmath202 . all these 3 quantities are functions of the energy . as for the case of the pendulum , the moduli @xmath203 and @xmath204 of the elliptic integrals are functions of the energy : [ dwmodulus ] @xmath205 the directional flux @xmath123 across half of the ds is then given by : [ dwflux ] @xmath206 \label{dw1flux } \\ \phi_2 ( e ) & = \frac{8\alpha_2^{3/2}}{3 } \label{dw2flux } \\ \phi_3 ( e ) & = \frac{2\sqrt{2\delta } b^2}{3k_3 ^ 2 } \left [ ( 2k_3 ^ 2 - 1 ) \mathcal{e}_1 ( k_3 ) + ( 1-k_3 ^ 2 ) \mathcal{k}_1(k_3 ) \right ] . \label{dw3flux}\end{aligned}\ ] ] a plot of the flux @xmath123 as a function of energy @xmath93 is shown in figure [ figfluxex2 ] . in this subsection we compute the gap time distribution of our two examples . before proceeding to the calculation of the gap time distribution , we need to determine the gap time for reactant trajectories . in our two examples the determination of the gap times of reacting orbits is straightforward since the gap time is just half of the period of the periodic orbits in the double - well degree - of - freedom ( @xmath155-@xmath207 ) for which the energy is greater than the energy of the saddle point ( @xmath165 , @xmath208 ) . the period of these periodic orbits as a function of the energy is given by : @xmath209 where @xmath210 stands for energy in the independent second degree - of - freedom , and the total energy of the two , uncoupled one degree - of - freedom systems is denoted by @xmath211 . the gap time as a function of the energy @xmath210 is then given by : @xmath212 for clarity here we will use the notations @xmath213 for the action associated with the first degree - of - freedom ( pendulum for example one and double - well for example 2 ) and @xmath214 for the action associated with the second degree - of - freedom . because of the direct relation between the energies of the different degrees - of - freedom and the actions associated with these degrees - of - freedom , at constant energy we have : @xmath215 so that we can formally write the gap time as a function of the total energy and the action @xmath213 : @xmath216 . again at constant @xmath217 , the number of trajectories having @xmath218 is proportional to the area enclosed by the curve @xmath219 in the ( @xmath182-@xmath183 ) plane , that is to say , it equals @xmath220 . the number of trajectories @xmath221 for which the gap time @xmath132 has the property that @xmath222 is : @xmath223 = 2 \pi \frac{\partial i}{\partial s } \delta s. \label{numtraj}\ ] ] the gap time distribution is now given by : [ gaptimedistrib ] @xmath224 figure [ figgaptimedistrib ] shows the gap time distribution for both examples . figure [ figgaptimedistrib ] shows that the gap time distributions for both examples possess a singularity . this singularity arises from the vanishing of the derivative of the action @xmath213 with respect to the gap time @xmath132 , which is equivalent to the singularity of the derivative of this action with respect to the energy @xmath225 . this singularity appears because the action variable @xmath213 is a piecewise defined function with respect to the energy @xmath225 . whereas the action @xmath213 is a continuous function with respect to the energy @xmath225 , its derivative ( the inverse of the associated frequency ) diverges at the separatrix ( homoclinic orbits ) , which is exactly where bifurcation occurs . interpreted geometrically , we see that just below the bifurcation the rate of growth of the area enclosed by the periodic orbits in the ( @xmath182-@xmath183 ) plane increases to infinity as we get closer to the homoclinic orbit . in the same way , just after the bifurcation this growth rate decreases from infinity . the appearance of homoclinic orbits is manifested in the flux @xmath226 , which is directly related to the action @xmath213 , through the appearance of an inflexion point at the energy of the bifurcation . in this section we will turn our attention to the question of normal hyperbolicity of the isoenergetic invariant manifold , which forms the boundary of the isoenergetic dividing surface . for both examples @xmath28 is either a periodic orbit or the union of homoclinic orbits and the saddle points that they connect . the two dimensional non - isoenergetic invariant manifold is defined by @xmath227 , which is a hyperbolic saddle point with associated growth rate @xmath228 . for @xmath148 tangent vectors normal to @xmath10 experience exponential growth and decay under the linearised dynamics . hence one way that normal hyperbolicity could be lost is for @xmath228 to go from positive to negative . however , this possibility is not interesting from the point of view of reaction dynamics since this would destroy the reactive ( double well ) nature of the problem . hence , we will require @xmath148 . the more interesting situation is the violation of the growth condition in the definition of a nhim ( definition [ def : nhim ] ) . when @xmath28 is a periodic orbit , this growth rate is zero , and @xmath28 is a nhim . when @xmath28 is a union of homoclinic orbits and the saddle points that they connect , whether or not @xmath28 satisfies the growth rate conditions of definition [ def : nhim ] depends on the nature of @xmath229 and @xmath228 . in particular , for example 1 @xmath230 are the eigenvalues associated with the saddle point on the @xmath231 plane and for example 2 @xmath232 are the eigenvalues associated with the saddle point in the @xmath233 plane . we therefore have the following situations : example 1 : : : for @xmath234 the growth rate conditions of definition [ def : nhim ] are not satisfied , and @xmath28 is not a nhim . example 2 : : : for @xmath235 the growth rate conditions of definition [ def : nhim ] are not satisfied , and @xmath28 is not a nhim . we now examine the consequences of this ` loss of normal hyperbolicity ' in more detail . we imagine the parameters @xmath234 fixed for example 1 and @xmath235 for example 2 , and we vary the energy in such a way that we pass through the energy value corresponding to the homoclinic orbits and saddle points . with @xmath229 and @xmath228 fixed as above , at this energy value the conditions of definition [ def : nhim ] do not hold . this situation could be referred to as ` loss of normal hyperbolicity ' . we examine the implication of this loss of normal hyperbolicity for the quantities that we have computed for our two examples . the dividing surface and the ( directional ) flux through the dividing surface . : : the dividing surface , as a function of energy , is shown in figure [ figdsex1 ] for example 1 and in figure [ figdsex2 ] for example 2 . while in both examples the geometry of the surface undergoes a qualitative change is we pass through the bifurcation , we are still able to define a dividing surface having the no recrossing property as we pass through the bifurcation , i.e. , even as the nhim experiences a loss of normal hyperbolicity ( as just defined ) . + the directional flux , as a function of energy , is shown in figure [ figfluxpendulum ] for example 1 and in figure [ figfluxex2 ] for example 2 . for both examples we see that the flux varies continuously as a function of the energy , even as we pass through the bifurcation , i.e. , even as the nhim experiences a loss of normal hyperbolicity . the gap time distribution . : : from figure [ figgaptimedistrib ] we have already noted note that the gap time distributions for both examples possess a singularity . this singularity is associated with the existence of a homoclinic orbit in the ds , and is _ not _ related to the loss of normal hyperbolicity of the nhim . in this paper we have studied the breakdown of normal hyperbolicity and its consequences for quantities related to reaction dynamics ; in particular , the dividing surface , the flux through the dividing surface , and the gap time distribution . our approach is to study these questions using simple , two degree - of - freedom hamiltonian models for which calculations for the different geometrical and dynamical quantities can be carried out exactly . for our examples , we showed that resonances ( homoclinic orbits ) within the normally hyperbolic invariant manifold may , or may not , lead to ` loss of normal hyperbolicity ' . moreover , we showed that for our examples the onset of such resonances results in a change in topology of the dividing surface , but it does not affect our ability to define a dividing surface ( ds ) , and that the flux through the ds varies continuously with energy , even as the energy is varied in such a way that normal hyperbolicity is lost . for both our examples we have shown that the gap time distribution exhibits a signature singularity at energies corresponding to emergence of a homoclinic orbit in the ds , but these singularities are not associated with loss of normal hyperbolicity . fm , pc , and sw acknowledge the support of the office of naval research ( grant no . n00014 - 01 - 1 - 0769 ) and the leverhulme trust . in this appendix we review the compution of action variables for the integrable simple pendulum system . the hamiltonian of the simple pendulum is : @xmath236 in order to determine the action variable for this system we have to consider three different cases : libration ( @xmath237 ) , separatrix ( @xmath238 ) , and rotation ( @xmath239 ) . the action for this case is given by : @xmath240 for this case the system does nt have enough energy to cover the full range of the angle @xmath241 and there should be two turning points where the momentum vanishes , @xmath242 . let @xmath243 be the positive value of @xmath241 at the turning point which depends on the energy : @xmath244 the momentum @xmath245 can be expressed as a function of @xmath243 and @xmath241 : @xmath246 substituting this expression into the eq . we obtain the integral : @xmath247 after an appropriate change of variable we can put the former integral into a form which makes it resemble an elliptic integral : @xmath248 where the modulus @xmath192 of the elliptic integral is given by @xmath249 . this integral can be expressed in terms of complete elliptic integrals as follows : @xmath250 , \label{app17}\ ] ] where @xmath190 and @xmath191 stand for the complete elliptic integrals of the first and second kinds respectively @xcite . for this case the system has enough energy to cover the full range of the angle @xmath241 and there is no point where the momentum vanishes . the momentum is given by : @xmath251 the corresponding action integral is : @xmath252 again using a change of variables we can transform this integral to : @xmath253 where the modulus @xmath193 is given by @xmath254 . this integral is just twice the complete elliptic integral of the second kind , so that the action for the rotation case is : @xmath255 for the separatrix case we can repeat the same kind of calculations as for the rotation case . noting that for this case we have @xmath256 and @xmath257 the action for the separatrix case is just : @xmath258 in this appendix we compute action variables for the integrable symmetric double - well system . the hamiltonian of the system is : @xmath259 in order to determine the action variable of this system we have to consider three different cases distinguished by the value of the energy @xmath93 : @xmath260 , the separatrix case where @xmath261 , and @xmath262 . for this case the energy is below the saddle point energy , @xmath261 . trajectories are confined to one of the two wells . because the double - well is symmetric the action variable is the same for the motions in either well.from eq . we can write the momentum @xmath245 as a function of the energy and @xmath241 : @xmath263 the associated action integral is : @xmath264 where @xmath199 and @xmath200 are the two roots of the equation @xmath242 for the well situated situated on the side where @xmath265 for example and with @xmath266 . using a change of variable we can rearrange the result into the form : @xmath267 where @xmath202 and @xmath268 . the latter integral can be evaluated by parts and we finally obtain : @xmath269 . \label{app25}\ ] ] for this case the energy is above the saddle point energy . in this case the equation @xmath270 has only two roots situated symmetrically with respect to the coordinate origin @xmath271 . denoting the two roots @xmath199 and @xmath200 , @xmath272 , the action integral is : @xmath273 using an appropriate change of variable we can transform this integral to the form : @xmath274 where @xmath202 and @xmath275 . inetgrals of this type are tabulated ( see for example ref . ) and one gets the result : @xmath276 . \label{app28}\ ] ] the action in this case is obtained by setting @xmath277 , @xmath278 , @xmath279 and @xmath280 in the previous case , and noting also that instead of having a factor 4 in eq . we have here a factor 2 . taking into account the fact that @xmath257 we get for the separatrix action : @xmath281 10 p. pechukas . transition state theory . , 32:159177 , 1981 . e. pollak and p. talkner . , 15:026116 , 2005 . k. j. laidler and m. c. king . the development of transition state theory . , 87:26572664 , 1983 . b. c. garrett . perspective on `` the transition state method , '' wigner e. ( 1938 ) trans . faraday soc . 34:29 - 41 . , 103:200204 , 2000 . g. a. petersson . perspective on `` the activated complex in chemical reactions , '' eyring , h. ( 1995 ) j. chem . phys . 3 : 107 . , 103:190195 , 2000 . e. wigner . the transition state method . , 34:2941 , 1938 . m. s. child and e. pollak . analytical reaction dynamics : origin and implications of trapped periodic trajectories . , 73:4365 , 1980 . e. pollak and m. s. child . classical mechanics of a collinear exchange reaction : a direct evaluation of the reaction probability and product distribution . , 73:4373 , 1980 . e. pollak , m. s. child , and p. pechukas . classical transition state theory : a lower bound to the reaction probability . , 72(3):16691678 , 1980 . p. pechukas and e. pollak . classical transition state theory is exact if the transition state is unique . , 71:2062 , 1979 . e. pollak and p. pechukas . transition states , trapped trajectories , and classical bound states embedded in the continuum . , 69:1218 , 1978 . p. pechukas and e. pollak . trapped trajectories at the boundary of reactivity bands in molecular collisions . , 67(12):59765977 , 1977 . e. pollak and p. pechukas . . , 70(1):325333 , 1979 . r. devogelare and m. boudart . contribution to the theory of reaction rates . , 23:12361244 , 1955 . s. wiggins . on the geometry of transport in phase space i. transport in @xmath43 degree - of - freedom hamiltonian systems , @xmath282 . , 44:471501 , 1990 . r. e. gillilan and g. s. ezra . transport and turnstiles in multidimensional hamiltonian mappings for unimolecular fragmentation : application to van der waals predissociation . , 94:26482668 , 1991 . gillilan and w.p reinhardt . . , 156:478482 , 1989 . invariant surfaces and phase space flux in three - dimensional surface diffusion . , 93:5300 , 1990 . s. wiggins , l. wiesenfeld , c. jaff , and t. uzer . impenetrable barriers in phase - space . , 86(24):54785481 , 2001 . t. uzer , c. jaff , j. palacin , p. yanguas , and s. wiggins . the geometry of reaction dynamics . , 15(4):957 , 2002 . h. waalkens , r. schubert , and s. wiggins . wigner s dynamical transition state theory in phase space : classical and quantum . , 21(1):r1 , 2008 . c. b. li , a. shoujiguchi , m. toda , and t. komatsuzaki . definability of no - return transition states in the high - energy regime above the reaction threshold . , 97(2):28302 , 2006 . a. allahem and t. bartsch . chaotic dynamics in multidimensional transition states . submitted to arxiv.org , 2012 . m. inarrea , j. f. palacian , a. i. pascual , and j. p. salas . bifurcations of dividing surfaces in chemical reactions . , 135:014110 , 2011 . d. g. yang . breakdown of normal hyperbolicity for a family of invariant manifolds with generalized lyapunov type numbers uniformly bounded below their critical values . submitted to arxiv.org , 2009 . n. fenichel . persistence and smoothness of invariant manifolds for flows . , 21(193 - 226):1972 , 1971 . n. fenichel . asymptotic stability with rate conditions . , 23:11091137 , 1974 . n. fenichel . asymptotic stability with rate conditions , ii . , 26:8193 , 1977 . m. w. hirsch , c. pugh , and m. shub . invariant manifolds . , 76(5):10151019 , 1970 . s. wiggins . , volume 105 . springer , 1994 . a. delshams , m. gidea , and p. roldan . . , april 2012 . s. v. bolotin and d. v. treschev . remarks on the definition of hyperbolic tori of hamiltonian systems . , 5(4):401412 , 2000 . p. collins , g. s. ezra , and s. wiggins . isomerization dynamics of a buckled nanobeam . , 2012 . e. thiele . comparison of the classical theories of unimolecular reactions . , 36(6):14661472 , 1962 . e. thiele . . , 38(8):19591966 , 1963 . g. s. ezra , h. waalkens , and s. wiggins . microcanonical rates , gap times , and phase space dividing surfaces . , 130(16):164118 , 2009 . v. i. arnold . , volume 60 . springer , 1989 . j. binney , o. e. gerhard , and p. hut . structure of surfaces of section . , 215:59 , 1985 . , 84(6):31473161 , 1986 . m. toller , g. jacucci , g. delorenzi , and c. p. flynn . theory of classical diffusion jumps in solids . , 32(4):2082 , 1985 . r. s. mackay . flux over a saddle . , 145(8):425427 , 1990 . h. waalkens and s. wiggins . direct construction of a dividing surface of minimal flux for multi - degree - of - freedom systems that can not be recrossed . , 37(35):l435 , 2004 . p. brumer , d. e. fitz , and d. wardlaw . time delay for bimolecular collisions : utility of the spectral theorem in the classical limit . , 72:386 , 1980 . e. pollak . a classical spectral theorem in bimolecular collisions . , 74:6763 , 1981 . h. waalkens , a. burbanks , and s. wiggins . efficient procedure to compute the microcanonical volume of initial conditions that lead to escape trajectories from a multidimensional potential well . , 95(8):84301 , 2005 . h. waalkens , a. burbanks , and s. wiggins . a formula to compute the microcanonical volume of reactive initial conditions in transition state theory . , 38(45):l759 , 2005 . n. b. slater . new formulation of gaseous unimolecular dissociation rates . , 24(6):12561257 , 1956 . n. b. slater . . cornell university press ithaca , new york , 1959 . r. s. dumont and p. brumer . dynamical theory of statistical unimolecular decay . , 90(16):35093516 , 1986 . b. k. carpenter . nonexponential decay of reactive intermediates : new challenges for spectroscopic observation , kinetic modeling and mechanistic interpretation . , 16(11):858868 , 2003 . d. l. bunker . , 37:393 , 1962 . d. l. bunker . . , 40:1946 , 1964 . d. l. bunker and w. l. hase . . , 59:4621 , 1973 . d. l. bunker . . pergamon press , 1966 . d. l. bunker and m. pattengill . . , 48(2):772776 , 1968 . w. l. hase . . in w. h. miller , editor , _ modern theoretical chemistry , volume 2 , pages 121170 . plenum , new york , 1976 . s. y. grebenshchikov , r. schinke , and w. l. hase . state - specific dynamics of unimolecular dissociation . , 39:105242 , 2003 . u. lourderaj and w.l . theoretical and computational studies of non - rrkm unimolecular dynamics . , 113(11):22362253 , 2009 . h. hancock . . john wiley & sons , inc . ,

we consider the bipartite version of the _ degree / diameter problem _ , namely , given natural numbers and , find the maximum number of vertices in a bipartite graph of maximum degree and diameter . in this context , the moore bipartite bound represents an upper bound for . bipartite graphs of maximum degree , diameter and order called _ moore bipartite graphs _ turned out to be very rare . therefore , it is very interesting to investigate bipartite graphs of maximum degree , diameter and order with small ; that is , bipartite-graphs . the parameter is called the _ defect_. this paper considers bipartite graphs of defect at most , and presents all the known such graphs . bipartite graphs of defect have been studied in the past ; if and , they may only exist for . however , when bipartite-graphs represent a wide unexplored area . the main results of the paper include several necessary conditions for the existence of bipartite-graphs ; the complete catalogue of bipartite-graphs with and ; the complete catalogue of bipartite-graphs with , ( ) and ; and a non - existence proof of all bipartite-graphs with and odd . finally , we conjecture that there are no bipartite graphs of defect for and , and comment on some implications of our results for the upper bounds of . * keywords : * moore bipartite bound ; moore bipartite graph ; degree / diameter problem for bipartite graphs ; defect ; repeat . * ams subject classification : * 05c35 , 05c75 .

1010.5651

due to the diverse features and applications of interconnection networks , it is possible to find many interpretations about network `` optimality '' in the literature . here we are concerned with the following ; see @xcite , @xcite , and ( * ? ? ? _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ an optimal network contains the maximum possible number of nodes , given a limit on the number of connections attached to a node and a limit on the distance between any two nodes of the network . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this interpretation has attracted network designers and the research community in general due to its implications in the design of large interconnection networks . in graph - theoretical terms , this interpretation leads to the _ degree / diameter problem _ ( the problem of finding the largest possible number of vertices in a graph with given maximum degree and diameter ) . if the graphs in question are subject to further restrictions such as being bipartite , planarity and/or transitivity , we can state the degree / diameter problem for the classes of graphs under consideration . in this paper we will consider only bipartite graphs , and in this case , the degree / diameter problem can be stated as follows . * _ degree / diameter problem for bipartite graphs _ : given natural numbers @xmath0 and @xmath1 , find the largest possible number @xmath27 of vertices in a bipartite graph of maximum degree @xmath3 and diameter @xmath4 . note that @xmath27 is well defined for @xmath28 and @xmath1 . an upper bound for @xmath27 is given by the _ moore bipartite bound _ @xmath29 , defined below : @xmath30 bipartite graphs of degree @xmath31 , diameter @xmath4 and order @xmath32 are called _ graphs_. moore bipartite graphs are rare ; for @xmath33 they are the cycles of length @xmath34 , while for @xmath35 moore bipartite graphs exist only for diameters 2 , 3 , 4 and 6 ; see @xcite . therefore , we are interested in studying the existence or otherwise of bipartite graphs of given maximum degree @xmath3 , diameter @xmath4 and order @xmath36 for @xmath37 ; that is , bipartite @xmath8-graphs , where the parameter @xmath9 is called the _ defect_. for notational convenience , we consider moore bipartite graphs as having defect @xmath38 . only a few values of @xmath27 are known at present . with the exception of @xmath39 , settled in @xcite , the other known values of @xmath27 are those for which there is a moore bipartite graph . the paper @xcite combined with @xcite almost settled the case of bipartite graphs of defect 2 ; if @xmath40 and @xmath13 , then such graphs may only exist for @xmath14 and certain values of @xmath31 . bipartite @xmath8-graphs with @xmath41 have been rarely considered in the literature so far . in this paper we consider bipartite @xmath42-graphs with @xmath43 and @xmath44 . by using combinatorial approaches we obtain several important results about bipartite graphs of defect 4 , including several necessary conditions for the existence of bipartite @xmath16-graphs ; the complete catalogue of bipartite @xmath17-graphs with @xmath18 and @xmath19 ; the complete catalogue of bipartite @xmath45-graphs with @xmath43 , @xmath21 ( @xmath22 ) and @xmath19 ; and a non - existence proof of all bipartite @xmath42-graphs with @xmath40 and odd @xmath25 . finally , we conjecture that there are no bipartite graphs of defect @xmath10 for @xmath35 and @xmath26 . the main results in this paper do not apply to bipartite @xmath42-graphs with @xmath46 and @xmath47 . some of our assertions , however , do offer a partial characterisation of all bipartite @xmath42-graphs with @xmath40 and @xmath13 . at the time of writing the paper we do not foresee a conclusive way to take on the diameters @xmath48 and @xmath10 . to deal with such graphs it would be necessary to either find different ideas or complement some of the ones presented here . section [ sub : diameters34 ] contains further comments on such diameters . the terminology and notation used in this paper is standard and consistent with that used in @xcite , so only those concepts that can vary from texts to texts will be defined . all graphs considered are simple . the vertex set of a graph @xmath49 is denoted by @xmath50 , and its edge set by @xmath51 . the _ difference _ between the graphs @xmath49 and @xmath52 , denoted by @xmath53 , is the graph with vertex set @xmath54 and edge set formed by all the edges with both endvertices in @xmath54 . the set of neighbours of a vertex @xmath55 in @xmath49 is denoted by @xmath56 . for an edge @xmath57 we write @xmath58 , or alternatively @xmath59 . the set of edges in a graph @xmath49 joining a vertex @xmath55 in @xmath60 to a vertex @xmath61 in @xmath62 is denoted by @xmath63 ; for simplicity , we write @xmath64 rather than @xmath65 . a path of length @xmath66 is called a _ @xmath66-path _ , and cycle of length @xmath66 is called a _ @xmath66-cycle_. a path from a vertex @xmath55 to a vertex @xmath61 is denoted by @xmath67 . whenever we refer to paths we mean shortest paths . we will use the following notation for subpaths of a path @xmath68 : @xmath69 , where @xmath70 . the distance between a vertex @xmath55 and a vertex @xmath61 is denoted by @xmath71 . the union of three independent paths of length @xmath4 with common endvertices is denoted by @xmath72 . in a graph @xmath49 a vertex of degree at least 3 is called a _ branch vertex _ of @xmath49 . for @xmath33 the moore bipartite graphs are the cycles on @xmath34 vertices , while for @xmath74 and each @xmath40 they are the complete bipartite graphs of degree @xmath31 and order @xmath75 . for @xmath76 moore bipartite graphs of degree @xmath31 have been constructed only when @xmath77 is a prime power @xcite . furthermore , singleton @xcite proved that the existence of a moore bipartite graph of diameter @xmath48 is equivalent to the existence of a projective plane of order @xmath77 . the question of whether moore bipartite graphs of diameter @xmath48 , @xmath10 or @xmath78 exist for other values of @xmath31 remains open , and represents one of the most famous problems in combinatorics . for other values of @xmath1 and @xmath35 there are no moore bipartite graphs ( see @xcite ) . when @xmath33 or @xmath74 bipartite @xmath45-graphs with @xmath79 can be obtained by simple observation . for a given @xmath18 there is only one bipartite @xmath80-graph with @xmath79 : the path of length @xmath4 , which has defect @xmath81 . for a given @xmath43 there are exactly @xmath77 bipartite @xmath82-graphs with @xmath79 ; they are the complete bipartite graphs with partite sets of size @xmath31 and @xmath83 , where @xmath84 . therefore , from now on we assume @xmath35 and @xmath44 . we continue with some conditions for the regularity of bipartite @xmath8-graphs , which were obtained in @xcite . [ prop : regularity1 ] for @xmath85 , @xmath86 and @xmath44 , a bipartite @xmath8-graph is regular . [ prop : regularity2 ] for @xmath87 , @xmath24 and odd @xmath44 , a bipartite @xmath8-graph is regular . by propositions [ prop : regularity1 ] and [ prop : regularity2 ] , bipartite @xmath45-graphs with @xmath40 , @xmath44 and @xmath88 must be regular , implying the non - existence of such graphs for @xmath40 , @xmath44 and @xmath89 . in the same way , bipartite @xmath42-graphs with @xmath40 and @xmath90 and bipartite @xmath91-graphs with @xmath46 must be regular . for @xmath35 and @xmath13 , the only known bipartite @xmath92-graphs are depicted in fig . [ fig : bipartite(d,3,-2 ) ] . recall that such graphs do not exist when @xmath35 and @xmath93 ; see @xcite . bipartite @xmath94-graphs may be irregular . figure [ fig : bipartite(3,3,-4 ) ] depicts all such graphs , which were obtained by using the program _ geng _ from the package _ nauty _ written by mckay @xcite . the unique bipartite @xmath95-graph is shown in fig . [ fig : bipartite(3,4,-4 ) ] . ) the unique bipartite @xmath96-graph and ( @xmath97 ) the unique bipartite @xmath98-graph . ] all the bipartite @xmath99-graphs are depicted in fig . [ fig : bipartite(4,3,-4 ) ] . these graphs were obtained computationally by meringer @xcite using the program _ genreg_. an alternative description of the graph in fig . [ fig : bipartite(4,3,-4 ) ] ( @xmath97 ) was communicated to the second author by charles delorme : take @xmath100 as the vertex set of the graph , and for each even @xmath55 , add the edges @xmath101 , @xmath102 , @xmath103 and @xmath104 . the only known bipartite @xmath105-graph is depicted in fig . [ fig : bipartite(5,3,-4 ) ] ; this graph was independently found by charles delorme and by the first author . charles delorme described this graph as follows : take @xmath106 as its vertex set , and for each even @xmath55 , add the edges @xmath102 , @xmath101 , @xmath107 , @xmath108 and @xmath109 . from now on we use the symbol @xmath110 rather than @xmath31 to denote the degree of a regular graph , as it is customary . recall that , unless @xmath111 and @xmath14 , a bipartite @xmath112-graph with @xmath35 and @xmath44 must be regular . therefore , when referring to a regular bipartite @xmath112-graph with @xmath35 and @xmath44 , we are actually referring to any bipartite @xmath112-graph with @xmath35 and @xmath44 other than the ones exhibited in fig . [ fig : bipartite(3,3,-4 ) ] @xmath113 and @xmath114 . in a bipartite @xmath112-graph we call a cycle of length at most @xmath115 a _ short cycle_. [ prop : girthbipartite(d , d,-4 ) ] the girth of a regular bipartite @xmath112-graph @xmath49 with @xmath116 and @xmath44 is @xmath115 . furthermore , any vertex @xmath55 of @xmath49 lies on the short cycles specified below and no other short cycle , and we have the following cases : * @xmath55 is contained in * exactly three @xmath117-cycles**. then * @xmath55 is a branch vertex of one @xmath118 , or * @xmath55 is contained in * two @xmath117-cycles**. then * @xmath55 lies on * exactly two @xmath117-cycles * , whose intersection is a @xmath119-path with @xmath120 . each case is considered as a type . for instance , a vertex satisfying case @xmath121 is called a vertex of type @xmath121 . note that , if @xmath55 is of type @xmath122 and @xmath123 , the two short cycles containing @xmath55 constitute a @xmath118 . * let @xmath124 be an edge of @xmath49 . let us use the standard decomposition for a bipartite graph of even girth with respect to the edge @xmath124 @xcite . for @xmath125 , the sets @xmath126 and @xmath127 are defined as follows : [ cols= " < , < , < " , ] the decomposition of @xmath49 into the sets @xmath126 and @xmath127 is called the _ standard decomposition for a graph of even girth with respect to the edge @xmath124_. since @xmath49 is bipartite , its girth is even and @xmath128 for @xmath129 . * claim 1 * @xmath130 . * proof of claim 1 . * since the assertion is trivial for @xmath14 , we suppose that @xmath131 for @xmath93 . assume that the edge @xmath124 lies on a cycle of length @xmath132 . then , @xmath133 for @xmath134 , and @xmath135 therefore , @xmath136 which is a contradiction . hence , @xmath137 . if @xmath138 then the order of @xmath49 would be at least @xmath139 @xcite . thus , @xmath130 and the claim follows . @xmath140 we now proceed to prove the second part of the proposition . for a given vertex @xmath55 , we use again the standard decomposition for a bipartite graph with respect to an edge @xmath124 in @xmath49 . suppose that there are at least three edges joining vertices at @xmath141 to vertices at @xmath142 ; that is , @xmath143 . in such case @xmath144 , @xmath145 , and therefore @xmath146 which is a contradiction . consequently , @xmath147 . suppose that @xmath148 . if the two edges are both incident to a common vertex of @xmath142 then @xmath55 is of type @xmath121 , otherwise @xmath55 is of type @xmath122 . if instead @xmath149 then @xmath150 . since @xmath151 , there is a vertex @xmath152 such that @xmath153 . therefore , it follows @xmath122 . finally , if @xmath154 then both types may occur indeed , if there is a vertex @xmath152 such that @xmath155 then @xmath55 is of type @xmath121 ( this case can only occur if @xmath156 ) , otherwise there must exist two vertices @xmath157 such that @xmath158 , in which case @xmath55 is of type @xmath122 . this completes the proof of the proposition . we continue with the following observation , which will be implicitly used throughout the paper : [ obs : bipartite1 ] let @xmath159 ( the sets @xmath160 and @xmath161 are called partite sets ) be any bipartite graph of even @xmath162 finite diameter @xmath4 . the distance between a vertex @xmath163 and any vertex @xmath164 @xmath165 is at most @xmath166 . in virtue of proposition [ prop : girthbipartite(d , d,-4 ) ] , we define the following concepts : if two short cycles @xmath167 and @xmath168 are non - disjoint we say that @xmath167 and @xmath168 are _ neighbours_. for a vertex @xmath55 lying on a short cycle @xmath169 , we denote by @xmath170 the vertex @xmath171 in @xmath169 such that @xmath172 . we say @xmath171 is the _ repeat _ of @xmath55 in @xmath169 and vice versa , or simply that @xmath55 and @xmath171 are _ repeats _ in @xmath169 . alternatively , and more generally , we say that @xmath171 is a repeat of @xmath55 with _ multiplicity _ @xmath174 if there are exactly @xmath175 different paths of length @xmath166 from @xmath55 to @xmath171 . proposition [ prop : girthbipartite(d , d,-4 ) ] tells us that a vertex in @xmath49 may have a repeat of multiplicity 2 . accordingly , we denote by @xmath176 the multiset of the repeats of a vertex @xmath55 in @xmath49 . the concept of repeat can be easily extended to paths . for a path @xmath177 of length at most @xmath178 contained in a short cycle @xmath169 , we denote by @xmath179 the path @xmath180 defined as @xmath181 . we say that @xmath182 is the _ repeat _ of @xmath183 in @xmath169 and vice versa , or simply that @xmath183 and @xmath182 are _ repeats _ in @xmath169 . often our arguments revolve around the identification of the elements in the set @xmath184 of short cycles containing a given vertex @xmath55 ; we call this process _ saturating _ the vertex @xmath55 . a vertex @xmath55 is called _ saturated _ if the elements in @xmath184 have been completely identified . the following lemma will help us in this cycle identification process . [ lemm : saturatinglemma ] let @xmath185 be a @xmath117-cycle in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 , and @xmath188 two vertices in @xmath185 such that @xmath189 . let @xmath190 be a neighbour of @xmath191 not contained in @xmath185 , and @xmath192 the neighbours of @xmath193 not contained in @xmath185 . suppose there is no short cycle in @xmath49 containing the edge @xmath194 and intersecting @xmath185 at a path of length greater than @xmath195 . then , in @xmath49 there exist a vertex @xmath196 and a short cycle @xmath197 such that @xmath190 and @xmath198 are repeats in @xmath197 , and @xmath199 . * proof . * let @xmath200 be the neighbours of @xmath193 contained in @xmath185 . for @xmath201 , consider the path @xmath202 . as @xmath203 , @xmath204 can not go through @xmath205 or @xmath206 . if @xmath204 went through certain @xmath207 @xmath208 , then a cycle @xmath209 would either have length smaller than @xmath115 or be a short cycle intersecting @xmath185 at a @xmath210-path . consequently , @xmath204 must go through one of the neighbours of @xmath211 other than @xmath193 , and must be a @xmath210-path ; see fig . [ single saturating figure ] @xmath212 . in addition , @xmath213 . let @xmath214 be one of the neighbours of @xmath190 other than @xmath191 , not contained in any of the paths @xmath204 ( there is at least one of such vertices ) . consider a path @xmath215 . if @xmath216 went through @xmath205 , then the closed walk @xmath217 would either contain a cycle of length smaller than @xmath115 or be a short cycle intersecting @xmath185 at a @xmath218-path . consequently , @xmath216 must go through a certain @xmath219 @xmath220 and @xmath221 ( fig . [ single saturating figure ] @xmath222 ) . note that @xmath216 must be a @xmath210-path , and that @xmath223 ; otherwise there would be a cycle in @xmath49 of length smaller than @xmath115 . thus , we have obtained a short cycle @xmath224 such that @xmath190 and @xmath219 are repeats in @xmath197 , and @xmath199 . by setting @xmath225 the lemma follows . @xmath140 [ cor : preneighborhoodlemma ] let @xmath226 be vertices in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 such that @xmath227 . then , for every @xmath228 it follows that @xmath229 contains a repeat of @xmath190 . * proof . * let @xmath185 be a short cycle containing @xmath191 and @xmath193 . if the vertex @xmath190 is contained in @xmath185 or the edge @xmath230 belongs to a short cycle in @xmath49 intersecting @xmath185 at a path of length @xmath178 or @xmath166 , then the corollary trivially follows . if we instead assume that @xmath231 and there is no short cycle in @xmath49 containing the edge @xmath230 and intersecting @xmath185 at a path of length greater than @xmath195 , then the corollary follows from the saturating lemma . @xmath140 in this section we extend the concept of repeat to short cycles ; see the repeat cycle lemma . [ lemm : repeatcyclelemma ] let @xmath169 be a short cycle in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 , @xmath232 the set of neighbours of @xmath169 , and @xmath233 for @xmath234 . suppose at least one @xmath235 , for @xmath236 , is a path of length smaller than @xmath178 . then there is an additional short cycle @xmath237 in @xmath49 intersecting @xmath238 at @xmath239 , where @xmath234 . * proof . * observe that , according to our premises and proposition [ prop : girthbipartite(d , d,-4 ) ] , @xmath240 and @xmath241 for @xmath242 . we assume the denotation of the neighbours @xmath243 of @xmath169 and the corresponding intersection paths @xmath244 is such that @xmath245 . for @xmath234 , we also denote the endvertices of @xmath246 by @xmath247 and @xmath248 , where @xmath249 and @xmath250 ( see fig . [ repeat cycle figure ] ( @xmath251 ) ) . for @xmath252 , consider the cycles @xmath238 and @xmath253 . first suppose that @xmath254 is a path of length smaller than @xmath178 . since @xmath255 is saturated , there can not be a short cycle in @xmath49 , other than @xmath169 , containing the edge @xmath256 . since @xmath254 is a path of length smaller than @xmath178 , we apply the saturating lemma ( mapping @xmath238 to @xmath185 , @xmath255 to @xmath191 , @xmath248 to @xmath257 and @xmath258 to @xmath190 ) and obtain an additional short cycle @xmath197 in @xmath49 such that @xmath258 is a repeat in @xmath197 of a neighbour @xmath259 of @xmath248 , and @xmath260 . since @xmath258 is saturated , we have that necessarily @xmath261 , which in turn implies @xmath262 . in other words , it follows that @xmath263 . if instead @xmath254 is a @xmath218-path then @xmath264 must be a path of length smaller than @xmath178 . therefore , we can apply the above reasoning and deduce that @xmath265 . this way we obtain a subgraph @xmath266 intersecting @xmath238 at @xmath246 for @xmath234 ( see fig . [ repeat cycle figure ] ( @xmath97 ) , where part of the subgraph @xmath267 is highlighted in bold ) . we next show that @xmath267 must be indeed a cycle . * claim 1 . * @xmath267 is a @xmath117-cycle . * proof of claim 1 . * if the paths @xmath246 are pairwise disjoint then @xmath267 is obviously a @xmath117-cycle . suppose the paths @xmath246 are not pairwise disjoint ; then @xmath268 and , according to proposition [ prop : girthbipartite(d , d,-4 ) ] , @xmath267 contains no cycle . since @xmath267 is clearly connected it is therefore a tree . let @xmath269 be an arbitrary leaf in @xmath267 . if the repeat path @xmath270 had length greater than 0 , then @xmath271 would have at least two neighbours in @xmath267 . therefore , @xmath272 contains exactly one vertex , and thus , @xmath273 and @xmath274 . recall we do addition modulo @xmath66 on the subscripts of the vertices and the superscripts of the cycles . since @xmath275 and @xmath276 are edges in @xmath267 , it holds that @xmath277 and @xmath278 denote the same vertex . let @xmath279 be the neighbours of @xmath277 in @xmath280 ; @xmath281 the neighbours of @xmath278 in @xmath282 ; and @xmath283 the neighbours of @xmath284 in @xmath285 . we have that @xmath286 , otherwise there would be a third short cycle in @xmath49 containing @xmath284 . in particular , the vertices in @xmath287 are pairwise distinct and @xmath288 . see fig . [ fig : repeatcyclepatch ] ( @xmath251 ) and ( @xmath97 ) for two drawings of this situation . let @xmath289 denote the vertices in @xmath290 ; see fig . [ fig : repeatcyclepatch ] ( @xmath291 ) . consider a path @xmath292 . recall that @xmath293 has length at most @xmath166 . since @xmath284 can not be contained in a further short cycle , @xmath293 must be a @xmath210-path and go through a neighbour of @xmath277 not contained in @xmath287 . therefore , we have that @xmath294 and , by the pigeonhole principle , that there are two paths @xmath295 and @xmath296 containing a common neighbour of @xmath277 . this way , @xmath284 would be contained in a third short cycle , a contradiction . as a result , we conclude that the repeat graph @xmath267 of @xmath169 is indeed a @xmath117-cycle @xmath237 as claimed . this completes the proof of claim 1 , and thus , of the lemma . @xmath140 while not of primary interest , it is not difficult to prove now that the cycles @xmath297 in the previous lemma are pairwise disjoint . we call the aforementioned cycle @xmath237 the _ repeat _ of the cycle @xmath169 in @xmath49 , and denote it by @xmath298 . next some simple consequences of the repeat cycle lemma follow . [ cor : repeatcycleuniqueness ] if a short cycle @xmath169 has a repeat cycle @xmath237 then @xmath237 is unique . [ cor : repeatcyclesymmetry ] if @xmath299 then @xmath300 . [ cor : repeatpath ] let @xmath169 and @xmath167 be two short cycles in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 which intersect at a path @xmath301 of length smaller than @xmath178 , and let @xmath302 . then , the repeat cycle of @xmath169 intersects @xmath167 at @xmath303 . [ cor : handycorollary ] let @xmath185 be a short cycle in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 , and @xmath304 repeat vertices in @xmath185 . let @xmath197 and @xmath305 be the other short cycles containing @xmath55 and @xmath171 , respectively . suppose that @xmath306 is a path of length smaller than @xmath178 . then , setting @xmath307 and @xmath308 , we have that @xmath61 and @xmath309 are repeat vertices in the repeat cycle of @xmath185 . * proof . * we denote the @xmath66 neighbour cycles of @xmath185 as @xmath310 and their respective intersection paths with @xmath185 as @xmath244 in such way that @xmath311 . for @xmath312 , we also denote @xmath313 , where @xmath314 and @xmath315 . obviously , for some @xmath316 @xmath317 we have that @xmath318 , @xmath319 , @xmath320 , @xmath321 , @xmath322 and @xmath323 . we may assume @xmath324 . by the repeat cycle lemma , the vertices @xmath61 and @xmath309 belong to the repeat cycle @xmath325 of @xmath185 . then the paths @xmath326 and @xmath327 are both @xmath210-paths in @xmath49 , and the corollary follows . @xmath140 [ prop : cyclepartition ] the set @xmath328 of short cycles in a bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 can be partitioned into sets @xmath329 , @xmath330 and @xmath331 , where * @xmath329 is the set of short cycles whose intersections with neighbour cycles are @xmath210-paths ; * @xmath330 is the set of short cycles whose intersections with neighbour cycles are @xmath218-paths ; and * @xmath331 is the set of short cycles whose intersections with neighbour cycles are paths of length at most @xmath195 . * proof . * if @xmath49 is one of the non - regular graphs in fig . [ fig : bipartite(3,3,-4 ) ] the result trivially follows . we then assume that @xmath49 is regular . let @xmath169 be a short cycle in @xmath49 . if @xmath169 is contained in a @xmath118 then , according to proposition [ prop : girthbipartite(d , d,-4 ) ] , all the intersections of @xmath169 with its neighbour cycles are @xmath210-paths , in which case @xmath332 . now suppose that , for some short cycle @xmath167 , @xmath333 is a path of length @xmath178 . note that all vertices in @xmath334 are saturated . let @xmath335 be an arbitrary vertex in @xmath334 , @xmath336 , and @xmath168 the short cycle other than @xmath169 containing @xmath337 . suppose that @xmath338 is not a @xmath218-path . then clearly @xmath339 can not be a @xmath210-path , so it has length at most @xmath195 . but according to corollary [ cor : repeatpath ] , the cycle @xmath340 intersects @xmath169 at exactly @xmath341 , a proper subpath of @xmath334 . this implies that @xmath340 is a third short cycle containing the vertex @xmath335 , a contradiction . consequently , the intersections of @xmath169 with its ( exactly two ) neighbour cycles are @xmath218-paths , and @xmath342 . finally , if there is a short cycle intersecting @xmath169 at a path of length at most @xmath195 then , by the above reasoning , the intersections of @xmath169 with all of its neighbour cycles are paths of length at most @xmath195 , and @xmath343 . @xmath140 the preceding result could be stated alternatively in term of vertices as follows : [ prop : vertexpartition ] the set @xmath50 of vertices in a regular bipartite @xmath112-graph @xmath49 with @xmath186 and @xmath187 can be partitioned into sets @xmath344 , @xmath345 and @xmath346 , where * @xmath344 is the set of vertices contained in cycles of @xmath329 ; * @xmath345 is the set of vertices contained in cycles of @xmath330 ; * @xmath346 is the set of vertices contained in cycles of @xmath331 ; and @xmath329 , @xmath330 , @xmath331 are defined as in proposition [ prop : cyclepartition ] . [ theo : theta ] a bipartite @xmath112-graph @xmath49 with @xmath347 and @xmath25 does not contain a subgraph isomorphic to @xmath118 . * proof . * suppose that @xmath49 has a subgraph @xmath348 isomorphic to @xmath118 , with branch vertices @xmath251 and @xmath97 . let @xmath349 and @xmath350 be as in fig . [ theta figure ] ( @xmath251 ) , and let @xmath351 be one of the neighbours of @xmath352 not contained in @xmath348 . since all vertices of @xmath348 are saturated , there can not be a short cycle in @xmath49 containing any of the incident edges of @xmath349 or @xmath350 which are not contained in @xmath348 . according to this and by applying the saturating lemma , there is an additional short cycle @xmath353 in @xmath49 such that @xmath351 and one of the neighbours of @xmath354 not contained in @xmath348 ( say @xmath355 ) are repeats in @xmath353 , and @xmath356 . analogously , in @xmath49 there is an additional short cycle @xmath357 such that @xmath355 and one of the neighbours of @xmath358 not contained in @xmath348 ( say @xmath359 ) are repeats in @xmath357 , and @xmath360 ; an additional short cycle @xmath361 such that @xmath359 and one of the neighbours of @xmath362 not contained in @xmath348 ( say @xmath363 ) are repeats in @xmath361 , and @xmath364 ; and an additional short cycle @xmath365 such that @xmath363 and one of the neighbours of @xmath350 not contained in @xmath348 ( say @xmath366 ) are repeats in @xmath365 , and @xmath367 . see fig . [ theta figure ] ( @xmath97 ) . note that @xmath368 is a path of length at most @xmath369 ; otherwise for some vertex @xmath370 the closed walk @xmath371 would contain a cycle of length at most @xmath115 to which the vertex @xmath97 would belong , a contradiction . for similar reasons , the intersection paths @xmath372 and @xmath373 all have length at most 2 , with 2<@xmath178 . we now apply the handy corollary . by mapping the cycle @xmath357 to @xmath185 , the vertex @xmath355 to @xmath55 , the vertex @xmath359 to @xmath171 , the cycle @xmath353 to @xmath197 , the cycle @xmath361 to @xmath305 , the vertex @xmath351 to @xmath61 and the vertex @xmath363 to @xmath309 , we obtain that @xmath351 and @xmath363 are repeat vertices in the repeat cycle of @xmath357 . therefore , since @xmath374 , it follows that @xmath357 and @xmath365 are repeat cycles and @xmath375 . this way , there would be a cycle @xmath376 in @xmath49 of length @xmath377 ( since @xmath25 ) , a contradiction to the fact that @xmath203 . @xmath140 [ prop : numberofcycles ] the number @xmath378 of short cycles in a bipartite @xmath112-graph @xmath49 with @xmath347 and @xmath25 is given by the expression @xmath379 . * by theorem [ theo : theta ] , @xmath49 does not contain a subgraph isomorphic to @xmath118 . then , according to proposition [ prop : girthbipartite(d , d,-4 ) ] , every vertex of @xmath49 is contained in exactly two short cycles . we then count the number @xmath378 of short cycles of @xmath49 . since the order of @xmath49 is @xmath380 , we have that @xmath381 , and the proposition follows . @xmath140 since the number of short cycles in a graph @xmath49 must be an integer , the expression obtained for @xmath378 in proposition [ prop : numberofcycles ] already suffices to prove the non - existence of bipartite @xmath112-graphs for infinitely many pairs @xmath382 . consider first the case in which @xmath383 is an odd prime power . let @xmath384 be the multiplicative group of the field @xmath385 , let @xmath386 , and let @xmath387 be the cyclic subgroup of @xmath388 generated by @xmath389 . we observe that the sum of the elements of @xmath387 is null @xmath390 . furthermore , since the order of @xmath387 divides the order of @xmath388 , it must also divide @xmath391 . thus , we have @xmath392 therefore , it immediately follows [ cor : primepower ] there is no bipartite @xmath112-graph with @xmath116 and @xmath26 such that @xmath166 is an odd prime power . @xmath140 more generally , if @xmath393 is an odd prime factor of @xmath166 and @xmath394 , then @xmath395 [ cor : residues01 ] there is no bipartite @xmath112-graph with @xmath116 and @xmath396 such that @xmath397 , where @xmath393 is an odd prime factor of @xmath166 . @xmath140 it is also possible to examine completely the case of some small odd prime factors of @xmath166 . for example , it is not difficult to verify that , if @xmath398 then @xmath48 does not divide @xmath380 ; thus , [ cor : dmod3 ] there is no bipartite @xmath112-graph with @xmath116 and @xmath26 such that @xmath399 . now we turn to structural arguments to obtain other non - existence results . [ lemm:(d-2)-intersection ] any two non - disjoint short cycles in a bipartite @xmath112-graph @xmath49 with @xmath347 and @xmath400 intersect at a path of length smaller than @xmath178 . * proof . * since @xmath49 does not contain a graph isomorphic to @xmath118 , it is only necessary to prove here that any two non - disjoint short cycles in @xmath49 can not intersect at a path of length @xmath178 . suppose , by way of contradiction , that there are two short cycles @xmath167 and @xmath168 in @xmath49 intersecting at a path @xmath401 of length @xmath178 . according to proposition [ prop : cyclepartition ] , @xmath168 is intersected by exactly two short cycles , namely @xmath167 and @xmath402 , at two independent @xmath218-paths . by repeatedly applying this reasoning and considering @xmath49 is finite , we obtain a maximal length sequence @xmath403 of pairwise distinct short cycles in @xmath49 such that @xmath238 intersects @xmath404 at a path @xmath254 of length @xmath178 ( @xmath405 ) , and @xmath406 for any @xmath407 such that @xmath408 . let us denote the paths @xmath409 @xmath410 in such way that , for @xmath411 , @xmath412 and @xmath413 are edges in @xmath49 . also , let @xmath414 , @xmath415 , @xmath416 , and @xmath417 ; see fig . [ intersection length1 figure ] @xmath212 . set @xmath418 and @xmath419 . since the sequence @xmath403 is maximal and all the vertices in @xmath420 are saturated , it follows that @xmath421 , and we have either @xmath422 and @xmath423 ( fig . [ intersection length1 figure ] @xmath222 ) , or @xmath424 and @xmath425 ( fig . [ intersection length1 figure ] @xmath113 ) . if @xmath422 and @xmath423 , then @xmath426 ; otherwise the cycle @xmath427 would have length at most @xmath115 , contradicting the saturation of @xmath428 . if conversely @xmath424 and @xmath425 then @xmath429 ; otherwise the cycle @xmath430 containing @xmath428 would have length at most @xmath115 , a contradiction as well . for our purposes , it is enough to state @xmath431 in any case . let @xmath352 be the neighbour of @xmath432 on @xmath401 , and @xmath433 for @xmath434 . also , let @xmath351 be a neighbour of @xmath352 not contained in @xmath401 ; see fig . [ intersection length2 figure ] ( @xmath251 ) . since all vertices on @xmath401 are saturated , the edge @xmath435 can not be contained in a further short cycle . we apply the saturating lemma ( by mapping @xmath436 to @xmath185 , @xmath352 to @xmath191 , @xmath437 to @xmath257 , and @xmath351 to @xmath190 ) , and obtain in @xmath49 an additional short cycle @xmath353 such that @xmath351 and one of the neighbours of @xmath437 not contained in @xmath438 ( say @xmath439 ) are repeats in @xmath353 , and @xmath440 . analogously , for @xmath441 we obtain an additional short cycle @xmath442 in @xmath49 such that @xmath443 and a neighbour of @xmath444 not contained in @xmath445 ( say @xmath446 ) are repeats in @xmath442 , and @xmath447 ; see fig . [ intersection length2 figure ] ( @xmath97 ) . for @xmath448 or @xmath48 , @xmath449 can not be a @xmath218-path ; otherwise for some vertex @xmath450 , there would be a cycle @xmath451 of length at most @xmath452 ( since @xmath453 ) , a contradiction to the fact that @xmath454 is saturated and @xmath203 . analogously , @xmath372 can not be a @xmath218-path . we now apply the handy corollary . by mapping the cycles @xmath357 to @xmath185 , @xmath353 to @xmath197 and @xmath361 to @xmath305 , and the vertices @xmath355 to @xmath55 , @xmath359 to @xmath171 , @xmath351 to @xmath61 , and @xmath363 to @xmath309 , it follows that the vertices @xmath351 and @xmath363 are repeat vertices in the repeat cycle of @xmath357 . since @xmath374 , @xmath357 and @xmath365 are repeat cycles and @xmath455 . this way , we obtain a cycle @xmath456 in @xmath49 of length @xmath457 , a contradiction . this completes the proof of the lemma . [ theo : odddiameter ] there are no bipartite @xmath112-graphs for @xmath347 and odd @xmath25 . * the case @xmath458 can be easily discarded by using proposition [ prop : numberofcycles ] , so we assume @xmath400 . suppose there is a bipartite @xmath112-graph @xmath49 with @xmath347 and odd @xmath400 . according to lemma [ lemm:(d-2)-intersection ] , any two non - disjoint short cycles in @xmath49 intersect at a path of length smaller than @xmath178 , which means that every short cycle @xmath169 in @xmath49 has a repeat cycle @xmath237 ( by the repeat cycle lemma ) . because of the uniqueness and symmetry of repeat cycles , the number @xmath378 of short cycles in @xmath49 must be even . however , since @xmath4 is odd , the number @xmath459 of short cycles in @xmath49 is odd , a contradiction . @xmath140 furthermore , using theorem [ theo : odddiameter ] and proposition [ prop : numberofcycles ] we complete the catalogue of bipartite @xmath112-graphs with @xmath21 . from section [ sec : knowngraphs ] recall that , for @xmath26 , the only bipartite @xmath460-graph is the path of length @xmath461 . [ theo:5 - 187 ] the path of length 5 is the only bipartite @xmath42-graph with @xmath28 and @xmath21 . in this section we complete the catalogue of bipartite @xmath462-graphs . specifically , we prove the non - existence of bipartite @xmath462-graphs with even @xmath396 . [ lemm : d-3int ] any two non - disjoint short cycles in a bipartite @xmath462-graph @xmath49 with @xmath347 and @xmath400 intersect at a path of length smaller than @xmath195 . * proof . * by lemma [ lemm:(d-2)-intersection ] , it is only necessary to prove here that any two short cycles @xmath169 and @xmath167 in @xmath49 can not intersect at a path @xmath463 of length @xmath195 . we proceed by contradiction . let @xmath171 and @xmath309 be the repeat vertices of @xmath55 and @xmath61 in @xmath167 , respectively . by corollary [ cor : repeatpath ] , the repeat cycle @xmath237 of @xmath169 intersects @xmath167 at @xmath464 ( the repeat path of @xmath301 in @xmath167 ) ; see fig . [ fig : d-3int ] . if we denote by @xmath271 the neighbour of @xmath55 on @xmath465 , then we have that the other short cycle containing @xmath271 would also contain at least one of the vertices in @xmath466 , which contradicts the fact that @xmath55 and @xmath309 are both saturated . @xmath140 [ theo:(3,d,-4 ) ] there are no bipartite @xmath462-graphs with even @xmath396 . * proof . * recall that theorem [ theo:5 - 187 ] covers the case @xmath467 . let @xmath49 be a bipartite @xmath462-graph with even @xmath468 , @xmath469 a short cycle in @xmath49 , and @xmath470 , @xmath471 two repeat vertices in @xmath469 . let @xmath428 and @xmath472 be the neighbours of @xmath470 and @xmath471 , respectively , not contained in @xmath469 . according to the saturating lemma , there is an additional short cycle @xmath167 containing @xmath428 and @xmath472 such that @xmath473 . let @xmath432 be one of the neighbours of @xmath428 contained in @xmath167 , and @xmath474 . denote by @xmath475 and @xmath476 the neighbours of @xmath432 and @xmath477 , respectively , not contained in @xmath167 . again by the saturating lemma , there is an additional short cycle @xmath168 such that @xmath478 and @xmath479 . since @xmath111 , we may assume that the other short cycle @xmath169 containing @xmath470 also contains @xmath428 and a neighbour of @xmath470 in @xmath480 ( say @xmath481 ) . we first prove that @xmath482 . * claim 1 . * @xmath482 . * proof of claim 1 . * let @xmath481 , @xmath483 , @xmath484 , @xmath485 and @xmath486 be as in fig . [ fig : cubicevendiameter1 ] ( @xmath251 ) . consider a path @xmath487 . if @xmath488 , then @xmath183 would go through a neighbour of @xmath477 contained in @xmath167 and there would be a cycle in @xmath49 of length at most @xmath489 . therefore , we may assume @xmath490 . if @xmath491 then there would be a short cycle intersecting the cycle @xmath168 at a path of length @xmath195 , a contradiction to lemma [ lemm : d-3int ] . similarly , we have that @xmath492 . also , @xmath183 must be a @xmath210-path and @xmath493 ; otherwise there would be a short cycle intersecting the cycle @xmath168 at a path of length @xmath178 , a contradiction to lemma [ lemm:(d-2)-intersection ] . for @xmath494 , let @xmath495 and @xmath247 be the neighbours of @xmath496 and @xmath497 , respectively , not contained in @xmath498 , and let @xmath238 be ( in virtue of the saturating lemma ) the additional short cycle disjoint from @xmath498 which contains @xmath495 and @xmath247 . since @xmath183 must go through @xmath247 , we denote by @xmath248 the neighbour of @xmath247 on @xmath499 and set @xmath500 . we now show that , if @xmath501 , @xmath502 . assume the contrary ; that is , @xmath503 ( since @xmath203 , @xmath504 ) . in such case , there would be a short cycle @xmath505 intersecting @xmath238 at a path of length @xmath195 , contradicting lemma [ lemm : d-3int ] ( see figures [ fig : cubicevendiameter1 ] ( @xmath97 ) and [ fig : sequence ] ) . consequently , @xmath502 and @xmath183 must go through a neighbour of @xmath248 not contained in @xmath238 . this way , for @xmath494 we have @xmath506 , which means that @xmath507 . since the cycle @xmath508 contains the vertices @xmath509 and @xmath510 , we have @xmath511 , which implies that @xmath482 . @xmath140 as the selection of @xmath469 and @xmath169 was arbitrary , basically as a corollary of claim 1 we have : * claim 2 . * any two non - disjoint short cycles in @xmath49 intersect at an edge . finally , suppose that @xmath469 and @xmath169 intersect at @xmath512 , as stated by claim 1 . let @xmath513 be the repeat vertex of @xmath481 in @xmath469 ; then , by corollary [ cor : repeatpath ] , the repeat cycle @xmath237 of @xmath169 intersects @xmath469 at @xmath514 ( the repeat path of @xmath512 in @xmath469 ) . setting @xmath515 , we have @xmath216 is a path of length @xmath178 with saturated endvertices ( see fig . [ theocubic ] ) . therefore , by claim 2 , there exists a sequence @xmath516 of short cycles such that @xmath517 . however , since @xmath4 is even , the other short cycle containing @xmath518 would also contain one of the vertices in @xmath519 , which contradicts the fact that @xmath520 and @xmath513 are both saturated . this completes the proof of theorem [ theo:(3,d,-4 ) ] . @xmath140 combining theorems [ theo : odddiameter ] and [ theo:(3,d,-4 ) ] , we have that the only bipartite @xmath462-graphs with @xmath18 are those depicted in figures [ fig : bipartite(3,3,-4 ) ] and [ fig : bipartite(3,4,-4 ) ] , completing in this way the catalogue of such graphs . the main results obtained in this paper are summarised below . first we stated important structural properties of bipartite @xmath112-graphs with @xmath347 and @xmath44 . we found necessary conditions for the existence of bipartite @xmath112-graphs with @xmath347 and @xmath25 , which allowed us to prove the non - existence of such graphs for infinitely many pairs @xmath382 ; this included the case in which @xmath166 is an odd prime power , and the case in which @xmath399 . afterwards , we went on to proving that bipartite @xmath112-graphs for @xmath347 and odd @xmath25 do not exist . we completed the catalogue of bipartite @xmath42-graphs with @xmath28 and @xmath21 , which in turn completed the catalogue of bipartite @xmath45-graphs with @xmath28 , @xmath21 , @xmath22 and @xmath73 . catalogue of bipartite @xmath521-graphs with @xmath28 and @xmath21 . : : for @xmath21 and @xmath33 the only moore bipartite graphs are the @xmath34-cycles , whereas for @xmath467 and @xmath35 they are incidence graphs of generalised polygons . for other values of @xmath21 and @xmath40 there are no moore bipartite graphs . catalogue of bipartite @xmath92-graphs with @xmath28 and @xmath21 . : : the results of @xcite combined with @xcite showed that there are no such graphs . catalogue of bipartite @xmath42-graphs with @xmath28 and @xmath21 . : : the path of length 5 is the only such graph . catalogue of bipartite @xmath522-graphs with @xmath18 . : : the cubic moore bipartite graphs are the complete bipartite graph @xmath523 for @xmath74 , the unique incidence graph of the projective plane of order @xmath11 for @xmath14 , the unique incidence graph of the generalised quadrangle of order @xmath11 for @xmath524 , and the unique incidence graph of the generalised hexagon of order @xmath11 for @xmath467 . catalogue of bipartite @xmath525-graphs with @xmath18 . : : there are only two non - isomorphic @xmath525-graphs with @xmath18 ; a unique bipartite @xmath526-graph ( the claw graph ) , and a unique @xmath96-graph , which is depicted in fig . [ fig : bipartite(d,3,-2 ) ] ( @xmath251 ) . catalogue of bipartite @xmath462-graphs with @xmath18 . : : there exist no bipartite @xmath527-graphs . when the diameter is 3 , there are four non - isomorphic bipartite @xmath94-graphs ; all of them are shown in figure [ fig : bipartite(3,3,-4 ) ] . for diameter 4 , there is a unique bipartite @xmath95-graph , which is depicted in fig . [ fig : bipartite(3,4,-4 ) ] . the results of this paper , combined with those of @xcite , assert that there are no bipartite @xmath462-graphs with @xmath26 , outcome that gives an alternative proof of the optimality of the known bipartite @xmath528-graph ( see @xcite ) . the main results in this paper did not include bipartite @xmath112-graphs with @xmath156 and @xmath47 . however , we believe that the structural properties of these graphs provided in section [ sec : preresults ] could bear more conclusive results on such diameters . for instance , by using proposition [ prop : girthbipartite(d , d,-4 ) ] , lemma [ lemm : saturatinglemma ] ( saturating lemma ) , lemma [ lemm : repeatcyclelemma ] ( repeat cycle lemma ) , proposition [ prop : cyclepartition ] , and proposition [ prop : vertexpartition ] , we were able to prove analytically the uniqueness of the two bipartite @xmath99-graphs depicted in fig . [ fig : bipartite(4,3,-4 ) ] . we also think there should be no major difficulty to complete as well in a very similar manner the catalogue of bipartite @xmath105-graphs , which has so far as a unique element , the graph in fig . [ fig : bipartite(5,3,-4 ) ] . unfortunately , the final ideas used in the paper can not be easily extended to cover bipartite @xmath112-graphs with @xmath156 and @xmath47 . with our current approach we can not have theorem [ theo : theta ] for @xmath47 . in theorem [ theo : theta ] the intersection paths @xmath368 , @xmath372 and @xmath373 have length at most 2 , and for us to apply the the repeat cycle lemma we need the lengths of such paths to be less than @xmath178 . indeed , the graph in fig . [ fig : bipartite(4,3,-4 ) ] ( @xmath251 ) offers a good illustration of this . even if we had theorem [ theo : theta ] , something similar would occur with lemma [ lemm:(d-2)-intersection ] ; see the graphs in fig . [ fig : bipartite(4,3,-4 ) ] ( @xmath97 ) and fig . [ fig : bipartite(5,3,-4 ) ] . our results improve the upper bound on @xmath529 for many combinations of @xmath31 and @xmath4 . recall that a bipartite @xmath530-graph @xmath49 with @xmath40 and @xmath531 must be regular ( by proposition [ prop : regularity1 ] ) and thus can not exist . n. l. biggs and t. ito , _ graphs with even girth and small excess _ , mathematical proceedings of the cambridge philosophical society * 88 * ( 1980 ) , no . 1 , 110 , http://dx.doi.org/10.1017/s0305004100057303[doi:10.1017/s0305004100057303 ] . c. delorme , l. k. jrgensen , m. miller , and g.pineda-villavicencio , _ on bipartite graphs of diameter 3 and defect 2 _ , journal of graph theory * 61 * ( 2009 ) , no . 4 , 271288 , http://dx.doi.org/10.1002/jgt.20378[doi:10.1002/jgt.20378 ] . c. delorme , l. k. jrgensen , m. miller , and g. pineda - villavicencio , _ on bipartite graphs of defect 2 _ , european journal of combinatorics * 30 * ( 2009 ) , no . 4 , 798808 , http://dx.doi.org/10.1016/j.ejc.2008.09.030[10.1016/j.ejc.2008.09.030 ] . m. c. heydemann , _ cayley graphs and interconnection networks _ , graph symmetry : algebraic methods and applications ( g. hahn and g. sabidussi , eds . ) , nato asi series , series c : mathematical and physical sciences , vol . 497 , kluwer academic publishers , dordrecht , 1996 . l. k. jrgensen , _ nonexistence of certain cubic graphs with small diameters _ , discrete mathematics * 114 * ( 1993 ) , 265273 , http://dx.doi.org/10.1016/0012-365x(93)90371-y[doi:10.1016/0012-365x(93)90371-y ] . g. pineda - villavicencio , _ non - existence of bipartite graphs of diameter at least 4 and defect 2 _ , journal of algebraic combinatorics ( 2010 ) , http://dx.doi.org/10.1007/s10801-010-0266-0[doi:10.1007/s10801-010-0266-0 ] . r. c. singleton , _ on minimal graphs of maximum even girth _ , journal of combinatorial theory * 1 * ( 1966 ) , no . 3 , 306332 , http://dx.doi.org/10.1016/s0021-9800(66)80054-6[doi:10.1016/s0021-9800(66)80054-6 ] .

we study the problem of quantum quench across a critical point in a strongly coupled field theory using ads / cft techniques . the model involves a probe neutral scalar field with mass - squared in the range in a charged black brane background . for a given brane background there is a critical mass - squared , such that for the scalar field condenses . the theory is critical when and the source for the dual operator vanishes . at the critical point , the radial operator for the bulk linearized problem has a zero mode . we study the dynamics of the order parameter with a time dependent source , or a null - time dependent bulk mass across the critical point . we show that in the critical region the dynamics for an initially slow variation is dominated by the zero mode : this leads to an effective description in terms of a landau - ginsburg type dynamics with a _ linear _ time derivative . starting with an adiabatic initial condition in the ordered phase , we find that the order parameter drops to zero at a time which is later than the time when or hits zero . in the critical region , , and the departure of the order parameter from its adiabatic value , scale with the rate of change , with exponents determined by static critical behavior . numerical results for the order parameter are consistent with these expectations . uk/11 - 06 + * quantum quench across a holographic critical point * + pallab basu and sumit r. das

1109.3909

an important class of problems in quantum systems is that of quantum quench , where a parameter in the hamiltonian varies with time , typically attaining constant values at early and late times . starting with some initial state , the problem is to determine the nature of the final state . this problem has recently attracted a lot of attention in several areas of many - body physics , particularly because of progress in cold atom experiments @xcite,@xcite,@xcite . among other things , this problem is interesting for two different reasons . the first relates to the question of thermalization . does the system evolve into some kind of steady state ? if so , is the state `` thermal '' in any sense ? the second question deals with the situation where the quench takes place across a value of the parameter where there is an equilibriium critical point . in this case , kibble - zurek type scaling arguments indicate that there are several physical quantities which are _ universal _ and determined by the critical exponents of the critical point . furthermore for two dimensional theories which are _ suddenly _ quenched to a critical point , powerful techniques of boundary conformal field theory have been used in @xcite to show that ratios of relaxation times of one point functions , as well as the length / time scales associated with the behavior of two point functions of different operators are given in terms of ratios of their conformal dimensions at the critical point , and hence universal . another related application of this phenomenon is in gravitational physics and cosmology . in this context , this is the phenomenon of particle production in time dependent backgrounds , which is relevant to many physical problems ranging from black hole evaporation to behavior of quantum fluctuations in an expanding universe . there are very few theoretical tools available to study such systems when they are strongly coupled . in this note we will explore the use of ads / cft correspondence @xcite - @xcite to this problem . ads / cft techniques have been in fact used in the past to study quantum quench , though not at critical points . in ads / cft the boundary values of bulk fields become coupling constants , so that this problem becomes that of determining a time dependent background with specified initial and boundary conditions . as is usual , a _ problem in the boundary theory becomes a classical problem in general relativity . one class of problems involve a vacuum initial state ( in the bulk this is pure @xmath11 in the distant past ) and a change of the boundary value of some field ( e.g. the dilaton or the boundary metric ) over some finite time interval , in a regime where supergravity is always valid . in the boundary field theory this corresponds to turning on a time dependent source for the dual operator . under suitable conditions , this leads to black hole formation in the bulk @xcite . the correlators at future time would then be thermal with a temperature characterized by the hawking temperature . the time scale after which this happens depends on the nature of the correlators , but turns out to be always smaller than what one would expect from a conformally invariant system evolving to a thermal state . thus , in this case thermalization of the field theory is signalled by black hole formation . another class of problems involve a similar setup , but a suitable variation of the coupling which prevents black hole formation in the supergravity regime . the coupling , however , becomes _ weak _ at some time and the bulk string frame curvature grows large , leading to a breakdown of the supergravity approximation - thus mimicking a space - like singularity @xcite . for the case of a slow variation of the coupling it turns out that the gauge theory remains well defined and may be used to show that a smooth passage through this region of small coupling is possible without formation of a large black hole . related scenarios appear in @xcite and @xcite . many interesting phenomena in gauge - gravity duality , in particular phase transitions , can be explored consistently in a _ probe approximation_. in this approximation , a certain set of bulk fields can be treated separately and independently of bulk gravity . one set of examples are probe branes in a background @xmath12 . this introduces hypermultiplets which in general live on a defect in the original field theory . when the number of such branes , @xmath13 is much smaller than the number of flux units which produce the background geometry , @xmath14 , the dynamics of these branes do not backreact on the bulk metric which remains @xmath12 . the dynamics is then described by a dbi action in a fixed background - the fields in this action are then bulk probe fields . if the boundary value of such a probe field is time dependent in a fashion similar to above , black hole formation in the bulk is invisible in the probe approximation . however , rather remarkably , thermalization of the hypermultiplet sector is still visible . this manifests itself by the formation of an apparent horizon ( which evolve into an event horizon in some situations ) in the induced metric on the brane worldvolume @xcite . fluctuations of the brane feel this induced metric and their correlators have thermal properties . none of the above works study quench dynamics near a phase transition . in this paper we take the first step towards doing that , in a simple model which displays an equilibrium critical point in a probe limit @xcite . the model involves a neutral scalar field whose mass lies in the range @xmath15 in the background of a charged @xmath2 black brane . the overall coupling is chosen to be large so that the scalar can be treated as a probe . for a given brane background ( i.e. for a given temperature and charge density in the dual field theory ) , and for a vanishing non - normalizable mode of the field ( corresponding to a vanishing source of the dual operator ) there is always a critical mass @xmath3 below which the trivial solution is unstable . in this regime the stable solution is nontrivial , signifying a nonzero expectation for the dual operator in the dual theory . @xmath16 with a vanishing source is then a critical point , which happens to have standard mean field exponents . this setup is in fact quite similar to models of holographic superconductors and fermi surfaces @xcite-@xcite . however , here the scalar field is neutral so that the dynamics of the gauge field is frozen . this model has been argued to model phase transitions in antiferromagnets . as is well known , in this mass range there are two inequivalent quantizations . our analysis is restricted to the conventional quantization of the bulk scalar , where the mode which vanishes slower at the boundary is treated as the source in the boundary theory . a similar analysis can be easily performed for the alternative quantization . we will study the time evolution of the order parameter starting from adiabatic initial conditions in the ordered phase . the passage through the critical point is done in two ways - ( i ) by keeping the bulk mass at its critical value and changing the leading term of the near - boundary expansion of the bulk field and ( ii ) by changing a bulk mass parameter with a vanishing leading term . the field theory meaning of ( i ) is clear - this corresponds to turning on a source dual to the bulk field : the critical point appears when the source vanishes and the mass is at its critical value . the meaning of ( ii ) is less direct - changing a bulk mass means changing the anomalous dimension of the corresponding operator in the field theory . as discussed in @xcite , one way to do this is in fact coupling this scalar field to another field and changing the source for this other field . away from the critical point , adiabaticity holds for a sufficiently low rate of change . we will show that the adiabatic corrections are proportional to _ linear _ time derivatives , even though the equations of motion involve only second order time derivatives . this is a well - known phenomenon and happens because of dissipation implied by the presence of a horizon . in the bulk , we need to impose regularity conditions at the horizon , which is easily done using ingoing eddington - finkelstein coordinate @xcite @xmath17 as time . @xmath17 coincides with the usual time coordinate @xmath18 on the boundary and @xmath19 , though @xmath20 . the bulk equations now involve first order derivatives @xmath21 . as expected , adiabaticity fails as one approaches the critical point at a time which we choose to be @xmath22 . the appearance of a first order time derivative is related to the fact that the dynamical critical exponent of similar models turn out to be @xmath23 @xcite . when adiabaticity fails , the system enters a scaling regime . a useful way to understand the dynamics is to decompose the bulk field in terms of eigenstates of the radial operator which appears in the linearized problem around the equilibrium solution . for any nonzero temperature of the background black brane , this operator has a zero mode which is regular at the horizon exactly at the critical point . we will show that in the critical region , the dynamics for small @xmath24 is dominated by that of this zero mode . the other modes remain adiabatic . the scaling properties can be then understood in terms of a landau - ginsburg type dynamics of the zero mode with a _ linear _ time derivative . the order parameter remains nonzero even when the instantaneous value of the coupling reaches the equilibrium critical value and first drops to zero at a time @xmath25 where @xmath24 denotes the rate of change of the coupling in the critical region . the departure of the value of the order parameter from the instantaneous equilibrium value also scales as @xmath26 . the exponents @xmath27 and @xmath28 are essentially determined by the static critical exponents . beyond the critical region , however , the other modes become important . in fact , at very late times the bulk equation may be approximated by its linearized form , so that the solutions are nothing but the usual quasinormal modes around the background black hole . as usual , the imaginary part of the lowest quasinormal mode frequency then determine the decay of late time behavior . we also present preliminary results of a numerical solution of the time dependent equations for the bulk field and extract the time dependence of the order parameter . the results are consistent with the above picture . our numerical results are not yet accurate enough to verify the scaling behavior . however this should be possible with further work . in section 2 we discuss spatially homogeneous classical quench in a landau - ginsburg type model as a prelude . in section 3 we review the main faetures of the equilibrium phase transition . in section 4 we discuss the failure of adiabaticity in quantum quench across this transition . we then show that in the critical region the dynamics for an initially slow quench is dominated by that of the zero mode , and discuss its relationship of the dynamics to the landau - ginsburg model of section 2 . in section 5 we present the results of a numerical solution of the bulk equation of motion and verify the expectations in the previous section . section 6 contains some concluding remarks . the three appendices provide details of derivation of the some of the results contained in the main text . as a prelude to the main discussion , in this section we will study classical spatially homogeneous quench in a landau - ginsburg model , which is described by the equation , + m^2(t)+^2 -1+j(t)=0 [ 22 - 2 ] as discussed above , holography converts the problem of a _ quantum _ quench to a problem of _ classical _ quench in one more dimension - which is the motivation behind studying ( [ 22 - 2 ] ) . in a later section we will show that this simple model describes some of the essential features of the dynamics in the critical region of the holographic phase transition which is the main subject of this paper . note that we do not have any noise term ( which is what is needed for many other applications ) . when @xmath0 and @xmath10 are time independent , this model has a critical point at @xmath29 . here we have kept generic @xmath27 and @xmath28 for convenience of later discussion . in most conservative models , i.e. where energy is conserved , @xmath30 . however as we will see @xmath31 is more appropriate for the holographic transition which we will discuss . in the following we will therefore use this value . we will also concentrate on the @xmath32 case , i.e. a quartic interaction . let us first discuss the case where @xmath6 is time varying , but a @xmath0 which is independent of time and positive . for a time independent @xmath10 we have , m^2 + ^3+j=0 denote the solution of this static equation by @xmath33 . we then introduce a @xmath6 which is slowly varying with time . the idea is to start with adiabatic initial conditions at some early enough time and study the time evolution of the order parameter . to study the adiabatic expansion we write , = _ 0(j(t),m)+_1(t)+ here @xmath34 is an adiabaticity parameter introduced by performing a scaling @xmath35 so that any @xmath36 comes with a factor of @xmath34 . the ellipsis denote higher order terms in @xmath34 . to lowest order in @xmath34 , the equation governing @xmath37 is , _ 1 ( m^2 + 3 _ 0 ^ 2(t))= _ 0(t ) = j(t ) the adiabatic expansion is good for a finite @xmath38 for a sufficiently slowly varying @xmath6 . the adiabatic expansion fails when & & _ 1 ~_0 + & & _ 0 ~_0 [ adbreak ] for a sufficiently slowly varying @xmath6 , this is possible if both @xmath0 and @xmath39 are small , i.e. when we are close to the critical point @xmath29 . approaching the critical point along the direction @xmath40 we have @xmath41 , and [ adbreak ] leads to the condition for breakdown of adiabaticity , ~j^. [ 22 - 2a ] if the quench is linear near the critical point , that is @xmath42 for small @xmath10 , we get vt^ ~1 as the condition for breakdown of adiabaticity . when adiabaticity breaks down , the system enters a scaling region , in the scaling region we use @xmath43 . we rescale the field as : @xmath44 and @xmath45 . choosing @xmath46 and @xmath47 , we get a @xmath24 independent equation for @xmath48 . the quadratic or higher order temporal contribution in @xmath6 only gives a sub - leading contribution in @xmath49 in the scaling limit . hence we have a scaling solution , ( t)=v^ ( v^t ) . this scaling solution leads to an estimate of the magnitude of fluctuations ( @xmath50 ) , i.e. the departure of @xmath51 from the equilibrium value at @xmath22 . outside the critical region @xmath52 , whereas in the critical region @xmath53 . this also defines the `` zero crossing time''(@xmath8 ) defined by @xmath54 . we have @xmath55 at the critical point . which diverges in the @xmath56 limit . out of criticality @xmath57 actually approaches a constant in the @xmath56 limit . it is instructive to study the dynamics of the order parameter for a profile of @xmath6 which becomes a constant at early and late times and behaves linearly with time in the critical region , for example j(t ) = j_0 ( vt ) with @xmath40 for all times . the numerical solution of the equation of motion with an adiabatic initial condition at early times is shown in figure ( [ lgsource1 ] ) for a typical value of @xmath24 . [ lgsource1 ] at @xmath40 with @xmath58 ( from the bottom on the left ) . the adiabatic solution ( dashed ) is also shown as a comparison . ] numerically we see that at sufficiently early times the order parameter tracks the adiabatic solution and overshoots the adaiabatic solution as one approaches the critical region at @xmath22 , falling down to zero at some positive time @xmath8 . at later times , the solution again approaches the adiabatic solution . the numerical solutions for different values of @xmath24 are consistent with the behavior @xmath53 and @xmath59 as predicted by the scaling analysis . for profiles of @xmath6 which behave as some nontrivial power of @xmath18 in the critical region , i.e. @xmath60 a similar analysis leads to ( t , v ) = v^ ( t v^,1 ) leading to the scaling properties ( 0 ) ~v^ t_~v^- . if there are more than one interaction terms , e.g. @xmath61 present in the equation of motion , then the lowest order interaction ( @xmath62 ) dominates the critical dynamics . in this case we have , ( t , v)=v^ ( t v^ ) . in this section we will study classical quench across the critical point along the line @xmath63 by changing @xmath0 from negative to positive values . for static @xmath64 , the model has a non - trivial time - independent solution @xmath65 . the static model has a second order phase transition near @xmath40 . now we introduce a time dependent mass function which asymptotes to some constant value of @xmath66 at early times and then rise to a constant value @xmath67 at late times , e.g. m^2(t ) = -m_1 ^ 2 ( vt ) [ 22 - 5 ] once again we start with adiabatic initial conditions at some early enough time and study the time evolution of the order parameter . to study the adiabatic expansion we write , = _ 0(t)+_1(t)+where , _ 0(t)= [ 0 - 1 ] to lowest order in the adiabatic expansion _ 1 = using the fact that @xmath68 , we get that the adiabatic expansion breaks down as @xmath69 , i.e. when ~1 . [ 22 - 3 ] as expected this breaks down as we approach the equilibrium critical point . if @xmath70 near the critical point breakdown of adiabaticity happens when , t v^- when adiabaticity breaks down the system enters a scaling regime . in this regime , we can approximate @xmath71 . here the scaling relations are ( t , v ) v^ ( t v^,1 ) [ 22 - 4 ] however , if the solution does not cross a phase transition point then fluctuation around a static solution is suppressed exponentially in @xmath72 . it is also to be noted that if @xmath73 is sufficiently slowly varying , the exact functional form of @xmath73 is not needed . the only information we use to derive ( [ 22 - 4 ] ) is that @xmath74 is a linear function of time near the phase transition and @xmath75 as @xmath76 . the model of @xcite has a neutral scalar field @xmath77 in the background of a charged @xmath2 black brane . the lagrangian is given by = [ -()^2 -(^2+m^2)^2- ] [ 1 - 1 ] the background metric is given by ( in @xmath78 units ) ds^2 = [ -r^2 f(r)dt^2+r^2 d^2]+ [ 1 - 2 ] where f(r ) = [ 1 + - ] 0 1 [ 1 - 3 ] the associated hawking temperature is then given by t = ( 1- ) [ 1 - 4 ] in the following we will replace @xmath79 . one should add the einstein - hilbert action and a cosmological constant term to ( [ 1 - 1 ] ) . in the limit of large @xmath80 , however , the field @xmath51 can be regarded as a probe field since its back reaction to the metric via einstein s equations is small . this is the approximation we adopt . in @xcite it was shown that when the mass lies in the range - < m^2 < - [ 1 - 5 ] there is a critical phase transition at some value of @xmath81 when the source to the dual operator vanishes . conversely , for a given @xmath82 there is a value of @xmath83 where the theory is critical . the upper limit in ( [ 1 - 5 ] ) is the bf bound for the near - horizon @xmath84 geometry which appears in the extremal ( @xmath85 ) metric . ( note that the ads scale for this infrared @xmath84 is given by @xmath86 in our units ) . the lower bound is the bf bound for the asymptotic @xmath2 . field configurations which are translationally invariant in the @xmath87 directions satisfy the equations of motion [ -_t^2+_r(r^2 f(r)_r)]-m^2 -^3 = 0 [ 1 - 6 ] near the @xmath2 boundary the asymptotic behavior of the solution to the linearized equation is of the form , it is a valid concern whether we will be able to neglect the non - linear term near the boundary . this can be done as long as @xmath88 or @xmath64 . ] ( r ) = j(t)r^-_-[1+o(1/r^2)]+<o>(t ) r^-_+[1+o(1/ r^2 ) ] [ 1 - 7 ] where @xmath89 is given by _ = [ 1 - 8 ] in the range of masses of interest , both the solutions are normalizable , so that there is a choice of quantization . the standard quantization considers the coefficient @xmath6 is the source in the dual field theory and @xmath90 then gives the expectation value of the dual operator . in the alternative quantization the expectation and source change the role . consider first the linearized problem , ignoring the cubic term . by a standard change of coordinates to tortoise coordinates @xmath91 and a field refinition to @xmath92 , d=- ( r , t ) = [ 1 - 9 ] the horizon is then at @xmath93 and the boundary is at @xmath94 . the equation becomes ( @xcite ) -_t^2 = -_^2 + v_0 ( ) _ [ 1 - 10 ] with v_0 ( ) = f(r)[(m^2 + 2 ) - + ] [ 1 - 11 ] where in @xmath95 we need to express @xmath96 in terms of @xmath91 using ( [ 1 - 9 ] ) . for solutions of the type @xmath97 , equation ( [ 1 - 11 ] ) is a schrodinger problem in a potential @xmath98 . the potential goes to zero at the horizon @xmath99 and behaves as @xmath100 near the boundary @xmath101 . note that for a brane background at any finite temperature , @xmath102 near the horizon , while @xmath103 so that @xmath104 as we approach the horizon . in contrast , for the extremal background @xmath105 while @xmath106 so that @xmath107 . this makes the analysis for the extremal background rather subtle . in this paper we will work with the non - extremal case . due to the vanishing of the potential near the horizon the time - independent problem has a continuum of modes starting at @xmath108 . the behavior of the solution near the horizon @xmath109 is of the form @xmath110 , the two signs representing ingoing and outgoing waves . we are eventually interested in solving the problem with _ ingoing _ boundary conditions at the horizon . generically for any value of @xmath0 there are quasi - normal modes with complex frequencies . these are normalizable modes with ingoing boundary condition . unlike the continuum discussed above these modes form a set of disconnected poles in the complex frequency plane . however , normalizable bound states appear when @xmath0 is sufficiently negative : this is what has been shown in @xcite . there is a critical value of @xmath83 when a zero energy bound state appears , which vanishes in an appropriate fashion at the boundary and is in addition purely ingoing at the horizon . in the complex frequency plane some quasinormal mode(s ) hit the origin at @xmath83 . this critical mass is @xmath111 when @xmath112 and decreases with decreasing @xmath113 or increasing temperature . the existence of such a bound state for @xmath114 has been shown by variational methods in @xcite . a similar demosntration should be possible for other values of @xmath0 . we have numerically verfied the existence of this zero mode . a typical plot the zero mode is shown in in fig [ zmd ] . note that the leading large - r behavior has been factored out to ensure that the source term indeed vanishes at the boundary . , where @xmath115 ) with @xmath116 and @xmath117 . ] the presence of a bound state in the schrodinger problem means that the solution @xmath118 of the full nonlinear problem is unstable , and one has to look for other nontrivial solutions . in @xcite it was shown that such a stable nontrivial static solution @xmath119 exists in the range of masses given by ( [ 1 - 5 ] ) , both for the standard and alternative quantizations . this means that the expectation value of the dual operator is nonzero , i.e. the field condenses , and @xmath120 is a critical point . one can approach this critical point either by fixing the mass and tuning the temperature , or by fixing the temperature and tuning the mass . the latter may appear to be somewhat strange from the point of view of the boundary field theory . however , it is only the ir mass which is relevant - and one may imagine obtaining this by coupling @xmath51 to another field @xmath121 by @xmath122 . the expectation value of @xmath121 which may results from changing a coupling in the boundary field theory then renders a mass to @xmath51 @xcite . the critical behavior has standard mean field exponents at any finite @xmath82 . if the operator dual to the field @xmath51 is @xmath123 and the source is @xmath10 then an analysis identical to that presented in @xcite leads to ( for @xmath124 ) _ j=0~(m^2-m_c^2)^1/2 |_j=0~(m^2-m_c^2)^-1 < > _ m = m_c~j^1/3 [ 1 - 12 ] exactly at zero temperature the phase transition is of bkt type and the order parameter depends exponentially _ j=0 ~[1 - 13 ] the zero mode of the linearized operator will play a key role in the following . this zero mode is known to exist for any non - zero temperature . at zero temperature the situation is less clear . for a similar zero temperature system involving the d3-d5 system @xcite such a zero mode does not exist @xcite . our aim is to study quench dynamics across this equlibrium phase transition . in a quench situation , some parameter of the boundary theory is made time dependent . the first issue we will study is the breakdown of adiabaticity as we approach the critical point . first , we argue that away from the critical point , the order parameter tracks the changing coupling adiabatically . we will then determine the regime near the critical point where adiabataicity breaks and show that in this region the order parameter and the time scale are proportional to some power of @xmath24 pretty much like the model studied in section 2 . as discussed above , a simple way to quench across the critical point is to remain at the critical value of the mass and consider a time dependent source for the dual operator in the boundary field theory . this corresponds to a nontrivial boundary value for the bulk field . our discussion is in the conventional quantization of the theory , i.e @xmath6 in ( [ 1 - 7 ] ) is treated as a source . the case of alternative quantization should follow along identical lines . it is well known that to study low frequency modes in the background of a black brane it is convenient to use ingoing edddington - finkelstein coordinates , u=- t , where @xmath91 is defined in ( [ 1 - 9 ] ) . in terms of these coordinates the equation of motion [ 1 - 6 ] ) becomes -2_u_= -_^2 + v ( , ) . [ 4 - 1 ] where v(,)= v_0()+ ^3 . this equation has to be solved with the boundary condition that the field is _ regular at the horizon _ , which at the linearized level is equivalent to requiring that the waves are purely ingoing at the horizon @xcite . at the linear level the analysis can be of course performed in @xmath125 coordinates as well . however , in this case one needs special care to extract a leading singular piece before performing the low energy expansion @xcite . at the full nonlinear level this procedure is possibly rather involved , but imposing regularity in ef coordinates remain simple . we need to solve ( [ 4 - 1 ] ) with the condition ( u , ) ^-1+_- j(u ) as 0 [ 4 - 1 - 1 ] where @xmath126 are defined in ( [ 1 - 8 ] ) . we have considered the source to be a function of @xmath17 rather than a function of @xmath18 . on the boundary @xmath127 @xmath91 and @xmath18 are the same ( and quite generally @xmath128 ) , so that this does correspond to a time dependent source in the boundary theory . however , as we will soon see , it is convenient to use ( [ 4 - 1 - 1 ] ) . this form makes bulk causality explicit . to perform the adiabatic expansion , let us decompose the field @xmath129 as @xmath130 where @xmath131 and @xmath132 as @xmath133 . we get , @xmath134 \chi_s + \frac{f(r)}{r^2 } ( \chi_s^3 + 3 \chi_l \chi_s^2 + 3 \chi_l^2 \chi_s)=- [ -\partial_\rho^2 + v_0(\rho ) ] \chi_l-\frac{f(r)}{r^2}\chi_l^3 \\ \nonumber - 2 \partial_u\partial_\rho \chi_l - 2 \partial_u\partial_\rho \chi_s\end{aligned}\ ] ] if @xmath135 is time independent we have a static solution @xmath136 . this has been discussed in the previous section . when we turn on a time dependent @xmath137 which is _ slowly varying _ , one expects that the time dependent solution of ( [ 1 - 6 ] ) may be constructed in an adiabatic expansion . to understand that we write , _ s(,u ) = _ 0 ( , j(u ) ) + _ 1(,u ) + . here @xmath34 is an adiabaticity parameter which keeps track of the adiabatic expansion . if we scale @xmath138 , each @xmath17 derivative is of order @xmath139 . the ellipsis in ( [ 1 - 15 ] ) are higher order terms in @xmath34 . the idea then is to insert ( [ 1 - 15 ] ) into the equations of motion and obtain equations for @xmath140 order by order in @xmath34 . to the lowest order one ^(1 ) _ 1 = \ { [ -_^2 + v_0 ( ) ] + ( 3 ^ 2_0 + 6 _ l_0 + 3_l^2 ) } _ 1=+2 _ u__l + 2 _ u__0 [ 1 - 16 ] in deriving ( [ 1 - 16 ] ) we have ignored any time dependence of @xmath141 , as is appropriate in an adiabatic expansion . generically , at each order in perturbation we may cast the adiabatic problem as , _ n = j_n(u , ) , [ 1 - 17 ] where @xmath142 is a source term contains time derivatives and determined by the lower order equations in an adiabatic expansion . as discussed before , near the black hole horizon the differential operator becomes a plane wave operator and the potential part vanishes . since the potential @xmath95 goes to zero at the horizon @xmath93 and the terms which involve @xmath143 in ( [ 1 - 16 ] ) also vanish at the horizon ( because of the overall factor of @xmath144 ) , the operator @xmath145 has a continuous spectrum which begins at zero . it is , therefore , a non - trivial fact that there is adiabaticity away from the critical point . we will now show why this is so and how adiabaticity breaks down near the critical point . these equations are most conveniently solved by green s function method , _ n = d g^(n)(, ) j_n(u, ) . [ 1 - 17a ] where @xmath146 is the green s function of the operator @xmath147 with the boundary conditions @xmath148 and @xmath149 is regular . let us concentrate on the lowest order terms in the adiabatic expansion , @xmath150 . all higher terms may be found out by similar procedure . the green s function is given by , @xmath151 where @xmath141 and @xmath152 are solutions of homogeneous part of eqn ( [ 1 - 17 ] ) satisfying appropriate boundary condition at the horizon @xmath153 and the boundary @xmath94 respectively , and @xmath154 is the wronskian which is independent of @xmath91 in this case . we have normalized @xmath155 and @xmath156 in such a fashion that @xmath157 at the horizon and @xmath158 near the boundary . given that we are looking for a solution which is regular at the horizon ( @xmath159 ) , @xmath141 is a constant at the horizon and so is the green s function . nevertheless , the integral in eqn ( [ 1 - 17a ] ) is is finite . this is because regularity of the functions @xmath160 and @xmath161 mean that @xmath162 are finite at the horizon . since @xmath163 , this implies that @xmath164 , damping out any divergence in ( [ 1 - 17a ] ) from the large @xmath91 region . in general , near the horizon @xmath152 can be expressed as a linear combination of a regular and irregular solution , i.e @xmath165 . it can be easily checked that @xmath166 . 1(,u)= - d ( 2 _ u _ _ 0 + 2 _ u__l ) this is finite as long as @xmath167 is finite , and small for sufficiently small @xmath168 and @xmath169 . at the phase transition point , however , the operator @xmath170 has a zero mode which is regular at the horizon . this is because at @xmath63 the factors involving @xmath161 and @xmath160 in the left hand side of ( [ 1 - 16 ] ) vanish , and the operator is identical to the operator acting on the linearized small fluctuations at @xmath83 around the trivial solution @xmath171 , i.e. the operator @xmath172 which appears on the right hand side of ( [ 1 - 10 ] ) . we know that this operator has a zero mode which is regular at the horizon _ and _ vanishes as @xmath173 at the boundary @xmath94 . this means that at this point @xmath174 . therefore , the first adiabatic correction diverges . to derive the precise condition for adiabaticity , it is sufficient to look at the operator @xmath170 for small @xmath137 . for small @xmath137 , the leading departure from the critical operator comes from the term which is proportional to @xmath175 . thus we can use perturbation theory in @xmath10 to estimate @xmath176 . as argued before , @xmath177 , while @xmath178 . hence the leading divergence in @xmath141 can be estimated as _ 1(,u ) ~ = j^-4/3 . [ 1 - 20a ] the breakdown of adiabaticity happens when , _ 1(,u ) ~_0 + ~j^5/3 , [ 1 - 20b ] exactly as in the @xmath179 dimensional model ( equation ( [ 22 - 2a ] ) . assuming a linear quench , i.e @xmath180 , we get vu^ ~1 as the condition for breakdown of adiabaticity . in this section we argue that the critical region dynamics is dominated by that of the zero mode for ( initially ) small quench rate . we will demonstrate this for the case where @xmath181 in the critical region . extension to non - linear quenches is straightforward . to understand this let us rescale @xmath182 . assuming a linear quench , i.e. @xmath183 , we have at the leading order in @xmath24 , @xmath184 \tch_s+v^{\frac{2}{5 } } [ \frac{f(r)}{r^2 } ( \tch_s)^3 + \tilde u [ -\partial_\rho^2 + v_0(\rho ) ] \tch_l+2 \partial_{\tilde u}\partial_{\rho } \tch_s ] + \cdots=0 \label{veq}\end{aligned}\ ] ] the ellipsis denote terms which contains higher powers of @xmath24 . now let us expand the sub - leading part of the scalar field in eigenfunctions of the operator @xmath172 ( defined in equation ( [ 1 - 10 ] ) at the critical point , @xmath185 where @xmath186 satisfy _ ^c _ k = [ -_^2 + v_0^c()]_k = k^2 _ k where @xmath187 denotes the potential in ( [ 1 - 11 ] ) at @xmath188 . the eigenfunctions @xmath189 are delta function normalized . the eigenfunctions obey the condition _ 0 [ ^-_- _ k ( ) ] = 0 in terms of the eigen - coefficients @xmath190 the equation ( [ veq ] ) becomes , @xmath191 where , @xmath192 \ch_l \ , d\rho \\ b_{kk'}=\int d \rho \ , \ch_k \partial_\rho \ch_k ' & \,\,\ , , \ , c_{k , k',k'',k'''}=\int d\rho \ , \ch_k \ch_{k ' } \ch_{k '' } \ch_{k ' '' } \frac{f(r)}{r^2 } .\end{aligned}\ ] ] the equation ( [ modeeqn ] ) suggests that there is a solution in a perturbation expansion of powers of @xmath193 . writing @xmath194 we get an equation for @xmath195 , @xmath196 and a set of equations for @xmath197 @xmath198 in the last line of the above equation we have subtracted the equation for @xmath199 to make the @xmath200 limit explicit . if the values of @xmath201 were discrete , the @xmath202 mode would clearly dominate the dynamics for small @xmath24 . however , @xmath201 is a continuous parameter starting from zero - this naively indicates that the perturbation expansion in @xmath193 could break down . whether this happens or not depends on the behavior of the numerator on the right hand side of ( [ scalingeqn ] ) . it turns out , however , that all the terms like @xmath203 as @xmath204 , rendering the perturbation expandion in @xmath205 well defined . this is shown in appendix ( [ kscaling ] ) . thus , in the critical region the solution @xmath206 is dominated by the zero mode piece proportional to @xmath207 . in terms of the original variables , we have finally @xmath208 since the equation for @xmath195 is identical to the equation for the quantity @xmath51 in the toy model ( [ 22 - 2 ] ) , we are led to scaling relations identical to the @xmath179 dimensional model . the one point function of the dual operator is given by ( u ) ~lim_0 [ ^1-_+ _ s ( , u ) ] therefore , the time dependence of @xmath123 is governed by the same scaling behavior . the above analysis can be carried out for @xmath0 away from the critical value as well . in this case , the terms like @xmath209 as @xmath204 , so that the expansion in @xmath210 is not valid . this means that away from the critical point , all the modes are important . another way to quench across the critical point is to consider a time dependent @xmath0 with no source , keeping the background geometry fixed . the mass function is taken to asymptote to some constant value of @xmath4 at early times and then rise to a constant value @xmath211 at late times , e.g. m^2(u ) = m_c^2+(m_c^2-m_1 ^ 2)(v u ) [ 2 - 1 ] note that we have a mass which depends on the ef coordinate @xmath17 rather than the time coordinate @xmath18 . the motivation for this is as follows . as mentioned earlier , we may consider the bulk mass term as arising from a coupling of the scalar field with some other field . this latter field acquires an expectation value in response to a source which leads to a mass of our field @xmath51 . in this scenario , a time dependent source of the second field would lead , by causality , to an effective mass - squared which has the form @xmath212 , where the ellipsis denote terms which involve @xmath17 derivatives of @xmath7 . the function @xmath213 vanishes at the boundary and is regular at the horizon . to leading order we therefore get an equation like ( [ 4 - 1 ] ) with the potential @xmath214 modified by replacing @xmath215 . the analysis in the previous sections can be now easily repeated for this problem . below we outline the main steps and results . let us consider the case of a time dependent mass which asymptotes to a sufficiently negative value , @xmath216 . if the mass remained constant at @xmath216 there is a nonzero static solution @xmath217 . now once we make the the mass time dependent we solve ( [ 4 - 1 ] ) adiabatically in @xmath218 . for a mass which is _ slowly varying _ one expects that the time dependent solution of ( [ 1 - 6 ] ) may be constructed around an adiabatic solution , ( , u;m(u ) ) = _ 0 ( , m(u ) ) + ^2 _ 1(,u ) + . [ 1 - 15 ] @xmath219 is the static solution obtained by ignoring the time dependence of the mass completely . here @xmath34 is an adiabaticity parameter which keeps track of the adiabatic expansion . following similar arguments like @xmath10 quench case , the divergence in @xmath141 can be estimated to yield _ 1(,u ) ~ _ u ( ) [ 1 - 20c ] where @xmath220 . the condition for break down of adiabaticity for linear quench ( @xmath221 near the critical point ) is similar to @xmath179 dim case , t v^ ~1 when adiabiticity breaks down , the system enters a scaling regime just like the @xmath179 dim case . following arguments similar to what discussed in the previous section we find , @xmath222 the first term defines the scaling relation . the expectation value of the scalar field in the scaling regime has the following properties , = < ( t v^ ) > v^ in this section we will solve the equations ( [ 1 - 10 ] ) numerically , although our analysis is quite preliminary . mainly we would like to understand how the system behaves away from the regime of slowly varying quench . it is generally hard to solve pde s numerically . however we will use the convenient method of lines . in this method the pde is at first discretized only in the @xmath96 direction . this gives rise to a set of odes . this set of odes is then treated in a ode solver . we fix the boundary condition at the two ends , at the boundary and at the horizon . we use a regularity boundary condition at the horizon and appropriate standard quantization condition at infinity . the initial radial profile of the scalar field and its derivative are given at @xmath22 . in most cases the initial value scalar profile would be a solution with @xmath223 . hence @xmath224 . then we will change @xmath0 to some other value higher than the critical mass . the first point to note is that at late times , i.e. out of the critical region , any finite energy excitation is gradually sucked up by the black hole . in fact at late times , the equation can be approximately linearized and the solution in a black hole background can be expressed as a sum of quasi normal modes . hence , the quasi normal frequency with smallest imaginary part determines the late time decay of the scalar field in a black hole geometry @xcite . we find a similar decaying behavior even in the full non - linear theory ( fig [ dasbasu ] ) . note that the scalar field falls in the black hole as expected . a time dependent @xmath73 changes the scalar field profile near the boundary . as the potential vanishes near the black hole horizon , changing the value of @xmath73 does not have an immediate effect . hence a time dependent @xmath73 creates disturbances near the boundary . the disturbances then propagate towards horizon along a light cone . in this work we have taken a first step in using ads / cft methods to study quantum quench across holographic critical points . the particular critical point we study is a mean field transition at any finite temperature , and it is not surprising that the critical behavior can be obtained by a landau - ginsburg type model of an appropriate mode . it may appear surprising that the dynamics involves a first order time derivative , even though the underlying theory is lorentz invariant . as discussed above , this is a feature of non - zero temperature which manifests itself as a dissipative horizon in the bulk . however , similar techniques can be used to study other interesting non - mean field theory transitions . in particular , the zero temperature limit of the transition which we study is of berezinski - kosterlitz - thouless type . other examples of such non - trivial transitions are discussed in @xcite . the situation is more subtle in this case : it would be worthwhile to study quench in these systems . it would be also interesting to study dynamics of holographic transitions associated with black branes whose ir geometry is of lifshitz type @xcite . as discussed above , ads / cft maps quantum quench in a strongly coupled theory to a classical quench in one higher dimension . the dynamics is completely deterministic , as opposed to e.g. langevin dynamics - the effect of a non - zero temperature being accounted by the presence of a black brane in the bulk . such systems have also been studied in the past @xcite . in these studies , however the interesting questions are rather different . they involve , for example , long time behavior of fluctuations and correlation functions when averaged over initial conditions . these questions can be addressed in our framework as well - they would correspond to @xmath225 corrections of the dual field theory and are left to future work . a related direction is the study of entanglement entropy during a quench which has been studied holographically in other contexts in @xcite-@xcite . finally , it would be interesting to extend our considerations to inhomogeneous quench . in this case one interesting question relates to defect formation as one crosses the phase transition . some aspects of this question has been studied numerically in @xcite . it would be interesting to see if expected scaling properties continue to hold , and if there is any analytical insight in terms of the eigenmodes of the radial operator . in this section we will study quantum quench in a simple _ relativistic _ landau - ginsburg model . the relevant lagranian is = - ^2 + m^2(t ) + ^4 + j(t ) [ 22 - 1 ] in our language this is a system with @xmath226 quench in a holographic model without any black hole horizon , e.g. a quench in global @xmath11 , is expected to obey eqn ( 22 - 1 ) as an effective description . this is suggested by the fact that in this latter situation the lowest order adiabatic correction is proportional to the second derivative of the source , as is consistent with lack of dissipation ( see e.g. the last reference in @xcite . this kind of model also appears in dynamics of phase transitions in cosmology @xcite a thorough discussion on those issues is left to the future . the lagrangian ( [ 22 - 1 ] ) leads to the equation of motion + m^2(t)+^3+j(t)=0 [ 22 - 2b ] when @xmath0 and @xmath10 are time independent , this model has a critical point at @xmath29 . we study quench by seting @xmath227 and considering a time dependent @xmath228 . for small @xmath24 , breakdown of adiabaticity occurs when , tv^ ~1 the scaling solution in the critical regime is given by , ( t)= v^ ( t v^ ) . as discussed in the main text this scaling relation fixes @xmath229 and zero crossing time @xmath230 . if we continue the quench we may ask how much excess energy ( or the equivalently the amplitude of oscillation ) remains in the system at late times . if the adiabatic approximation were valid , the excess energy would be suppressed exponentially . however , here the system goes through a critical point where adiabaticity is broken and at late times the energy turns out to scale as a power of @xmath24 . the purpose of the next two appendices is to justify the perturbation expansion in @xmath231 used to derive the critical region scaling in section 4.2 . the idea is to examine the behavior of the eigenfunctions for small values of @xmath201 . before tackling the problem at hand we discuss a toy model in this appendix . the next appendix discusses the eigenfunctions of the real problem . the toy model involves the eigenvalue equation -_^2 _ k + v ( ) _ k = k^2 _ k [ a1 - 1 ] where the potential is a step function @xmath232 we want to find the eigenfunctions @xmath233 with continuous spectrum , which clearly starts from zero . this potential has features which are similar to the potential @xmath95 in the text . the solution for @xmath234 is given by @xmath235 the solution for @xmath236 is @xmath237 the @xmath201-independent coefficient in ( [ rhogtzero ] ) follows from normalization of the wave function . this may be verified by solving the problem with a ir cutoff at some large value of @xmath238 , imposing dirichlet boundary conditions there , and finally taking the limit @xmath239 . even though there is a continuous spectrum starting at @xmath202 the green s function for the operator on the left hand side of ( [ a1 - 1 ] ) exists for generic values of @xmath214 . however for = ( n + ) n=0,1,2,there is a zero mode solution which vanishes at @xmath94 _ and _ is regular at @xmath93 , @xmath240 , \quad \rho < 1 \\ \nonumber & = \frac{1}{\sqrt{\pi } } , \quad \rho > 1 \end{aligned}\ ] ] for these values of the potential , the green s function diverges . the potential for @xmath241 above corresponds to a critical point . matching two solutions at @xmath242 we get @xmath243 leading to @xmath244 we now examine the behavior of the wavefunctions in the small @xmath201 limit . for a generic potential @xmath245 , the leading order behavior for small @xmath201 is given by @xmath246 now consider a function @xmath247 , with finite support . the convolution of @xmath247 with the eigenfunction @xmath233 , in the @xmath200 limit becomes @xmath248 now let us discuss the critical case @xmath249 . in this case we get , upto an overall @xmath201-independent constant , @xmath250 this leads to @xmath251 in this appendix we will prove the scaling of quantities like @xmath252 which appear in ( [ scalingeqn ] ) . the eigenvalue problem we wish to solve is , @xmath253 @xmath201 is a continuous parameter which may be chosen to be positive . we will assume some simple properties for the potential @xmath98 near @xmath254 , i.e. , @xmath255 as @xmath159 . near the boundary @xmath94 the potential @xmath98 is assumed to be sufficiently well behaved so that the above eom has a solution of the form , _ k(0)~a ^_++ b ^_- . where @xmath256 and @xmath257 are coefficients of leading and subleading modes . the potential @xmath98 in the problem of interest ( i.e. the equation in a black brane background ) satisfies these properties . we use the boundary condition , @xmath258 the normalization we use is , @xmath259 . to solve the equation ( [ appeom ] ) for small @xmath201 , we divide the range of @xmath91 in two overlapping parts : ( i ) in the `` far '' region @xmath260 we may use @xmath261 ( ii ) in the `` near '' region we can treat @xmath262 purturbatively . then we match these two solutions in an overlapping region to find a small @xmath201 solution . in the perturbative `` far '' region : @xmath263 here @xmath264 is the solution of the @xmath202 equation with the correct boundary conditions at @xmath94 and @xmath265 is a constant to be determined . near @xmath159 , @xmath266 . the boundary condition we choose for the necessary derivative of @xmath267 is such that @xmath268 . for @xmath260 the equation ( [ appeom ] ) may be approximated as , @xmath269 the real solution of the above equation is given by , @xmath270 for @xmath159 , we have : @xmath271 demanding that the above solution reaches a delta function normalizable sinusoidal solution we have , @xmath272 this condition is needed for the normalization we use and is similar to the choice of coefficient in ( [ rhogtzero ] ) in the toy problem considered in the previous appendix . in the small @xmath201 limit we have a matching region ( latexmath:[$1\ll \rho_0 \ll small argument and match with the perturbative solution . we get at lowest order in @xmath201 , @xmath274 hence generically for small @xmath201 , @xmath275 ( note @xmath268 ) . things are special at the critical point , as @xmath276 there . at the critical point @xmath277 and @xmath278 . now consider a function @xmath247 with a compact support such that , @xmath279 is finite . now generically , @xmath280 in contrast , at the critical point , @xmath281 the subleading piece comes from the perturbative @xmath262 corrections . we would like to thank karl landsteiner , satya majumdar , gautam mandal , shiraz minwalla , takeshi morita , ganpathy murthy , tatsuma nishioka , omid saremi , alfred shapere , tadashi takayanagi , sandip trivedi and especially kristan jensen and krishnendu sengupta for disucssions . s.r.d . would like to thank institut de fisica teorica at madrid , tata institute of fundamental research at mumbai and indian association for the cultivation of science at kolkata for hospitality during the final stages of this work . this work is partially supported by national science foundation grants phy-0970069 and phy-0855614 . p. calabrese and j. l. cardy , j. stat . mech . * 0504 * ( 2005 ) p010 [ arxiv : cond - mat/0503393 ] ; p. calabrese and j. l. cardy , phys . * 96 * ( 2006 ) 136801 [ arxiv : cond - mat/0601225 ] p. calabrese and j. cardy , [ arxiv:0704.1880 [ cond-mat.stat-mech ] ] ; s. sotiriadis and j. cardy , j. stat . mech . ( 2008 ) p11003 , [ arxiv:0808.0116 [ cond-mat.stat-mech ] ] ; s. sotiriadis , p. calabrese and j. cardy , epl * 87 * ( 2009 ) 20002 , [ arxiv:0903.0895 [ cond-mat.stat-mech ] ] ; s. sotiriadis and j. cardy , arxiv:1002.0167 [ quant - ph ] . o. aharony , s. s. gubser , j. m. maldacena , h. ooguri and y. oz , phys . rept . * 323 * , 183 ( 2000 ) [ arxiv : hep - th/9905111 ] . r. a. janik and r. b. peschanski , phys . d * 74 * , 046007 ( 2006 ) [ arxiv : hep - th/0606149 ] ; r. a. janik , phys . lett . * 98 * , 022302 ( 2007 ) [ arxiv : hep - th/0610144 ] ; p. m. chesler and l. g. yaffe , phys . lett . * 102 * , 211601 ( 2009 ) [ arxiv:0812.2053 [ hep - th ] ] ; p. m. chesler , l. g. yaffe , phys . * d82 * , 026006 ( 2010 ) . [ arxiv:0906.4426 [ hep - th ] ] ; s. bhattacharyya and s. minwalla , jhep * 0909 * ( 2009 ) 034 [ arxiv:0904.0464 [ hep - th ] ] . j. abajo - arrastia , j. aparicio , e. lopez , jhep * 1011 * , 149 ( 2010 ) . [ arxiv:1006.4090 [ hep - th]];h . ebrahim , m. headrick , [ arxiv:1010.5443 [ hep - th ] ] ; v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer _ et al . _ , phys . lett . * 106 * , 191601 ( 2011 ) . [ arxiv:1012.4753 [ hep - th ] ] ; v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer _ et al . _ , phys . rev . * d84 * , 026010 ( 2011 ) . [ arxiv:1103.2683 [ hep - th ] ] . t. albash , c. v. johnson , new j. phys . * 13 * , 045017 ( 2011 ) . [ arxiv:1008.3027 [ hep - th ] ] ; t. takayanagi , t. ugajin , jhep * 1011 * , 054 ( 2010 ) . [ arxiv:1008.3439 [ hep - th ] ] ; c. t. asplund , s. g. avery , [ arxiv:1108.2510 [ hep - th ] ] . s. r. das , j. michelson , k. narayan and s. p. trivedi , phys . d * 74 * , 026002 ( 2006 ) [ arxiv : hep - th/0602107 ] ; a. awad , s. r. das , a. ghosh , j. h. oh and s. p. trivedi , phys . d * 80 * ( 2009 ) 126011 [ arxiv:0906.3275 [ hep - th ] ] . t. hertog and g. t. horowitz , jhep * 0407 * , 073 ( 2004 ) [ arxiv : hep - th/0406134 ] ; t. hertog and g. t. horowitz , jhep * 0504 * , 005 ( 2005 ) [ arxiv : hep - th/0503071 ] ; n. turok , b. craps and t. hertog , arxiv:0711.1824 [ hep - th ] ; b. craps , t. hertog and n. turok , arxiv:0712.4180 [ hep - th ] . j. mcgreevy and e. silverstein , jhep * 0508 * , 090 ( 2005 ) [ arxiv : hep - th/0506130 ] ; e. silverstein , phys . rev . d * 73 * , 086004 ( 2006 ) [ arxiv : hep - th/0510044 ] ; g. horowitz , a. lawrence and e. silverstein , jhep * 0907 * , 057 ( 2009 ) [ arxiv:0904.3922 [ hep - th ] ] . s. r. das , t. nishioka , t. takayanagi , jhep * 1007 * , 071 ( 2010 ) . [ arxiv:1005.3348 [ hep - th ] ] . k. hashimoto , n. iizuka , t. oka , [ arxiv:1012.4463 [ hep - th ] ] ; k. -y . kim , j. p. shock , j. tarrio , jhep * 1106 * , 017 ( 2011 ) . [ arxiv:1103.4581 [ hep - th ] ] ; s. prem kumar , phys . * d84 * , 026003 ( 2011 ) . [ arxiv:1104.1405 [ hep - th ] ] . s. janiszewski , a. karch , [ arxiv:1106.4010 [ hep - th ] ] ; c. hoyos , t. nishioka , a. obannon , [ arxiv:1106.4030 [ hep - th ] ] . u. gursoy , e. kiritsis , l. mazzanti , f. nitti , nucl . phys . * b820 * , 148 - 177 ( 2009 ) . [ arxiv:0903.2859 [ hep - th ] ] . u. gursoy , e. kiritsis , l. mazzanti , f. nitti , jhep * 1012 * , 088 ( 2010 ) . [ arxiv:1006.3261 [ hep - th ] ] . j. g. russo , p. k. townsend , class . grav . * 25 * , 175017 ( 2008 ) . [ arxiv:0805.3488 [ hep - th ] ] ; m. chernicoff and a. guijosa , jhep * 0806 * ( 2008 ) 005 [ arxiv:0803.3070 [ hep - th ] ] ; a. paredes , k. peeters and m. zamaklar , jhep * 0904 * , 015 ( 2009 ) [ arxiv:0812.0981 [ hep - th ] ] ; c. athanasiou , p. m. chesler , h. liu , d. nickel and k. rajagopal , arxiv:1001.3880 [ hep - th ] ; e. caceres , m. chernicoff , a. guijosa and j. f. pedraza , arxiv:1003.5332 [ hep - th ] ; t. hirata , s. mukohyama and t. takayanagi , jhep * 0805 * ( 2008 ) 089 [ arxiv:0804.1176 [ hep - th ] ] ; t. hirayama , p. w. kao , s. kawamoto and f. l. lin , arxiv:1001.1289 [ hep - th ] ; a. obannon , jhep * 0901 * , 074 ( 2009 ) . [ arxiv:0811.0198 [ hep - th ] ] ; n. evans , e. threlfall , [ arxiv:0807.3679 [ hep - th ] ] ; n. evans , e. threlfall , phys . rev . * d79 * , 066008 ( 2009 ) . [ arxiv:0812.3273 [ hep - th ] ] . s. a. hartnoll , c. p. herzog , g. t. horowitz , jhep * 0812 * , 015 ( 2008 ) . [ arxiv:0810.1563 [ hep - th ] ] ; s. a. hartnoll , c. p. herzog , g. t. horowitz , phys . * 101 * , 031601 ( 2008 ) . [ arxiv:0803.3295 [ hep - th ] ] . k. jensen , a. karch , d. t. son , e. g. thompson , phys . lett . * 105 * , 041601 ( 2010 ) ; [ arxiv:1002.3159 [ hep - th]];k . jensen , phys . rev . * d82 * , 046005 ( 2010 ) . [ arxiv:1006.3066 [ hep - th]];n . evans , k. jensen , k. -y . kim , phys . rev . * d82 * , 105012 ( 2010 ) . [ arxiv:1008.1889 [ hep - th ] ] .

* purpose : * optoacoustic tomography ( oat ) is inherently a three - dimensional ( 3d ) inverse problem . however , most studies of oat image reconstruction still employ two - dimensional ( 2d ) imaging models . one important reason is because 3d image reconstruction is computationally burdensome . the aim of this work is to accelerate existing image reconstruction algorithms for 3d oat by use of parallel programming techniques . * methods : * parallelization strategies are proposed to accelerate a filtered backprojection ( fbp ) algorithm and two different pairs of projection / backprojection operations that correspond to two different numerical imaging models . the algorithms are designed to fully exploit the parallel computing power of graphic processing units ( gpus ) . in order to evaluate the parallelization strategies for the projection / backprojection pairs , an iterative image reconstruction algorithm is implemented . computer - simulation and experimental studies are conducted to investigate the computational efficiency and numerical accuracy of the developed algorithms . * results : * the gpu implementations improve the computational efficiency by factors of , , and for the fbp algorithm and the two pairs of projection / backprojection operators , respectively . accurate images are reconstructed by use of the fbp and iterative image reconstruction algorithms from both computer - simulated and experimental data . * conclusions : * parallelization strategies for 3d oat image reconstruction are proposed for the first time . these gpu - based implementations significantly reduce the computational time for 3d image reconstruction , complementing our earlier work on 3d oat iterative image reconstruction .

1304.2017

optoacoustic tomography ( oat ) , also known as photoacoustic computed tomography , is an emerging imaging modality that has great potential for a wide range of biomedical imaging applications @xcite . in oat , a short laser pulse is employed to irradiate biological tissues . when the biological tissues absorb the optical energy , acoustic wave fields can be generated via the thermoacoustic effect . the acoustic wave fields propagate outward in three - dimensional ( 3d ) space and are measured by use of ultrasonic transducers that are distributed outside the object . the goal of oat is to obtain an estimate of the absorbed energy density map within the object from the measured acoustic signals . to accomplish this , an image reconstruction algorithm is required . a variety of analytic image reconstruction algorithms have been proposed @xcite . these algorithms generally assume an idealized transducer model and an acoustically homogeneous medium . also , since they are based on discretization of continuous reconstruction formulae , these algorithms require the acoustic pressure to be densely sampled over a surface that encloses the object to obtain an accurate reconstruction . to overcome these limitations , iterative image reconstruction algorithms have been proposed @xcite . although the optoacoustic wave intrinsically propagates in 3d space , when applying to experimental data , most studies have employed two - dimensional ( 2d ) imaging models by making certain assumptions on the transducer responses and/or the object structures @xcite . an important reason is because the computation required for 3d oat image reconstruction is excessively burdensome . therefore , acceleration of 3d image reconstruction will facilitate algorithm development and many applications including real - time 3d pact @xcite . a graphics processing unit ( gpu ) card is a specialized device specifically designed for parallel computations @xcite . compute unified device architecture ( cuda ) is an extension of the c / fortran language that provides a convenient programming platform to exploit the parallel computational power of gpus @xcite . the cuda - based parallel programming technique has been successfully applied to accelerate image reconstruction in mature imaging modalities such as x - ray computed tomography ( ct ) @xcite and magnetic resonance imaging ( mri ) @xcite . in oat , however , only a few works on utilization of gpus to accelerate image reconstruction have been reported @xcite . for example , the k - wave toolbox employs the nvidia cuda fast fourier transform library ( cufft ) to accelerate the computation of 3d fft @xcite . also a gpu - based sparse matrix - vector multiplication strategy has been applied to 3d oat image reconstruction for the case that the system matrix is sparse and can be stored in memory @xcite . however , there remains an important need to develop efficient implementations of oat reconstruction algorithms for general applications in which the system matrix is too large to be stored . in this work , we propose parallelization strategies , for use with gpus , to accelerate 3d image reconstruction in oat . both filtered backprojection ( fbp ) and iterative image reconstruction algorithms are investigated . for use with iterative image reconstruction algorithms , we focus on the parallelization of projection and backprojection operators . specifically , we develop two pairs of projection / backprojection operators that correspond to two distinct discrete - to - discrete ( d - d ) imaging models employed in oat , namely the interpolation - based and the spherical - voxel - based d - d imaging models . note that our implementations of the backprojection operators compute the exact adjoint of the forward operators , and therefore the projector pairs are ` matched ' @xcite . the remainder of the article is organized as follows . in section [ sect : background ] , we briefly review oat imaging models in their continuous and discrete forms . we propose gpu - based parallelization strategies in section [ sect : gpumethods ] . numerical studies and results are described in section [ sect : nummethods ] and section [ sect : results ] respectively . finally , a brief discussion and summary of the proposed algorithms are provided in section [ sect : summary ] . a continuous - to - continuous ( c - c ) oat imaging model neglects sampling effects and provides a mapping from the absorbed energy density function @xmath3 to the induced acoustic pressure function @xmath4 . here , @xmath5 is the temporal coordinate , @xmath6 and @xmath7 denote the locations within the object support @xmath8 and on the measurement surface @xmath9 , respectively . a canonical oat c - c imaging model can be expressed as @xcite : @xmath10 where @xmath11 is the dirac delta function , @xmath12 , @xmath13 , and @xmath14 denote the thermal coefficient of volume expansion , ( constant ) speed - of - sound , and the specific heat capacity of the medium at constant pressure , respectively . we introduce an operator notation @xmath15 to denote this c - c mapping . alternatively , eqn . can be reformulated as the well - known spherical radon transform ( srt ) @xcite : @xmath16 where the function @xmath17 is related to @xmath4 as @xmath18 the srt model provides an intuitive interpretation of each value of @xmath17 as a surface integral of @xmath3 over a sphere centered at @xmath19 with radius @xmath20 . based on c - c imaging models , a variety of analytic image reconstruction algorithms have been developed @xcite . for the case of a spherical measurement geometry , an fbp algorithm in its continuous form is given by @xcite : @xmath21_{t=\frac{|\mathbf r-\mathbf r^s|}{c_0}},\ ] ] where @xmath22 denotes the radius of the measurement surface @xmath9 . when sampling effects are considered , an oat system is properly described as a continuous - to - discrete ( c - d ) imaging model @xcite : @xmath23_{qk+k } = h^e(t ) * _ t \frac{1}{s_q } \int_{s_q}\!\ ! d \mathbf r^s\ , p(\mathbf r^s , t)\big|_{t = k\delta_t},\quad \substack { q=0,1,\cdots , q-1\\ k=0,1,\cdots , k-1 } , \ ] ] where @xmath24 and @xmath25 denote the total numbers of transducers ( indexed by @xmath26 ) and the time samples ( indexed by @xmath27 ) respectively . @xmath28 is the surface area of the @xmath26-th transducer , which is assumed to be a subset of @xmath9 ; @xmath29 denotes the acousto - electric impulse response ( eir ) of each transducer that , without loss of generality , is assumed to be identical for all transducers ; ` @xmath30 ' denotes a linear convolution with respect to time coordinate ; and @xmath31 is the temporal sampling interval . the vector @xmath32 represents the lexicographically ordered measured voltage signals whose @xmath33-th element is denoted by @xmath34_{qk+k}$ ] . in order to apply iterative image reconstruction algorithms a d - d imaging model is required , which necessitates the discretization of @xmath3 . the following @xmath35-dimensional representation of the object function can be employed @xcite : @xmath36_n \psi_n(\mathbf r),\ ] ] where @xmath37 is a coefficient vector whose @xmath38-th element is denoted by @xmath39_n$ ] and @xmath40 is the expansion function . on substitution from eqn . into eqn . , where @xmath4 is defined by eqn . , one obtains a d - d mapping from @xmath37 to @xmath32 , expressed as @xmath41 where each element of the matrix @xmath42 is defined as @xmath43_{qk+k , n } = \big [ h^e * _ t \frac{1}{s_q } \int_{s_q}\!\!d \mathbf r^s \mathcal h_{\rm cc } \psi_n \big]_{t = k\delta_t}.\ ] ] here , @xmath42 is the d - d imaging operator also known as system matrix or projection operator . note that the ` @xmath44 ' in eqn . is due to the use of the finite - dimensional representation of the object function ( i.e. , eqn . ) . no additional approximations have been introduced . below we describe two types of d - d imaging models that have been employed in oat @xcite : the interpolation - based imaging model and the spherical - voxel - based imaging model . the quantities @xmath32 , @xmath42 , and @xmath45 ( or @xmath46 ) in the two models will be distinguished by the subscripts ( or superscripts ) ` int ' and ` sph ' , respectively . the interpolation - based d - d imaging model defines the coefficient vector as samples of the object function on the nodes of a uniform cartesian grid : @xmath47_n = \int_v\!\!d \mathbf r\ , \delta(\mathbf r-\mathbf r_n ) a(\mathbf r),\quad n=0,1,\cdots , n-1,\ ] ] where , @xmath48 specifies the location of the @xmath38-th node of the uniform cartesian grid . the definition of the expansion function depends on the choice of interpolation method @xcite . if a trilinear interpolation method is employed , the expansion function can be expressed as @xcite : @xmath49 where @xmath50 is the distance between two neighboring grid points . in principle , the interpolation - based d - d imaging model can be constructed by substitution from eqns . and to eqn . . in practice , however , implementation of the surface integral over @xmath28 is difficult for the choice of expansion functions in eqn . . also , implementations of the temporal convolution and @xmath51 usually require extra discretization procedures . therefore , utilization of the interpolation - based d - d model commonly assumes the transducers to be point - like . in this case , the implementation of @xmath52 is decomposed as a three - step operation : @xmath53 where @xmath54 , @xmath55 , and @xmath56 are discrete approximations of the srt ( eqn . ) , the differential operator ( eqn . ) , and the operator that implements a temporal convolution with eir , respectively . we implemented @xmath54 in a way @xcite that is similar to the ` ray - driven ' implementation of radon transform in x - ray ct @xcite , i.e , for each data sample , we accumulated the contributions from the voxels that resided on the spherical shell specified by the data sample . by use of eqns . , , , and , one obtains : @xmath57_{qk+k } = \delta_s^2 \sum_{n=0}^{n-1 } \big[\boldsymbol \alpha_{\rm int}\big]_n \sum_{i=0}^{n_i-1 } \sum_{j=0}^{n_j-1 } \psi_n^{\rm int}(\mathbf r_{k , i , j } ) \equiv \big[\mathbf g\big]_{qk+k } , \ ] ] where @xmath58_{qk+k } \approx g(\mathbf r^s_q , t)|_{t = k\delta_t}$ ] with @xmath59 specifying the location of the @xmath26-th point - like transducer , and @xmath60 and @xmath61 denote the numbers of divisions over the two angular coordinates of a local spherical coordinate system shown in fig . [ fig : geo]-(b ) . a derivation of eqn . is provided in appendix . the differential operator in eqn . is approximated as @xmath62_{qk+k } = \frac{\beta}{8\pi c_p\delta_t^2 } \big(\frac{[\mathbf g]_{qk+k+1}}{k+1 } -\frac{[\mathbf g]_{qk+k-1}}{k-1 } \big ) \equiv \big[\mathbf p_{\rm int}\big]_{qk+k},\ ] ] where @xmath63_{qk+k } \approx p(\mathbf r^s_q , t)|_{t = k\delta_t}$ ] . finally , the continuous temporal convolution is approximated by a discrete linear convolution as @xcite @xmath64_{qk+k } = \sum_{\kappa = 0}^{k-1 } [ \mathbf h^e ] _ { k-1-\kappa } [ \mathbf p_{\rm int}]_{qk+\kappa } \equiv [ \mathbf u_{\rm int}]_{qk+k},\ ] ] where @xmath65_{k } = \delta_t h^e(t)|_{t = k\delta_t}$ ] . the spherical - voxel - based imaging model is also widely employed in oat @xcite . it employs the expansion functions @xmath66 where @xmath67 is defined as in eqn . . the @xmath38-th expansion function @xmath68 is a uniform sphere that is inscribed by the @xmath38-th cuboid of a cartesian grid . the @xmath38-th component of the coefficient vector @xmath69 is defined as : @xmath70_n = \frac{v_{\text{cube}}}{v_{\text{sph } } } \int_{v}\!\ ! d\mathbf{r } \ ; \psi^{\rm sph}_n ( \mathbf{r})a(\mathbf{r}),\ ] ] where @xmath71 and @xmath72 are the volumes of a cubic voxel of dimension @xmath50 and of a spherical voxel of radius @xmath73 respectively . unlike the interpolation - based imaging model , by use of the expansion functions defined in eqn . , the surface integral over @xmath28 and @xmath74 in eqn . can be converted to a temporal convolution and calculated analytically @xcite . to avoid utilizing excessively high sampling rate to mitigate aliasing , the spherical - voxel - based imaging model can be conveniently implemented in the temporal frequency domain as @xcite : @xmath75_{ql+l}=\tilde p_0(f ) \sum_{n=0}^{n-1}\big[\boldsymbol\alpha_{\rm sph}\big]_n \frac{1}{s_q } \tilde h_q^s(\mathbf r_n , f ) \big|_{f = l\delta_f } , \quad { \rm for}\ , l=0 , 1 , \cdots , l-1,\ ] ] where @xmath76 is the frequency sampling interval , and @xmath77 denotes the total number of temporal - frequency samples indexed by @xmath78 . a derivation of eqn . can be found in ref . . the function @xmath79 represents the temporal fourier transform of the spatial impulse response ( sir ) of the @xmath26-th transducer for the source located at @xmath67 , expressed as : @xmath80 also , @xmath81 is defined as @xmath82 \tilde{h}^e(f),\ ] ] where @xmath83 is the eir in temporal - frequency domain . in summary , the imaging model can be expressed in matrix form as : @xmath84 iterative image reconstruction algorithms employ numerical implementations of the projection operator , i.e. , the system matrix @xmath42 , as well as its adjoint , denoted by @xmath85 @xcite . the adjoint is also referred to as the backprojection operator . note that for most practical applications , @xmath42 and @xmath85 are too large to be stored in the random access memory of currently available computers . therefore , in practice , the actions of @xmath42 and @xmath85 are almost always calculated on the fly . the same strategy was adopted in this work . according to the definition of the adjoint operator @xcite , @xmath86 , where @xmath87_{qk+k } = \sum_{\kappa=0}^{k-1 } [ \mathbf h^e]_{\kappa-1-k } [ \mathbf u]_{qk+\kappa } \equiv [ \mathbf p'_{\rm int}]_{qk+k},\ ] ] @xmath88_{qk+k}=\frac{\beta}{8\pi c_p\delta_t^2k } \big ( \big[\mathbf p'_{\rm int}\big]_{qk+k-1 } -\big[\mathbf p'_{\rm int}\big]_{qk+k+1}\big ) \equiv \big[\mathbf g'\big]_{qk+k},\ ] ] and @xmath89_{n } = \delta_s^2 \sum_{q=0}^{q-1 } \sum_{k=0}^{k-1 } \big[\mathbf g'\big]_{qk+k } \sum_{i=0}^{n_i-1}\sum_{j=0}^{n_j-1}\psi_n^{\rm int } ( \mathbf r_{k , i , j } ) \equiv [ \boldsymbol\alpha'_{\rm int}]_n.\ ] ] it can also be verified that the adjoint operator @xmath90 is given by : @xmath91_{n } = \sum_{q=0}^{q-1}\sum_{l=0}^{l-1 } \big[\tilde{\mathbf u}_{\rm sph}\big]_{ql+l } \tilde{p}^*_0(f ) \tilde{h}^*_q(\mathbf r_n , f)\big|_{f = l\delta_f},\ ] ] where the superscript ` * ' denotes the complex conjugate . unlike the unmatched backprojection opertors @xcite that are obtained by discretization of the continuous adjoint operator , utilization of the exact adjoint operator facilitates the convergence of iterative image reconstruction algorithms . the key features of gpu architecture and the basics of cuda programming are briefly summarized in this section . we refer the readers to refs . for additional details . a gpu card contains multiple streaming multiprocessors . each streaming multiprocessor is configured with multiple processor cores . for example , the tesla c1060 possesses @xmath92 streaming multiprocessors with @xmath93 processor cores on each ; and the tesla c2050 possesses @xmath94 streaming multiprocessors with @xmath95 processor cores on each @xcite . the processor cores in each multiprocessor execute the same instruction on different pieces of data , which is referred to as `` single instruction , multiple data '' ( simd ) model of parallel programming . in order to fully exploit the computing power of gpus , one of the major challenges is to design a parallelization strategy fitting in the simd framework such that the largest number of processor cores can execute the computation simultaneously @xcite . a gpu card has six types of memory that have varying capacities and different access rules and efficiencies : ( 1 ) registers are assigned for each thread and have the fastest access . ( 2 ) shared memory is assigned for each block and can be efficiently accessed by all threads in the block if designed appropriately . ( 3 ) constant memory is read - only and can be accessed by all threads efficiently . ( 4 ) texture memory is also read - only and is optimized for interpolation operations . ( 5 ) global memory has the slowest access that takes hundreds times more clock cycles than does the computation of basic arithmetic operations . ( 6 ) local memory is assigned for each thread but has a slow access as does the global memory . therefore , an efficient gpu - based implementation in general requires a limited number of global and local memory access . cuda is a platform and programming model developed by nvidia that includes a collection of functions and keywords to exploit the parallel computing power of gpus @xcite . a cuda parallel program is composed of a host program and kernels . the host program is executed by cpus and launches the kernels , which are custom - designed functions executed by gpus . a general parallel programming strategy is to launch multiple instances of a kernel and to run the multiple instances concurrently on gpus . in cuda , each instance of the kernel is named as a thread and processes only a portion of the data . a hierarchy of threads is employed : threads are grouped into blocks , and blocks are grouped into a grid . therefore , each thread is specified by a multi - index containing a block index and a thread index within the block . in this section , we propose gpu - based parallelization strategies for the fbp algorithm and the projection / backprojection operations corresponding to the interpolation - based and the spherical - voxel - based d - d imaging models . we employed a spherical measurement geometry shown in fig . [ fig : geo]-(a ) . the measurement sphere was of radius @xmath22 centered at the origin of the cartesian coordinate system ( or the equivalent spherical coordinate system ) . the polar angle @xmath96 $ ] was equally divided with interval @xmath97 , starting from @xmath98 . at each polar angle , a ring on the sphere that was parallel to the plane @xmath99 can be specified , resulting @xmath100 rings . on each ring , @xmath101 ultrasonic transducers were assumed to be uniformly distributed with azimuth angle interval @xmath102 . hereafter , each azimuth angle will be referred to as a tomographic view . at each view , we assumed that @xmath103 temporal samples were acquired and the first sample corresponded to time instance @xmath104 . for implementations in temporal - frequency domain , we assumed that @xmath105 temporal - frequency samples were available and the first sample corresponded to @xmath106 . the region to be reconstructed was a rectangular cuboid whose edges were parallel to the axes of the coordinate system and the left - bottom - back vertex was located at @xmath107 . the numbers of voxels along the three coordinates will be denoted by @xmath108 , @xmath109 , and @xmath110 , respectively , totally @xmath111 voxels . we also assumed the cuboid was contained in another sphere of radius @xmath112 that was concentric with the measurement sphere shown in fig . [ fig : geo]-(b ) . central processing unit ( cpu)-based implementations of continuous fbp formulae have been described in refs . . though the discretization methods vary , in general , three approximations have to be employed . firstly , the first - order derivative term @xmath113 has to be approximated by a difference scheme up to certain order @xcite . secondly , the measurement sphere has to be divided into small patches , and the surface integral has to be approximated by a summation of the area of every patch weighted by the effective value of the integrand on the patch . finally , the value of the integrand at an arbitrary time instance @xmath114 has to be approximated by certain interpolation method . in this study , we approximated the surface integral by use of the trapezoidal rule . as described earlier , the spherical surface was divided into @xmath115 patches . for the transducer indexed by @xmath26 that was located at @xmath116 , the area of the patch was approximated by @xmath117 . the value at time instance @xmath118 was approximated by the linear interpolation from its two neighboring samples as : @xmath119_{qk+k } + \big(\tilde k - k\big)\big[\mathbf p\big]_{qk+k+1},\ ] ] where @xmath120 , and @xmath27 is the integer part of @xmath121 . here @xmath122 is a vector of lexicographically ordered samples of the pressure function @xmath4 , which is estimated from the measured voltage data vector @xmath32 . also , the first - order derivative term was approximated by : @xmath123_{qk+k+1 } - \big[\mathbf p\big]_{qk+k } \big).\ ] ] by use of these three numerical approximations , the discretized fbp formula was expressed as : @xmath124_n = -\frac{c_pr^s\delta_{\theta^s}\delta_{\phi^s}}{\pi\beta c_0 ^ 3\delta_t } \sum_{n_r=0}^{n_r-1}\sin \theta^s_q & \sum_{n_v=0}^{n_v-1 } \bigg\ { \big(1.5-\frac { k+t_{\rm min}/\delta_t } { \tilde k + t_{\rm min}/\delta_t}\big)\big[\mathbf p\big]_{qk+k+1}\\ & + \big(\frac { k+1+t_{\rm min}/\delta_t } { \tilde k + t_{\rm min}/\delta_t}-1.5\big)\big[\mathbf p\big]_{qk+k } \bigg\}. \end{split}\ ] ] unlike the implementations of fbp formulas in x - ray cone beam ct @xcite , we combined the filter and the linear interpolation . this reduced the number of visits to the global memory in the gpu implementation described below . we implemented the fbp formula in a way that is similar to the ` pixel - driven ' implementation in x - ray ct @xcite , i.e. , we assigned each thread to execute the two accumulative summations in eqn . for each voxel . we bound the pressure data @xmath122 to texture memory because it is cached and has a faster accessing rate . therefore our implementation only requires access to texture memory twice and to global memory once . the pseudo - codes are provided in algs . [ alg : fbp ] and [ alg : k_fbp ] for the host part and the device part respectively . note that the pseudo - codes do not intend to be always optimal because the performance of the codes could depend on the dimensions of @xmath122 and @xmath125 . for example , we set the block size to be @xmath126 because for our applications , @xmath110 was bigger than @xmath108 and @xmath109 and smaller than the limit number of threads that a block can support ( i.e. , 1024 for the nvidia tesla c2050 ) . if the values of @xmath108 , @xmath109 , and @xmath110 change , we may need to redesign the dimensions of the grid and blocks . however , the general simd parallelization strategy remains . the forward projection operation @xmath128 is composed of three consecutive operations @xmath129 , @xmath130 , and @xmath131 that are defined in eqns . , , and , respectively . both the difference operator @xmath55 and the one - dimensional ( 1d ) convolution @xmath56 have low computational complexities while the srt operator @xmath54 is computationally burdensome . hence , we developed the gpu - based implementation of @xmath54 while leaving @xmath55 and @xmath56 to be implemented by cpus . the srt in oat shares many features with the radon transform in x - ray ct . thus , our gpu - based implementation is closely related to the implementations of radon transform that have been optimized for x - ray ct@xcite . the surface integral was approximated according to the trapezoidal rule . firstly , the integral surface was divided into small patches , which is described in the appendix . secondly , each patch was assigned an effective value of the object function by trilinear interpolation . the trilinear interpolation was calculated by use of the texture memory of gpus that is specifically designed for interpolation . finally , gpu threads accumulated the areas of patches weighted by the effective values of the object function and wrote the final results to global memory . the pseudo - codes for implementation of @xmath54 are provided in algs . [ alg : inth ] and [ alg : k_inth ] for the host part and the device part , respectively . note that we employed the `` one - level''-strategy @xcite , i.e. , each thread calculates one data sample . higher level strategies have been proposed to improve the performance by assigning each block to calculate multiple data samples @xcite , which , however , caused many thread idles in oat mainly because the amount of computation required to calculate a data sample varies largely among samples for srt . implementation of the backprojection operator @xmath132 was very similar to the implementation of @xmath52 . the operators @xmath133 and @xmath134 were calculated on cpus while @xmath135 was calculated by use of gpus . the pseudo - codes are provided in algs . [ alg : intht ] and [ alg : k_intht ] . we made use of the cuda function ` atomicadd ' to add weights to global memory from each thread . implementation of the forward projection operation for the spherical - voxel - based imaging model is distinct from that of the interpolation - based model . the major difference is that calculation of each element of the data vector for the spherical - voxel - based imaging model requires the accumulation of the contributions from all voxels because the model is expressed in the temporal frequency domain . because of this , the amount of computation required to calculate each data sample in the spherical - voxel - based imaging model is almost identical , simplifying the parallelization strategy . we proposed a parallelization strategy that was inspired by one applied in advanced mri reconstruction @xcite and is summarized as follows . discrete samples of @xmath81 defined in eqn . were precalcualted and stored as a vector @xmath138 in constant memory . because the size of the input vector @xmath139 is often too large to fit in the constant memory , we divided @xmath139 into sub - vectors that matched the capacity of the constant memory . we employed a cpu loop to copy every sub - vector sequentially to the constant memory and call the gpu kernel function to accumulate a partial summation . the major advantage of this design is that the total number of global memory visits to calculate one data sample is reduced to the number of sub - vectors . implementation of the projection operator for the spherical - voxel - based imaging model generally involves more arithmetic operations than does the interpolation - based imaging model . moreover , the spherical - voxel - based imaging model has been employed to compensate for the finite aperture size effect of transducers @xcite , which makes the computation even more burdensome . because of this , we further developed an implementation that employed multiple gpus . the pseudo - codes of the projection operation are provided in algs . [ alg : sphh ] , [ alg : fwd_pthread ] , and [ alg : kfwdsph ] . we created @xmath140 pthreads on cpus by use of the ` pthread.h ' library . here , we denote the threads on cpus by ` pthread ' to distinguish from threads on gpus . we divided the input vector @xmath69 into @xmath140 sub - vectors ( denoted by @xmath141 s ) of equal size and declared an output vector @xmath142 of dimension @xmath143 . by calling the pthread function ` fwd_pthread ' , @xmath144 pthreads simultaneously calculated the projection . each pthread projected an @xmath141 to a partial voltage data vector @xmath145 that filled in the larger vector @xmath142 . once all pthreads finished filling their @xmath145 into @xmath142 , the projection data @xmath146 were obtained by a summation of the @xmath140 @xmath145 s . implementation of the backprojection operator was similar except the dividing and looping were over the vector @xmath146 instead of @xmath139 . the pseudo - codes for the backprojection operation are provided in algs . [ alg : sphht ] , [ alg : bwd_pthread ] , and [ alg : kbwdsph ] . the computational efficiency and accuracy of the proposed gpu - based implementations of the fbp algorithm and projection / backprojection operators for use with iterative image reconstruction algorithms were quantified in computer simulation and experimental oat imaging studies . _ numerical phantom : _ the numerical phantom consisted of @xmath147 uniform spheres that were blurred by a 3d gaussian kernel possessing a full width at half maximum ( fwhm ) of @xmath148-mm . the phantom was contained within a cuboid of size @xmath149-mm@xmath150 . a 2d image corresponding to the plane @xmath151 through the phantom is shown in fig . [ fig : fbp]-(a ) . .5 cm _ simulated projection data : _ the measurement surface was a sphere of radius @xmath152-mm . corresponding to an exsiting oat imaging system @xcite . as described in section [ sect : gpumethods ] , ideal point - like transducers were uniformly distributed over @xmath153 rings and @xmath154 tomographic views . the @xmath153 rings covered the full @xmath155 polar angle , i.e. , @xmath156 , while the @xmath154 views covered the full @xmath157 azimuth angle . the speed of sound was set at @xmath158-mm/@xmath159s . we selected the grneisen coefficient as @xmath160 of arbitrary units ( a.u . ) . for each transducer , we analytically calculated @xmath161 temporal samples of the pressure function at the sampling rate of @xmath162-mhz by use of eqn . . because we employed a smooth object function , the pressure data were calculated by the following two steps : firstly , we calculated temporal samples of pressure function @xmath163 that corresponds to the @xmath147 uniform spheres by @xcite @xmath164_{t = k\delta_t } , & { \rm if } \ ; \big|c_0k\delta_t-|\mathbf r^s -\mathbf r_i| \big| \leq r_i \\ 0 , & { \rm otherwise } \end{array}\right.\ ] ] where @xmath165 , @xmath166 and @xmath167 denote the center location , the radius and the absorbed energy density of the @xmath168-th sphere , respectively . subsequently , we convolved @xmath169 with a one - dimensional ( 1d ) gaussian kernel with @xmath170-@xmath171 @xcite to produce the pressure data . from the simulated pressure data , we calculated the temporal - frequency spectrum by use of fast fourier transform ( fft ) , from which we created an alternative data vector that contained @xmath172 frequency components occupying @xmath173$]-mhz for each transducer . the simulated projection data in either the time domain or the temporal frequency domain will hereafter be referred to as `` @xmath174''-data . by undersampling the `` @xmath174''-data uniformly over rings and tomographic views , we created three subsets that contained varying number of transducers . these data sets will be referred to as `` @xmath175''-data , `` @xmath176''-data , and `` @xmath177''-data , where the two numbers specify the number of rings and the number of tomographic views , respectively . .5 cm _ reconstruction algorithms : _ the gpu accelerated fbp algorithm was employed to reconstruct the object function sampled on a 3d cartesian grid with spacing @xmath178-mm . the dimension of the reconstructed images @xmath125 was @xmath179 . we employed an iterative image reconstruction algorithm that sought to minimize a penalized least - squares ( pls ) objective @xcite . two versions of the reconstruction algorithm were developed that utilized the interpolation - based imaging model and the spherical - voxel - based imaging model respectively . the two versions sought to solve the optimization problems by use of the linear conjugate gradient ( cg ) method @xcite : @xmath180 and @xmath181 respectively , where @xmath182 is a regularizing penalty term whose impact is controlled by the regularization parameter @xmath159 . the penalty term was employed only when processing the experimental data as described in section [ sect : nummethods]-b . the reconstruction algorithms required computation of one projection and one backprojection operation at each iteration . hereafter , the two reconstruction algorithms will be referred to as pls - int and pls - sph algorithms , respectively . we set @xmath178-mm . therefore , both the dimensions of @xmath183 and @xmath184 were @xmath179 . .5 cm _ performance assessment : _ we compared the computational times of 3d image reconstruction corresponding to the gpu- and cpu - based implementations . the cpu - based implementations of the pls - int and pls - sph algorithms take several days to complete a single iteration even for the `` @xmath177''-data . therefore , we only recorded the computational time for the cpu - based implementations to complete a single iteration when the data vector contained a single transducer . we assumed that the computational times were linearly proportional to the number of transducers in the data sets because the cpu - based implementations are sequential . the gpu - based implementations employed the single - precision floating - point format rather than the conventional double - precision utilized by cpu - based implementations . in order to quantify how the single - precision floating - point format would degrade the image accuracy , we calculated the root mean square error ( rmse ) between the reconstructed image and the phantom defined by : @xmath185 where @xmath37 and @xmath186 are the samples of the phantom and the coefficients of the reconstructed images respectively . .5 cm _ hardware specifications : _ all implementations were tested on the platform consisted of dual quad - core intel(r ) xeon ( r ) cpus with a clock speed @xmath187-ghz . the gpu - based implementations of the fbp and the pls - int algorithms were tested on a single tesla c2050 gpu , while the pls - sph algorithm was tested on @xmath93 tesla c1060 gpus . the fbp , pls - int and pls - sph algorithms were investigated by use of an existing data set corresponding to a live mouse @xcite . the scanning geometry and dimensions were the same as those employed in the computer - simulation studies except that only @xmath188 rings were uniformly distributed over the polar angle ranging from @xmath189 to @xmath190 . the transducers were of size @xmath191-mm@xmath192 . the raw data were acquired at @xmath193 tomographic views , which are referred to as `` full data '' . we undersampled the `` full data '' uniformly over the tomographic views , constructing a subset containing @xmath194 tomographic views . the subset will be referred to as `` quarter data '' . unlike in the idealized computer - simulation studies , the transducer response has to be compensated for when processing the experimental data . when implementing the fbp algorithm , the eir was compensated for by a direct fourier deconvolution , expressed in temporal frequency domain as @xcite : @xmath195 where @xmath196 is a window function for noise suppression . in this study , we adopted the hann window function defined as : @xmath197,\ ] ] where the cutoff frequency was chosen as @xmath198-mhz . when applying iterative image reconstruction algorithms , the transducer effects were implicitly compensated for during iteration by employing imaging models that incorporates the transducer charactertics @xcite . we incorporated the eir into the interpolation - based imaging model while incorporating both the eir and the sir into the spherical - voxel - based imaging model . for both pls - int and pls - sph algorithms , we employed a quadratic smoothness penalty to mitigate measurement noise @xcite : @xmath199_n - [ \boldsymbol\alpha]_{n_x}\big)^2 + \big([\boldsymbol\alpha]_n - [ \boldsymbol\alpha]_{n_y}\big)^2 + \big([\boldsymbol\alpha]_n - [ \boldsymbol\alpha]_{n_z}\big)^2,\ ] ] where @xmath200 , @xmath201 and @xmath202 were the indices of the neighboring voxels before the @xmath38-th voxel along the three cartesian axes , respectively . as shown in table [ tab : time ] , the gpu - based implementations took less than @xmath203 , @xmath204 and @xmath205 of the computational times required by corresponding cpu - based implementations for the fbp , the pls - int , and the pls - sph algorithms , respectively . the relative computational times for the gpu - based implementations are nearly linearly proportional to the amount of data . note that the `` @xmath206''-data and the `` quarter data '' are of the same size . however , the computational times of the `` quarter data '' are more than @xmath207 times those of the `` @xmath206''-data . this is because the calculation of the sir increases the computational complexity of the reconstruction algorithm . images reconstructed by use of the cpu- and gpu - based implementations of the fbp algorithm are almost identical . from the `` @xmath208''-data , in which case , transducers were densely distributed over the measurement surface , both implementations reconstructed accurate images , as shown fig . [ fig : fbp]-(b ) and -(c ) . the profiles along the three arrows in fig . [ fig : fbp ] are plotted in fig . [ fig : profile]-(a ) , suggesting a nearly exact reconstruction . as expected , when the amount of measurement data are reduced , the reconstructed images contain more artifacts as shown in fig . [ fig : fbp2 ] . however , the images reconstructed by use of gpu- and cpu - based implementations remain indistinguishable . the plots of the rmse versus the amount of measurement data employed in fig . [ fig : error ] overlap , also suggesting the single - precision floating - point format employed by the gpu - based implementation has little impact on the computational accuracy . the gpu - based implementations of the pls - int and pls - sph algorithms both reconstructed accurate images as displayed in fig . [ fig : pls ] . as expected , the images reconstructed by use of both iterative algorithms contain fewer artifacts than do those reconstructed by use of the fbp algorithm from the same amount of data . unlike the images reconstructed by use of the fbp algorithm from the `` @xmath206''-data ( fig . [ fig : fbp2]-(a ) or -(d ) ) , the images reconstructed by use of both iterative algorithms ( fig . [ fig : pls]-(a ) and -(d ) ) appear to be identical to the numerical phantom . the profiles along the two arrows in fig . [ fig : pls]-(a ) and -(d ) are plotted in fig . [ fig : profile]-(b ) , further confirming the computational accuracy of iterative image reconstruction algorithms . the plots of the rmse versus the amount of measurement data employed in fig . [ fig : error ] suggest the iterative image reconstruction algorithms in general outperform the fbp algorithm from the same amount of data . the maximum intensity projection ( mip ) of the 3d mouse images reconstructed by use of the gpu - based implementations reveal the mouse body vasculature as shown in fig . [ fig : mousev1803d ] . images reconstructed by use of both the pls - int and the pls - sph algorithms appear to have cleaner background than do the images reconstructed by use of the fbp algorithm from the same amount of data . all images reconstructed by iterative algorithms were obtained by @xmath209-iterations starting with uniform zeros as the initial guess . the pls - int algorithm took approximately a half day and @xmath210 days to process the `` quarter data '' and the `` full data '' respectively . the pls - sph algorithm took approximately one day and @xmath211 days to process the `` quarter data '' and the `` full data '' respectively . alternatively , if the cpu - based implementations were utilized , the pls - int algorithm would take an estimated @xmath212 days and @xmath213 days to process the `` quarter data '' and the `` full data '' respectively . the pls - sph algorithm would take an estimated @xmath214 days and @xmath215 days to process the `` quarter data '' and the `` full data '' respectively . in this study , we developed and investigated gpu - based implementations of the fbp algorithm and two pairs of projection / backprojection operators for 3d oat . our implementation of the fbp algorithm improved the computational efficiency over @xmath0 times compared to the cpu - based implementation . more importantly , our implementations of the projection / backprojection operators demonstrate the feasibility of 3d iterative image reconstruction in practice . this work complements our earlier studies that demonstrated the feasibility of 3d iterative image reconstruction in practice @xcite . our current implementations of the iterative image reconstruction algorithms still require several days to process the densely sampled data set , which , however , can be further improved . firstly , the amount of measurement data required for accurate image reconstruction can be further reduced by developing advanced image reconstruction methods @xcite . secondly , the number of iterations required can be reduced by developing fast - converging optimization algorithms @xcite . the proposed parallelization strategies by use of gpus are of general interest . the implementation of the fbp algorithm @xcite can be adapted to other analytic image reconstruction algorithms , including those described in refs . . we demonstrated the feasibility of pls algorithm that utilized the proposed gpu - based implementations of the projection / backprojection operators . by use of these implementations , many advanced image reconstruction algorithms may also be feasible in practice @xcite . though we described our parallelization strategies for the projection / backprojection operators that utilized two discrete - to - discrete imaging models , these strategies can also be applied to other d - d imaging models @xcite . therefore , the proposed algorithms will facilitate the further investigation and application of advanced image reconstruction algorithms in 3d oat . this research was supported in part by nih award eb010049 and ca167446 . _ derivation of equation _ the integrated data function @xmath216 in eqn . , evaluated at the @xmath26-th transducer and the @xmath27-th time instance , can be expressed as : @xmath217 where @xmath218 denotes the location of the @xmath26-th point - like transducer . we defined a local coordinate system , distinguished by a superscript ` tr ' , centered at the @xmath26-th transducer with the @xmath219-axis pointing to the origin of the global coordinate system as shown in fig . [ fig : geo]-(b ) . assuming the object function @xmath3 is compactly supported in a sphere of radius @xmath112 , the integral surface is symmetric about the @xmath219-axis . thus , the orientations of the @xmath220- and @xmath221-axes can be arbitrary within the @xmath222 plane . representing the right - hand side of eqn . in the local spherical coordinate system , one obtains @xmath223 where @xmath224 is half of the apex angle of the cone that corresponds to the intersectional spherical cap as shown in fig . [ fig : geo]-(b ) . the polar angle @xmath225 and the azimuth angle @xmath226 were discretized with intervals @xmath227 and @xmath228 that satisfied @xmath229 therefore , eqn . can be approximated by @xmath230 where @xmath231 , @xmath232 , @xmath233 , and @xmath234 . we denoted by @xmath235 the location in the global coordinate system corresponding to the location vector @xmath236 in the local coordinate system in eqn . . on substitution from the finite - dimensional representation eqn . into eqn . with @xmath37 and @xmath237 defined by eqns . and , respectively , we obtained : @xmath238_n \sum_{i=0}^{n_i-1 } \sum_{j=0}^{n_j-1}\psi_n^{\rm int}(\mathbf r_{k , i , j } ) \equiv \big[\mathbf g\big]_{qk+k}.\ ] ] 60ifxundefined [ 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 in @noop _ _ , ( , ) chap . @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1364/ao.45.001866 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * , ( ) link:\doibase 10.1121/1.1501898 [ * * , ( ) ] link:\doibase 10.1118/1.2409234 [ * * , ( ) ] link:\doibase 10.1117/1.2992131 [ * * , ( ) ] link:\doibase 10.1109/tmi.2009.2024082 [ * * , ( ) ] link:\doibase 10.1109/tmi.2008.2007825 [ * * , ( ) ] link:\doibase 10.1109/tmi.2010.2072514 [ * * , ( ) ] link:\doibase 10.1117/1.3381187 [ * * , ( ) ] ( , ) p. link:\doibase 10.1117/1.3443793 [ * * , ( ) ] link:\doibase 10.1117/1.3584847 [ * * , ( ) ] link:\doibase 10.1118/1.3556916 [ * * , ( ) ] link:\doibase 10.1109/tbme.2012.2187649 [ * * , ( ) ] ( , ) p. http://stacks.iop.org/0031-9155/57/i=17/a=5399 [ * * , ( ) ] link:\doibase 10.1117/1.jbo.17.6.066016 [ * * , ( ) ] link:\doibase 10.1109/tmi.2012.2208471 [ * * , ( ) ] link:\doibase 10.1117/1.jbo.17.6.061211 [ * * , ( ) ] link:\doibase 10.1364/boe.3.001427 [ * * , ( ) ] link:\doibase 10.1364/oe.20.022712 [ * * , ( ) ] link:\doibase 10.1109/mm.2008.31 [ * * , ( ) ] @noop _ _ ( ) link:\doibase 10.1155/2009/149079 [ * * , ( ) ] link:\doibase 10.1016/j.parco.2010.01.004 [ * * , ( ) ] link:\doibase 10.1118/1.3591994 [ * * , ( ) ] link:\doibase 10.1016/j.jpdc.2008.05.013 [ * * , ( ) ] link:\doibase 10.1117/1.3360308 [ * * , ( ) ] @noop _ _ ( , ) @noop _ _ ( , , ) @noop * * ( ) in @noop _ _ , ( , ) chap . link:\doibase 10.1109/tuffc.2011.1809 [ * * , ( ) ] @noop _ _ ( , ) @noop * * , ( ) link:\doibase 10.1109/tmi.2005.852055 [ * * , ( ) ] @noop _ _ ( , , ) @noop * * , ( ) link:\doibase 10.1117/1.2992131 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1109/42.870265 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1117/1.3259361 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop `` , '' ( , ) link:\doibase 10.1109/83.760336 [ * * , ( ) ] link:\doibase 10.1364/oe.20.016510 [ * * , ( ) ] link:\doibase 10.1109/tip.2009.2028250 [ * * , ( ) ] link:\doibase 10.1561/2200000016 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1137/070682137 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1109/tmi.2010.2044584 [ * * , ( ) ] \1 . ( a ) schematic of the 3d oat scanning geometry . ( b ) schematic of the local coordinate system for the implementation of interpolation - based d - d imaging model . .3 cm \2 . slices corresponding to the plane @xmath151 of ( a ) the phantom and the images reconstructed by use of ( b ) the cpu - based and ( c ) the gpu - based implementations of the fbp algorithm from the `` @xmath208''-data . .3 cm \3 . slices corresponding to the plane @xmath151 of the images reconstructed by use of the fbp algorithm with ( a ) the cpu - based implementation from the `` @xmath206''-data , ( b ) the cpu - based implementation from the `` @xmath239''-data , ( c ) the cpu - based implementation from the `` @xmath240''-data , ( d ) the gpu - based implementation from the `` @xmath206''-data , ( e ) the gpu - based implementation from the `` @xmath239''-data , and ( f ) the gpu - based implementation from the `` @xmath240''-data . .3 cm \4 . slices corresponding to the plane @xmath151 of the images reconstructed by use of the gpu - based implementations of ( a ) the pls - int algorithm from the `` @xmath206''-data , ( b ) the pls - int algorithm from the `` @xmath239''-data , ( c ) the pls - int algorithm from the `` @xmath240''-data , ( d ) the pls - sph algorithm from the `` @xmath206''-data , ( e ) the pls - sph algorithm from the `` @xmath239''-data , and ( f ) the pls - sph algorithm from the `` @xmath240''-data . .3 cm \5 . profiles along the line @xmath241-mm of the images reconstructed by use of ( a ) the cpu- and gpu - based implementations of the fbp algorithm from the `` @xmath208''-data , and ( b ) the gpu - based implementations of the pls - int and the pls - sph algorithms from the `` @xmath206''-data . .3 cm \7 . mip renderings of the 3d images of the mouse body reconstructed by use of the gpu - based implementations of ( a ) the fbp algorithm from the `` full data '' , ( b ) the pls - int algorithm from the `` full data '' with @xmath242 , ( c ) the pls - sph algorithm from the `` full data '' with @xmath242 , ( d ) the fbp algorithm from the `` quarter data '' , ( e ) the pls - int algorithm from the `` quarter data '' with @xmath243 , and ( f ) the pls - sph algorithm from the `` quarter data '' with @xmath243 . the grayscale window is [ 0,12.0 ] . @xmath247 , @xmath250 , @xmath251 @xmath251 @xmath252 ; @xmath253 ; @xmath254 @xmath255 @xmath256 ; @xmath257 ; @xmath258 ; @xmath259 @xmath260 ; @xmath261 ; @xmath262 @xmath263 @xmath264 ; @xmath265 @xmath266 { \rm t}\!\_\mathbf p[n_r][n_v][n_t ] + \big [ 1.5 - ( n_t\delta_t+t_{\rm min})/(t_n\delta_t+t_{\rm min})\big ] { \rm t}\!\_\mathbf p[n_r][n_v][n_t+1 ] \big\}$ ] @xmath267[{\rm blockidx.x}][{\rm threadidx.x}]=\sigma$ ] @xmath272 , @xmath274 @xmath272 @xmath275 ; @xmath276 ; @xmath277 @xmath278 @xmath255 ; @xmath279 @xmath280 ; @xmath281 ; @xmath282 @xmath283;@xmath284 @xmath285 ; @xmath286 ; @xmath287 @xmath288 ; @xmath289 ; @xmath290 @xmath291 @xmath292 @xmath293 @xmath294[{\rm blockidx.x}][{\rm blockidx.y}]=\sigma \delta_s^2 $ ] @xmath300 , @xmath301 @xmath300 @xmath275 ; @xmath276 ; @xmath277 @xmath278 ; @xmath279 @xmath280 ; @xmath281 ; @xmath282 @xmath283;@xmath284 @xmath302 ; @xmath286 ; @xmath303 @xmath288 ; @xmath289 ; @xmath290 @xmath304 ; @xmath305 ; @xmath306 @xmath307[n_y][n_x ] \operatorname{+\!\!=}\delta_s^2(n_x+1-x_n)(n_y+1-y_n)(n_z+1-z_n ) { \rm t\_}\mathbf g'_{\rm int}[\rm { threadidx.x}][{\rm blockidx.x}][{\rm blockidx.y}]$ ] @xmath292 @xmath293 @xmath69 , @xmath308 @xmath309 parm_fwdarg[@xmath310.@xmath311 = @xmath311 parm_fwdarg[@xmath310.@xmath312 = & @xmath313 $ ] parm_fwdarg[@xmath310.@xmath314=&@xmath315 $ ] parm_fwdarg[@xmath310.@xmath316=@xmath317 $ ] pthread_create(&pthreads[@xmath310 , null , fwd_pthread , ( void * ) ( parm_fwdarg+@xmath311 ) ) @xmath318 \,+\!\!= \tilde{\mathbf u}'_{\rm sph}[n+n_{\rm pth}n_rn_vn_f]$ ] @xmath328 , @xmath329 , d_@xmath145 , c_@xmath141 , c_@xmath138 d_@xmath145 @xmath330 ; @xmath277 ; @xmath331 @xmath257 ; @xmath332 ; @xmath333 @xmath334 ; @xmath335 @xmath336 @xmath337 @xmath338 ; @xmath339 @xmath340\big ( \tilde{h}^r { \rm c}\!\_\tilde{\mathbf p}_0[{\rm threadidx.x}].r -\tilde{h}^i { \rm c}\!\_\tilde{\mathbf p}_0[{\rm threadidx.x}].i \big)$ ] @xmath341\big ( \tilde{h}^r { \rm c}\!\_\tilde{\mathbf p}_0[{\rm threadidx.x}].i + \tilde{h}^i { \rm c}\!\_\tilde{\mathbf p}_0[{\rm threadidx.x}].r \big)$ ] @xmath342[{\rm blockidx.x}][{\rm threadidx.x}].r\,+\!\!= \sigma^r$ ] @xmath342[{\rm blockidx.x}][{\rm threadidx.x}].i\,+\!\!= \sigma^i$ ] @xmath343 , @xmath308 @xmath344 parm_bwdarg[@xmath310.@xmath311 = @xmath311 parm_bwdarg[@xmath310.@xmath312 = & @xmath313 $ ] parm_bwdarg[@xmath310.@xmath345=&@xmath346 $ ] parm_bwdarg[@xmath310.@xmath347=@xmath348 $ ] pthread_create(&pthreads[@xmath310 , null , bwd_pthread , ( void * ) ( parm_bwdarg+@xmath311 ) ) @xmath349\,+\!\!= \boldsymbol\alpha''_{\rm sph}[n+n_{\rm pth}n_xn_yn_z]$ ] @xmath357 , @xmath358 , @xmath359 , d_@xmath355 , c_@xmath360 , c_@xmath138 d_@xmath355 @xmath252 ; @xmath253 ; @xmath361 @xmath337 ; @xmath255 @xmath338 ; @xmath339 @xmath362.r\big ( \tilde{h}^r { \rm c}\!\_\tilde{\mathbf p}_0[n_f].r -\tilde{h}^i { \rm c}\!\_\tilde{\mathbf p}_0[n_f].i \big ) + { \rm c}\!\_\tilde{\mathbf u}_{\rm pth}[n_f].i\big ( \tilde{h}^i { \rm c}\!\_\tilde{\mathbf p}_0[n_f].r + \tilde{h}^r { \rm c}\!\_\tilde{\mathbf p}_0[n_f].i \big)$ ] @xmath363[{\rm blockidx.x}][{\rm threadidx.x}]\,+\!\!= \sigma$ ]

being motivated by recent experimental studies , we investigate magnetic structures of the mo pyrochloresmoo ( y , nd , and gd ) and their impact on the electronic properties . the latter are closely related with the behavior of twelve mo( ) bands , located near the fermi level and well separated from the rest of the spectrum . we use a mean - field hartree - fock approach , which combines fine details of the electronic structure for these bands , extracted from the conventional calculations in the local - density approximation , the spin - orbit interaction , and the on - site coulomb interactions amongst the mo( ) electrons , treated in the most general rotationally invariant form . the coulomb repulsion plays a very important role in the problem , and the semi - empirical value- ev accounts simultaneously for the metal - insulator ( m - i ) transition , the ferromagnetic ( fm ) - spin - glass ( sg ) transition , and for the observed enhancement of the anomalous hall effect ( ahe ) . the m - i transition is mainly controlled by . the magnetic structure at the metallic side is nearly collinear fm , due to the double exchange mechanism . the transition into the insulating state is accompanied by the large canting of spin and orbital magnetic moments . the sign of exchange interactions in the insulating state is controlled by the mo - mo distances . smaller distances favor the antiferromagnetic coupling , which preludes the sg behavior in the frustrated pyrochlore lattice . large ahe is expected in the nearly collinear fm state , near the point of m - i transition , and is related with the unquenched orbital magnetization at the mo sites . we also predict large magneto - optical effect in the same fm compounds .

cond-mat0211615

the pyrochlores with the chemical formula @xmath0mo@xmath1o@xmath2 ( @xmath3 being the divalent element ) exhibit a number of interesting and not completely understood phenomena,@xcite which present the main motivation for the analysis of electronic and magnetic structure of these compounds , which we undertake in this work . @xmath11 the pyrochlores @xmath0mo@xmath1o@xmath2 have rather puzzling magnetic phase diagram as a function of averaged ionic radius of @xmath3-sites , @xmath12 ( fig . [ fig.tcexp]).@xcite large @xmath12 ( @xmath13@xmath14@xmath8@xmath15 ) stabilizes the ferromagnetic ( fm ) ground state . typical examples of the fm pyrochlores are nd@xmath1mo@xmath1o@xmath2 and gd@xmath1mo@xmath1o@xmath2 . the curie temperature ( @xmath16 ) is of the order of 80 k and slowly increases with @xmath12 . smaller @xmath12 ( @xmath17@xmath14 ) give rise to the spin - glass ( sg ) behavior . the characteristic transition temperature into the sg state is of the order of 20 k. the canonical example of the sg compounds is y@xmath1mo@xmath1o@xmath2.@xcite now it is commonly accepted that the exchange coupling between nearest - neighbor ( _ nn _ ) mo spins changes the sign around @xmath14 and becomes antiferromagnetic ( afm ) in the sg region . the pyrochlores with afm interactions in the lattice present a typical example of geometrically frustrated systems with infinitely degenerate magnetic ground state.@xcite according to the recent experimental data,@xcite the sg behavior itself is caused by the combination of the geometrical frustrations and the disorder of the local lattice distortions . the latter is required in order to produce an inhomogeneity in the distribution of the _ nn _ interactions and freeze the random spin configuration . the origin of this behavior itself is very complicated , and we will not be able to address it properly in our work . however , our main concern will be in some sense more general : we will try to understand which parameter of the crystal structure controls the sign of the _ nn _ magnetic interactions , and which part of the electronic structure is responsible for the fm and afm interactions in these compounds ? contrary to the widespread point of view that the magnetic ground state is controlled by the mo - o - mo angle , we will argue that the key parameter is in fact the mo - mo distance , which is directly related with the unit cell volume and controls the strength of direct interactions between extended mo(@xmath6 ) orbitals . according to the experimental data , all mo pyrochlores which show the sg behavior are small - gap insulators . however , an opposite statement is generally incorrect and in different compounds the ferromagnetism is known to coexist with the metallic as well as insulating behavior.@xcite we will show that this trend can be naturally understood in terms of the mo - mo distances and the strength of the on - site coulomb interaction @xmath7 amongst the mo(@xmath6 ) electrons . @xmath11 the spin - orbit interaction ( soi ) in the pyrochlore lattice should generally lead to a non - collinear magnetic ordering associated with the modulation of single - ion anisotropy axes.@xcite the magnitude and the main source of this non - collinearity ( whether it is primarily associated with the mo or @xmath3 sublattices , and whether it reflects the ordering of spin or orbital magnetic moments ) is largely unresolved problem . in the present work we will try to rationalize this question by focusing on the magnetic structure of the mo sublattice . we will show that as long as the system is metallic ( that by itself depends on the strength of the coulomb interaction @xmath7 ) , the non - collinearity is very small . however , the transition into the insulating state is accompanied by the abrupt change of the spin and orbital magnetic structures , which become essentially non - collinear . @xmath11 some ferromagnetic ( or nearly ferromagnetic ) pyrochlores exhibit rather unusual anomalous hall effect ( ahe ) . a non - vanishing hall conductivity ( @xmath18 ) emerges near @xmath16 , then monotonously increases with the decrease of the temperature @xmath19 , and reaches the maximal value ( @xmath18@xmath8@xmath20 @xmath21 for nd@xmath1mo@xmath1o@xmath2 ) near @xmath19@xmath4@xmath22 . both magnitude and the temperature dependence of the ahe is in a drastic contrast with other oxide materials , such as colossal - magnetoresistive manganites , for example.@xcite the theory of the ahe ( originally called as the extraordinary hall effect ) has very long and rather controversial history.@xcite many controversies were related with the lack of the general transport theory and partly with the lack of good computer facilities in 1950s and 1960s , so that many discussions went around very simplified and not always justified approximations , such as the free - electron and weak scattering limit , the single - orbital tight - binding approach , the classical boltzmann transport equation , etc.@xcite the modern way to address the problem is to use the kubo formalism and relate @xmath18 with the real part of the antisymmetric off - diagonal element of the conductivity tensor.@xcite from this point of view there are no rigid constraints which would forbid the existence of @xmath18 , even for periodic systems , once the time - reversal symmetry is violated macroscopically.@xcite the main question is which mechanism leads to the finite value of @xmath18 . the first one was proposed by karplus and luttinger,@xcite who stressed the importance of soi in the problem . although the concrete scenario considered by karplus and luttinger was not correct,@xcite they were certainly right by arguing that the soi is essential in order to couple the spin polarization of the conduction electrons to the lattice and produce the right - left asymmetry of the stationary bloch states , which may lead to the hall current , perpendicular to both the electric field and the magnetization . however , an exceptionally large @xmath18 observed in the pyrochlore compounds urged several authors to search for a new and unconventional mechanism for the ahe . such mechanism was proposed to be due to the spin chirality ( basically , the solid angle subtended by three non - coplanar spins ) in essentially non - collinear magnetic structure.@xcite the chirality acts as an effective field , which in the combination with the peculiar lattice geometry is believed to yield a finite @xmath18 . although the mechanism itself is rather exotic and there is a great deal of cancellations between different elements of the non - collinear spin structure,@xcite these authors were certainly right by stressing the crucial role of the inter - band transitions in the problem of ahe . in this work we will go back to a more conventional picture and argue that the large value of @xmath18 in some mo pyrochlores can be explained by rather unique combination of three factors : ( i ) large soi associated with the mo(@xmath6 ) states in the nearly collinear fm structure ; ( ii ) strong enhancement of the soi by the on - site coulomb interactions in the @xmath5 electron systems,@xcite and ( iii ) half - metallic electronic structure of the fm pyrochlores.@xcite regarding the coulomb interaction itself , which in our work will be treated as a parameter , we will be able to present a very consistent picture by arguing that @xmath7@xmath8@xmath9-@xmath10 ev in mo pyrochlores accounts simultaneously for ( i ) metal - insulator transition ; ( ii ) fm - sg transition ; ( iii ) required enhancement for the ahe in the fm pyrochlore compounds . the rest of the paper is organized as follows . in sec . [ sec : main_details ] we remind the basic details of the crystal and electronic structure of @xmath0mo@xmath1o@xmath2 . we consider three characteristic compounds corresponding to @xmath3@xmath4 y , gd and nd . in sec . [ sec : model ] we describe the model hartree - fock ( hf ) approach , which combines the ideas of _ ab - initio _ electronic structure calculations with the physics of degenerate hubbard model on the pyrochlore lattice . [ sec : dos ] and [ sec : exchange ] summarize results of calculations without soi ( correspondingly , the basic changes of the electronic structure and the orbital ordering , and the behavior of inter - atomic magnetic interactions ) . secs . [ sec : nc ] and [ sec : hall ] are devoted to the analysis of soi related properties ( the non - collinear magnetic structure , the hall conductivity , and the magneto - optical effect ) . finally , in sec . [ sec : summary ] we discuss some subtle points related with the interpretation of experimental data and give the summary of the entire work . the pyrochlores @xmath0mo@xmath1o@xmath2 crystallize in a face - centered cubic structure with the space group @xmath23 , in which @xmath3 and mo occupy correspondingly @xmath24 and @xmath25 positions , and form interpenetrating sublattices of the corner - sharing tetrahedra . there are two non - equivalent types of oxygen sites , and only one internal parameter which may control the properties of @xmath0mo@xmath1o@xmath2 . that is the coordinate @xmath26 of the o @xmath27 sites . the single mo tetrahedron is shown in fig . [ fig.tetrahedron ] . four mo sites are located at @xmath28@xmath4@xmath29 , @xmath30@xmath4@xmath31 , @xmath32@xmath4@xmath33 , and @xmath34@xmath4@xmath35 , in units of cubic lattice parameter @xmath36 . each mo site has sixfold o @xmath27 coordination . the oxygen atoms specify the local coordinate frame around each mo site . around the site 1 , it is given by @xmath37 where @xmath38@xmath4 @xmath39 , @xmath40 , or @xmath41 . @xmath26@xmath4@xmath42 corresponds to the perfect octahedral environment . in this case @xmath43@xmath4@xmath44 is the standard @xmath45 rotation around the cubic @xmath46 $ ] axis . @xmath26@xmath13@xmath42 gives rise to an additional trigonal contraction of the local coordinate frame . similar matrices associated with the sites 2 , 3 , and 4 can be obtained by the @xmath47 rotations of @xmath43 around @xmath39 , @xmath40 , and @xmath41 , respectively . structural parameters of @xmath0mo@xmath1o@xmath2 are listed in table [ tab : structure ] . corresponding densities of states , obtained in the local - spin - density approximation ( lsda ) are shown in figs . [ fig.doslsda ] and [ fig.dost2g ] . in the local coordinate frame , the mo(@xmath6 ) orbitals are split into the triply - degenerate @xmath5 and doubly - degenerate @xmath48 states . the splitting is of the order of 4 ev . twelve @xmath5 bands are located near the fermi level and well separated from the rest of the spectrum , which consists of the broad o(@xmath49 ) band spreading from -8.5 to -2.5 ev and either y(@xmath6 ) or gd / nd(@xmath50 ) bands located just above the group of the @xmath5 states . the mo(@xmath48 ) states are located in the higher part of the spectrum . interestingly enough that all three compounds are ferromagnetic , even in the lsda , that is rather unusual for the @xmath6 oxides , perhaps except the well know example of srruo@xmath51.@xcite the trigonal distortion and different hybridization with the o(@xmath49 ) states will further split the mo(@xmath5 ) states into the one - dimensional @xmath52 and two - dimensional @xmath53 representation.@xcite the crystal structure affects the mo(@xmath5 ) bandwidth via two mechanisms ( see table [ tab : structure ] ) . 1 . the mo - o - mo angle , which decreases in the direction nd@xmath54gd@xmath54y . therefore , the interactions between mo(@xmath5 ) orbitals which are mediated by the o(@xmath49 ) states will also decrease . this effect is partly compensated by the decrease of the mo - o bondlength for @xmath3@xmath4 gd and y. 2 . the lattice parameter @xmath36 and the mo - mo distance , which decrease in the direction nd@xmath54gd@xmath54y by 2.6% . this will increase direct interactions between extended mo(@xmath6 ) orbitals . generally , these two effects act in the opposite directions and partly compensate each other . for example , the width of the @xmath53 band is practically the same for all three compounds ( fig . [ fig.dost2g ] ) . on the other hand , the @xmath52 orbitals , whose lobes are the most distant from all neighboring oxygen sites are mainly affected by the second mechanism , and the @xmath52 bandwidth will _ increase _ in the direction nd@xmath54gd@xmath54y . as we will show in sec . [ sec : exchange ] , this effect plays a crucial role in the stabilization of afm interactions in y@xmath1mo@xmath1o@xmath2 , which preludes the sg behavior . thus , despite an apparent complexity of the crystal structure , the pyrochlores @xmath0mo@xmath1o@xmath2 present rather simple example of the electronic structure and in order to understand the nature of fascinating electronic and magnetic properties of these compounds , we need to concentrate on the behavior of twelve well isolated mo(@xmath5 ) bands.@xcite in this and subsequent sections we will further elaborate this picture , by cutting the mo(@xmath5 ) bands from the rest of the spectrum , including the on - site coulomb interactions amongst the mo(@xmath6 ) electrons , and solving this problem on the level of hf approximation . hence , our model hamiltonian has the following form , in the basis of non - relativistic lda eigenfunctions @xmath55 ( lda is the spin - restricted version of lsda ) : @xmath56 the spin indices are already included to @xmath57 and @xmath58 , which are running from @xmath59 to @xmath60@xmath61@xmath62 . the first term in @xmath63 is the non - magnetic lda part and @xmath64 are the lda eigenvalues . @xmath65 is the matrix of relativistic soi , which is computed in the conventional way , @xmath65@xmath4@xmath66@xmath4@xmath67 , @xmath68 is the operator of angular momentum , @xmath69 is the vector of pauli matrices , @xmath70 is the self - consistent lda potential , and @xmath71 is the velocity of light ( @xmath71@xmath72@xmath73 in ry units ) . the inclusion of @xmath65 into the model hamiltonian is optional , that will be always specified in the text . all magnetic effects are caused by the on - site coulomb interactions , @xmath74 , which can be written in terms of projections onto an orthonormal set of atomic - like mo(@xmath6 ) orbitals @xmath75 : @xmath76 where @xmath77 are the translation vectors in the face - centered cubic pyrochlore lattice , and @xmath78 specify different mo sites in the unit cell . @xmath79@xmath80@xmath81 are the joint indices including the spin ( @xmath82@xmath4 @xmath83 or @xmath84 ) and orbital ( @xmath85@xmath4 @xmath86 , @xmath87 , @xmath88@xmath89@xmath90 , @xmath91 , or @xmath92@xmath89@xmath93 ) degrees of freedom . matrix elements of the hf potential , @xmath94 , can be expressed in the most general rotationally - invariant form as:@xcite @xmath95 where @xmath96@xmath4@xmath97 are the matrix elements of the coulomb interactions . the latter are fully specified by three radial slater integrals : @xmath98 , @xmath99 , and @xmath100 , which are assumed to be _ renormalized _ from the bare atomic values . equivalently , one can introduce parameters of the averaged coulomb interaction , @xmath7@xmath4@xmath98 , the intra - atomic exchange coupling , @xmath101@xmath4@xmath102@xmath103@xmath104 , and the non - sphericity of the coulomb interactions between orbitals with the same spin , @xmath105@xmath4@xmath106@xmath89@xmath107 . generally , both @xmath108@xmath4@xmath7@xmath89@xmath101 and @xmath105 contribute to the orbital polarization in solids and affect the properties caused by violation of the time - inversion symmetry.@xcite the density matrix @xmath109@xmath80@xmath110 can be obtained from the green function @xmath111@xmath80@xmath112 , using the standard relation @xmath113 where @xmath114 is the spectral representation for the green function in terms of the eigenfunctions @xmath115 and the eigenvalues @xmath116 of the hamiltonian ( [ eqn : hamiltonian ] ) . @xmath117 is the volume of the first brillouin zone . ( [ eqn : hamiltonian ] ) , ( [ eqn : dmatrix ] ) and ( [ eqn : green ] ) are solved self - consistently . then , the total energy is given by @xmath118 according to the constraint - lsda calculations,@xcite @xmath101 for mo is of the order of @xmath119 ev , and is not sensitive to details of the crystal environment in solids . @xmath105 can be estimated from @xmath101 using the ratio @xmath120@xmath121@xmath122 , which holds for the slater integrals in the atomic limit . this yields @xmath105@xmath121@xmath123 ev . the coulomb @xmath7 is treated as the parameter in order to consider different scenarios , covering both metallic and insulating behavior of @xmath0mo@xmath1o@xmath2 . the constraint - lsda calculations for mo compounds typically yield @xmath7@xmath72@xmath124 ev.@xcite this value can be further reduced by allowing for the ( proper ) @xmath48 electrons to participate in the screening of the on - site coulomb interactions between @xmath5 electrons.@xcite for the practical calculations along this line , it is convenient to use a short atomic - like basis set . for these purposes , @xmath64 and @xmath55 have been first calculated using the nearly - orthogonal version of the linearized - muffin - tin orbital method in the atomic - spheres approximation ( asa - lmto).@xcite the atomic orbitals @xmath75 have been identified with the basis orbitals of the lmto method . atomic spheres radii have been chosen so to reproduce the electronic structure obtained by more accurate flapw method,@xcite especially for @xmath5 bands located near the fermi level . we have also added a number of empty spheres at the @xmath125 and @xmath126 positions of the @xmath127 lattice . all calculations have been performed on the mesh of 12@xmath6112@xmath6112 @xmath128-points in the first brillouin zone . formally , our approach is similar to the rotationally - invariant lda@xmath103@xmath7,@xcite except that we apply it only to a limited number of bands , directly responsible for the properties of @xmath0mo@xmath1o@xmath2 compounds . this will certainly simplify the calculations . in addition , we get rid of several artifacts of the conventional lda@xmath103@xmath7 method , such as the substraction of the double - counting terms , which is a very ambiguous procedure by itself , and in the number of cases leads to a systematic error.@xcite let us first discuss results without soi , assuming fm ordering of the mo spins . according to the lsda calculations ( fig . [ fig.dost2g ] ) , the majority ( @xmath83 ) spin @xmath52 band is fully occupied and the fermi level crosses the doubly - degenerate @xmath53 band . therefore , it is clear that at some point the coulomb @xmath7 will split the @xmath53 band and form an insulating solution with the spontaneously broken @xmath23 symmetry . such situation occurs between @xmath7@xmath4 2.0 and 2.5 ev for all considered compounds ( fig . [ fig.gap ] , the afm alignment of the mo spins may open the band gap even for smaller @xmath7 , that well correlates with the fact that all sg compounds are insulators ) . in the metallic regime realized for small @xmath7 , the densities of states are similar to those in lsda ( figs . [ fig.doslsda ] and [ fig.dost2g ] ) , and we will not show them here again . we only mention that the major effect of @xmath7 is the shift of the @xmath83-spin @xmath52 band to the low - energy part of the spectrum relative to the @xmath53 band . corresponding orbital ordering ( the distribution of mo(@xmath6 ) electron densities ) is shown in fig . [ fig.ooyf05 ] . it comes exclusively from the local trigonal distortions of the oxygen octahedra and represents the alternating @xmath52 orbital densities in the background of degenerate @xmath53 orbitals . typical densities of states in the insulating phase are shown in fig . [ fig.dos30 ] . the @xmath52 band has characteristic for pyrochlores three - peak structure.@xcite smaller band gap in the case of y@xmath1mo@xmath1o@xmath2 ( fig . [ fig.gap ] ) is related with broad minority ( @xmath84 ) spin @xmath52 band , whose tail spreads below the unoccupied @xmath83-spin @xmath53 band . the distribution of the @xmath53 states is very similar for all three compounds . contrary to the metallic state , the orbital ordering is determined not only by local trigonal distortions , but also the form of the superexchange ( se ) interactions between _ nn _ mo sites , and minimizes the energy of these interactions.@xcite generally , it depends on the magnetic state . two typical examples for the fm and afm ( obtained after the flip of the spin moments at the sites 2 and 3 ) phases are shown in figs . [ fig.ooyf30 ] and [ fig.ooya30 ] , respectively . the building blocks of the afm phase are the fm chains of the atoms ... -1 - 4 - 1 - 4- .. and ... -2 - 3 - 2 - 3- ... , which are coupled antiferromagnetically . as expected for the fm spin ordering,@xcite the @xmath53 orbitals tends to order antiferromagnetically and form two mo sublattices . corresponding eigenfunctions of the occupied @xmath53 orbitals , obtained after the diagonalization of the density matrix _ in the local coordinate frame _ , are given by @xmath129@xmath121@xmath130@xmath103@xmath131@xmath103@xmath132 for the sites 1 and 4 , and @xmath133@xmath121@xmath130@xmath103@xmath134@xmath103@xmath135 for the sites 2 and 3 , with the coefficients @xmath136@xmath121@xmath137 , @xmath138@xmath121@xmath139 , and @xmath140@xmath121@xmath89@xmath141 . clearly , this orbital ordering breaks the @xmath23 symmetry : if the sites belonging to the same sublattice can still be transformed to each other using the symmetry operations of the @xmath23 group , the sites belonging to different sublattices can not . this will generally lead to the anisotropy of electronic properties , including the anisotropy of _ nn _ magnetic interactions , which will be discussed in the next section . the afm spin ordering coexists with the fm orbital ordering for which @xmath136@xmath121@xmath89@xmath142 , and @xmath138@xmath4@xmath140@xmath121@xmath143 for all mo sites , in the local coordinate frame.@xcite it breaks the @xmath23 symmetry in the spin sector , but not in the orbitals one . parameters of exchange interactions between sites @xmath78 and @xmath144 can be calculated as@xcite @xmath145 where @xmath146 is the block of matrix elements of the green function with the spin @xmath83 or @xmath84 : @xmath147 and @xmath148 is the magnetic part of the hf potential : @xmath149 @xmath150 is the trace running over the spin ( orbital ) indices . results for _ nn _ magnetic interactions for all considered compounds are shown in fig . [ fig.exchange ] , as a function of @xmath7 . we note the following . 1 . @xmath151 , which are ferromagnetic for small @xmath7 , exhibit a sharp drop at the point of transition into the insulating state ( see fig . [ fig.gap ] for the behavior of the band gap ) . 2 . there is a significant difference between nd / gd and y based compounds : in the y case , the exchange parameters are almost rigidly shifted towards negative values , so that the _ nn _ coupling become antiferromagnetic in the insulating phase , while it remains ferromagnetic in the case of nd and gd . similar conclusions can be obtained from the analysis of total energies shown in fig . [ fig.etot].@xcite the behavior can be easily understood by considering partial @xmath52 and @xmath53 contributions to the _ nn _ exchange coupling , calculated after transformation to the local coordinate frame at each site of the system ( fig . [ fig.jpartial ] ) . then , the main interactions are @xmath52-@xmath52 and @xmath53-@xmath53 . the @xmath52-@xmath53 interaction is small and can be neglected . the interaction between @xmath5 and proper @xmath48 orbitals is of the order of 1.5 - 1.8 mev and only weakly depends on @xmath7 and the type of the material . therefore , it can be regarded as the weak fm background . large fm @xmath53-@xmath53 interaction in the metallic regime is related with the double exchange ( de ) mechanism , which is the measure of the kinetic energy for the itinerant @xmath83-spin @xmath53 electrons ( note that for small @xmath7 all compounds are either half - metallic or nearly half - metallic fig . [ fig.doslsda]).@xcite some increase of @xmath151 for small @xmath7 is caused by the reduction of afm se contributions , which are proportional to @xmath152 . as long as the system is metallic , the de interactions are not sensitive to the value of @xmath7 , and the fm coupling dominates . the transition into the insulating state is caused by the localization of the @xmath53 electrons on the atomic orbitals . this will reduce the kinetic energy and suppress the de interactions , that explain the sharp drop of @xmath151 ( fig . [ fig.exchange ] ) . the main difference between y and nd / gd based compounds is related with the @xmath52-@xmath52 interactions . since the @xmath83-spin @xmath52 band is fully occupied and the @xmath84-spin band is empty , the interactions are antiferromagnetic and the underlying microscopic mechanism is the superexchange . since the se coupling is proportional to the square of the @xmath52 bandwidth , this interaction is the largest in the case of y , that explains the afm character of the total coupling realized in this compound for large @xmath7.@xcite the anisotropy of the _ nn _ interactions , @xmath153@xmath4@xmath154@xmath89@xmath155 , caused by the orbital ordering in the fm state is of the order of 1 - 2 mev ( fig . [ fig.exchange ] ) . this is a small value in comparison with the sharp drop of @xmath151 caused by the coulomb interaction @xmath7 . however , it can play an important role in the insulating regime , where @xmath153 becomes comparable with @xmath151 . for example , one interesting question is whether this @xmath153 , of a purely coulombic origin , is sufficient to explain the appearance of the sg state . in addition , @xmath153 has the same order of magnitude as the single - ion anisotropy energy ( sec . [ sec : nc ] ) and the mo - gd exchange interaction ( ref . ) . therefore , in the insulating state there is no unique interaction which would control the magnetic behavior of the @xmath0mo@xmath1o@xmath2 compounds . all interactions are equally important , that represents the main complexity on the way of theoretical description of these materials . the afm spin ordering changes the orbital ordering in the direction which further stabilizes the _ nn _ magnetic interactions,@xcite and in a number of cases may lead to the charge disproportionation within the fm chains.@xcite for example , @xmath7@xmath4@xmath156 ev appears to be sufficient to open the band gap ( @xmath8@xmath157 ev ) in the afm phase of y@xmath1mo@xmath1o@xmath2 . the solution is accompanied by the small charge disproportionation , and the values of the spin magnetic moments at the four mo sites are @xmath158@xmath4@xmath89@xmath159@xmath4@xmath160 @xmath161 and @xmath89@xmath162@xmath4@xmath163@xmath4@xmath9 @xmath161 . corresponding values of the inter - atomic exchange interactions are @xmath154@xmath4@xmath89@xmath164 , @xmath165@xmath4@xmath89@xmath166 and @xmath155@xmath4@xmath167 mev , i.e. the interactions are antiferromagnetic in the bonds 1 - 2 and 1 - 3 , and ferromagnetic in the bond 1 - 4 . therefore , the obtained solution is locally stable . the difference between @xmath154 and @xmath165 is caused by the charge disproportionation . further increase of @xmath7 stabilizes the afm coupling in all mo - mo bonds , giving rise to the geometrical frustrations . typical values of the exchange parameters obtained in the hf solution for @xmath7@xmath4@xmath124 ev , which does not show any sign of the charge disproportionation , are @xmath154@xmath4@xmath165@xmath4@xmath89@xmath168 and @xmath154@xmath4@xmath89@xmath169 mev . in this and subsequent sections we will discuss the effects of soi on the electronic and magnetic structure of the mo pyrochlores . the matrix element of @xmath170 calculated in the basis of mo(@xmath6 ) orbitals is of the order of @xmath171@xmath4@xmath172 mev . it defines the characteristic energy gain after adding the soi into the model hf hamiltonian . the equilibrium magnetic ordering depends on the competition between the isotropic exchange interactions , considered in the previous section , and the single - ion anisotropies energy . the isotropic exchange interactions tend to align mo spins parallel to each other ( at least in the case of the fm coupling ) . however , due to the peculiarities of the pyrochlore structure , there is no unique direction of the spin magnetization which would satisfy the single - ion anisotropy energies simultaneously for all mo sites . these anisotropies lead to the distortion of the collinear fm ordering.@xcite in the case of fm coupling between _ nn _ spins , there are two magnetic structures , which are typically discussed in the literature and can be obtained by distorting the collinear fm ordering along either @xmath173 or @xmath174 direction ( see fig . [ fig.fstructures ] ) . 1 . the two - in , two - out structure , where the directions of the magnetic moments at the four mo sites are given by : @xmath175@xmath4@xmath176@xmath89@xmath177 , @xmath178@xmath4@xmath179@xmath89@xmath180@xmath89@xmath177 , @xmath181@xmath4@xmath182@xmath89@xmath183@xmath89@xmath177 , and @xmath184@xmath4@xmath179@xmath89@xmath185@xmath89@xmath183@xmath89@xmath177 , in terms of the distortion @xmath186 from the perfect fm alignments . 2 . the umbrella structure : @xmath175@xmath4@xmath187 , @xmath178@xmath4@xmath188@xmath89@xmath189@xmath103@xmath183@xmath103@xmath190 , @xmath181@xmath4@xmath188@xmath103@xmath183@xmath89@xmath189@xmath103@xmath190 , and @xmath178@xmath4@xmath188@xmath103@xmath183@xmath103@xmath183@xmath89@xmath191 . since the mo(@xmath6 ) states are strongly polarized by the on - site coulomb interactions , the main contribution to the single - ion anisotropy energy comes from the @xmath83-spin electrons . in this case there is an one - by - one correspondence between the single - ion anisotropy energy and the magnitude of the orbital magnetization , so that the direction which maximizes the magnetization automatically minimizes the magnetic anisotropy energy.@xcite for the @xmath5 electrons affected by the trigonal distortion , the spontaneous orbital magnetization is determined by the following factors:@xcite 1 . the interaction between occupied @xmath52 and unoccupied @xmath53 states favors the in - plane geometry of the orbital magnetization ( i.e. , perpendicular to the trigonal axes ) . the latter is proportional to the ratio @xmath192 , where @xmath193 is the distance between the position of the @xmath52 states and the unoccupied part of the spectrum . clearly , this interaction will vanish when the coulomb @xmath7 is large . 2 . on the contrary , the interaction between @xmath53 orbitals favors the easy axes geometry . in the pure ionic limit , such solution is the subject for the degenerate perturbation theory . therefore , the wave functions and the orbital magnetization do not depend on @xmath171 . the soi in such situation only lifts the orbital degeneracy and defines the symmetry of the @xmath53 orbitals which will be further split by @xmath7 . the solution is accompanied by the large orbital moment of the order of @xmath59 @xmath161 , which is parallel to the trigonal axis . thus , when @xmath7 is sufficiently large , the second mechanism will dominate and one can naturally expect the easy axes scenario for the single - ion anisotropies . in the combination with the fm coupling between _ nn _ spins , it yields the umbrella ground state.@xcite the qualitative analysis is supported by results of direct calculations shown in fig . [ fig.canting ] . in the metallic regime , the isotropic double exchange interaction between mo spins clearly dominates over the single - ion anisotropy energies and the canting of spin magnetic moments is small ( both for the two - in , two - out and umbrella structure ) . the orbital moments are not particularly strong ( @xmath17@xmath194 @xmath161 per mo site ) . they are canted off the global fm direction by about @xmath195 . in the insulating state , the isotropic exchange interactions are suppressed and become comparable with the single - ion anisotropy energies , which are enhanced . this results in the large canting of the magnetic moments , which can be very different for the spin and orbital counterparts ( in the other words , the spin and orbital moments are not collinear with respect to each other).@xcite the reason for this difference is the following . directions of the orbital magnetic moments are determined mainly by the single - ion anisotropy terms , whereas the interactions between moments located at different sites are typically much smaller . therefore , the orbital moments always tend to align first along the crystallographic directions ( the trigonal axes , in our case ) . on the contrary , alignment of the spin magnetic moments is strongly affected by the inter - site interactions , which generally lead to a smaller canting . for large @xmath7 ( @xmath83 ev ) , the single - ion anisotropies start to dominate over the isotropic exchange interactions . in this limit , the orbital magnetic moments are almost perfectly aligned along the trigonal axes ( that corresponds to the 55@xmath196 and 70@xmath196 canting for the two - in , two - out and umbrella structure , respectively ) . the value of the local orbital magnetic moment is about 0.6 @xmath161 , which is smaller than 1 @xmath161 expected in the pure atomic limit . the umbrella structure has lower energy ( perhaps except the region of very small @xmath7 ) . the energy difference between the two - in , two - out and umbrella structures is small in the metallic regime ( @xmath8@xmath197 mev per formula unit , which corresponds to the coercive field of 1 tesla ) . however , it becomes large in the insulating state , where the interactions between the @xmath53 orbitals clearly dominate . in the case of afm coupling between _ nn _ spins , the single - ion anisotropies lift , at least partially , the high degeneracy of the magnetic ground state in the frustrated pyrochlore lattice . the equilibrium magnetic structure realized in model hf calculations for y@xmath1mo@xmath1o@xmath2 can be obtained by the inversion of the magnetic moments at the sites 2 and 3 of the two - in , two - out structure ( fig . [ fig.astructure ] ) . for @xmath7@xmath4@xmath124 ev , the values of spin and orbital magnetic moments at the mo sites are @xmath198 and @xmath199 @xmath161 , respectively . they are canted off the cubic ( 0,0,1 ) axis by correspondingly @xmath200 and @xmath201 . the afm interactions are compromised with the tetrahedral geometry of the pyrochlore lattice by forming the afm coupling between two of the three projections of the _ nn _ spins in each of the mo - mo bonds , while the remaining projection is coupled ferromagnetically . the conductivity tensor , @xmath202@xmath4@xmath203@xmath103@xmath204 , can be calculated from the eigenvalues , @xmath205 , and the eigenfunctions , @xmath206 , of the hf hamiltonian using the kubo formula,@xcite where the inter - band ( regular ) and intra - band ( drude ) contributions are given by ( in ry units ) @xmath207 \label{eqn : sinter}\ ] ] and @xmath208 respectively . in these equations , @xmath209 is the zero - temperature fermi distribution function , @xmath210@xmath4@xmath211 is the matrix element of momentum operator , @xmath212@xmath4@xmath213@xmath89@xmath214 , @xmath215 is the unit cell volume , and @xmath216 is the phenomenological relaxation time ( @xmath217@xmath121@xmath218 mev ) introduces in order to perform summation over the discrete mesh of @xmath128-points . the diagonal elements of @xmath204 can be also expressed through the plasma frequencies , @xmath219 , as @xmath220@xmath4@xmath221 . the conductivity tensor has the form @xmath222 both for the two - in , two - out and umbrella structure , if the direction of the @xmath41 axis is chosen as @xmath173 and @xmath174 , respectively . the hall conductivity , defined as @xmath18@xmath4@xmath223 , is shown in fig . [ fig.hall ] as a function of @xmath7 . as expected,@xcite the effect is caused by the inter - band transitions , while all intra - band contributions vanish , even in the metallic state @xmath18 takes the maximal value around @xmath7@xmath4@xmath9 ev , which compromises between the value of the orbital magnetization and the metallic behavior of the system . on the one hand , @xmath7 is large enough to unquench the orbital magnetic moment of the order of 0.08 @xmath161 per mo site . on the other hand it is small enough in order to keep the system in the metallic regime ( see fig . [ fig.gap ] ) . the factor - two increase of @xmath18 from @xmath7@xmath4@xmath119 to @xmath9 ev well correlates with similar increase of the orbital magnetic moment ( fig . [ fig.canting ] ) , suggesting that both effects have relativistic origin . the behavior coexists with nearly collinear fm ordering realized in the metallic regime . the transition into the insulating state , which is accompanied by the large canting of the magnetic moments , results in the abrupt drop of @xmath18.@xcite therefore , we conclude that the large @xmath18 observed in the experiment is due to the orbital magnetic moments induced by the soi in the nearly collinear fm state.@xcite the absolute value of @xmath18 , @xmath8@xmath224 @xmath21 , obtained in the calculations is in a good agreement with the experimental data for nd@xmath1mo@xmath1o@xmath2.@xcite however , we would like to emphasize again that the soi alone is not sufficient to obtain such a large values of @xmath18 , and it is essential to consider the on - site coulomb interactions beyond lsda , which play a very important role in the problem and enhance the effect of the soi.@xcite according to our calculations , @xmath225 , in the geometry which specify the form of the tensor ( [ eqn : tensor ] ) , only weakly depends on the magnetic structure and the direction of the net magnetization . large anisotropy of @xmath18 observed in the experiment@xcite is presumably related with the details of the experimental set - up and reflects the @xmath226-law . for example , if the electric filed is directed along the @xmath227 axis , and the current is measures along the @xmath228 axis , the effect is proportional to @xmath229 in the case of the umbrella structure , that roughly corresponds to the experimental situation . similar arguments apply for the magneto - optical kerr rotation which will be discussed below . if our scenario is correct , and large ahe in the mo pyrochlores can be explained by the inter - band transitions in the presence of the soi ( i.e. , by the conventional means ) , it is natural to expect similar enhancement for other phenomena related with the soi . below we consider the complex kerr effect , which can be also regarded as a possible check for our theory in the future . the kerr rotation ( @xmath230 ) and kerr ellipticity ( @xmath231 ) can be found from the expression,@xcite @xmath232 which is formally exact and valid for arbitrary @xmath230 and @xmath231 . @xmath233@xmath4@xmath234 are the complex refractive indices , in terms of matrix elements of the dielectric tensor , @xmath235@xmath4@xmath236@xmath103@xmath237 . results of these calculations for gd@xmath1mo@xmath1o@xmath2 are shown in fig . [ fig.conductivity ] , together with the @xmath238-dependence of the matrix elements of the optical conductivity . we consider the umbrella structure obtained for @xmath7@xmath4@xmath9 ev , which corresponds to the maximal rotation . calculations for nd@xmath1mo@xmath1o@xmath2 and the two - in , two - out structure yield very similar results . we also discuss partial contributions of the @xmath52 and @xmath53 states , and show the spectra obtained after removing the @xmath52 orbitals from the basis.@xcite apart from the hf model itself , we do not use any adjustable parameters . for example , the intra - band drude contribution to the optical conductivity has been obtained in the calculations . the diagonal part of @xmath202 comes mainly from the inter - band @xmath53@xmath54@xmath53 transitions . the contribution of the @xmath52 states is rather small . the absorptive ( re ) part of @xmath239 has a maximum around 1 ev . both the peak position and its intensity roughly agree with the experimental data,@xcite tough the latter reveal an additional structure in the low-@xmath238 energy part of the spectrum , which can be due to the quantum fluctuation effects,@xcite beyond the mean - field approximation considered in our work.@xcite the contribution of the @xmath52 states to @xmath225 is very important , and can bring up to 50% of the spectral weight at certain values of @xmath238 . however , the low-@xmath238 part of @xmath225 comes mainly from the inter - band @xmath53@xmath54@xmath53 transitions , which are responsible for almost 90% of the hall conductivity . the large intensity of @xmath225 is due to the combination of three factors : ( i ) large soi at the mo sites ; ( ii ) the enhancement of the soi by the on - site coulomb interactions , which is especially efficient for the degenerate @xmath53 states ; ( iii ) the half - metallic electronic structure of gd@xmath1mo@xmath1o@xmath2 and nd@xmath1mo@xmath1o@xmath2 : since only @xmath83-spin states contribute to the low-@xmath238 part of @xmath225 , the cancellation between different spin channels does not take place.@xcite the kerr rotation spectrum , calculated using eg . ( [ eqn : kerr ] ) , has two structures . the first one is close to the plasma edge ( @xmath240@xmath121@xmath157 ev ) , and is directly related with the plasma enhancement effect.@xcite the exceptionally large value of @xmath230 ( @xmath8@xmath241 ) is also related with large @xmath225 in this region , which correlates with the large value of the hall conductivity . the second structure ( @xmath230@xmath8@xmath89@xmath242 ) near @xmath243 ev is related with the form of @xmath244 , which has a deep in this region as a joint effect of both @xmath53@xmath54@xmath53 and @xmath52@xmath54@xmath53 inter - band transitions . smaller @xmath7 would yield somewhat smaller estimates for the rotation angles ( correspondingly @xmath245 and @xmath89@xmath246 for @xmath7@xmath4@xmath119 ev ) . nevertheless , the fm pyrochlores @xmath0mo@xmath1o@xmath2 seem to be rather promising materials for the magneto - optical applications . the anomalous hall and kerr effects simultaneously vanish in the afm phase , due to the cancellation of contributions coming from different mo sublattices in the macroscopic sample.@xcite we have presented results of model hf calculations for pyrochlores @xmath0mo@xmath1o@xmath2 ( @xmath3@xmath4 y , gd , and nd ) . the model combines details of lda electronic structure for the mo(@xmath5 ) bands located near the fermi level , the spin - orbit interaction , and the on - site coulomb interactions amongst the mo(@xmath6 ) electrons . the main results can be summarized as follows . @xmath11 the magnetic ground state of the @xmath0mo@xmath1o@xmath2 compounds depends on the mo - mo distance , which controls the strength of afm se interactions between the @xmath52 orbitals . the latter competes with the fm interactions between the @xmath53 orbitals , and dominates when the mo - mo distance is small . in the frustrated pyrochlore lattice , this corresponds to the transition into the sg state , though we can not specify precisely which mechanism freeze the random configuration of the mo spins . it could be the disorder of the mo - mo bondlengths proposed in refs . and . another possibility is the anisotropy of nearest - neighbor magnetic interactions caused by the orbital ordering.@xcite it is also important to note that the coulomb repulsion @xmath7 should be sufficiently strong in order to open the band gap and suppress the fm double exchange interactions between the @xmath53 orbitals . @xmath11 the metal - insulator transition in these systems does not necessary coincide with the fm - sg transitions . all metallic compounds are expected to be nearly collinear ferromagnets , due to the double exchange mechanism . however , the inter - atomic exchange coupling in the insulating state can be both ferromagnetic and antiferromagnetic , depending on the mo - mo bondlength . the finding is qualitatively consistent with the experimental data.@xcite the magnetic structure in the insulating state is expected to be strongly non - collinear , due to the competition between the isotropic exchange interactions and the single - ion anisotropies . @xmath11 the large hall conductivity can be explained by the combination of three factors in the nearly collinear fm state : large soi at the mo sites , which unquenches the orbital magnetization ; strong enhancement of this effect by the on - site coulomb interaction ; and the half - metallic electronic structure of the fm pyrochlores . we also expect large magneto - optical effect in the same compounds . a possible way to enhance these properties is to use heavy @xmath50 elements , which have large spin - orbit coupling . unfortunately , all known @xmath50 pyrochlores are non - magnetic , though some of them are presumably close to the point of magnetic instability.@xcite nevertheless , one can try to exploit an old idea known from the physics of intermetallic alloys,@xcite and replace part of the mo sites in @xmath0moo@xmath2 by the @xmath50 elements . then , the main question is whether the remaining mo sites will be magnetic in the new environment or not . if they will , they will also magnetize the @xmath50 states at the neighboring sites , and in this way enhance the non - reciprocal characteristics of the considered pyrochlore compounds . as a purely model example , we have considered the ordered gd@xmath1mowo@xmath2 alloy ( using the lattice parameters of gd@xmath1mo@xmath1o@xmath2 ) . according to the lsda calculations , the system is expected to be ferromagnetic . the values of ( spin only ) magnetic moments at the mo and w sites are 2.1 and 1.5 @xmath161 , respectively . the situation is highly unusual , because the examples when @xmath6 and @xmath50 elements form a magnetic solution are rather rare . although gd@xmath1mowo@xmath2 itself may be rather unrealistic example from the technological point of view , the idea to use mo atoms in order to magnetize the heavy @xmath50 elements in the pyrochlore lattice seems to be promising . in conclusion , the simple hf calculations allow to rationalize many unusual properties of the pyrochlore compounds @xmath0mo@xmath1o@xmath2 . it is also important that all these properties can be explained by the universal value of @xmath7@xmath8@xmath9-@xmath10 ev , which is of the same order of magnitude as the characteristic bandwidth for the mo(@xmath5 ) electrons . this will certainly place the considered compounds to the direct proximity of the metal - insulator transition , meaning that the quantum fluctuations beyond the mean - field hf approach will play an important role too . this is certainly a very important direction for the extension of the present analysis . finally , we would like to make several comments on the comparison between our calculations and the experimental magnetization@xcite and neutron diffraction@xcite data . \(1 ) the experimental value of the mo moment ( @xmath8@xmath247-@xmath248 @xmath161 ) seems to be smaller than @xmath8@xmath160 @xmath161 obtained in our mean - field calculations for the nearly collinear fm state , even after taking into account the orbital part ( @xmath8@xmath194 @xmath161 ) , which according to the third hund s rule is antiparallel to the spin one and partly compensate the latter . the small difference is presumably related with the quantum fluctuations at the proximity of the metal - insulator transition.@xcite \(2 ) the experimental data for nd@xmath1mo@xmath1o@xmath2 and sm@xmath1mo@xmath1o@xmath2 typically indicate at the existence of two antiferromagnetically coupled sublattices . these sublattices are usually regarded as mo and @xmath3 ones . however , according to the electronic structure point of view , the _ spin _ coupling between these two sublattices should be ferromagnetic,@xcite being actually consistent with the experimental data for gd@xmath1mo@xmath1o@xmath2.@xcite the controversies can be reconciled with each other if the local magnetic moment at the nd / sm sites has a large orbital counterpart , which is larger than the spin one . the afm orientation of the spin and orbital moments at the nd / sm sites is again the subject of the third hund s rule . then , the magnetization and neutron diffraction experiments presumably probe the afm coupling between _ spin _ magnetic moments at the mo sites and the _ orbital _ magnetic moments at the nd / sm sites . gd@xmath249 has no orbital moment , and therefore the experimental data for this compound indicate at the conventional fm coupling between two spin sublattices . then , the sharp decrease of the magnetization observed in nd@xmath1mo@xmath1o@xmath2 at low @xmath19,@xcite which is frequently tempted to connect with the anomaly of the hall conductivity is due to the orbital magnetic structure , not the spin one . the change of the _ orbital _ magnetic structure seems to be the right subject to concentrate at when trying to explain , for example , the temperature dependence of the ahe . in this respect , another interesting problem seems to be the behavior of _ spin and orbital _ magnetic moments at the mo sites , because already in the ground state these two sublattices from mutually non - collinear magnetic structures . therefore , it is reasonable to expect very unconventional spin and orbital dynamic at the elevated temperatures . i would like to thank lars nordstrm for collaboration on early stages of this project and kind hospitality during my stay at uppsala university . i also benefited from discussions with n. nagaosa , y. tokura , k. terakura , and a. b. granovski . j. ye , y. b. kim , a. j. millis , b. i. shraiman , p. majumdar , and z. teanovish , phys . lett . * 83 * , 3737 ( 1999 ) . y. lyanda - geller , s. h. chun , m. b. salamon , p. m. goldbart , p. d. han , y. tomioka , a. asamitsu , and y. tokura , phys . b * 63 * , 184426 ( 2001 ) . j. smith , physica * 21 * , 877 ( 1955 ) ; j. smith , _ ibid . _ * 24 * , 39 ( 1958 ) ; j. m. luttinger , phys . rev . * 112 * , 739 ( 1958 ) ; j. kondo , prog . theor . phys . * 27 * , 772 ( 1962 ) ; f. e. maranzana , phys . rev . * 160 * , 421 ( 1967 ) ; l. berger , phys . b * 2 * , 4559 ( 1970 ) ; l. berger , _ ibid . _ * 5 * , 1862 ( 1972 ) ; s. k. lyo and t. holstein , phys . rev . * 29 * , 423 ( 1972 ) ; j. smith , phys . rev . b * 17 * , 1450 ( 1978 ) . we would not like to go here very deeply into all details of these controversies . briefly , karplus and luttinger in their pioneering work@xcite considered the contributions proportional to @xmath250 , where @xmath251 is the fermi distribution function . these contributions have the form of intra - band transitions in the conductivity tensor depicted by eq . ( [ eqn : sintra ] ) . all of them vanish in the periodic crystal , as it was correctly pointed out by smith.@xcite however , if the product of two momentum operators , @xmath252 , has an imaginary part , one can expect another contribution to @xmath18 coming from eq . ( [ eqn : sinter ] ) . this contribution is caused by the inter - band transitions , and in this sense the considered mechanism has the same origin as the magneto - optical effect . this term was intensively studied recently by nagaosa _ _ in the contexy of the spin - chirality problem.@xcite see also comments in ref . . a. b. granovskiy and ye . i. kondorskiy , fiz . metal . metalloved . * 39 * , 718 ( 1975 ) [ english transl . : the physics of metals and metallography * 39 * , no.4 , p.36 ( 1975 ) ] ; ye . i. kondorskiy , a. v. vedyayev , and a. b. granovskiy , _ ibid . _ * 40 * , 455 ( 1975 ) [ english transl . : the physics of metals and metallography * 40 * , no.3 , p.1 ( 1975 ) ] ; n. ryzhanova , a. vedyayev , a. crpieux , and c. lacroix , phys . b * 57 * , 2943 ( 1998 ) . using expression ( [ eqn : sinter ] ) for the conductivity tensor , it is easy to show that @xmath253@xmath4@xmath254@xmath89@xmath255 . therefore , @xmath256@xmath4@xmath22 , whereas @xmath18@xmath80@xmath257 can be finite ( the author thank n. nagaosa for pointing out this property ) . on the other hand , the onsager constraint @xmath258@xmath4@xmath259@xmath89@xmath260 ( @xmath79 being the net magnetization ) requires @xmath18 to be an even function of @xmath79.@xcite strictly speaking , the effect is masked by the diagonal elements of the conductivity tensor , and since @xmath261@xmath54@xmath262 , the resistivity @xmath263@xmath4@xmath89@xmath264 , which is the measured quantity , will vanish anyway in the perfectly periodic metallic state . in this sence , additional scattering mechanisms@xcite are indispensable in order to have finite @xmath261 and @xmath263 . however , it does not mean that @xmath265 itself will vanish in the periodic state . in our work we will consider such , perhaps somewhat idealistic , situation . in the local coordinate frame , the @xmath52 and two @xmath53 orbitals have the following form : @xmath266@xmath4@xmath267@xmath103@xmath268@xmath103@xmath269 , @xmath270@xmath4@xmath271@xmath89@xmath269 , and @xmath272@xmath4@xmath273@xmath89@xmath274@xmath103@xmath268@xmath103@xmath275 . by considering the _ nn _ hoppings between mo(@xmath52 ) orbitals in the pyrochlore lattice , @xmath276 , it is easy to show that the electronic structure for the @xmath52 states will consist of three bands ( in units of @xmath276 ) : @xmath277@xmath4@xmath89@xmath62@xmath278 , the so - called bonding and antibonding bands , and the doubly - degenerate non - bonding band @xmath279@xmath4@xmath62 , where @xmath280@xmath4@xmath281@xmath282@xmath283@xmath103@xmath284 . in fact , there is a non - spherical ( proportional to the coulomb parameter @xmath105 ) interaction beween occupied @xmath53 and @xmath52 orbitals , which will deforms the density of the @xmath52 electrons . this interaction has an appreciable effect in the afm state . for example , for y@xmath1mo@xmath1o@xmath2 and @xmath7@xmath4@xmath124 ev , the occupied @xmath52 orbital at the site 1 has the form @xmath285@xmath4@xmath130@xmath103@xmath131@xmath103@xmath135 , where @xmath136@xmath121@xmath286@xmath121@xmath287 . this will also lead to a small ( less than 1 mev ) anisotropy of the @xmath52-@xmath52 superexchange interactions . note that @xmath151 defined by eq . ( [ eqn : exchange ] ) , which is basically the second derivative of the total energy,@xcite characterizes the _ local stability _ of the magnetic state , and provides a _ complementary _ information to the standard total energy calculations . generally , the parameters @xmath151 depend on the magnetic state in which they are calculated and there is no simple relation between @xmath151 and the total energy . on the other hand , the total energy difference calculated for a limited number of magnetic states does not necessary guarantee that these states are locally stable to be the local minima of the total energy . the gd(@xmath288 ) core spin polarizes the valence , mainly gd(@xmath50 ) , electrons , which can interact with the mo(@xmath6 ) electrons . according to our estimates based on eq . [ eqn : exchange ] , the interaction @xmath289 is ferromagnetic and is of the order of 2 mev ( in lsda , and treating the gd(@xmath288 ) states as the core states ) , in a good agreement with the experimantal data.@xcite similar conclusion can be obtained from the analysis of the total energy difference , assuming the fm and afm coupling between mo and gd spins . results of these calculations have very transparent expanation , because gd@xmath1mo@xmath1o@xmath2 has half - metallic electronic structure and the fm coupling between mo and gd spins results in the additional energy gain for the @xmath83-spin mo(@xmath5 ) electrons ( see fig . [ fig.doslsda ] ) . for comparison , the direct gd(@xmath50)-gd(@xmath50 ) interaction is much smaller ( @xmath8@xmath290 mev , which is formally beyond the accuracy verge of the present calculations ) . presumably , other factors can also play some role in explaining the large difference of @xmath52-@xmath52 interactions obtained for y@xmath1mo@xmath1o@xmath2 and nd@xmath1mo@xmath1o@xmath2 . the standard se interaction is proportional to the ( minus ) second moment of the density of states , that actually explains the afm sign of the coupling.@xcite the @xmath83-spin @xmath52 bandwidths in y@xmath1mo@xmath1o@xmath2 and nd@xmath1mo@xmath1o@xmath2 differ by factor two ( fig . [ fig.dos30 ] ) , that should lead to the factor four difference for corresponding se interactions . however , the @xmath52 states are not fully isolated and interact with some other states , so that the expression for the se interaction should be modified as @xmath291 , where @xmath292 is the shift of the @xmath52 states due to the hybridization with other states . this shift can be different for the @xmath83- and @xmath84-spin electrons and also depend on the material , that presumably explains an additional difference between y@xmath1mo@xmath1o@xmath2 and nd@xmath1mo@xmath1o@xmath2 . note that at least in the mean - field approximation for the one - orbital model , the effects of the inter - site coulomb interactions , which are responsible for the charge disproportionation , can be included as a renormalization of the local ( on - site ) interaction , in the spirit of model hf calculations performed in this work : i. v. solovyev , phys . b * 60 * , 8550 ( 1999 ) . the conclusion can be easily verified by considering the spin hamiltonian @xmath293@xmath4@xmath89@xmath294@xmath103@xmath295 , consisting of the _ nn _ exchange and anisotropy parts . @xmath296@xmath4@xmath297 is the single - ion anisotropy tensor at the site 1 . @xmath298 , @xmath299 , and @xmath300 are obtained by the 180@xmath196 rotations of @xmath301 around @xmath39 , @xmath40 , and @xmath41 , respectively . then if @xmath101@xmath13@xmath22 ( the fm coupling ) and @xmath302@xmath13@xmath22 ( the easy axes scenario ) , the umbrella structure has lower energy in comparison with the two - in , two - out structure due to the additional energy gain coming from the _ nn _ exchange interactions . note that @xmath18@xmath303@xmath304 , where @xmath305 is the band gap . in addition , the non - collinear alignment reduces the net magnetization and the conventional contribution to the ahe caused by the soi.@xcite c. s. wang and j. callaway , phys . rev . b * 9 * , 4897 ( 1974 ) ; yu . a. uspenski and s. v. khalilov , zh . eksp . 95 * , 1022 ( 1989 ) [ english transl . jetp * 68 * , 588 ( 1989 ) ] ; p. m. oppeneer , t. maurer , j. sticht , and j. kbler , phys . b * 45 * , 10924 ( 1992 ) . another possibility could be the sample purity . however , according to the experimental estimates the sample quality seems to be good ( for example in y@xmath1mo@xmath1o@xmath2 the concentration of anti - site defects and deviation from the nominal oxygen stoichiometry was estimated as 4% and 1% , respectively).@xcite .structural parameters of @xmath0mo@xmath1o@xmath2 ( @xmath3@xmath4 y , nd , and gd ) : the cubic lattice parameter @xmath36 ( in ) , positions of o @xmath27 sites @xmath310 $ ] ( units of @xmath36 ) , the distances mo - mo and mo - o ( in ) , and the angles mo - o - mo and o - mo - o ( in degrees ) . [ cols="^,^,^,^,^,^,^ " , ]

the disk complex of a surface in a 3-manifold is used to define its _ topological index_. surfaces with well - defined topological index are shown to generalize well known classes , such as incompressible , strongly irreducible , and critical surfaces . the main result is that one may always isotope a surface with topological index to meet an incompressible surface so that the sum of the indices of the components of is at most . this theorem and its corollaries generalize many known results about surfaces in 3-manifolds , and often provides more efficient proofs . the paper concludes with a list of questions and conjectures , including a natural generalization of hempel s _ distance _ to surfaces with topological index .

0901.0208

let @xmath0 be a properly embedded , separating surface with no torus components in a compact , orientable 3-manifold @xmath5 . then the _ disk complex _ , @xmath6 , is defined as follows : 1 . vertices of @xmath6 are isotopy classes of compressions for @xmath0 . 2 . a set of @xmath7 vertices forms an @xmath8-simplex if there are representatives for each that are pairwise disjoint . here we explore what information is contained in the topology of @xmath6 . to this end , we define : [ d : indexn ] the _ homotopy index _ of a complex @xmath9 is defined to be 0 if @xmath10 , and the smallest @xmath1 such that @xmath11 is non - trivial , otherwise . we say a surface @xmath0 is _ topologically minimal _ if its disk complex @xmath6 is either empty or non - contractible . when @xmath0 is topologically minimal , we say its _ topological index _ is the homotopy index of @xmath6 . for example , a surface @xmath0 has topological index 1 if and only if @xmath12 is non - trivial , i.e. its disk complex is disconnected . when @xmath0 is the boundary of a handlebody then the disk complex was first defined by mccullough in @xcite , who showed that in this case @xmath6 is contractible . it follows that such surfaces are not topologically minimal . the goal of the present paper is to show that topologically minimal surfaces are a natural generalization of several well - known classes of surfaces in 3-manifolds , and that the results that hold for each of these classes also hold true for all topologically minimal surfaces . as an added benefit , proofs involving the set of all topologically minimal surfaces are often much shorter than existing proofs involving just , say , index 2 surfaces . this is largely owing to the inductive nature of the arguments . by definition , incompressible surfaces have topological index 0 . in the next section we show that the _ strongly irreducible _ surfaces of casson and gordon @xcite are precisely those that have topological index 1 . we also show that _ critical surfaces _ , previously defined by the author in @xcite and @xcite , have topological index 2 . one important property shared by these types of surfaces is that they may always be isotoped to meet an incompressible surface in a collection of loops that are essential on both . we show here that this is in fact a corollary of a powerful result about all topologically minimal surfaces . this is given by theorem [ c : maincorollary ] , which asserts that a topologically minimal surface @xmath0 and an incompressible surface @xmath2 can be isotoped so that @xmath3 is topologically minimal in @xmath13 . section [ s : heegaard ] contains corollaries to theorem [ c : maincorollary ] . we show there that if @xmath5 contains a topologically minimal heegaard surface then @xmath14 is incompressible . it then follows that if a closed 3-manifold @xmath5 contains any topologically minimal surface @xmath0 then either it is a heegaard surface , @xmath5 is haken , or @xmath0 is contained in a ball . ( in the final section we conjecture that this last possibility can not happen . ) finally , we show that if the disjoint union of surfaces is topologically minimal then so are its components , and its topological index is the sum of the indices of its components . combining this with theorem [ c : maincorollary ] , we find that a surface @xmath0 with topological index @xmath1 can be isotoped to meet an incompressible surface @xmath2 in such a way so that the sum of the indices of the components of @xmath3 is at most @xmath1 . this is a generalization of known results about topological index 0 and 1 surfaces . in any new theory , the questions raised are as important as the new results . in the final section of this paper we list a few tantalizing questions and conjectures about topologically minimal surfaces . these include conjectures about the possible indices of topologically minimal surfaces in various kinds of 3-manifolds , a natural generalization of hempel s _ distance _ invariant @xcite to surfaces of arbitrary topological index , and conjectures which relate geometric minimal surfaces to topologically minimal surfaces . much of the motivation for this work comes from ideas of hyam rubinstein . in the late 1990 s rubinstein pioneered the viewpoint that strongly irreducible heegaard splittings were the right class of surfaces within which to search for unstable ( geometrically ) minimal surfaces of index 1 , as well as their pl analogues , the so - called almost normal " surfaces . one often finds such surfaces by minimax arguments involving 1-parameter sweepouts . many of the topological arguments involving strongly irreducible surfaces also use 1-parameter sweepouts , so it became natural to think about such surfaces as being topologically minimal , " in a very imprecise sense . in later work the author defined critical surfaces as an attempt to find some topological analogue to geometrically minimal surfaces that have index 2 . as one would expect from such an analogue , arguments involving critical surfaces often involve 2-parameter sweepouts . in this paper we make precise the idea of topological index , demonstrate its usefulness , and conjecture its relation to geometric minimal surfaces . the present work is the first in a sequence of papers on this topic . in @xcite we define a relative version of topological index for surfaces with non - empty boundary . the main result of that paper is that complicated amalgamating surfaces act as barriers to low index , low genus , topologically minimal surfaces . this is the key technical tool necessary for the author s construction of a counter - example to the _ stabilization conjecture _ for heegaard splittings @xcite . further applications are given in @xcite , where we prove several results about amalgamation and isotopy of heegaard splittings . finally , in joint work with jesse johnson , we produce examples of 3-manifolds containing surfaces with high topological index @xcite . the author thanks several people for helpful comments during the preparation of this paper . jesse johnson had helpful suggestions regarding the construction of the family @xmath15 defined in the proof of theorem [ t : maintheorem ] . cameron gordon , daryl cooper , and andrew casson provided advice necessary for the proof of corollary [ c : indexsum ] . general helpful comments were made by martin scharlemann and yoav moriah . finally , it was saul schleimer and eric sedgiwck who first brought the index 1 case of theorem [ c : componentsintersection ] to the attention of the author , which was the beginings of the paper @xcite . in some sense this work is an extension of the main result of that paper . in this section we show that the concept of topological index generalizes several well known classes of surfaces in 3-manifolds . [ d : h / d ] let @xmath0 be a properly embedded surface in a 3-manifold @xmath5 . a loop @xmath16 on @xmath0 is _ essential _ if it does not bound a subdisk of @xmath0 . a disk @xmath17 is a _ compression _ for @xmath0 if @xmath18 is an essential loop on @xmath0 . the surface @xmath0 is _ incompressible _ if there are no compressions for it . if @xmath17 is a compression for @xmath0 then we construct the surface @xmath19 as follows . let @xmath20 denote the manifold obtained from @xmath5 by cutting open along @xmath0 . let @xmath21 denote a neighborhood of @xmath17 in @xmath20 . the surface @xmath19 is obtained from @xmath0 by removing @xmath22 and replacing it with the frontier of @xmath21 in @xmath20 . it follows immediately from the definitions that a surface has topological index 0 if and only if it is incompressible . we now show that surfaces with topological index 1 and 2 are also familiar . let @xmath23 and @xmath24 denote the sides of a heegaard surface @xmath0 , and @xmath25 and @xmath26 the subspaces of @xmath6 spanned by compressions in @xmath23 and @xmath24 . mccullough has called these complexes the _ disk complexes _ of @xmath23 and @xmath24 . mccullough proved that such disk complexes are contractible @xcite . it follows that the topology of @xmath6 is entirely determined by the simplices that connect @xmath25 to @xmath26 . with this in mind , it is natural to introduce special terminology when there are no edges connecting @xmath25 to @xmath26 . the following definition is due to casson and gordon @xcite . @xmath0 is _ strongly irreducible _ if there are compressions on opposite sides of @xmath0 , but each compression on one side meets all compressions on the other . the main result of @xcite is that if the minimal genus heegaard splitting of a 3-manifold is not strongly irreducible , then the manifold contains an incompressible surface . [ l : ttstrongirreducibility ] @xmath0 has topological index 1 if and only if it is strongly irreducible . by definition , a surface has topological index 1 when @xmath12 is non - trivial . hence , in this case @xmath6 is disconnected . however , by mccullough s result @xmath25 and @xmath26 are contractible , so the only way for @xmath6 to be disconnected is if both @xmath25 and @xmath26 are non - empty , and there are no edges connecting them . there are thus compressions on both sides , but any pair of such compressions intersect . in @xcite the author introduced the idea of a _ critical _ surface . the main result of that paper is that if the minimal genus common stabilization of a pair of heegaard splittings is not critical , then the manifold contains an incompressible surface . critical surfaces were also instrumental in the author s proof of a conjecture of c. gordon @xcite . [ d : critical ] @xmath0 is _ critical _ if the compressions for @xmath0 can be partitioned into sets @xmath27 and @xmath28 such that : 1 . for each @xmath29 there is at least one pair of disks @xmath30 on opposite sides of @xmath0 such that @xmath31 . 2 . if @xmath32 and @xmath33 are on opposite sides of @xmath0 then @xmath34 . [ l : critical ] @xmath0 has topological index 2 if and only if it is critical . we first establish that if @xmath0 has topological index 2 then it is critical . let @xmath35 be the subspace of @xmath6 consisting of those cells spanned by vertices in both @xmath25 and @xmath26 . [ c : betterpath ] any path in @xmath35 which connects two vertices representing disks on the same side of @xmath0 is homotopic in @xmath35 to a path in either @xmath25 or @xmath26 . let @xmath36 be such a path , where @xmath37 and @xmath38 are disks in @xmath23 , and the path contains the fewest possible number of disks in @xmath24 . let @xmath39 be the first disk in this path in @xmath24 . then @xmath40 is in @xmath23 . there are now two cases , depending on whether @xmath41 in @xmath23 or @xmath24 . consider first the case that @xmath42 . we now produce a path @xmath43 in @xmath35 from @xmath40 to @xmath41 that consists entirely of disks in @xmath23 . by replacing @xmath39 with this path , we get a new path from @xmath37 to @xmath38 which contains fewer disks in @xmath24 , a contradiction . first , note that by an innermost disk argument we may assume that each component of @xmath44 is an arc . now we construct the new path from @xmath40 to @xmath41 by induction as follows : 1 . let @xmath45 2 . assume we have constructed a sequence of disks @xmath46 in @xmath35 of disks that lie in @xmath23 , such that for all @xmath47 , @xmath48 . if @xmath49 then we let @xmath50 and we have produced the desired path . otherwise , let @xmath16 be an arc of @xmath51 that is outermost on @xmath41 . the arc @xmath16 then cuts off a subdisk disk @xmath52 of @xmath41 whose interior is disjoint from @xmath53 . the arc @xmath16 also cuts the disk @xmath46 into two subdisks . one of these , together with the disk @xmath52 , forms a compressing disk @xmath54 which meets @xmath41 fewer times than @xmath46 did . note that every disk in the path @xmath55 is disjoint from @xmath39 , a compressing disk in @xmath24 . hence , the entire path @xmath55 lies in @xmath35 , as desired . we now move on to the case that @xmath56 . as the edge @xmath57 is in @xmath35 , it must be part of a simplex @xmath58 whose vertices represent disks on both sides of @xmath0 . therefore , there must be a vertex @xmath59 of @xmath58 which represents a disk on the opposite side of @xmath0 as both @xmath39 and @xmath41 , i.e. a disk in @xmath23 . now insert @xmath59 in the path between @xmath39 and @xmath41 , and apply the above argument to again get rid of @xmath39 . since @xmath0 has topological index 2 is follows immediately that @xmath60 is non - trivial . since both @xmath25 and @xmath26 are contractible we conclude there is a non - trivial loop @xmath61 in @xmath6 that passes from @xmath25 to @xmath26 and back , crossing through @xmath35 exactly twice . let @xmath62 and @xmath63 be the two edges of @xmath35 traversed by this path , where @xmath64 and @xmath65 . [ c : gammadisconnected ] the edges @xmath62 and @xmath63 are in different components of @xmath35 . suppose not . then there is a path in @xmath35 connecting @xmath66 to @xmath67 . by claim [ c : betterpath ] there is such a path @xmath68 consisting entirely of disks in @xmath23 . as @xmath25 is simply connected , the path @xmath69 is homotopic to @xmath68 . the edges @xmath62 and @xmath63 , together with @xmath68 , now form a path in @xmath35 from @xmath70 to @xmath71 . by claim [ c : betterpath ] this is homotpic to a path @xmath72 which lies entirely in @xmath73 . we have thus homotoped the original loop @xmath61 entirely into @xmath26 , a simply connected space , contradicting its non - contractibility . claim [ c : gammadisconnected ] immediately implies that @xmath35 is disconnected . we may therefore partition its components into two non - empty sets , @xmath27 and @xmath28 , where @xmath74 . since @xmath27 and @xmath28 are a partition of the components of @xmath35 , there are no edges @xmath75 that connect them , where @xmath32 and @xmath33 . the sets @xmath27 and @xmath28 thus satisfy the conditions of definition [ d : critical ] . ( note that any vertex of @xmath6 that is not in @xmath35 can be added to either @xmath27 or @xmath28 , and the conditions of definition [ d : critical ] will still be satisfied . ) to complete the proof of the theorem , we must now establish that if @xmath0 is critical then it has topological index 2 . let @xmath76 , and @xmath77 be as in definition [ d : critical ] . we must produce a non - trivial loop in @xmath6 . since @xmath25 is contractible , there is a path of compressions in @xmath25 from @xmath66 to @xmath67 . similarly , there is a path from @xmath70 to @xmath71 in @xmath26 . these two paths , together with the edges @xmath78 , form a loop @xmath16 in @xmath6 . by way of contradiction , suppose @xmath16 is trivial in @xmath60 . then there is a map @xmath79 of a disk @xmath17 into @xmath6 such that @xmath80 . for some triangulation @xmath81 of @xmath17 , we may assume @xmath79 is simplicial . we now assume that all choices have been made so that the number of 2-simplices in @xmath81 is minimal . let @xmath58 denote the triangle in @xmath81 that has @xmath62 as one of its edges . without loss of generality we assume the third vertex of @xmath58 represents a compression in @xmath23 , and denote it as @xmath82 . since @xmath83 is an edge of @xmath58 , it follows that @xmath84 . hence , by criticality @xmath32 . if @xmath82 is in the interior of @xmath17 then remove @xmath58 from @xmath17 and replace @xmath66 with @xmath82 . this increases the combinatorial length of @xmath85 , but reduces the number of 2-simplices in @xmath81 , a contradiction . the remaining case is when @xmath82 is in @xmath85 . then the edge @xmath83 cuts @xmath17 into two smaller disks . one of these , @xmath86 , contains the edge @xmath63 . if we now replace @xmath17 with @xmath86 and @xmath66 with @xmath82 , we again contradict our minimality assumption . in this section we show that a topologically minimal surface can always be isotoped so that it meets the complement of an incompressible surface in a topologically minimal surface . let @xmath0 and @xmath2 be properly embedded surfaces in a 3-manifold @xmath5 . let @xmath17 be a compression for @xmath0 . we say @xmath17 has a _ shadow _ ( with respect to @xmath2 ) if there is a disk @xmath86 where @xmath87 , @xmath88 , and the interior of @xmath86 meets @xmath0 in loops that are inessential on @xmath0 . the disk @xmath86 is said to be a _ shadow _ of @xmath17 . see figure [ f : shadow ] . , for @xmath0 and its shadow , @xmath86.,width=3 ] the main idea behind this paper is to exploit relationships between the homotopy indices of various complexes that depend on a specific position of @xmath0 . the first of these is the disk complex @xmath6 of @xmath0 . the complex @xmath89 is the subset of @xmath6 such that each vertex has a shadow . later we will encounter a third complex , @xmath90 . the relationship between the homotopy indices of the complexes @xmath6 and @xmath89 is given presently in theorem [ t : maintheorem ] . later in this section we will use this theorem to prove that when @xmath0 is topologically minimal , then it can be isotoped so that it is topologically minimal in the complement of @xmath2 . we then show that many of the standard results in 3-manifold topology , presently known for surfaces with low topological index , generalize to surfaces with arbitrary topological index . [ t : maintheorem ] let @xmath0 and @xmath2 be properly embedded surfaces in @xmath5 , where @xmath0 has topological index @xmath1 . then @xmath0 may be isotoped so that 1 . @xmath0 meets @xmath2 in @xmath91 points of tangency , for some @xmath92 . away from these tangencies @xmath0 is transverse to @xmath2 . the complex @xmath89 has homotopy index @xmath93 . when @xmath0 has topological index 0 the result is immediate , as @xmath94 . we will assume , then , that @xmath0 has topological index @xmath95 . it follows that @xmath96 is non - trivial , and thus there is a map @xmath97 of an @xmath98-sphere @xmath99 into the @xmath98-skeleton of @xmath6 which is not homotopic to a point . let @xmath21 be the cone on @xmath99 to a point @xmath100 . hence , @xmath21 is an @xmath1-ball . our first challenge is to define a continuous family of surfaces @xmath15 in @xmath5 isotopic to @xmath0 , where @xmath101 . let @xmath81 be a triangulation of @xmath102 so that the map @xmath103 is simplicial . let @xmath104 denote the set of vertices of @xmath6 that are contained in @xmath105 . for each @xmath106 choose a representative @xmath39 from the isotopy class of disks represented by @xmath107 so that if @xmath108 is an edge of @xmath6 , then @xmath109 . for each @xmath106 , let @xmath110 be a small enough neighborhood of @xmath39 in @xmath5 so that @xmath111 whenever @xmath108 is an edge of @xmath6 , and let @xmath112 be a homeomorphism that takes @xmath110 to the standard unit ball in @xmath113 . choose @xmath112 so that @xmath114 is the graph of @xmath115 ( in cylindrical coordinates ) , and @xmath116 is a disk in the @xmath117-plane . for each disk @xmath39 we now define a family of surfaces @xmath118 in @xmath110 , parameterized by a variable @xmath119 $ ] . these surfaces are given by the images of the graphs of @xmath120 , under the map @xmath121 ( see figure [ f : h(t ) ] ) . @xmath122{hxfamily.eps}\ ] ] extend @xmath81 to a triangulation @xmath123 on @xmath21 by coning each simplex of @xmath81 to the point @xmath100 . suppose @xmath124 is the image of an @xmath98-simplex @xmath58 of @xmath81 under the map @xmath103 . we now identify the @xmath1-simplex of @xmath123 which is the cone on @xmath58 with the unit cube in @xmath125 . label the axes of @xmath125 with the variables @xmath126 . place @xmath100 at the origin , and the vertex @xmath127 of @xmath58 such that @xmath128 at the point with @xmath129 and @xmath130 for all @xmath131 . if @xmath91 is at the barycenter of a face @xmath132 of @xmath58 then place it at the vertex of the cube where the coordinates corresponding to the vertices of @xmath132 are @xmath133 and the other coordinates are 0 . we now linearly extend over the entire simplex to complete the identification with the cube . now , if @xmath134 is in this @xmath1-simplex then @xmath134 has coordinates @xmath135 . let @xmath15 be the surface obtained from @xmath0 by replacing @xmath136 with the surface @xmath137 , for each @xmath106 between @xmath138 and @xmath139 . see figure [ f : hxdefn ] . repeating this for each @xmath1-simplex of @xmath123 gives us the complete family of surfaces @xmath15 . of @xmath123 , and a few of the surfaces @xmath15 for @xmath140 . the union of the faces of the cube that do not meet @xmath100 is a simplex of @xmath81.,width=5 ] we assume @xmath0 is initially transverse to @xmath2 . for each @xmath106 , the surface @xmath141 is tangent to @xmath2 for finitely many values @xmath142 of @xmath143 . hence , for each @xmath101 the surface @xmath15 is tangent to @xmath2 at finitely many points , and each such point is in a distinct ball @xmath110 . note also that if @xmath144 , then @xmath15 and @xmath145 agree inside of @xmath110 . hence , if @xmath15 is tangent to @xmath2 in @xmath110 then the surface @xmath145 will also be tangent to @xmath2 , for all @xmath146 in the plane where @xmath147 . it follows that each @xmath1-simplex of @xmath123 is cubed by the points @xmath134 where @xmath15 is tangent to @xmath2 . see figure [ f : cubing ] . hence , @xmath21 is cubed by the @xmath1-simplices of @xmath123 , together with this cubing of each such simplex . we denote this cubing of @xmath21 as @xmath148 . it follows that if @xmath134 is in a codimension @xmath91 cell of @xmath148 then the surface @xmath15 is tangent to @xmath2 in at most @xmath91 points . of @xmath123 is cut up by planes into subcubes . each such plane is determined by the points @xmath134 in which @xmath15 is tangent to @xmath2 in @xmath110 , for some @xmath106.,width=3 ] we now produce a contradiction by defining a continuous map @xmath149 from @xmath21 into @xmath6 . the map @xmath150 will be equal to @xmath103 on the barycenters of the @xmath98-cells of @xmath81 , which will in turn imply that @xmath149 maps @xmath99 onto @xmath105 with the same degree as @xmath103 . a contradiction follows as @xmath105 is not homotopic to a point . for each @xmath101 let @xmath151 . if @xmath152 is a cell of @xmath148 , then we define @xmath153 to be the set @xmath154 , for any choice of @xmath134 in the interior of @xmath152 . note that if @xmath134 and @xmath146 are in the interior of the same cell @xmath152 of @xmath148 , then the pair @xmath155 is isotopic to @xmath156 . hence , @xmath157 , and thus @xmath153 is well defined . the map @xmath149 defined below will take each cell @xmath152 of @xmath148 into @xmath153 . first , we establish a few properties of @xmath153 . [ c : subset ] suppose @xmath132 is a cell of @xmath148 which lies on the boundary of a cell @xmath152 . then @xmath158 . pick @xmath159 and @xmath160 . if @xmath161 then @xmath17 is isotopic to a compression for @xmath15 that has a shadow @xmath86 . to show @xmath162 we must show that @xmath17 is isotopic to a compression for @xmath145 that has a shadow . note that @xmath163 is obtained from @xmath164 by resolving some tangency . hence , any loop of @xmath165 is isotopic to a loop of @xmath166 . it follows that since @xmath167 was a loop on @xmath15 disjoint from @xmath2 , then @xmath167 will be a loop on @xmath145 that is disjoint from @xmath2 . furthermore , as the interior of @xmath86 meets @xmath15 in a collection of loops that are inessential on @xmath15 , it follows that the interior of @xmath86 meets @xmath145 in a collection of loops that are inessential on @xmath145 . we conclude @xmath86 is a shadow for @xmath17 , both as a compression for @xmath15 and as a compression for @xmath145 . hence , @xmath168 . [ c : contradiction ] for each cell @xmath152 of @xmath148 , @xmath169 let @xmath134 be in the interior of a codimension @xmath91 cell @xmath152 of @xmath148 . then the dimension @xmath170 is @xmath171 . the surface @xmath15 is tangent to @xmath2 in at most @xmath91 points , and is transverse to @xmath15 elsewhere . recall @xmath151 . thus , if the theorem is false then @xmath154 is non - empty , and @xmath172 for all @xmath173 . we now define @xmath149 on the @xmath138-skeleton of @xmath148 . for each @xmath138-cell @xmath174 , we will choose a point in @xmath154 to be @xmath175 . if @xmath134 is in the interior of @xmath21 then @xmath175 may be chosen to be an arbitrary point of @xmath154 . if @xmath134 is a point of @xmath102 then @xmath134 is contained in ( perhaps more than one ) @xmath98-simplex @xmath176 of @xmath81 . let @xmath177 denote the face of @xmath176 spanned by the vertices @xmath127 such that @xmath178 if @xmath179 , and @xmath180 otherwise . ( note that if @xmath134 was on the boundary of @xmath176 , so that it was also contained in some other @xmath98-simplex of @xmath81 , then we still end up with the same simplex @xmath181 of @xmath81 . ) so , for example , if @xmath134 is at the barycenter of @xmath176 then @xmath182 . by construction , for each vertex @xmath127 of @xmath177 the surface @xmath15 is pinched to a point along a disk @xmath17 in the isotopy class of @xmath183 . hence , for all @xmath146 near @xmath134 the disk @xmath17 is a compression for @xmath145 that is disjoint from @xmath2 . it follows that the entire simplex @xmath184 is contained in @xmath154 , and thus we may choose the barycenter of @xmath184 to be the image of @xmath175 . in particular , if @xmath134 is the barycenter of @xmath176 then @xmath185 . we now proceed to define the rest of the map @xmath149 by induction . let @xmath152 be a @xmath186-dimensional cell of @xmath148 . by induction , assume @xmath149 has been defined on the @xmath187-skeleton of @xmath148 . in particular , @xmath149 has been defined on @xmath188 . suppose @xmath132 is a face of @xmath152 . by claim [ c : subset ] @xmath158 . by assumption @xmath189 is defined and @xmath190 . we conclude @xmath191 for all @xmath192 , and thus @xmath193 since @xmath194 it follows from claim [ c : contradiction ] that @xmath195 . since @xmath196 is the dimension of @xmath188 , we can thus extend @xmath149 to a map from @xmath152 into @xmath153 . what remains to be shown is that if @xmath152 is in @xmath102 then the extension of @xmath149 from @xmath188 to @xmath152 may be done so that @xmath197 . let @xmath198 be the simplex of @xmath81 whose interior contains the interior of @xmath152 . we need only show that @xmath199 . since @xmath200 will be a subsimplex of @xmath201 , it follows that @xmath149 can be extended over @xmath152 to this subsimplex . by equation [ e : boundary ] , @xmath202 . so all we must do now is to show @xmath203 . let @xmath132 denote a face of @xmath152 , and @xmath204 the simplex of @xmath81 whose interior contains the interior of @xmath132 . then @xmath205 is contained in @xmath206 . by induction we may assume @xmath207 . putting this together we conclude @xmath208 for each @xmath192 , and thus @xmath203 . let @xmath2 be a properly embedded surface in a 3-manifold @xmath5 . then we let @xmath209 denote the complement of a neighborhood of @xmath2 in @xmath5 . for each subset @xmath210 of @xmath5 , let @xmath211 . we define the complex @xmath90 precisely as above , where the vertices of @xmath90 correspond to the compressions for @xmath212 in @xmath209 . the relationship between the complexes @xmath6 , @xmath90 , and @xmath89 is depicted in figure [ f : complexes ] . , @xmath89 , and @xmath90 overlap.,width=4 ] we now use theorem [ t : maintheorem ] to show that when @xmath0 is topologically minimal and @xmath2 is incompressible , then @xmath0 may be isotoped so that @xmath212 is topologically minimal in @xmath209 . in section [ s : heegaard ] we explore the implications of this when @xmath0 is a heegaard surface . [ l : noncompressioneffect ] let @xmath2 be a properly embedded , incompressible surface in an irreducible 3-manifold @xmath5 . let @xmath0 be a properly embedded surface in @xmath5 which meets @xmath2 transversally , with the exception of a finite number of center and saddle tangencies , such that @xmath89 has well defined homotopy index . let @xmath17 be a compression for @xmath212 in @xmath209 that is not a compression for @xmath0 . then @xmath213 . let @xmath20 and @xmath21 be as given in definition [ d : h / d ] . then @xmath19 is obtained from @xmath0 by removing @xmath22 from @xmath0 and replacing it with the frontier @xmath214 of @xmath21 in @xmath20 . as @xmath17 is not a compression for @xmath0 , @xmath85 bounds a subdisk @xmath215 . we first show @xmath216 . suppose @xmath217 . then @xmath218 can be isotoped off of @xmath214 . if @xmath59 now meets the ball @xmath21 then it can be further isotoped so that @xmath219 is a collection of disks parallel to @xmath17 . but then each component of @xmath219 can be swapped with a disk parallel to @xmath220 . the resulting disk has the same boundary as @xmath59 , but is disjoint from @xmath0 . by the irreducibility of @xmath5 this disk must therefore be properly isotopic to @xmath59 . see figure [ f : htoh / dpart2 ] . we conclude that @xmath59 was a compression for @xmath0 that persisted as a compression for @xmath19 . @xmath59 is therefore a compression for @xmath0 that is disjoint from @xmath17 . is not a compression for @xmath0 , any compression @xmath59 for @xmath19 ( right figure ) is always isotopic to a compression for @xmath0 ( left figure ) . if @xmath221 is a shadow for @xmath59 as a compression for @xmath19 ( right figure ) , then @xmath221 is a shadow for @xmath59 as a compression for @xmath0 ( left figure).,width=4 ] now let @xmath221 be a shadow for @xmath59 as a compression for @xmath19 . as @xmath222 , it follows that @xmath223 . so , if @xmath221 meets the ball @xmath21 , then it meets it in disks parallel to @xmath17 . the disk @xmath221 thus meets @xmath0 in loops isotopic to @xmath224 , together with loops parallel to @xmath225 . it follows that the interior of @xmath221 meets @xmath0 in inessential loops , and thus @xmath221 is a shadow for @xmath59 as a compression for @xmath0 , i.e. @xmath226 . see figure [ f : htoh / dpart2 ] . we now show @xmath227 . let @xmath59 now denote an element of @xmath89 . thus , @xmath228 . we assume @xmath59 has been chosen so that @xmath229 is minimal . first we suppose @xmath230 . if the interior of @xmath59 meets @xmath17 then we may surger it off by a standard innermost disk argument . so in this case we may assume @xmath231 . since @xmath59 is a compression for @xmath0 but @xmath17 is not , it now follows that @xmath59 is a compression for @xmath19 . any shadow for @xmath59 as a compression for @xmath0 will be a shadow for @xmath59 as a compression for @xmath19 , and thus @xmath217 . finally , we consider the case @xmath232 . our goal is to isotope @xmath59 to a compression @xmath233 such that @xmath234 , contradicting our minimality assumption . let @xmath61 denote an arc of @xmath235 that is outermost on @xmath220 . then @xmath61 cuts a disk @xmath236 off of @xmath220 whose interior does not meet @xmath59 . we can use the disk @xmath236 to guide an isotopy of both @xmath59 and its shadow @xmath221 to a compression @xmath237 for @xmath0 and a disk @xmath238 with @xmath239 . see figure [ f : etoe0 ] . note that @xmath234 . if @xmath240 , then it follows from the fact that @xmath221 was a shadow of @xmath59 that @xmath238 will be a shadow of @xmath237 . thus , @xmath233 as desired . to obtain @xmath237 and @xmath238 from @xmath59 and @xmath221.,width=4 ] if @xmath241 then the disk @xmath238 will not be a shadow for @xmath237 , since @xmath242 . what remains then is to show that nonetheless , @xmath237 has a shadow . let @xmath243 denote a small product neighborhood of @xmath2 . since @xmath242 , it follows that @xmath244 . let @xmath245 denote a loop of @xmath246 that is outermost on @xmath238 . as @xmath2 is incompressible , @xmath245 bounds a subdisk @xmath247 of @xmath248 . see figure [ f : e0toshadow ] . although @xmath2 may not be transverse to @xmath0 , the surface @xmath248 thus , the disk @xmath247 meets @xmath0 in a collection of loops . we claim these loops are inessential on @xmath0 , and thus @xmath247 can be used to surger @xmath238 to a disk which meets @xmath2 fewer times . the new disk will meet @xmath0 more times , but each new intersection introduced will be inessential on @xmath0 . thus , by repeating this process we transform @xmath238 to a shadow for @xmath237 , as desired . to obtain a contradiction , suppose at least one loop of @xmath249 is essential on @xmath0 . let @xmath16 be a such loop that is innermost on @xmath247 . the loop @xmath16 bounds a subdisk @xmath250 of @xmath247 whose interior may meet @xmath0 in inessential loops . see figure [ f : e0toshadow ] . we claim @xmath250 is the shadow of a compression @xmath251 for @xmath0 , and thus @xmath252 . , @xmath253 , and @xmath245 , and the disks @xmath250 , @xmath21 , and @xmath247.,width=4 ] let @xmath253 denote a loop of @xmath254 that is innermost on @xmath250 . as @xmath253 is inessential on @xmath0 it bounds a subdisk @xmath21 of @xmath0 . see figure [ f : e0toshadow ] . the disk @xmath21 can be used to surger @xmath250 , lowering @xmath255 . continuing in this way we arrive at a disk @xmath251 with the same boundary as @xmath250 but whose interior is disjoint from @xmath0 . as @xmath256 is essential on @xmath0 , we conclude @xmath251 is a compression for @xmath0 . the disk @xmath250 is then a shadow for @xmath251 , and thus @xmath252 . finally , suppose @xmath210 is any other element of @xmath89 . as @xmath257 and @xmath258 , it follows that @xmath259 . by a standard innermost disk argument ( and the irreducibility of @xmath5 ) we may isotope @xmath210 to remove any intersections of its interior with the interior of @xmath251 . thus , we may assume @xmath260 . the disk @xmath210 is therefore connected to the disk @xmath251 by an edge in @xmath89 . as this holds for all disks @xmath261 , we conclude @xmath89 is contractible to @xmath251 . as @xmath89 is not contractible , we have reached a contradiction . [ c : maincorollary ] let @xmath2 be a properly embedded , incompressible surface in an irreducible 3-manifold @xmath5 . let @xmath0 be a properly embedded surface in @xmath5 with topological index @xmath1 . then @xmath0 may be isotoped so that 1 . @xmath0 meets @xmath2 in @xmath91 saddle tangencies , for some @xmath92 . away from these tangencies @xmath0 is transverse to @xmath2 . @xmath212 has topological index @xmath106 , for some @xmath93 . we begin by isotoping @xmath0 so as to satisfy the conclusion of theorem [ t : maintheorem ] . hence , we assume @xmath0 is tangent to @xmath2 in @xmath91 points , and the homotopy index of @xmath89 is at most @xmath171 . let @xmath17 be a compression for @xmath212 that is not a compression for @xmath0 . then @xmath85 bounds a subdisk @xmath220 of @xmath0 . by lemma [ l : noncompressioneffect ] , @xmath213 . the surface @xmath19 contains a component @xmath262 isotopic to @xmath0 ( by the irreducibility of @xmath5 ) , and a surface isotopic to @xmath263 . note that as @xmath264 , @xmath262 meets @xmath2 fewer times than @xmath0 did . thus , we may repeat the above procedure only finitely many times . note also that this procedure will remove all center tangencies of @xmath0 with @xmath2 . we arrive at a surface @xmath265 isotopic to @xmath0 with @xmath266 , such that every compression for @xmath267 is also a compression for @xmath265 . as such compressions lie in the complement of @xmath2 , they are their own shadows . hence , such compressions are elements of @xmath268 . we conclude @xmath269 . we claim the opposite inclusion is true as well , and thus @xmath270 . suppose now @xmath271 . let @xmath221 be a shadow of @xmath59 . let @xmath253 be a loop of @xmath272 that is innermost on @xmath221 . then @xmath253 bounds subdisks @xmath273 and @xmath274 . if @xmath275 , then @xmath276 is a compression for @xmath267 that is not a compression for @xmath265 , a contradiction . we conclude @xmath277 . since @xmath278 and @xmath273 , we conclude @xmath279 . the sphere @xmath280 thus bounds a ball in the complement of @xmath2 that we can use to guide an isotopy of @xmath276 to @xmath281 . ( this may remove other components of @xmath282 as well . ) we thus transform the disk @xmath221 to a disk @xmath283 such that @xmath284 , @xmath285 , and @xmath286 . continuing in this way we arrive at a compression for @xmath265 with the same boundary as @xmath59 , which is disjoint from @xmath2 . thus @xmath287 . we have now produced a surface @xmath265 , isotopic to @xmath0 , such that @xmath288 thus , the homotopy index of @xmath289 is equal to the homotopy index of @xmath89 . [ c : essentialintersection ] let @xmath2 be a properly embedded , incompressible surface in an irreducible 3-manifold @xmath5 . let @xmath0 be a properly embedded surface in @xmath5 with topological index @xmath1 . then @xmath0 and @xmath2 may be isotoped so that any loop of @xmath290 is essential on both surfaces . when @xmath0 is a heegaard surface whose topological index is one this is a well - known result that has been used extensively in the literature . it first appears in the literature in the above form as lemma 6 of @xcite , although it is implicit in theorem 1 of @xcite and is directly implied by theorem 1.1 of @xcite . the first step is to use theorem [ c : maincorollary ] to isotope @xmath0 so that @xmath212 is topologically minimal . the manifold @xmath209 is obtained from @xmath5 by removing a submanifold @xmath291 . let @xmath292 and @xmath293 denote the copies of @xmath2 on the boundary of @xmath243 . each component of @xmath294 is a loop or arc of @xmath295 . hence , we must show that every loop of @xmath295 that is inessential on @xmath292 is inessential on @xmath212 . if there is a loop of @xmath295 that is inessential on @xmath292 then there is such a loop @xmath16 that bounds a subdisk @xmath276 of @xmath292 whose interior is disjoint from @xmath212 . if @xmath16 is essential on @xmath212 then @xmath276 is a compression for @xmath212 . now suppose @xmath17 is some other element of @xmath90 . as @xmath296 , the disks @xmath276 and @xmath17 can be made disjoint in @xmath209 , and hence @xmath297 is an edge of @xmath90 . we conclude @xmath276 is connected by an edge to every other element of @xmath90 . it follows that @xmath90 is contractible to @xmath276 , a contradiction . we conclude that all loops of @xmath294 that are inessential on @xmath292 are also inessential on @xmath0 . any such loop thus bounds a disk component of @xmath212 that can be isotoped into @xmath243 , without affecting @xmath90 . by successively performing this operation we thus arrive at the desired position of @xmath0 with respect to @xmath292 , a surface isotopic to @xmath2 . in this section we give some applications of topological index theory to heegaard splittings of 3-manifolds . we also show that the topological index of a surface is the sum of the topological indices of its components . [ l : negativebdy ] let @xmath0 be a properly embedded surface which separates @xmath5 into @xmath23 and @xmath24 . let @xmath298 be a surface obtained from @xmath0 by a sequence of compressions into @xmath23 . then @xmath298 is incompressible in the submanifold cobounded by @xmath0 and @xmath298 . let @xmath299 denote the union of the compressions used to obtain @xmath298 from @xmath0 . let @xmath59 denote a compression for @xmath298 that lies between @xmath0 and @xmath298 . by an innermost disk argument , we may surger @xmath59 off of each disk @xmath39 . but the complement of a neighborhood of @xmath300 in this submanifold is a product . as the boundary of a product does not admit compressions , we have thus reached a contradiction . [ l : topminsubmfld ] let @xmath0 be a properly embedded surface which separates @xmath5 into @xmath23 and @xmath24 . let @xmath298 and @xmath301 be surfaces obtained from @xmath0 by maximal sequences of compressions into @xmath23 and @xmath24 . let @xmath302 be the submanifold of @xmath5 cobounded by @xmath298 and @xmath301 . if @xmath0 is topologically minimal in @xmath5 then @xmath0 is topologically minimal in @xmath302 . it suffices to show that every compression for @xmath0 in @xmath5 is isotopic to a compression in @xmath302 . let @xmath17 be such a compression , and assume @xmath303 . isotope @xmath17 so that it meets @xmath298 minimally . if @xmath304 , then the conclusion of the lemma follows . hence , we assume there is a subdisk @xmath86 of @xmath17 , cut off by @xmath298 , whose interior is disjoint from @xmath298 . if @xmath305 is not essential , then we contradict our assumption that @xmath306 is minimal . hence , @xmath305 is essential and we conclude @xmath86 is a compression for @xmath298 . if @xmath86 lies outside of @xmath302 then we contradict the maximality of the sequence of compressions used to obtain @xmath298 . but if @xmath86 lies in @xmath302 then it is in the submanifold cobounded by @xmath0 and @xmath298 . this contradicts lemma [ l : negativebdy ] . [ t : incompressibleboundary ] let @xmath0 be a properly embedded surface which separates @xmath5 into @xmath23 and @xmath24 . let @xmath298 be a surface obtained from @xmath0 by a maximal sequence of compressions into @xmath23 . if @xmath0 is topologically minimal then @xmath298 is incompressible in @xmath5 . let @xmath301 be the surface obtained from @xmath0 by a maximal sequence of compressions into @xmath24 , and @xmath302 the submanifold of @xmath5 cobounded by @xmath298 and @xmath301 . by lemma [ l : topminsubmfld ] the surface @xmath0 is topologically minimal in @xmath302 . we now claim that if either @xmath298 or @xmath301 is compressible , then there is a compression for one that misses the other . assume there is no such compression for @xmath301 . let @xmath17 be a compression for @xmath298 in @xmath5 . isotope @xmath17 so that it meets @xmath301 minimally . if @xmath17 misses @xmath301 then we establish our claim . assume then that @xmath17 meets @xmath301 . let @xmath86 be a subdisk of @xmath17 cut off by @xmath301 . if @xmath307 is inessential on @xmath301 , then we contradict our assumption that @xmath308 is minimal . but if @xmath307 is essential on @xmath301 then @xmath86 is a compression for @xmath301 that misses @xmath298 , a contradiction . we conclude there is a compression @xmath17 for either @xmath298 or @xmath301 that misses the other . that is , @xmath17 is a compression for @xmath309 . if @xmath17 lies outside of @xmath302 then we contradict the minimality of the sequence of compressions used to obtain @xmath298 or @xmath301 . hence , @xmath310 . note that @xmath17 is itself a properly embedded , incompressible surface in @xmath302 . we may thus apply corollary [ c : essentialintersection ] to isotope @xmath0 in @xmath302 to meet @xmath17 in a collection of loops that are essential on both surfaces . since @xmath17 does not contain any essential loops , we conclude @xmath311 . the disk @xmath17 now lies either between @xmath0 and @xmath298 , or between @xmath0 and @xmath301 . in either case we contradict lemma [ l : negativebdy ] . let @xmath0 be a topologically minimal heegaard surface in a 3-manifold , @xmath5 . then @xmath14 is incompressible . in the topological index one case this follows also from a celebrated lemma of haken @xcite . in the topological index two case it was established by the author in @xcite . let @xmath23 , @xmath24 , @xmath298 , and @xmath301 be as in theorem [ t : incompressibleboundary ] . since @xmath0 is a heegaard surface , every component of @xmath14 is parallel to a component of either @xmath298 or @xmath301 . the result is thus an immediate application of theorem [ t : incompressibleboundary ] . [ c : topminheegaard ] let @xmath0 be a closed topologically minimal surface in an irreducible 3-manifold , @xmath5 . then either 1 . @xmath5 contains a non - boundary parallel , incompressible surface , 2 . @xmath0 is a heegaard surface in @xmath5 , 3 . @xmath0 is contained in a ball , or 4 . @xmath0 is isotopic into a neighborhood of @xmath14 . in the next section we conjecture that the third possibility does not happen . in particular , if @xmath5 is a non - haken 3-manifold then it would follow that every topologically minimal surface in @xmath5 is a heegaard surface . let @xmath23 , @xmath24 , @xmath298 , and @xmath301 be as in theorem [ t : incompressibleboundary ] . suppose first some component of @xmath309 is a sphere . by the irreducibility of @xmath5 , this sphere bounds a ball . if the ball contains @xmath0 , then the result follows . otherwise , we may remove each such sphere component from @xmath309 . if the resulting surfaces are boundary parallel , then either @xmath0 is contained in a neighborhood of some boundary component of @xmath5 , or @xmath0 is a heegaard splitting of @xmath5 . if some component of @xmath309 is not boundary parallel then by theorem [ t : incompressibleboundary ] it is incompressible , and the result follows . [ c : join ] suppose @xmath2 and @xmath312 are disjoint surfaces in an irreducible 3-manifold @xmath5 , and @xmath313 is topologically minimal . then @xmath314 is the join of @xmath315 and @xmath316 . let @xmath317 . let @xmath23 , @xmath24 , @xmath298 , and @xmath301 be as in theorem [ t : incompressibleboundary ] . by theorem [ t : incompressibleboundary ] the surfaces @xmath298 and @xmath301 are incompressible in @xmath5 . if @xmath59 is a compression for @xmath2 then , as @xmath298 and @xmath301 are incompressible , we may isotope @xmath59 so that it is disjoint from both of these surfaces . it follows that @xmath59 is entirely contained in the component of @xmath302 that contains @xmath2 . but the surfaces @xmath2 and @xmath312 lie in different components of @xmath302 . thus , @xmath59 must be disjoint from the surface @xmath312 . hence , any compression for @xmath2 is isotopic to a compression for @xmath313 . we conclude there is a one - to - one correspondence between the vertices of @xmath6 and the vertices of @xmath318 . as every compression for @xmath2 will be disjoint from every compression for @xmath312 , we conclude that @xmath6 is the join of @xmath315 and @xmath316 . [ c : indexsum ] suppose @xmath2 and @xmath312 are disjoint separating surfaces in an irreducible 3-manifold @xmath5 , and @xmath313 is topologically minimal . then @xmath2 and @xmath312 are topologically minimal and @xmath319 note that the hypothesis that @xmath313 is topologically minimal is extremely important . for example , let @xmath2 and @xmath312 be parallel surfaces in @xmath5 that each have topological index one . then all of the compressing disks for @xmath317 are on the same side " of @xmath0 . hence , by mccullough s result @xcite , @xmath320 is contractible . thus @xmath0 does not have topological index two , as one might expect . we first show that @xmath2 and @xmath312 are topologically minimal . if not , then @xmath315 ( say ) is non - empty and contractible . but the join of a contractible space with any other space is also contractible . it thus follows from corollary [ c : join ] that @xmath313 is not topologically minimal . if either @xmath2 or @xmath312 has topological index 0 then the result is immediate . we assume , then , that the topological index of @xmath2 is @xmath95 and the topological index of @xmath312 is @xmath321 . by definition , @xmath98 is the smallest @xmath106 such that @xmath322 , and @xmath323 is the smallest @xmath47 such that @xmath324 . our goal is to show that @xmath325 is the smallest @xmath326 such that @xmath327 . by corollary [ c : join ] , this is equivalent to showing that @xmath325 is the smallest @xmath326 such that @xmath328 . when @xmath329 then @xmath330 . suppose @xmath2 seperates @xmath5 into @xmath23 and @xmath24 . let @xmath331 and @xmath332 denote the subsets of @xmath315 spanned by the compressions that lie in @xmath23 and @xmath24 , respectively . by an argument identical to the one given by mccullough in @xcite , @xmath331 and @xmath332 are contractible . if we contract these to points @xmath333 and @xmath334 , then the remaining 1-simplices of @xmath315 join these two points . the fundamental group @xmath335 is generated by these 1-simplices . the remaining 2-simplices have become bigons that run once over each of two 1-simplices . hence , each such 2-simplex gives rise to a relation in @xmath335 that kills one generator . it follows that @xmath335 is free , and hence the non - triviality of @xmath335 implies @xmath336 is also non - trivial . similarly , if @xmath337 we conclude @xmath338 is non - trivial . for @xmath339 the non - triviality of @xmath340 follows from the hurewicz theorem . by lemma 2.1 from @xcite : @xmath341 in particular , it follows from the fact that @xmath98 is the smallest @xmath106 such that @xmath342 is non - trivial , and @xmath323 is the smallest @xmath47 such that @xmath343 is non - trivial , that @xmath325 is the smallest @xmath326 such that @xmath344 is non - trivial . as an immediate corollary we obtain : [ c : indexncomponents ] if the topological index of @xmath0 is @xmath1 , then the sum of the indices of the components of @xmath0 is exactly @xmath1 . combining theorem [ t : maintheorem ] with corollary [ c : indexncomponents ] implies : [ c : componentsintersection ] let @xmath2 be a properly embedded , incompressible surface in an irreducible 3-manifold @xmath5 . let @xmath0 be a properly embedded surface in @xmath5 with topological index @xmath1 . then @xmath0 may be isotoped so that 1 . @xmath0 meets @xmath2 in @xmath91 saddles , for some @xmath92 , and 2 . the sum of the topological indices of the components of @xmath212 , plus @xmath91 , is at most @xmath1 . when @xmath0 is a heegaard surface whose topological index is one , this result says that @xmath2 cuts @xmath0 up into incompressible pieces , along with at most one index one piece . versions of this result were obtained by schultens for graph manifolds @xcite , and the author , sedgwick , and schleimer for more general haken manifolds @xcite . note also the similarity to the classification of _ almost normal _ surfaces given by rubinstein . such surfaces are cut up by the 2-skeleton of a triangulation into triangles and quadrilaterals , and exactly one special " piece . rubinstein @xcite and stocking @xcite proved that topological index 1 surfaces can always be isotoped to be almost normal . we believe the analogy is not a coincidence ; a relative ( with respect to the 1-skeleton of a triangulation ) version of theorem [ c : componentsintersection ] should recover the rubinstein - stocking result , and generalize it to arbitrary topological index . in any new theory , the questions raised are as important as the new results . here we compile a list of questions and conjectures that we hope will stimulate further research on topologically minimal surfaces . in @xcite we produce a fairly generic family of 3-manifolds @xmath345 such that @xmath346 contains a heegaard surface @xmath347 whose topological index is precisely @xmath1 . the manifold @xmath346 is the @xmath1-fold cover of @xmath348 , and the surface @xmath347 is the lift of @xmath349 . suppose @xmath5 contains unstabilized heegaard surfaces @xmath2 and @xmath312 that do not have topological index 1 . suppose further that the minimal genus common stabilization of @xmath2 and @xmath312 does not have topological index 2 . then @xmath5 contains a surface that has topological index @xmath351 . [ cnj : product ] let @xmath2 be a surface of positive genus . then the only connected , topologically minimal surfaces in @xmath354 are a single copy of @xmath2 and two copies of @xmath2 connected by an unknotted tube . by the argument given in the proof of corollary [ c : topminheegaard ] , any topologically minimal surface in @xmath354 would be a heegaard surface or would be contained in a ball . the only incompressible ( i.e. index 0 ) surface in @xmath354 is a copy of @xmath2 . by @xcite , the only strongly irreducible ( i.e. index 1 ) heegaard surface is two copies of @xmath2 connected by an unknotted tube . so , if conjecture [ cnj : s3 ] is true , then conjecture [ cnj : product ] is equivalent to the assertion that @xmath354 contains no topologically minimal surfaces whose index is larger than one . rubinstein and scharlemann have shown @xcite that corollary [ c : essentialintersection ] holds when @xmath0 and @xmath2 both have topological index 1 . this was instrumental in their proof that there is an upper bound on the smallest genus of a common stabilization of heegaard surfaces @xmath2 and @xmath312 , in terms of the genera of @xmath2 and @xmath312 . the index 0 case was proved by freedman , hass and scott @xcite , and the index 1 case by pitts and rubinstein @xcite . if true , it would indicate that topologically minimal surfaces are truly special . one would not expect , for example , a random " surface in a 3-manifold to be isotopic to a minimal surface . [ generalized hempel distance ] for each surface @xmath0 there is a natural map of @xmath6 into @xmath356 , its curve complex , where the image of a compression @xmath17 is @xmath85 . by @xcite , @xmath356 has the homotopy type of a wedge of spheres . it follows that for low values of @xmath1 ( in relation to the genus of @xmath0 ) , each map @xmath357 can be extended to a map @xmath358 . if we make all choices so that the number @xmath359 of @xmath1-dimensional simplices in @xmath360 is minimal , then we get an interesting invariant when @xmath361 is not homotopic to a point in @xmath6 . when @xmath0 is a heegaard surface that has topological index 1 , hempel called the invariant @xmath362 the distance of @xmath0 @xcite . many interesting results have been obtained about hempel s distance . what can be said about the invariant @xmath359 for larger values of @xmath1 ? j. pitts and j. h. rubinstein . applications of minimax to minimal surfaces and the topology of 3-manifolds . in _ miniconference on geometry and partial differential equations , 2 ( canberra 1986)_. proc . centre math . austral . univ . , * 12 * , austral . univ . , canberra , 1987 .

this paper studies three types of functions arising separately in the analysis of algorithms that we analyze exactly using similar mellin transform techniques . the first is the solution to a multidimensional divide - and - conquer ( mdc ) recurrence that arises when solving problems on points in-dimensional space . the second involves weighted digital sums . write in its binary representation and set . we analyze the average . the third is a different variant of weighted digital sums . write as with and set . we analyze the average . we show that both the mdc functions and ( with ) have solutions of the form where are constants and s are periodic functions with period one ( given by absolutely convergent fourier series ) . we also show that has a solution of the form where is a constant , and s are again periodic functions with period one ( given by absolutely convergent fourier series ) .

1003.0150

in this paper we use mellin transform techniques to analyze three types of functions arising separately in the analysis of algorithms : multidimensional divide - and - conquer and two different types of weighted digital sums . + the multidimensional divide - and - conquer ( mdc ) recurrence first appeared in the description of the running time of algorithms for finding maximal points in multidimensional space . previous analyses by monier @xcite gave only first order asymptotic , showing that the running time for the @xmath0-dimensional version of the problem is ( @xmath19 ) @xmath20 for some constant @xmath21 . we will extend the mellin transform techniques for solving divide - and - conquer problems originally developed in @xcite ( see @xcite for a review of more recent innovations ) to derive _ exact _ solutions , which will be in the form of @xmath22 where @xmath12 are constants and @xmath13 s are periodic functions with period one given by absolutely convergent fourier series . + the second type of function we study is a generalization of weighted digital sums ( wds ) . start by representing integer @xmath1 in binary as @xmath2 . define @xmath23 i.e. , _ weight _ the @xmath24 digit by its location in the representation . one can also view this as @xmath25 times @xmath26 , which is analogous to the _ derivative _ of the binary representation of @xmath1 . this sum arises naturally in the analysis of binomial queues where brown @xcite gave upper and lower bounds @xmath27 generalizing @xmath28 allows the `` weights '' to be any polynomial of @xmath29 . set @xmath30 and @xmath31 , define @xmath32 where @xmath33 is the @xmath34 rising factorial of @xmath29 . the function @xmath35 is not smooth ( see figure [ fig : smn - second - asymptotic ] ) . we will instead analyze its _ average _ @xmath36 we will show that , surprisingly , @xmath9 has an exact formula , which is in exactly the same form ( [ eq : tsmn - closed - form ] ) derived above for the mdc problem ( with @xmath10 and different constants ) . + the third type of function we study is another wds variant . its simplest form arises when analyzing the worst - case running time of bottom - up mergesort . but , any mergesort uses exactly @xmath37 merges , so the running time derived with cost @xmath38 is exactly @xmath37 more than the real worst - case running time . ] that the worst - case running time to merge two sorted lists of sizes @xmath39 and @xmath40 into one sorted list is @xmath41 . define @xmath42 to be the worst - case running time of bottom - up mergesort with @xmath1 elements . bottom - up mergesort essentially splits a list of @xmath1 items into two sublists , sorts each recursively , and then merges them back together . if @xmath1 is a power of @xmath25 , then it splits the list into two even parts . if @xmath1 is not a power of @xmath25 though , i.e. , @xmath43 with @xmath44 , then the algorithm splits the items into one list of size @xmath45 , one list of size @xmath46 . thus it is known that @xmath42 satisfies the recurrences : @xmath47 panny and prodinger @xcite derived an exact solution for @xmath42 containing a term @xmath48 , where @xmath49 , defined by a fourier series , is periodic with period one . however , the fourier series is only cesro summable . furthermore , @xmath49 is discontinuous at all dyadic points ( points of the form @xmath50 , where @xmath1 is integer , @xmath51 is non - negative integer ) , which are exactly the points of interest . in this paper we will decompose @xmath42 into two different types of wds and analyze the ( smoothed ) average of each part . we will then generalize the functions found and analyze the generalizations . starting with the binary representation of @xmath1 , ignore the @xmath52 bits and write @xmath1 as the sum of descending powers of @xmath25 , i.e. @xmath5 with @xmath6 . iterating the above recurrence for @xmath42 gives @xmath53 where @xmath28 is the wds of the first type defined previously . this motivates the introduction of another variant of wds : @xmath54 as with the wds of the first type , @xmath55 is not smooth enough to be analyzed directly ( see figure [ fig : wmn - first - asymptotic ] ) , so we instead study its average @xmath56 similar to the wds of the first type , this problem may be generalized by weighting the powers of @xmath25 with polynomial weights ; for wds of the second type the weights are of the form @xmath57 . the difference is due to ease of analysis . both types of span the space of polynomials and hence our study allows _ any _ polynomial weights . ] , i.e. by defining @xmath58 and , @xmath31 , @xmath59 and then introducing the average functions @xmath60 we will show that @xmath14 has an exact closed - form formula , which is in the form of @xmath61 where @xmath16 is a constant , @xmath17 and @xmath18 s are periodic functions with period one given by absolutely convergent fourier series . our approach to solving all three problems will be similar . we first use the mellin - perron formula and problem - specific facts to reduce the analysis to the calculation of an integral of the form @xmath62 for some problem specific kernel @xmath63 . we then identify the singularities and residues of @xmath63 and use the cauchy residue theorem in the limit to evaluate the integral . the main tools used in this paper are dirichlet generating functions and the mellin - perron formula . for more background see , @xcite , @xcite and @xcite . [ thm : mp ] let @xmath64 , @xmath65 be a sequence and @xmath66 lie in the half - plane of absolute convergence of @xmath67 . then for any @xmath68 , @xmath69 in particular , when @xmath70 and @xmath71 , @xmath72 we will analyze the mdc functions and the two types of wds by rewriting them to summations in the form of the left hand side of ( [ eq : mpf - m=1 ] ) . wds of the second type will also need a summation as in the left hand side of ( [ eq : mpf - m=2 ] ) . the mellin - perron formula will then enable us to evaluate the associated line integrals instead . the line integrals will be evaluated _ exactly _ via the cauchy residue theorem by considering integrations over some special contours . in the right hand side of ( [ eq : mpf - general - m ] ) , @xmath73 , the _ dirichlet generating function _ ( dgf ) of @xmath64 , is the only factor depending upon @xmath64 . the cauchy residue theorem relates the value of the line integral to the residues at the poles of the kernel in the line integral , thus understanding the locations and associated residues of the dgf s singularities will be essential to evaluating the line integral . define the _ backward difference function _ @xmath74 by @xmath75 for any function @xmath76 . the following lemma will be needed later in the analysis of wds . [ thm : key ] let @xmath76 be a function with @xmath77 and @xmath78 then @xmath79 where @xmath66 lies in the half - plane of absolute convergence of @xmath80 . note that @xmath81 and then apply ( [ eq : mpf - m=1 ] ) . a similar lemma , previously proven by flajolet and golin @xcite , will be needed to analyze the mdc functions . for any sequence @xmath82 , define its _ double difference sequence _ @xmath83 by @xmath84 for all @xmath1 . [ lem : solution - f - integral ] consider the recurrence @xmath85 with boundary conditions @xmath86 and @xmath87 . then @xmath88 and @xmath89 where @xmath90 lies in the half - plane of absolute convergence of @xmath91 . the riemann - zeta function is defined by @xmath92 when @xmath93 . since it will appear in the integral kernels in the analyses of the wds , we list some basic facts concerning the riemann - zeta function @xcite that we will need . first , @xmath94 can be analytically continued to be analytic in the whole complex plane with the exception of a simple pole at @xmath95 with residue @xmath96 . next , in @xcite , flajolet et . al . proved the identity @xmath97 by mimicking their proof , we prove the similar formula ( for completeness the proof is provided in appendix [ app : left - zero ] ) : @xmath98 when integrating @xmath94 the following asymptotic bounds @xcite will be useful : [ lem : zeta - bound ] if @xmath99 , where @xmath100 , the riemann - zeta function satisfies the bound @xmath101 where @xmath102 to understand the locations and associated residues of the integral kernels , we will need closed - form formulae of their associated dgfs . we start with some basic definitions . [ def : v - and - v2 ] express @xmath2 in its binary representation . + set @xmath103 to be the number of `` 1 ' 's in the binary representation of @xmath1 and @xmath104 to be the number of trailing `` 0 ' 's in the binary representation of @xmath1 . + for example , + if @xmath105 then @xmath106 and @xmath107 ; + if @xmath108 then @xmath109 and @xmath110 . we can now introduce two useful dgfs . [ def : vm - zm ] @xmath111 , denote the dgfs of @xmath112 and @xmath113 by @xmath114 the analysis of these dgfs will require the following facts . [ lem : property - v - and - v_2 ] let @xmath1 be a positive integer . then 1 . @xmath115 and @xmath116 ; 2 . @xmath117 ; 3 . if @xmath1 is odd , @xmath110 ; 4 . @xmath118 . if @xmath2 then @xmath119 and @xmath120 , so @xmath115 and @xmath116 . if the binary representation of @xmath1 has @xmath29 trailing `` @xmath52 ' 's , then the binary representation of @xmath121 will have @xmath122 trailing `` @xmath52 ' 's . this proves @xmath117 . if @xmath1 is odd , the rightmost digit of the binary representation of @xmath1 must be @xmath96 , i.e. there is no trailing `` 0 '' in the representation . hence @xmath110 for odd integer @xmath1 . slightly rewriting @xmath1 as @xmath123 where @xmath124 shows that @xmath125 . therefore @xmath126 . from lemma [ lem : property - v - and - v_2 ] , it is straightforward to prove the following lemma , which includes formulae expressing some special dgfs in terms of @xmath94 and @xmath127 . for completeness , we provide its proof in appendix [ app : dgf ] . [ lem : dgf - closed ] for @xmath128 , @xmath129 the following dgfs have closed - form formulae in terms of @xmath94 : @xmath130 @xmath131 in all three problems , evaluating the line integrals of the kernels will reduce to summations of residues at poles regularly spaced along a vertical line . these summations will best be expressed as fourier series . to be useful , we will need to show that these fourier series converge absolutely . our major tools will be the following two lemmas . [ thm : bound - derivative ] let @xmath132 , @xmath133 , @xmath134 and @xmath135 be a complex function . if 1 . @xmath135 is analytic in @xmath136 and 2 . @xmath137 such that @xmath138 , @xmath139 , then , for every fixed integer @xmath140 , latexmath:[\[\forall \sigma\geq\sigma_0,\ , \forall |t|\geq t_0 , \quad from the cauchy integral formula , for all @xmath99 with @xmath142 and @xmath143 , @xmath144 where @xmath145 . hence @xmath146 for fixed @xmath147 and @xmath148 . before stating the next lemma , we clarify that the statement `` @xmath149 has a pole of order at most @xmath150 at @xmath151 '' , allows the possibility that @xmath149 is analytic at @xmath151 ( and might even have a zero there ) . [ thm : sum - residues ] let @xmath152 . @xmath153 , set @xmath154 . if 1 . @xmath155 , @xmath135 is analytic at @xmath156 , 2 . @xmath157 , such that for all integers positive integers @xmath158 @xmath159 ( where the constant in the big @xmath160 may depend upon @xmath147 ) 3 . @xmath161 , @xmath162 has a pole of order at most @xmath39 at @xmath156 ; + furthermore , the coefficients of the laurent series of @xmath162 are identical at each @xmath156 , 4 . @xmath163 has a pole of order at most @xmath40 at @xmath164 , then the sum of residues at @xmath156 can be written in the form @xmath165 where the @xmath166 s are constants and @xmath167 s are periodic functions with period one given by their fourier series @xmath168 . furthermore , all the fourier series @xmath167 are absolutely convergent . we first introduce a notation . when @xmath169 is clear from the context , @xmath170\ ] ] represents the laurent series @xmath171 . we start by stating the laurent series of each factor of @xmath172 at @xmath156 , where @xmath155 : @xmath173\\ f(s ) & = & \left[0 \,\big|\ , f(\theta_j ) , f'(\theta_j ) , \frac{1}{2!}f''(\theta_j ) , \frac{1}{3!}f^{(3)}(\theta_j),\cdots\right]\\ n^s & = & n^\sigma e^{2\pi ij\lg n}\left[0 \,\big|\ , 1 , \ln n , \frac{1}{2!}\ln^2 n , \frac{1}{3!}\ln^3 n , \cdots\right]\\ \frac{1}{s } & = & \left[0 \,\big|\ , \frac{1}{\theta_j } , -\frac{1}{(\theta_j)^2 } , \frac{1}{(\theta_j)^3},\cdots\right]\\ \frac{1}{s+1 } & = & \left[0 \,\big|\ , \frac{1}{\theta_j+1 } , -\frac{1}{(\theta_j+1)^2 } , \frac{1}{(\theta_j+1)^3},\cdots\right]\end{aligned}\ ] ] the residue of @xmath172 at @xmath156 is obtained by multiplying all these series together and extracting the coefficient of the term @xmath174 . the residue will therefore be the sum of terms , each term of the form @xmath175 where @xmath176 and @xmath177 . hence the sum of these residues , when sorted according to the variable @xmath178 , is @xmath179 where @xmath180 the relevant laurent series of @xmath172 at @xmath164 are : @xmath173\\ n^s & = & n^\sigma\left[0 \,\big|\ , 1 , \ln n , \frac{1}{2!}\ln^2 n , \frac{1}{3!}\ln^3 n , \cdots\right]\\ \frac{f(s)}{s(s+1 ) } & = & \left[-n_2 \,\big|\ , h_0 , h_1 , h_2 , h_3 , \cdots\right]\end{aligned}\ ] ] by multiplying all these series together and extracting the coefficient of the term @xmath181 , the residue at @xmath164 is found to be of the form @xmath182 the second summation in ( [ eq : residue - theta_0 ] ) combines with ( [ eq : residue - theta_j ] ) to give the second summation in ( [ eq : sum - residues ] ) and the first summation in ( [ eq : residue - theta_0 ] ) gives the first summation in ( [ eq : sum - residues ] ) . we now prove the absolute convergence of the fourier series . take @xmath183 . note that for @xmath184 , @xmath159 . thus @xmath185 since @xmath186 , the fourier series is absolutely convergent . to conclude , we note that as we only upper bound the order of poles but do not know their exact order , @xmath166 may be zero and the @xmath167 s may be constant functions , or even zero functions . _ multidimensional divide - and - conquer _ ( mdc ) was first introduced by bentley and shamos @xcite in the context of solving multidimensional computational geometry problems . the generic idea is to solve a problem on @xmath1 @xmath0-dimensional points by ( i ) first splitting the points into two almost equal subsets and solving the problem seperately on each subset , then ( ii ) taking all @xmath1 points , projecting them down to @xmath187 dimensional space and solving the problem on the projected set , and finally ( iii ) constructing a solution to the complete problem by intelligently combining the solutions to the 3 previously solved ones . the recursion bottoms out when the dimension @xmath188 , in which case a straightforward solution is given , or when @xmath189 , which has a trivial solution . the methodology can be applied to give good solutions for many problems , including the empirical cumulative distribution function ( ecdf ) problem , maxima , range searching , closest pair , and the all nearest neighbour problem . of particular interest to us is the _ all - points ecdf problem in @xmath190 _ ( ecdf-@xmath191 ) . for two points @xmath192 , @xmath193 , we say @xmath194 _ dominates _ @xmath195 if @xmath196 for all @xmath197 . given a set @xmath198 of @xmath1 points in @xmath190 , the _ rank _ of a point @xmath194 is the number of points in @xmath198 dominated by @xmath194 . the ecdf-@xmath191 problem is to compute the rank of each point in @xmath198 . when @xmath199 , a slight modification of bottom - up mergesort will solve ecdf-@xmath25 in @xmath200 time . monier @xcite proposed an mdc algorithm for solving ecdf-@xmath191 for larger @xmath191 , based on the description of bentley @xcite . monier analyzed the worst - case running time of this algorithm , @xmath201 , described by the following recurrence : @xmath202 by translation into a combinatorial path - counting problem he derived the first order asymptotic of @xmath201 . more specifically , he showed that , for fixed @xmath191 , @xmath203 we will derive exact solutions for the ecdf-@xmath191 running time using lemma [ lem : solution - f - integral ] from @xcite . to do so , we will have to slightly modify the case @xmath199 to have a more precise initial condition . in what follows we will denote @xmath201 by @xmath204 . the recurrences corresponding to ( [ eq : t_mdc_def ] ) will be : @xmath205 where @xmath206 to use lemma [ lem : solution - f - integral ] first requires a better understanding of the dgf of @xmath207 , which we denote by by @xmath208 . start by noting that , directly from the lemma , @xmath209 one can work out directly that @xmath210 while , for @xmath211 , @xmath212 . thus , @xmath213 for @xmath214 , @xmath215 hence @xmath216 iterating the above recurrence with initial condition ( [ eq : df2 ] ) yields @xmath217 from lemma [ lem : solution - f - integral ] , for @xmath218 , @xmath219 we note that flajolet and golin @xcite explicitly solve the @xmath199 boundary case : @xmath220 where , setting @xmath221 , @xmath222 we now evaluate the integral in ( [ eq : f_n^k - recurrence ] ) : @xmath223 fix some real @xmath224 and consider the counterclockwise rectangular contour @xmath225 , where ( see figure [ fig : contour_upsilon ] ) @xmath226 denote the kernel of the integral in ( [ eq : def - i_fk(s ) ] ) by @xmath227 : @xmath228 + note that @xmath229 . we now show that , for @xmath230 , @xmath231 . thus @xmath232 by the cauchy residue theorem , @xmath233 will be equal to @xmath1 times the sum of the residues at the poles inside @xmath234 as @xmath235 . the poles of @xmath227 inside @xmath234 are : 1 . a pole of order @xmath191 at @xmath236 ; 2 . poles of order @xmath237 at @xmath238 , where @xmath239 ; 3 . a simple pole at @xmath240 . to avoid poles of @xmath227 on @xmath234 , we only consider values of @xmath241 . now consider the horizontal paths @xmath242 . then @xmath243 for the leftmost path it is easy to see @xmath244 hence @xmath233 is @xmath1 times the sum of the residues at the poles of @xmath227 inside @xmath234 , taking @xmath245 . [ thm : mdc_equation ] for @xmath246 , @xmath247 where @xmath248 s are periodic functions with period one , which are given by absolutely convergent fourier series @xmath249 whose coefficients @xmath250 can be determined explicitly . in particular , the average value of @xmath251 is @xmath252 furthermore , if @xmath191 is even , @xmath253 ; if @xmath191 is odd , @xmath254 . we proceed by induction on @xmath191 . as previously mentioned , for @xmath199 this theorem was already proved by flajolet and golin @xcite . now assume that ( [ eq : fnk - general - exact ] ) is true for @xmath255 . the residue of @xmath256 at @xmath240 is @xmath257 we can now apply lemma [ thm : sum - residues ] , by taking @xmath258 , @xmath259 ( its laurent series coefficients at each @xmath260 are identical ) and @xmath261 . since @xmath262 when @xmath263 , we may take @xmath264 . the order of poles of @xmath162 at @xmath260 is @xmath265 and the order of pole of @xmath163 at @xmath266 is @xmath96 . the sum of residues at @xmath260 , where @xmath267 , is given by @xmath268 where @xmath269 s are periodic functions with period one which are given by absolutely convergent fourier series . @xmath270 can be explicitly calculated to be @xmath271 . hence by ( [ eq : f_n^k - recurrence ] ) , @xmath272\\ & = & \frac{n \lg^{k_0 - 2 } n}{(k_0 - 2 ) ! } + \sum_{m=0}^{k_0 - 3 } \left(n \lg^m n\right ) a_m^{k_0 - 1}(\lg n ) + c_{k_0 - 1}\\ & & \qquad + \frac{n\lg^{k_0 - 1 } n}{(k_0 - 1 ) ! } + \sum_{m=0}^{k_0 - 2}\left(n \lg^m n\right ) b_m^{k_0}(\lg n ) + ( -1)^{k_0}\\ & = & \frac{n\lg^{k_0 - 1 } n}{(k_0 - 1 ) ! } + \left(\frac{1}{(k_0 - 2 ) ! } + b_{k_0 - 2}^{k_0}(\lg n)\right)n\lg^{k_0 - 2 } n \\ & & \qquad + \sum_{m=0}^{k_0 - 3}\left(a_m^{k_0 - 1}(\lg n ) + b_m^{k_0}(\lg n)\right)n\lg^m n + c_{k_0 - 1 } + ( -1)^{k_0}.\end{aligned}\ ] ] letting @xmath273 and @xmath274 for @xmath275 proves ( [ eq : fnk - general - exact ] ) . the average value of @xmath251 is found by expressing all the laurent series ( in the proof of lemma [ thm : sum - residues ] ) explicitly . finally , since @xmath276 , @xmath277 alternates between being even and odd with @xmath278 . we now analyze @xmath279 as defined in ( [ eq : def - smn ] ) and ( [ eq : def - tsmn ] ) . by lemma [ thm : key ] , this reduces to evaluating @xmath280 we start by deriving a closed form for @xmath281 recall that @xmath3 . observe that if @xmath2 , then @xmath282 in particular , when @xmath128 , the weight @xmath283 for the rightmost digit @xmath284 is always zero , so @xmath285 next , observe that @xmath286 and for @xmath287 , @xmath288 these facts lead to : [ lem : dgf - nabla - sm - exact ] @xmath289 the proof is by induction on @xmath290 . when @xmath291 , by ( [ eq : sm(2n+1)-recurrence ] ) and ( [ eq : s1(2n)-recurrence ] ) , we get @xmath292 hence @xmath293 then @xmath294 and the lemma is proved for @xmath291 . now assume the lemma is true for @xmath295 . iterating ( [ eq : sm(2n)-recurrence ] ) gives @xmath296 where @xmath297 is the @xmath298 falling factorial of @xmath191 . appling ( [ eq : sm(2n+1)-recurrence ] ) gives @xmath299 substituting the above two formulae into ( [ eq : def - ams ] ) yields @xmath300\\ & = & \frac{1}{2^{s-1}}a_k(s ) + \frac{k!\zeta(s)}{2^{s-1}}\left(1 + \sum_{i=1}^{k-1 } \frac{2^{(k - i-1)(s-1)}}{(2^{s-1}-1)^{k - i}}\right)\\ & = & \frac{1}{2^{s-1}}a_k(s ) + \frac{k!\zeta(s)}{2^{s-1}}\left(1 + \frac{1}{2^{s-1 } } \sum_{i=1}^{k-1}\left(\frac{2^{s-1}}{2^{s-1}-1}\right)^{k - i}\right)\\ & = & \frac{1}{2^{s-1}}a_k(s ) + \frac{k!\zeta(s)}{2^{s-1}}\left(\frac{2^{s-1}}{2^{s-1}-1}\right)^{k-1}\end{aligned}\ ] ] and hence @xmath301 substituting the result of lemma [ lem : dgf - nabla - sm - exact ] into the integral of ( [ eq : tsmn - integral - raw - dgf ] ) gives @xmath302 fix some real @xmath224 and consider the counterclockwise rectangular contour @xmath303 , where ( see figure [ fig : contour_gamma1 ] ) @xmath304 denote the kernel of the integral in ( [ eq : tsmn - integral - exact - dgf ] ) by @xmath305 : @xmath306 note that @xmath307 . as in the mdc case , we now show that @xmath308 for @xmath230 . thus @xmath309 hence by the cauchy residue theorem , @xmath9 will be equal to @xmath310 times the sum of the residues at the poles inside @xmath311 as @xmath235 . we know that @xmath94 has a simple pole at @xmath95 . the poles of @xmath305 inside @xmath311 are : 1 . a pole of order @xmath312 at @xmath95 ; 2 . poles of order @xmath290 at @xmath313 , where @xmath314 ; 3 . a simple pole at @xmath236 . to avoid poles of @xmath305 on @xmath311 , we again only consider values of @xmath315 . to show that @xmath316 for @xmath230 as @xmath235 , we need the following two lemmas . [ thm : bound_top_bottom_zeta_general ] consider integral @xmath317 where @xmath318 . furthermore , suppose that for @xmath99 with @xmath319 , @xmath320 . then , both as @xmath235 and @xmath321 , @xmath322 . along the path of the integral , @xmath323 . lemma [ lem : zeta - bound ] gives the bound @xmath324 together with the given fact that @xmath325 , @xmath326 as @xmath235 . [ thm : bound_left_zeta_general ] suppose @xmath327 for some real sequence @xmath328 and positive integer sequence @xmath329 . if this series is uniformly convergent for @xmath330 then @xmath331 if the series is uniformly convergent for @xmath332 then @xmath333 for the first integral , note that @xmath334 the first equality is the definition of @xmath335 the second follows from the uniform convergence of the series and the last equality follows from ( [ eq : left=0-m=1 ] ) . the second integral is evaluated similarly , using ( [ eq : left=0-m=2 ] ) in place of ( [ eq : left=0-m=1 ] ) . to evaluate the integrals along @xmath336 and @xmath337 , note that @xmath338 is bounded as @xmath339 and @xmath340 . thus , by lemma [ thm : bound_top_bottom_zeta_general ] , as @xmath245 , @xmath341 to evaluate the integral along @xmath342 , note that @xmath343 along @xmath342 , so we may write @xmath344 the series is both absolutely convergent and uniformly convergent on @xmath345 , so we may write ( see @xcite ) @xmath346 for some @xmath347 , where this new series is again uniformly convergent on @xmath345 . by lemma [ thm : bound_left_zeta_general ] , @xmath348 we have successfully shown that the integrals along @xmath349 @xmath342 and @xmath337 vanish as @xmath245 , and hence @xmath9 is @xmath310 times the sum of the residues at the poles of @xmath305 inside @xmath311 , after taking @xmath245 . [ thm : tsn - general ] for @xmath128 , @xmath350 where @xmath351 s are periodic functions with period one , which are given by absolutely convergent fourier series @xmath352 whose coefficients @xmath353 can be determined explicitly . in particular , the average value of @xmath354 is @xmath355\approx \frac{m^2}{4}-0.915648 m.\ ] ] as shown , @xmath9 is @xmath310 times the sum of residues at the poles of @xmath305 inside @xmath311 as @xmath235 . the residue of @xmath305 at @xmath236 is @xmath356 by lemma [ lem : zeta - bound ] , we have the bound @xmath357 when @xmath358 for some sufficiently small @xmath148 . by lemma [ thm : bound - derivative ] , @xmath359 for any fixed positive integer @xmath147 . in lemma [ thm : sum - residues ] , take @xmath360 , @xmath361 ( its laurent series coefficients at each @xmath362 are identical ) and @xmath363 . from last paragraph we can take @xmath364 and @xmath365 . the order of poles of @xmath162 at @xmath362 is @xmath290 , and the order of pole of @xmath163 at @xmath366 is @xmath96 . the sum of residues at @xmath362 , where @xmath267 , is given by @xmath367 where @xmath368 s are periodic functions with period one which are given by absolutely convergent fourier series . @xmath369 can be explicitly calculated to be @xmath370 . hence @xmath371\\ & = & \frac{1}{2}n\lg^m n + \sum_{d=0}^{m-1 } \left(n\lg^d n\right)m!\overline{f}_{m , d}(\lg n ) + ( -1)^{m+1}m!.\end{aligned}\ ] ] letting @xmath372 for @xmath373 , proves ( [ eq : tmn - general - exact ] ) . the average value of @xmath354 is found by expressing all the laurent series ( in the proof of lemma [ thm : sum - residues ] ) explicitly . we now analyze @xmath374 as defined by ( [ eq : def - wn ] ) and ( [ eq : def - twn ] ) . the analysis will be extended to @xmath14 for @xmath375 in the next section . the general methodology used to analyze @xmath376 is the same as in the previous sections ; use lemma [ thm : key ] to rewrite @xmath377 the main difficulty that will be encountered is that the dgf here will not be `` nice '' enough to permit integrating the kernel directly . we will have to split the dgf into two parts , using the @xmath70 case of ( [ eq : mpf - general - m ] ) to evaluate the first part and the @xmath71 case to evaluate the second part . set @xmath378 to be the dgf of @xmath379 . we start by deriving , for all @xmath128 , a formula for @xmath380 in terms of dgfs @xmath127 and @xmath381 introduced in definition [ def : vm - zm ] . we will then analyze the case @xmath291 in this section , and leave the cases @xmath287 to the next section . [ lem : dgf - nabla - wm - exact ] @xmath382 observe that if @xmath1 is expressed as @xmath5 with @xmath6 , then @xmath383 recalling from lemma [ lem : property - v - and - v_2 ] that @xmath118 and @xmath116 gives @xmath384 then , ( [ eq : dgf - odd - v ] ) in lemma [ lem : dgf - closed ] permits writing @xmath385 solving for @xmath380 proves the lemma . for @xmath291 , applying ( [ eq : dgf - v2 ] ) from lemma [ lem : dgf - closed ] to lemma [ lem : dgf - nabla - wm - exact ] gives @xmath386 substituting this into ( [ eq : twn - integral - raw - dgf ] ) yields @xmath387 the second integral can be evaluated exactly by the method used in section [ subsect : wds1-evaluation ] . evaluating the first integral requires more work . historically , @xmath388 was one of the first digital functions to be analyzed using the mellin transform techniques . the original analysis in 1975 by delange @xcite used a combinatorial decomposition of the binary representations of integers to directly derive an exact fourier series formula for @xmath389 . in 1994 , flajolet et . @xcite reproved delange s result using the mellin transform techniques . however , @xmath390 , the dgf of @xmath388 , does not seem to have been explicitly studied before hwang s analysis @xcite in 1998 . first , denote @xmath391 by @xmath392 standard algebraic manipulations , e.g. in @xcite , let us rewrite a summation of this form as an integral : @xmath393 where @xmath394 hwang @xcite derived the following formula of @xmath390 , revealing its singularities in @xmath395 : @xmath396 substituting ( [ eq : dgf - v(s)-hwang ] ) into ( [ eq : b1s - closed ] ) yields @xmath397 from the integral form of @xmath398 , we know that it is analytic in @xmath395 . hence , ( [ eq : b1s ] ) ( together with the fact that @xmath94 has no zero on the line @xmath399 ) shows that at @xmath400 , @xmath401 possesses simple poles . depending upon the values of @xmath402 and @xmath403 , @xmath401 may either possess simple poles at @xmath404 or be analytic at @xmath404 . these are all possible poles of @xmath401 inside @xmath311 which we defined in ( [ eq : gamma ] ) . using hwang s representation _ would _ yield a closed - form formula for @xmath376 by considering contour @xmath311 . unfortunately , the residues appearing in the resulting fourier coefficients would be expressed in terms of the value of @xmath398 at various poles , something which is not well understood . in the next subsection , we will show how to use the higher order version of the mellin - perron formula to sidestep this issue and express the fourier coefficients in terms of the riemann - zeta function . we now see how to manipulate the first integral in ( [ eq : tw - integral - exact ] ) to yield a formula in terms of values of the riemann - zeta function . the general approach is to note that @xmath390 is the dgf of @xmath405 , so the first integral in ( [ eq : tw - integral - exact ] ) , when transformed from integral back to summation by ( [ eq : mpf - m=1 ] ) , is a double summation of @xmath405 . a double summation of @xmath405 is also a triple summation of @xmath406 , and we can write a closed - form formula for the dgf of @xmath406 in terms of @xmath94 . equation ( [ eq : mpf - m=2 ] ) then provides an exact formula of the triple summation of @xmath406 , and we can evaluate the first integral in ( [ eq : tw - integral - exact ] ) . we now present the details . define @xmath407 algebraic manipulations permit writing @xmath408 in two different ways : @xmath409 and @xmath410.\ ] ] applying ( [ eq : mpf - m=1 ] ) to ( [ eq : ttvn - double - summation ] ) , yields @xmath411 where the right side is exactly the first integral in ( [ eq : tw - integral - exact ] ) . applying ( [ eq : mpf - m=2 ] ) and ( [ eq : mpf - m=1 ] ) to ( [ eq : ttvn - triple - summation ] ) gives the alternate expression @xmath412 setting ( [ eq : ttvn - double - summation - integral ] ) equal to ( [ eq : ttvn - triple - summation - integral ] ) and using the closed - form formula for @xmath413 in lemma [ lem : dgf - closed ] gives @xmath414 substituting the above equality into ( [ eq : tw - integral - exact ] ) yields a `` nicer '' integral representation for @xmath376 . @xmath415 the three integrals in ( [ eq : tw - integral - exact - zeta - only ] ) can be evaluated almost exactly as in section [ subsect : wds1-evaluation ] . that is , for the first integral consider contour @xmath416 , where @xmath417 for the second and the third integrals consider contour @xmath311 defined in ( [ eq : gamma ] ) . next , prove that the integrals along the left , top and bottom paths tend to zero ( using lemma [ thm : bound_top_bottom_zeta_general ] and lemma [ thm : bound_left_zeta_general ] ) . finally , evaluate the sum of residues at the poles inside @xmath418 or @xmath311 . since these are almost exactly the same as in section [ subsect : wds1-evaluation ] , we leave out the details , only stating the results . see figure [ fig : contour_wds2_higher ] for the contours . the poles of the kernel of the the first integral inside @xmath418 are a double pole at @xmath236 and simple poles at @xmath260 ( where @xmath314 ) and @xmath240 . by summing the residues at all these poles , the first integral evaluates to @xmath419 where @xmath420 the poles of the kernel of the second integral inside @xmath311 are a double pole at @xmath236 and simple poles at @xmath260 ( where @xmath314 ) . by summing the residues at all these poles , the second integral evaluates to @xmath421 where @xmath422 the poles of the kernel of the third integral inside @xmath311 are a double pole at @xmath95 , simple poles at @xmath362 ( where @xmath314 ) , a double pole at @xmath236 and simple poles at @xmath260 ( where @xmath314 ) . by summing the residues at all these poles , the third integral evaluates to @xmath423 where @xmath424 and @xmath425 combining the three integrals above yields : [ thm : twn - general ] @xmath426 where @xmath427 and @xmath428 are two absolutely convergent fourier series , whose coefficients are given by @xmath429 the average value of @xmath427 is @xmath430 substituting ( [ eq : tw - first - integral - exact ] ) , ( [ eq : tw - second - integral - exact ] ) and ( [ eq : tw - third - integral - exact ] ) into ( [ eq : tw - integral - exact - zeta - only ] ) yields @xmath431-\frac{1}{4}\lg n + \left[h_2(\lg n)-h_{3,1}(\lg n)+\frac{1}{4}\right].\ ] ] setting @xmath432 and @xmath433 gives ( [ eq : twn - general ] ) and the fourier series . lemma [ lem : zeta - bound ] gives @xmath434 hence , as @xmath435 the terms in ( [ eq : w1f ] ) , ( [ eq : w2f ] ) and ( [ eq : w31f ] ) are @xmath436 and the terms in ( [ eq : w32f ] ) are @xmath437 , implying the absolute convergences of @xmath438 , @xmath439 , @xmath440 and @xmath441 , and thus @xmath427 and @xmath428 . we now analyze @xmath442 as defined by ( [ eq : def - wmn ] ) and ( [ eq : def - twmn ] ) . again , by lemma [ thm : key ] , @xmath443 where @xmath380 is the dgf of @xmath379 as defined in ( [ eq : def - bms ] ) . as before , we will integrate along contour @xmath311 we defined in ( [ eq : gamma ] ) , and prove that the integrals along the top , bottom and left contours vanish as @xmath235 , while that on the right contour equals ( [ eq : twmn - integral - raw - dgf ] ) and then apply the cauchy residue theorem . the dgf @xmath380 for @xmath444 is much more complicated than the dgfs previously encountered in this paper . we will therefore have to introduce new techniques to study it . we saw from ( [ eq : b1s ] ) that the order of the poles of @xmath445 at @xmath95 and @xmath362 were all at most @xmath96 . by lemma [ thm : sum - residues ] , this implied that `` coefficient '' of the first order term of @xmath376 , i.e. the @xmath1 term in ( [ eq : twn - general ] ) , is the fourier series @xmath446 . analogously , for all @xmath375 we will prove that the orders of poles of @xmath447 at @xmath95 and @xmath362 are all at most @xmath96 . this will again imply that for @xmath444 , @xmath14 has a `` periodic first - order coefficient '' . to start , we will need the following semi - recursive formula of @xmath380 . [ lem : bms - funeq ] for @xmath375 , @xmath380 satisfies @xmath448 where @xmath449 is defined as @xmath450.\ ] ] lemma [ lem : dgf - nabla - wm - exact ] gives @xmath451 we now derive two seperate functional equations for @xmath127 and @xmath381 and combine them to yield ( [ eq : bms - key3 ] ) . recalling from lemma [ lem : property - v - and - v_2 ] that @xmath452 and @xmath453 gives @xmath454 solving for @xmath127 yields @xmath455 next , from lemma [ lem : property - v - and - v_2 ] it is easy to show @xmath456 and @xmath457 . also , using ( [ eq : dgf - odd - v ] ) in lemma [ lem : dgf - closed ] , gives @xmath458 since @xmath459 , this solves to @xmath460 substituting ( [ eq : vms - funeq ] ) and ( [ eq : zms - funeq ] ) into ( [ eq : bms - closed - recall ] ) yields @xmath461 finally , using ( [ eq : bms - closed - recall ] ) to simplify the internal terms in ( [ eq : bms - key2 ] ) yields ( [ eq : bms - key3 ] ) . the next lemma gives a `` closed - form '' formula for @xmath380 , in terms of @xmath462 , previously defined in ( [ eq : def - irs ] ) , and @xmath94 . @xmath463 where @xmath464 and @xmath465 are two polynomials , with @xmath466 , @xmath467 for @xmath287 and @xmath468 for @xmath469 . we prove by induction , using lemma [ lem : bms - funeq ] . first , note that ( [ eq : b1s ] ) is just the special case of this lemma for @xmath291 . it also gives @xmath466 . now assume the lemma is true for @xmath470 . then , @xmath471 where @xmath472 and @xmath473 s are polynomials satisfying @xmath474 and @xmath475 for @xmath476 . now set @xmath477}{i^s}.\ ] ] from the definition of @xmath391 in ( [ eq : def - irs ] ) and some algebraic manipulation , @xmath478 grabner and hwang @xcite proved @xmath479 where @xmath480 are the stirling numbers of the second kind . noting that @xmath481 , @xmath449 can be rewritten as @xmath482 where @xmath483 and @xmath484 s are polynomials satisfying @xmath485 . this permits writing @xmath486 where @xmath487 and @xmath488 s are polynomials satisfying @xmath489 . substituting ( [ eq : bms - form - step1 ] ) and ( [ eq : bms - form - step2 ] ) back into ( [ eq : bms - key3 ] ) gives @xmath490 the lemma is proved for @xmath491 by setting @xmath492 , @xmath493 for @xmath476 and @xmath494 . we can now find the poles of @xmath380 inside @xmath311 . see figure [ fig : contour_gamma2 ] for locations . [ thm : bms - asym - behaviour ] for @xmath128 , the singularities of @xmath380 inside @xmath311 are + ( i ) poles of order at most @xmath96 at @xmath95 and @xmath362 ; and + ( ii ) poles of order @xmath290 at @xmath236 and @xmath260 . hence , the singularities of @xmath447 inside @xmath311 are + ( i ) poles of order at most @xmath96 at @xmath95 and @xmath362 ; + ( ii ) a pole of order @xmath495 at @xmath236 ; and + ( iii ) poles of order @xmath290 at @xmath260 . ( [ eq : bms - form ] ) permits us to identify the singularities by working through the various terms and recalling that @xmath462 is analytic when @xmath496 the recurrence relations @xmath467 with initial condition @xmath497 give @xmath498 for @xmath128 . hence at @xmath400 , @xmath499 has poles of order _ exactly _ @xmath290 , while @xmath94 is analytic ( but is not zero ) . at @xmath500 @xmath499 and @xmath94 are all analytic . at @xmath95 , @xmath499 is analytic , but @xmath94 has a simple pole . at @xmath400 , the order of poles of @xmath501 is _ at most _ @xmath502 . at @xmath503 , @xmath501 has poles of order at most @xmath96 ( due to the term @xmath504 ) . as in the previous problems , we must again first show that the integrals along the top , bottom and left contours vanish as @xmath235 . we need two basic observations . suppose @xmath505 , where @xmath506 is a polynomial and @xmath507 are non - negative integers . + + * * when @xmath508 , @xmath509 can be expressed as a power series of @xmath510 , and this series is absolutely and uniformly convergent on the line @xmath511 . + furthermore , if @xmath512 , i.e. the constant term of @xmath506 is zero , then the constant term of the power series is also zero . + * * @xmath509 is bounded along the line segment @xmath513 independently of @xmath46 . [ thm : bound_top_bottom_bms ] @xmath514 for @xmath515 , grabner and hwang @xcite proved that @xmath516 furthermore , lemma [ lem : zeta - bound ] gives @xmath517 by ( [ eq : bms - form ] ) and fact 2 , @xmath518 is bounded by @xmath519 along @xmath520 . hence @xmath521 as @xmath339 . [ thm : left=0-bms ] for any positive integer @xmath290 , @xmath522 grabner and hwang @xcite proved the bound @xmath523 for any @xmath524 . this upper bound allows us to use a theorem from hwang @xcite to prove hwang @xcite proved the following theorem : [ thm : left=0-general ] suppose @xmath525 for some nonnegative , real arithmetic function @xmath526 . if 1 . @xmath527 converges for @xmath528 , where @xmath529 , 2 . @xmath530 for some @xmath531 , then we have @xmath532 for all integers @xmath533 . grabner and hwang @xcite proved the bound @xmath523 for any @xmath524 , which enables us to use theorem [ thm : left=0-general ] to get @xmath534 for positive integers @xmath535 . ( [ eq : bms - form ] ) shows that @xmath380 can be expressed in the form of @xmath536 while @xmath537 . by fact 1 , when @xmath538 , @xmath539 and @xmath540 can be expressed as power series of @xmath510 , and the power series for @xmath540 have zero constant terms . hence , when @xmath538 , we may rewrite @xmath380 to be @xmath541 for some @xmath542 and @xmath543 . hence @xmath544 however , the power series @xmath545 and @xmath546 are uniformly convergent on @xmath547 , by fact 1 . this allows interchange of the integral sign and the summation signs . hence , @xmath548 can be expressed as a series , in which each term is either a constant multiplied by an integral in the form of ( [ eq : bound_left_irs_general ] ) , or a constant multiplied by an integral in the form of ( [ eq : left=0-m=1 ] ) . we can now state our final result . @xmath549 where @xmath16 is a constant , @xmath17 and @xmath18 s are periodic functions with period one given by absolutely convergent fourier series . consider the contour @xmath311 in figure [ fig : contour_gamma2 ] , taking @xmath235 . lemma [ thm : bound_top_bottom_bms ] and lemma [ thm : left=0-bms ] show that @xmath550 for @xmath230 . hence @xmath551 is the sum of residues at the poles of @xmath447 inside @xmath311 , by the cauchy residue theorem . by lemma [ lem : zeta - bound ] , we have the bound @xmath552 when @xmath553 for sufficiently small @xmath148 . grabner and hwang @xcite also proved that @xmath554 when @xmath553 for sufficiently small @xmath148 . hence by lemma [ thm : bound - derivative ] , @xmath555 and @xmath556 for any fixed positive integer @xmath147 . @xmath380 can be expressed in the form of ( [ eq : bms - form ] ) . knowing that each function in the form of @xmath557 will have a laurent series with identical coefficients at @xmath558 for any fixed @xmath559 , togather with the results from the last paragraph and corollary [ thm : bms - asym - behaviour ] , we use lemma [ thm : sum - residues ] when @xmath560 to derive @xmath561 and @xmath562 where @xmath17 and @xmath18 s are periodic functions with period one given by absolutely convergent fourier series . mellin transform techniques have previously been extensively used to analyze various divide - and - conquer algorithms and digital sums . a common theme in those analyses is the appearance of a ( usually second order ) periodic term , usually expressed in terms of a fourier series . this fourier series is the sum of residues of a complex function which has singularities regularly spaced along a vertical line . in this paper we pushed the technique further to derive exact analyses of the solution to multidimensional divide - and - conquer recurrences and various , more complicated , weighted digital sums . our closed form solutions had the properties that _ all _ terms were either polylogarithmic or @xmath1 times a polylogarithm , with all coefficients either being constant or a periodic function given by a fourier series . our analysis of the multidimensional divide - and - conquer recurrence was a straightforward extension of the use of mellin transform techniques for the analysis of simple divide - and - conquer recurrences . our analyses of weighted digital sums , though , required developing a better understanding of various dirichlet generating functions of differences of digital functions . 99 tom m. apostol , _ `` mathematical analysis '' _ , addison - wesley publishing company , 1974 . tom m. apostol , _ `` introduction to analytic number theory '' _ , springer - verlag new york , inc . , 1976 . joseph bak and donald j. newman , _ `` complex analysis '' _ , springer - verlag new york , inc . , 1996 . jon louis bentley , _ `` multidimensional divide - and - conquer '' _ acm 23(4 ) : 214 - 229 ( 1980 ) jon louis bentley and michael ian shamos , _ `` divide - and - conquer in multidimensional space '' _ , stoc 1976 : 220 - 230 mark r. brown , _ `` implementation and analysis of binomial queue algorithms '' _ , siam j. computing 7(3 ) : 298 - 319 ( 1978 ) j. coquet , _ `` power sums of digital sums '' _ , j. number theory 22 : 161 - 176 ( 1986 ) h. delange , _ `` sur la fonction sommatoire de la fonction `` somme des chiffres '' '' _ , enseignement math . 21(2 ) : 31 - 47 ( 1975 ) philippe flajolet , peter j. grabner , peter kirschenhofer , helmut prodinger and robert f. tichy , _ `` mellin transforms and asymptotics : digital sums '' _ , theoretical computer science 123(2 ) : 291 - 314 ( 1994 ) philippe flajolet and mordecai j. golin , _ `` mellin transforms and asymptotics : the mergesort recurrence '' _ , acta informatica 31(7 ) : 673 - 696 ( 1994 ) philippe flajolet and lyle ramshaw , _ `` a note on gray code and odd - even merge '' _ , siam journal on computing 9(1 ) : 142 - 158 ( 1980 ) philippe flajolet and robert sedgewick , _ `` analytic combinatorics '' _ , cambridge university press , 2008 . peter j. grabner and hsien - kuei hwang , _ `` digital sums and divide - and - conquer recurrences : fourier expansions and absolute convergence '' _ , constructive approximation 21(2 ) : 149 - 179 ( 2005 ) hsien - kuei hwang , _ `` asymptotics of divide - and - conquer recurrences : batcher s sorting algorithm and a minimum euclidean matching heuristic '' _ , algorithmica 22(4 ) : 529 - 546 ( 1998 ) peter kirschenhofer , helmut prodinger and robert f. tichy , _ `` ber die ziffernsumme natrlicher zahlen und verwandte probleme '' _ , in _ zahlentheoretische analysis _ , lecture notes in mathematics 1114 : 55 - 65 ( 1985 ) chun yu james lee , _ `` the analysis of multidimensional divide - and - conquer recurrences '' _ , master thesis , the hong kong university of science and technology , 2007 . benoit b. mandelbrot , _ `` the fractal geometry of nature '' _ , w. h. freeman , 1982 . louis monier , _ combinatorial solutions of multidimensional divide - and - conquer recurrences _ , journal of algorithms 1(1 ) : 60 - 74 ( 1980 ) wolfgang panny and helmut prodinger , _ `` bottom - up mergesort - a detailed analysis '' _ , algorithmica 14(4 ) : 340 - 354 ( 1995 ) marko r. riedel , _ `` applications of the mellin - perron formula in number theory '' _ , master thesis , university of toronto , 1996 . walter rudin , _ `` principles of mathematical analysis '' _ , mcgraw - hill , inc . , 1976 . elias m. stein and rami shakarchi , _ `` fourier analysis '' _ , princeton university press , 2003 . edward c. titchmarsh , _ `` the theory of the riemann zeta - function '' _ , oxford science publications , 1986 . george n. watson and edmund t. whittaker , _ `` a course of modern analysis '' _ , cambridge university press , 1963 . our main technique for solving the multidimensional divide - and - conquer recurrence is a generalization of lemma [ lem : solution - f - integral ] for basic divide - and - conquer recurrences , originally proved in @xcite by flajolet and golin . in order to make this paper self - contained , we provide the proof of that lemma here . 1 . a simple pole at @xmath95 . the residue at this pole is @xmath587 . 2 . a simple pole at @xmath236 . the residue at this pole is @xmath588 . 3 . a simple pole at @xmath240 . the residue at this pole is @xmath589 .

we use crossing statistics and its generalization to determine the anisotropic direction imposed on a stochastic fields indimension . this approach enables us to examine not only the rotational invariance of morphology but also we can determine the gaussianity of underlying stochastic field in various dimensions . theoretical prediction of up - crossing statistics ( crossing with positive slope at a given threshold of height fluctuation ) , , and generalized roughness function , , for correlation length ( ) and / with scaling exponent ( ) anisotropies are calculated . the strategy to examine the anisotropy nature and to determine its direction is as follows : we consider a set of normal axes , and sign them ( parallel ) and ( normal ) with respect to unknown anisotropic direction . then we determine and in both directions . the directional dependency of difference between computed results in mentioned directions are clarify . finally we systematically recognize the anisotropy direction at confidence interval using p - value approach . in order to distinguish between nature of anisotropies , after applying a typical method in determining the scaling exponents in both mentioned directions with respect to the recognized anisotropy direction using up - crossing statistics , the kind and the ratio of correlation length anisotropy are specified . our algorithm can be mounted with a simple software on various instruments for surface analysis , such as afm , stm and etc . * keywords * : crossing statistics , stochastic field , anisotropy , correlation length , scaling exponent .

1508.01409

in various researches and scientific activities ranging from theoretical and simulation to technological approaches , surfaces and interfaces are ubiquitous . characterization of rough surfaces is necessary through the simulation and making stochastic field for further applications . the electromagnetic and mechanical as well as chemical features of rough surfaces and interfaces are affected by morphological evolution during growth or etching processes . subsequently , investigation of local ( geometrical ) and global ( topological ) properties of rough surfaces play crucial role in determining the behavior of a rough surfaces as elements and their impact on incident beam of ray @xcite . during various aspects and processes such as evaporation / condensation @xcite , surface diffusion during the growth , geophysical studies @xcite , kinetic roughening on a nearby surface @xcite , exotic growth processes @xcite and etc . , it is possible to form anisotropic rough surfaces . many important properties of rough surfaces such as friction , diffusivity of particles of surface , contact , and wear can be influenced by topography of the underlying surface such as anisotropy . propagation of cracks @xcite , formation of ripple patterns @xcite , turbulent flow @xcite , depinning in anisotropic media @xcite , elastic - plastic contact model @xcite , lubricant film flow @xcite , anderson localization in anisotropic random media @xcite , diffusion , transport @xcite , transmission @xcite , turbulent boundary layer @xcite , cosmic microwave background radiation @xcite and large scale structures in the cosmos @xcite are some phenomena affected by roughness and anisotropies . in addition , from machinery and workpiece processes , many associated parameters influence on morphology of constructed rough surfaces @xcite ( see @xcite for experimental setup ) . from experimental , theoretical as well as technological points of view , it is important to have a measure for anisotropy probe , which can exhibit different properties . in principle , to determine whether underlying rough surface is anisotropic or not , following questions must be answered : first of all , what is the degree of anisotropy ? , what is the direction of anisotropy ? and finally , what is the kind of anisotropy ? @xcite . in the presence of anisotropy , one can divide the anisotropies into two categories : anisotropy in scaling exponents and / with in correlation length scale @xcite . the applicability of fractal properties and scaling concepts have been considered in @xcite , but if there is no scaling behavior represented by underlying surface , therefor one can not find reliable results by implementation of mentioned method based on scaling behavior . many given rough surfaces , even without any scaling properties , have also anisotropic nature and it is important to provide a robust and feasible algorithm for characterization of probable anisotropy . especially , in stochastic rough interfaces , the anisotropic features could be screened by the random nature of the surface , and it can induce additional and/or spurious properties . for instance , in the growth of anisotropic organic thin films , or erosion and growth with incident angle , anisotropic recognition and determining the kinds of anisotropies are of interest . lateral ordering of nano - dots has been reported in many experiments @xcite . in many cases , the underlying order in the distribution of nano - dots is not directly observable without further analysis on the surface profile , such as fourier analysis with especial wavelet @xcite and curvelet analysis . also in cases , where nano - ripples form , the exact experimental parameters which lead to ripple formation can not be easily measured unless one is able to detect and track the formation of asymmetric ripples precisely . since usual methods to detect anisotropies , e.g. , fourier transform , have their numerical and technical limits , especially in situations , where having large number of samples to make a proper statistical ensemble not to be possible . however many researchers have tried to examine the anisotropic surfaces , but a little part of them has been devoted to give a quantitative measurements to quantify anisotropy @xcite . an extensive quantitative analysis has been done by zhao et . al . by means of diffraction from anisotropic rough surfaces e. vivo et . al . , used the so - called height power spectral density analysis to examine the scaling anisotropic rough surface @xcite . recently , g. guillemot et . al . , have introduced a so - called regularity parameter to quantify the degree of anisotropy @xcite . according to field theoretic renormalization , there is an extensive work done by b. schmittmann et . a trivial and well - known approach is calculating the height - height correlation function and check whether the roughness exponent is directionally dependent or not ? . even though , previous researches give a good approach to find the direction of anisotropy , but only in some cases a method to discriminate between various natures of anisotropy has been established . in this paper , by notion of anisotropic feature in surface erosion and self - affine rough surfaces , we introduce and apply crossing statistics as a benchmark of anisotropy measure . we show that this method makes a feasible measure to quantify the existence of anisotropy and to discriminate isotropic and anisotropic pattern in real space . it can be mounted on the experimental devices , such as atomic force microscopy and scanning tunnelling microscopy . for the first time , rice introduced level crossing statistics @xcite . after that this method has been improved and used to investigate up - crossing and down - crossing of a typical stochastic field . during the last decades , many researches have been done to examine @xmath10 and @xmath0dimensional stochastic processes from the level crossing point of view @xcite . some advantages of this approach are as follows : in many of previous researches with the same purpose , there is no well - defined approach to quantify the degree of probable anisotropy at different thresholds while in the crossing statistics method there is a systematic framework to examine anisotropic nature for various values of thresholds ( see section [ method ] ) . in addition , this method enables us to predict theoretical crossing statistics even in the presence of more complicated correlation function as well as probability density function of underlying fluctuation functions . in other words , the non - gaussianity of underlying stochastic fields can be characterized by using this method , simultaneously . in the presence of isotropy and homogeneity , by calculating crossing statistics in @xmath10d , one can determine extension of given results for quantifying @xmath0d and @xmath11d without doing statistical analysis , directly in higher dimensions since they are much more complicated and much more time consuming @xcite . as we will show , crossing statistics can offer a new measure for characteristics length scales for a riven thresholds . the rest of paper is organized as follows : in sec . [ surf ] we will give a brief explanation regarding the synthetic generation of isotropic and anisotropic rough surfaces . we implement crossing statistics to investigate the height fluctuation of isotropic and anisotropic rough surfaces in sec . [ method ] . simulations of isotropic and anisotropic rough surfaces and analysis based on crossing statistics by means of theoretical and numerical computations will be given in detail in sec . [ result ] . summary and conclusions are presented in sec . in order to study the capability of our method to distinguish the anisotropy of a given rough surface , we use two different methods for preparing synthetic rough surfaces . we utilize fractional brownian motion ( fbm ) for generating synthetic self - affine rough surfaces explaining a wide range of growth models . for the second approach , monte carlo ( mc ) simulation model for pattern formation by ion - beam sputtering ( ibs ) is used @xcite . there are many methods introduced to simulate synthetic rough surface in @xmath0d ( `` 2 '' refers independent parameters , here are @xmath12 and `` 1 '' is devoted to dependent parameter , here is height of stochastic surface and the rest represents ) . irrespective to multifractality nature of surface , some of models to generate rough surfaces are : multiplicative cascading process @xcite , random measure @xmath13-model @xcite , @xmath1-model @xcite , log - stable models , log - infinitely divisible cascade models @xcite , and @xmath14model @xcite . in addition , the so - called successive random addition method @xcite , the weierstrass - mandelbrot function @xcite , as well as the optimization method @xcite have been introduced and applied for field generators . a very efficient way to generate rough surface is the modified fourier filtering method @xcite . in order to characterize the anisotropic properties of the system under study , in this paper , we use the modified fourier filtering method . to simulate gaussian anisotropic rough surface in @xmath0d with correlation length anisotropic model , following power spectrum is considered : @xmath15^{\gamma+1}}\end{aligned}\ ] ] here @xmath16 and @xmath17 are the correlation lengths in @xmath18 and @xmath19 directions , respectively . @xmath20 is the cutoff wave vector and @xmath21 is scaling exponent . @xmath22 is variance of surface . @xmath23 is the size of rough surface . for scaling anisotropic model , we have following power spectrum : @xmath24^{\gamma_u+1/2}\left[k_c^2+\xi_w^2k_w^2 \right]^{\gamma_w+1/2}}\nonumber\\\end{aligned}\ ] ] here @xmath25 and @xmath26 are the scaling exponents in direction @xmath18 and @xmath19 , respectively . both of above power spectrums represent the so - called fractional brownian motion approach . two points on stochastic surface separated with distance @xmath27 are correlated and correlation is diminished for @xmath28 @xcite . other quantities in eqs . ( [ spectrumcoraniso ] ) and ( [ spectrumexponent ] ) guarantee to have @xmath29 . sketch showing the monte carlo modeling set - up for an ion - beam sputtering . as described in the text , an ion beam trajectory makes an angle @xmath30 with the axes @xmath31 , and the projection of the ion - beam direction on the @xmath32 plane , makes an angle of @xmath33 relative to the @xmath34 axis . anisotropic direction is perpendicular to the @xmath32 projection of the ion - beam . ] upper : isotropic simulated rough surface for @xmath35 and @xmath36 . model : an preferred direction for @xmath37 and @xmath38 exists for simulated surface . lower panel corresponds to simulated anisotropic rough surface for @xmath39 and @xmath40,title="fig : " ] upper : isotropic simulated rough surface for @xmath35 and @xmath36 . model : an preferred direction for @xmath37 and @xmath38 exists for simulated surface . lower panel corresponds to simulated anisotropic rough surface for @xmath39 and @xmath40,title="fig : " ] upper : isotropic simulated rough surface for @xmath35 and @xmath36 . model : an preferred direction for @xmath37 and @xmath38 exists for simulated surface . lower panel corresponds to simulated anisotropic rough surface for @xmath39 and @xmath40,title="fig : " ] surface sputtering by energetic ions ( ne@xmath41 , ar@xmath41 , xe@xmath41 , etc ) as an efficient method to manufacture nano - scale structures on surface of solids ( glass , metals , semiconductors , etc ) is widely applied and examined in the last five decades @xcite . the base of an ion - beam sputtering ( ibs ) experiment is shooting energetic ions in the range of kev toward the prepared surface of the solid . etching the surface due to atomic collision cascades initiated by the energetic ions , along with enhanced surface diffusion of lateral ad - atoms leads to formation of regular patterns with typical size of 10 - 100 nm , in most cases and isotropic roughened or smoothed surfaces in other cases of experiments and computer simulations @xcite . nano - ripples , quantum dots , and nano - holes with symmetric and amorphous lateral distributions are among different types of patterns , froming in ibs experiments . though such patterns are highly desirable for practical and technological applications in many different areas @xcite , there is not much known about the underlying mechanisms of formation and anomalous features of them . coarsening of the patterns in time , presence of symmetries in unexpected directions , and complete phase diagram of type of the patterns forming in different experimental conditions are the most important and puzzling challenges in theoretical studies as well as experimental investigations . the mc modelling set - up , which is based on the theoretical model of bradley - harper @xcite , includes two main parts . erosion of surface atoms due to collisions of ions and diffusion of lateral atoms of the solid , enhanced by the heat released by collision cascades . we consider a @xmath0dimensional cubic lattice of atoms of size @xmath42 , with periodic boundary conditions and solid - on - solid restriction ( see fig . [ ibs ] ) . ions navigate to the surface from random starting points at a plane parallel to the initially flat solid surface ( i.e. @xmath43 plane ) . as indicated in fig . [ ibs ] , an ion beam follows a straight trajectory that makes an angle @xmath30 with the normal of this plane . the projection of the ion - beam direction on the plane target surface ( ( @xmath32 ) plane ) , makes an azimuthal angle of @xmath33 relative to the @xmath34 axis . all ions penetrate into the bulk in a typical distance and release their energy modelled by a 3d gaussian distribution @xcite . the share of energy for each lateral atom of the solid is calculated based on the gaussian distribution and each lateral atom is eroded with a probability proportional to that energy . in each diffusion sweep , hops to nearest neighbour sites are checked for all atoms with empty neighbours . here , the probability of acceptance of a possible hope is calculated based on arrhenius hoping rate , @xmath44 , where @xmath45 is a temperature dependent and material specific attempt rate , @xmath46 is an energy barrier assign to the different possible local configuration of the lattice before and after the hop , @xmath47 is the boltzmann constant and @xmath48 is the surface temperature . upon varying the parameters values and irradiation time length , different kind of isotropic and anisotropic surface profiles emerge @xcite . here , we fix all parameters except @xmath30 and @xmath33 . examples of surface profile in size of @xmath49 after shooting 30 atom / lateral atom at different beam directions are depicted in fig . [ aniso surf ] . typical surface with a cell represented by a square . the size of mesh equates to resolution of underlying rough surface . ] as explained in the introduction , we are interested in finding a criterion to distinguish the isotropic and anisotropic rough surfaces , consequently , the crossing statistics method will be carried out in the @xmath0dimensional framework @xcite . this method has been used to examine cosmological stochastic fields and many aspects of it have been derived in @xcite . in order to make more sense for further usage , we summarize mentioned method with some modifications in the following steps : * step1 : definition of variables : * suppose that for a rough surface in 2d , height of the fluctuations is represented by @xmath50 at coordinate @xmath51 with resolution @xmath52 and size @xmath42 ( see fig . [ surface1 ] ) . it is not compulsory that the size of width and height of underlying rough surface to be same . for convenience , suppose that the origin of the coordinate system is located at the centre of rough surface . we assign height fluctuations by @xmath53 , where @xmath54 and @xmath55 demonstrate the coordinate position through the basis vectors namely , @xmath34 and @xmath56 , respectively . in this case we have @xmath57 for the square shape of rough surface . for 1d field as indicated in upper panel of fig . [ level11 ] , crossing points with positive slope at arbitrary threshold , @xmath58 , are so - called up - crossing indicated by @xmath59-symbols in this figure . here @xmath1 and @xmath22 are the value of surface height and variance of height fluctuations , respectively . we can extend this crossing statistics for 2d and 3d stochastic fields . for example for ( 2 + 1)d rough surface , the iso - height contours at given threshold represent crossing statistics for a ( 2 + 1)d surface , while for ( 3 + 1)d stochastic field , iso - density surfaces recognize crossing statistics in this dimension @xcite . according to mathematical framework introduced for crossing statistics , it is straightforward to demonstrate that crossing statistics in higher dimensions are proportional to crossing statistics in 1d slice of underlying field . in this paper we use the up - crossing statistics through a line taken in arbitrary direction as a criterion to pick up anisotropy imposed on a stochastic field in 2d . positive slope crossing at the level @xmath60 . ] sketch of joint probability density function of a typical fluctuation and its derivative with respect to corresponding dynamical parameter ( position ) in the level crossing theory . the shaded area indicates the total probability of finding crossing with positive slope at level @xmath61 . the symbol @xmath62 can be replaced for each direction . ] * step2 : preparing data sets : * we consider two categories of slices for height fluctuation in two separate and orthogonal directions labeled by @xmath18 and @xmath19 . it must point out that these two directions are produced by rotation counterclockwise with respect to the origin of coordinate through the angle @xmath33 . for @xmath63 the common axes to be retrieved . we indicate the @xmath10d fluctuations along these slices by @xmath64 and @xmath65 . here @xmath66 refers to the @xmath66th slice through the @xmath18 or @xmath19 directions . the size of these @xmath10d signals depend on the resolution and the direction of slicing of underlying rough surface . the upper panel of fig . [ level11 ] shows a schematic of @xmath10d slice of underlying rough surface . if @xmath67 to be invariant under eulerian rotation , consequently the statistical isotropy will be valid , and in this case each of mentioned slices belong to the so - called isotropic gibbs ensemble . for an isotropic and homogenous process , in long run , the up - crossing and down - crossing are statistically equivalent @xcite . here in this study , we only take up - crossing events in order to find a benchmark for anisotropy . * step3 : theoretical approach : * probability distribution function ( pdf ) of the height of rough surface is represented by @xmath68 and the corresponding conditional pdf is defined by @xmath69 , here @xmath70 . the gradient of the height can be written as : @xmath71 . for both direction @xmath18 and @xmath19 , we construct one dimensional fluctuations as @xmath72 , in which @xmath62 symbol can be replaced by @xmath18 and @xmath19 and @xmath73 runs from @xmath74 to @xmath75 , here @xmath75 is sample size as : @xmath76 . let @xmath77 denotes the number of up - crossing ( crossing with positive slope ) the height fluctuations at an arbitrary level @xmath1 . where the additional parameter @xmath78 corresponds to the angle between the direction of a fixed frame and positive axis of e.g. @xmath18 ( see fig . [ aniso surf ] for more details ) . for convenient , we set @xmath79 . the ensemble averaging for level crossing with positive slope is also given by : @xmath80 to evaluate the conditions of up - crossing , we consider a 1d interval with length @xmath52 started from position @xmath81 . the sample fluctuations , @xmath82 will intersect the level @xmath1 in direction @xmath83 if and only if two following necessary and sufficient conditions are satisfied ( see the lower panel of fig . [ level11 ] ) : @xmath84 and the slope of signal should be larger or at least equal to the slope of line which is plotted by using the point at the beginning of interval and point located at the level @xmath1 , namely : @xmath85 } { \delta}\le\eta_{\diamond}(\phi;n , m_1).\ ] ] for each slicing we set the mean value to zero . according to joint pdf of height fluctuations and its derivative @xmath86 , the region corresponding to @xmath87 and @xmath88 , in the plane ( @xmath89 ) is related to the probability of up - crossing in direction @xmath90 at level @xmath1 . in fig . [ proba ] , we have indicated the shaded volume corresponding to joint probability density function to satisfy the mentioned conditions @xcite . subsequently , the probability of having up - crossing in the interval @xmath52 is given by : @xmath91 here @xmath92 is step function . therefore , @xmath93 here @xmath94 has been marginalized over other component of @xmath95 vector ( hereafter we remove bar symbol for convenience ) and @xmath96 is the average of up - crossing at level @xmath97 or @xmath98 corresponds to wavelength of having an up - crossing at level @xmath1 through the direction @xmath99 , statistically . the most familiar form of eq . ( [ levelmain ] ) is @xmath100 . in addition , if @xmath101 as for homogeneous and isotropic gaussian field then eq . ( [ levelmain ] ) indicates that @xmath102 . from theoretical point of view , one can calculate up - crossing statistic using the functional form of joint pdf of relevant variables . in the case of multivariate gaussian joint pdf of relevant variables of rough surface , we have : @xmath103 where @xmath104 and @xmath105 is the inverse of the covariance matrix of underlying variables : @xmath106.\ ] ] each element of covariant matrix can be computed using power spectrum of underlying 2d rough surface , @xmath107 . we derived these element for @xmath108dimensional isotropic stochastic field in appendix . we suppose that @xmath109 , therefore , the analytical form of up - crossing statistics for isotropic rough surface for arbitrary slice ( eq . ( [ levelmain ] ) ) reads as ( see the appendix for more details ) : @xmath110 where @xmath22 and @xmath111 are spectral parameters defined in appendix . additional 2-factor in the denominator of eq . ( [ theory for nu gaussain ] ) in comparison to that of introduced in ref . @xcite is in fact due to the up - crossing enumeration rather than taking into account both up- and down - crossings . however , we wrote up - crossing statistics for gaussian stochastic field , but it is possible to rewrite it for general case as : @xmath112 here @xmath113 is dirac delta function . in addition to above definition for up - crossing , @xcite introduced another kind of crossing for which a condition is taken into account for perpendicular direction of up - crossing . in another word , the value of field in perpendicular of at each crossing point should be extrema . therefore the mathematical form of this circumstance is : @xmath114 the perturbation formula for eqs . ( [ theory for nu nongaussain ] ) up to @xmath115 has been given in ref . @xcite and for an isotropic gaussian field in @xmath0-d closed form of eq . ( [ theory for nu nongaussaincond ] ) has been indicated in @xcite . as we are going to use this method for probing anisotropy imposed on a typical 2d rough surface , we can rewrite up - crossing in arbitrary direction based on 1d power spectrum , @xmath116 , as @xcite @xmath117 where @xmath118 and 1d power spectrum is given by : @xmath119 for isotropic rough surface we have @xmath120 , consequently : @xmath121 . upper panel : simulated isotropic gaussian rough surface . middle panel : up - crossing analysis for the isotropic gaussian rough surface . lower panel is @xmath122 for isotropic gaussian rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : simulated isotropic gaussian rough surface . middle panel : up - crossing analysis for the isotropic gaussian rough surface . lower panel is @xmath122 for isotropic gaussian rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : simulated isotropic gaussian rough surface . middle panel : up - crossing analysis for the isotropic gaussian rough surface . lower panel is @xmath122 for isotropic gaussian rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel shows the up - crossing analysis for completely isotropic rough surface for two arbitrary directions . lower panel corresponds to @xmath123 as a function of level for anisotropic rough surface for two anisotropic directions.,title="fig : " ] upper panel shows the up - crossing analysis for completely isotropic rough surface for two arbitrary directions . lower panel corresponds to @xmath123 as a function of level for anisotropic rough surface for two anisotropic directions.,title="fig : " ] for gaussian anisotropic rough surface we use power spectrum given in eq . ( [ spectrumcoraniso ] ) belonging to correlation length anisotropic model . the up - crossing in direction @xmath62 is : @xmath124 therefore for anisotropic gaussian rough surface , the ratio of up - crossing in two anisotropic direction is @xmath125 representing the inverse ratio of corresponding correlation length scales . for scaling anisotropic model , power spectrum introduced in eq . ( [ spectrumexponent ] ) is implemented . the up - crossing in direction @xmath62 is : @xmath126 in this case we have : @xmath127 another useful parameter based on @xmath96 is introduced by : @xmath128 obviously , for @xmath129 the above quantity specifies the total number of up - crossing for height fluctuations with positive slope at all levels in direction @xmath62 . for a typical rough fluctuation , @xmath130 can be a criterion to recognize the roughness of processes . for a typical long - range correlated process , @xmath130 is smaller than the same just it ordering has been destroyed , while for anti - correlated data set @xmath130 is larger than that of for which its memories have been destroyed by shuffling method . for isotropic gaussian rough surface we have : @xmath131 for correlation length anisotropic gaussian surface , by using eqs . ( [ spectrumcoraniso ] ) and ( [ theory for nu gaussain ] ) , eq . ( [ ntq ] ) reads as : @xmath132 while for scaling exponent anisotropic gaussian surface , we consider power spectrum according to eq . ( [ spectrumexponent ] ) , therefore , eq . ( [ ntq ] ) becomes : @xmath133 the upper panel of fig . [ gff ] shows isotropic gaussian rough surface . the filled circle symbols in the middle panel of this figure correspond to the numerical computation of @xmath2 while the solid line is calculated by eq . ( [ theory for nu gaussain ] ) , which is the theoretical prediction for up - crossing as a function of @xmath1 . the generalized crossing statistics , @xmath122 , has been shown in lower panel . as we expect , our results do not depend on special direction . our results demonstrate that there exists a good consistency between numerical and theoretical prediction . upper panel : simulated anisotropic gaussian rough surface in which , its power spectrum is given by eq . ( [ spectrumcoraniso ] ) middle panel : up - crossing analysis for the correlation length anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : simulated anisotropic gaussian rough surface in which , its power spectrum is given by eq . ( [ spectrumcoraniso ] ) middle panel : up - crossing analysis for the correlation length anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : simulated anisotropic gaussian rough surface in which , its power spectrum is given by eq . ( [ spectrumcoraniso ] ) middle panel : up - crossing analysis for the correlation length anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : synthetic anisotropic gaussian rough surface with correlation as well as scaling exponent anisotropies ( eq . ( [ spectrumexponent ] ) ) . middle panel : up - crossing analysis anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : synthetic anisotropic gaussian rough surface with correlation as well as scaling exponent anisotropies ( eq . ( [ spectrumexponent ] ) ) . middle panel : up - crossing analysis anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] upper panel : synthetic anisotropic gaussian rough surface with correlation as well as scaling exponent anisotropies ( eq . ( [ spectrumexponent ] ) ) . middle panel : up - crossing analysis anisotropic gaussian rough surface . lower panel is @xmath122 for the mentioned simulated rough surface . the red line represents the theoretical prediction and filled circles correspond to the numerical computation.,title="fig : " ] in the following sections , we are going to derive the @xmath134 for height fluctuations in two distinct directions and then we try to find a robust criterion to distinguish the isotropic and anisotropic surfaces due to various growth mechanisms . after generating a typical @xmath0-dimensional stochastic field via synthetic method or preparing a rough surface in an experiment , an important question is that , whether a preferred direction has been imposed on the underlying stochastic field . suppose we indicate an arbitrary feature on the rough surface @xmath135 ( in this paper we consider @xmath136 ) . statistical isotropy causes @xmath135 to be invariant under eulerian transformation : @xmath137 here @xmath138 represents the rotation matrix . in order to quantify the probable anisotropy existed in rough surface , we apply the crossing statistics method to calculate @xmath139 and @xmath140 of our synthetic rough surfaces . we expect the up - crossing statistics for various direction on an isotropic rough surface to be statistically identical , while in anisotropic case , @xmath141 gets different values at least for mean value of threshold . upper panel of fig . [ nu for anisotropic surf ] confirms this statement . the lower panel corresponds to the same quantity for an anisotropic rough surface . in this plot , we select @xmath18 and @xmath19 axes , for which , we have maximum anisotropy direction imposed on the synthetic rough surface . in fig . [ model1 ] , we used power spectrum for correlation anisotropy ( eq . ( [ spectrumcoraniso ] ) ) for typical valued for free parameters and simulated anisotropic rough surface . then we computed crossing statistics for direction parallel and perpendicular to known anisotropic direction . the solid lines in middle and lower panels of mentioned figure indicates theoretical prediction . fig . [ model2 ] contains same information except for scaling exponent anisotropy . to use the efficient capability of crossing statistics to detect the direction of anisotropy , we use an anstz as : @xmath142 ^ 2}{[\sigma_{w}^2(n;\phi , q)+\sigma_{u}^2(n;\phi , q)]}\ ] ] here @xmath143 denotes the error bar of generalized total up - crossing and @xmath66 runs from 1 to the total number of samples profiles . since we are looking for @xmath78 , for which the maximum difference is detected , then we compute @xmath144 as a function of @xmath78 for each value of @xmath145 and finally , by computing the p - value for this quantity , degree of reliability can be quantified . the presence of @xmath145 , enables us to quantify the contribution of various values of feature fluctuations in the anisotropy . [ qvalue ] shows @xmath146 for @xmath129 as a function of @xmath78 for synthetic anisotropic rough surface simulated by ibs method with @xmath147 . it has demonstrated that there is peak for @xmath146 around @xmath148 . the value of @xmath149 as a function of @xmath78 for anisotropic rough surface shown in fig . [ aniso surf ] with @xmath147 . ] in order to quantify degree of anisotropy in underlying rough surface and find reliable results we should investigate the statistical deviation between @xmath150 and @xmath151 . the significance of mentioned deviation , is systematically performed by calculating student s @xmath152test for equal sample size and unequal mean and variance for each @xmath145 s and @xmath78 according to : @xmath153\nonumber\\ & & \qquad\times \sqrt{\frac{n_{run}}{\sigma_{u}^2(\phi , q)+\sigma_{w}^2(\phi , q)}}\end{aligned}\ ] ] here @xmath154 indicates the number of ensemble which is equal to @xmath66 introduced in section [ method ] . the @xmath155value , corresponding to @xmath156 for @xmath157 degrees of freedom is determined by two - tailed hypothesis : @xmath158 . the chi - square for the mentioned p - value reads as : @xmath159 finally , by using chi - square distribution function for final p - value , @xmath160 , associated with the @xmath161 and with @xmath162 degrees of freedom , is computed as : @xmath163 for @xmath9 significance level , @xmath164 , we can conservatively say that there exists a significance difference between two generalized up - crossing quantities for two directions , @xmath18 and @xmath19 at given @xmath78 . fig . [ final_pvalue ] represents the p - value for anisotropic rough surface indicated in fig . [ qvalue ] . the significance of difference given by @xmath165-value analysis for anisotropic rough surface shown in fig . [ aniso surf ] with @xmath147 . ] beside the capability of crossing statistics to determine the direction of anisotropy , there is another advantage for the mentioned method in distinguishing the kind of anisotropy imposed in the underlying stochastic field . in principle , correlation length anisotropy and/or scaling exponent anisotropies are ubiquitous in simulations and experiments . in practice , if we are going to discriminate between the two mentioned kinds of anisotropies , firstly we should compute eq . ( [ pvalue22 ] ) . after determining the direction of anisotropy , generally , based on the ratio @xmath166 , we can determine the left hand side of eq . ( [ ratio2 ] ) and/or eq . ( [ ntq3 ] ) . according to widely - used methods such as spectral analysis @xcite , fluctuation analysis @xcite , detrended fluctuation analysis ( dfa ) @xcite , wavelet transform module maxima ( wtmm ) @xcite and discrete wavelets @xcite , the value of scaling exponents in @xmath18 and @xmath19 direction are determined and finally by means of eq . ( [ ratio2 ] ) and/or eq . ( [ ntq3 ] ) , the kind of anisotropy and the ratio of correlation length anisotropy is determined . it is worth noting that , methods which are implemented for determining scaling exponent are usually give accurate value for scaling exponent while methods established for computing characteristic correlation length scale encounter to finite size effects of the underlying stochastic field . statistical periodicity of anisotropic patterns at given threshold can be examined by up - crossing statistics . as mentioned in section [ method ] and can be found from eq . ( [ levelmain ] ) , @xmath141 represents wave number at threshold @xmath58 , consequently , @xmath167 shows the statistical characteristic length scale for up - crossing at threshold @xmath58 . in addition , the generalized total number of crossing statistics ( eq . ( [ ntq ] ) ) is useful criterion to measure the kind of roughness for all threshold , @xmath1 , in various directions . for example for @xmath129 , eqs . ( [ ntq ] ) , ( [ ntq1 ] ) and ( [ ntq2 ] ) represent somehow total roughness of surface in given direction . anisotropy and non - gaussianity are two important properties of stochastic fields and should be well addressed to investigate from theoretical and experimental points of view . several methods have been implemented to explore exotic features and the mentioned properties of stochastic fields , but systematic and other limitations in theoretical and computational approaches cause this field of research to remain under debate . based on previous works regarding crossing statistics in various dimensions @xcite , in this paper we relied on crossings at a given threshold , @xmath168 and introduced them as a robust benchmark for anisotropy detection imposed in stochastic fields in @xmath0d . in addition , we showed that this method can examine the gaussianity nature of @xmath0dimensional rough surfaces . according to an extensive study by b. s. ryden @xcite , the crossing statistics for anisotropic field in @xmath73d is related to that of computed from @xmath108dimensional slices of underlying field . subsequently , we used @xmath169 for prepared slices parallel and perpendicular to a typical direction , @xmath62 , and compared them to find probable anisotropic direction . the so - called generalized total crossing , @xmath170 with positive slope has been investigated for complementary test . the characteristic wavelength , @xmath171 , or characteristics length scale , 1/ @xmath171 , at an arbitrary threshold , can be introduced in the context of crossing statistics for further evaluations . from theoretical point of view , according to multivariate probability density function written for vector field of underlying stochastic field , we showed that crossing statistics for arbitrary slice in an isotropic gaussian rough surface is given by eq . ( [ theory for nu gaussain ] ) if 2dimensional power spectrum to be as prior . also theoretical prediction is introduced by eq . ( [ nu1d ] ) if 1dimensional marginalized power spectrum ( eq . ( [ spectrum1d ] ) ) is determined . this robust theoretical approach enables us to derive conditional crossing statistics for various stochastic fields . for example we derived perturbations expansion for up - crossing for @xmath108dimensional isotropic stochastic field . applying additional condition for up - crossing from computational approach is slightly time consuming and also one can setup theoretical approach for it as introduced in eq . ( [ theory for nu nongaussaincond ] ) @xcite . we simulated synthetic isotropic and anisotropic rough surfaces in the context of modified fourier filtering method with anisotropy imposed on the rough surface due to correlation length scale ( eq . ( [ spectrumcoraniso ] ) ) and/or due to scaling anisotropic model ( eq . ( [ spectrumexponent ] ) ) . up - crossing enumeration of simulated isotropic gaussian rough surface for two perpendicular axes ( @xmath18 and @xmath19 ) were consistent with that of predicted by theoretical predictions ( see fig . [ gff ] ) . beside determining the direction of anisotropy , the kinds of anisotropy existed in rough surfaces has many motivations from experimental point of view , subsequently we used two familiar method to simulate anisotropic rough surfaces . [ model1 ] indicates our results for simulated correlation length anisotropic surface . our results confirmed that theoretical prediction for the ratio of up - crossing statistics for @xmath18 and @xmath19 directions are compatible with that of determined by computation . therefore , we are not only able to determine the direction of anisotropy but also one can determine the ratio of correlation length scales for @xmath18 and @xmath19 directions by using the quantity @xmath172 . for anisotropy produced by different scaling exponents , we found consistent results indicated in fig [ model2 ] . to distinguish between correlation length and scaling exponent anisotropies , according to eq . ( [ ratio2 ] ) , we should use prior information about the value of @xmath173 s or @xmath21 s . using a method to determine the scaling exponent , one can break this degeneracy and then determine the kinds of anisotropy and the ratio of correlation lengths in @xmath18 and @xmath19 directions . it is worth noting that methods for determining scaling exponents are much well established while because of various definitions for correlation length scale computation of the mentioned characteristic scale , is most challenging . consequently , up - crossing analysis can determine correlation length scale in a more robust approach . for the second synthetic rough surfaces , we implemented ion - beam sputtering method to construct anisotropic rough surface with known anisotropic parameters . we introduced @xmath144 in eq . ( [ anstz1 ] ) and agin by determining two independent directions on the rough surface and by rotating them , the mentioned quantity has been computed as a function of @xmath78 . our results confirmed that for anisotropic case , difference between up - crossing analysis parallel and perpendicular to anisotropic direction is reliable and to quantify this statement , we computed p - value and at @xmath9 confidence interval we could determine the anisotropic direction , @xmath78 ( figs . [ qvalue ] and [ final_pvalue ] ) . before finishing this paper , some advantages of up - crossing statistics as anisotopic probe are listed below : + 1 ) crossing statistics is well - established theoretical as well as computational approaches . namely , from computational algorithm , we are working in real space and it is almost not - affected by boundary effects . + 2 ) as this method is working in real space , so one can say that initial information is not modulated with other auxiliaries quantities in phase space . + 3 ) it is simple to add arbitrary condition for enumeration crossing statistics . this is also straightforward from theoretical approach . + 4 ) from computational point of view , one can apply this method for even anisotropic non - gaussian fields for arbitrary dimension . in some cases , one can find analytical perdition such as presented in ref . + 5 ) one can determine various characteristic length ( time ) scale for arbitrary threshold . + 6 ) for some cases , e.g. isotropic gaussian stochastic field , up - crossing statistics for higher dimensions can be written in terms of crossing statistics in lower dimensions . + 7 ) this method is able to determine the non - gaussianity nature accompanying anisotropy . the contribution of different scales in anisotropy can be examined by adopting the various values for @xmath145 . + 8) the generalized up - crossing , @xmath174 , gives also a criterion for determining roughness @xcite and exotic features @xcite . + it could be interesting to apply the above mentioned pipeline to real stochastic fields in condensed matter , cosmology and astrophysics and etc . and examine the results for further applications . also curve - crossing method is another useful method for this purpose @xcite . * acknowledgments : * s.m.s . movahed is grateful to a. vafaei sadr and s. bazmi for their comments on the perturbation approach . movahed is tankful to associate and federation office of ictp for their support and the hospitality of hecap section of ictp , where some parts of this analysis were done . in this appendix we will give derivation of some important equations used in this paper . for a stochastic field in @xmath108d , we consider a covariant vector field containing most relevant quantities for underlying stochastic field as : @xmath175 , where @xmath1 represents the value of stochastic field , @xmath176 , @xmath177 s are first derivative and @xmath178 s correspond to second derivative with respect to independent parameter in @xmath179 and @xmath180 directions . correlation function of stochastic field becomes : @xmath181 the so - called spectral parameters are : @xmath182 for isotropic rough surface , we can write : @xmath183 the correlation functions of derivatives of stochastic field in isotropic case are : @xmath184 where @xmath185 using correlation function we can write : @xmath186 to compute up - crossing statistics we should also determine the statistical average of absolute value of derivative of underlying stochastic field , so for gaussian pdf , we have : @xmath187 because @xmath188 and @xmath189 , so @xmath190 . for @xmath191 , one can write : @xmath192 subsequently , for @xmath193 : @xmath194 . plugging them in eq . ( [ theory for nu nongaussain ] ) , one can simply get theoretical prediction for gaussian rough surface in arbitrary direction , namely eq . ( [ theory for nu gaussain ] ) . in the presence of weak non - gaussianity , there is a perturbative approach to setup theoretical model for every desired feature ( see also @xcite ) . here to make more complete our explanation , we will give perturbative equations up to @xmath195 , for up - crossing ( eq . ( [ theory for nu nongaussain ] ) ) . the so - called characteristics function which is related to the free energy of underlying field is defined by @xcite : @xmath196 using the definition of cumulants , @xmath197 ( here @xmath198 is written to emphasize that here we have cumulants rather than moments . as examples @xmath199 and @xmath200 . if the mean value of underlying stochastic field to be zero , consequently , cumulants are identical to moments . ) , one can expand @xmath201 as : @xmath202 so above equation becomes : @xmath203 by using inverse fourier transform one can read the probability density function as follows : @xmath204 by inserting eq . ( [ parti1 ] ) in eq . ( [ parti2 ] ) , we find : @xmath205}\nonumber\\ & \times&\sqrt{\frac{{\rm det } \mathcal{m}}{(2\pi)^{n } } } \ { \bf e}^{-\frac{1}{2}({\bf a}^{t}.\mathcal{m}.{\bf a})}\end{aligned}\ ] ] here @xmath105 is inverse of covariance @xmath206 matrix and for @xmath207 it is the same as eq . ( [ cov1 ] ) . it is clear that the last term in above equation equates to multivariate gaussian probability density function introduced in eq . ( [ jpdf11 ] ) . by using eq . ( [ parti3 ] ) , the statistical average of a typical feature , @xmath208 , in the general case reads @xcite : @xmath209 by taking into account up to @xmath195 in the context of perturbative approach , the probability density function of @xmath210 reads as @xcite : @xmath211\end{aligned}\ ] ] in which , @xmath212 also @xmath213 and @xmath214 are hermite polynomials of orders @xmath215 and @xmath216 , respectively . now we are ready to compute crossing statistics represented in eq . ( [ theory for nu nongaussain ] ) in @xmath108d @xcite : @xmath217\end{aligned}\ ] ] where @xmath218 also @xmath219 j. t. johnson , phd thesis , massachusetts institute of technology , ( 1996 ) . y. p. zhao , g. c. wang , and t. m. lu , phys . b , * 58 * , 7300 ( 1998 ) . t.r . thomas , b.g . rosn and n. amini , wear * 232 * , 41 ( 1999 ) . z. moktadir and k. sato , surface science , * 470 * , l57 ( 2000 ) . a. arnodo , n. decoster and s. g. roux , eur . j. b , * 15 * , 567 ( 2000 ) . c. bourlier and g. berginc , waves in random media , * 14 * , 229 ( 2004 ) ; waves in random media , * 14 * , 253 ( 2004 ) . l. simonot and g. obein , applied optics , * 46 * , 2615 ( 2007 ) . i. simonsen , j. b. kryvi , a. a. maradudin and t. a. leskova , computer physics communications , * 182 * , 1904 ( 2011 ) . g. guillemot , m. bigerelle and z. kawaja , scanning , * 36 * , 127 ( 2014 ) . g. badano , x. baudry , p. ballet , p. duvaut , a. million , e. micoud , s. kaismoune , p. fougres , s. mibord , p. tran - van and a. etcheberry , journal of electronic materials , * 37 * , 1369 ( 2008 ) . y. chen , b. he , j. lee and n. a. patankar , journal of colloid and interface science , * 281 * , 458 ( 2005 ) . e.vivo , m. nicoli , m. engler , t. michely , l. v zquez and r. cuerno , phys . b , * 86 * , 245427 ( 2012 ) . a. ballestad , b. j. ruck , m. adamcyk , t. pinnington and t. tiedje , phys . lett . , * 86 * , 2377 ( 2001 ) . c. g. fox and d. e. hayes , rev . geophys . , * 23 * , 1 ( 1985 ) ; b. klinkenberg and m. f. goodchild , earth surface processes and landforms , * 17 * , 217 ( 1992 ) . j. villain , journal de physique i , * 19 * , 19 ( 1991 ) ; d. e. wolf , phys . * 67 * , 1783 ( 1991 ) . m. ansari - rad , s. m. vaez allaei , and m. sahimi , phys . e , * 85 * , 021121 ( 2012 ) . h. nishimori and n. ouchi , phys . lett . , * 71 * , 197 ( 1993 ) . l. chau and k. bhaganagar , phys . fluids , * 24 * , 115102 ( 2012 ) . l. h. tang , m. kardar and d. dhar , phys . * 74 * , 920 ( 1995 ) . j. a. ogilvy , j. phys . phys . , * 25 * , 1798 ( 1992 ) . c. y. poon and r. s. sayles , journal of tribology , * 116 * , 194 ( 1994 ) . f. plourabou , m. prat , and n. letalleur , phys . e , * 64 * , 011202 ( 2001 ) . z. q. zhang , q. j. chu , w. xue and p. sheng , phys . b , * 42 * , 4613 ( 1990 ) . m. e. thompson and s. r. brown , journal of geophysical research : solid earth ( 1978 - 2012 ) , * 96 * , 21923 ( 1991 ) . a. adrover , w. schwalm , m. giona , and d. bachand , phys . e , * 55 * , 7304 ( 1997 ) . jongwoo lim , kedar hippalgaonkar , sean c. andrews , arun majumdar and peidong yang , nano lett.,12 ( 5 ) , pp 2475 - 2482 ( 2012 ) g. palasantzas , j. barnas and th . m. de hosson . 88 , 927 ( 2000 ) . l. keirsbulck , l. labraga , a. mazouz and c. tournier , experiments in fluids , * 33 * , 497 ( 2002 ) . p. a. r. ade , _ et al . _ , astronomy and astrophysics , * 571 * ( 2014 ) . s. codis , c. pichon , d. pogosyan , f. bernardeau and t. matsubara , mnras volume 435 , issue 1pp . 531 - 564 ( 2013 ) m. tejedor , ja garca , j. carrizo and l. elbaile , ieee trans . * 34 * ( 1998 ) . l. bing - chi , l. kai , k. xiao - li , z. ji - qiang , h. yu - dan , l. jiang - shan , w. wei - dong and t. yong - jian , chinese physics b , * 23 * , 066804 ( 2014 ) . w. liu , l. lu , y. li , y. cai and d. p. sekulic , epl , * 97 * , 46003 ( 2012 ) . a. s. zapevalov , acoustical physics , * 53 * , 75 ( 2007 ) . p. g. benardos and g. c. vosniakos , international journal of machine tools and manufacture , * 43 * , 833 ( 2003 ) . b. ziberi , f. frost , t. hche , and b. rauschenbach , phys . b , * 72 * , 235310 ( 2005 ) . b. ziberi , f. frost , and b. rauschenbach , appl . * 88 * , 173115 ( 2006 ) . b. ziberi , m. cornejo , f. frost and b. rauschenbach , j. phys . : condens . matter * 21 * , 224003 ( 2009 ) . e. j. heller and m. g. lagally , appl . lett . , * 60 * , 2675 ( 1992 ) . r. maboudian , v. bressler - hill , k. pond , x. s. wang , p. m. petroff and w. h. weinberg , surf . , * 302 * , 269 ( 1994 ) . c. orme , m. d. johnson , j. l. sudijono , k. t. leung and b. g. orr , appl . phys . lett . , * 64 * , 860 ( 1994 ) . m. a. cotta , r. a. hamm , t. w. staley , s. n. g. chu , l. r. harriott , m. b. panish and h. temkin , phys . lett . , * 70 * , 4106 ( 1993 ) . r. l. headrick , j. m. baribeau and y. e. strausser , appl . , * 66 * , 96 ( 1995 ) . m. t. sinn , j. a. del alamo , b. r. bennett , k. haberman and f. g. celii , j. electron . mater . , * 25 * , 313 ( 1996 ) . n. e. lee , d. g. cahill and j. e. greene , phys . b , * 53 * , 7876 ( 1996 ) . y. p. zhao , h. n. yang , g. c. wang and t. m. lu , phys . rev . b , * 57 * , 1922 ( 1998 ) . b. schmittmann , g. pruessner and h. k. janssen , phys . e , * 73 * , 051603 ( 2006 ) . s. o. rice , bell system tech . j. , * 23 * , 282 ( 1944 ) ; bell system tech . j. , * 24 * , 46 ( 1945 ) . b. s. ryden , the astrophysical journal , * 333 * , 41 ( 1988 ) . p. h. brill , cors bulletin , * 34 * , 4 ( 2000 ) . j. r. bond and g. efstathiou , mon . not . r. astr . , * 226 * , 655 ( 1987 ) . t. matsubara , the astrophysical journal , * 584 * , 1 ( 2003 ) . m. s. movahed and s. khosravi , jcap , * 1103 * , 012 ( 2011 ) . a. k. hartmann , r. kree , u . geyer and m. klbel , phys . b , * 65 * , 193403 ( 2002 ) . e. o. yewande , a. k. hartmann and r. kree , phys . rev . b , * 71 * , 195405 ( 2005 ) . e. o. yewande , r. kree an a. k. hartmann , phys b , * 73 * , 115434 ( 2006 ) . a. k. hartmann , r. kree and t. yasseri , j. phys . : condens . matter , * 21 * , 224015 ( 2009 ) . for the full description and documentation of all the versions of the model see t. yasseri , how to make nano - waves on solid surfaces , saarbrcken , vdm verlag , isbn:978 - 3639292602 ( 2010 ) . r. kree , t. yasseri and a.k . hartmann , nucl . ins . meth . in phys . b 267 , 1407 ( 2009 ) j. feder , _ fractals _ , plenum press new york and london , ( 1988 ) . g. f. gu and w. x. zhou , phys . e , * 74 * , 061104 , ( 2006 ) . j. w. kantelhardt [ arxiv:0804.0747v1 ] . c. meneveau and k. r. sreenivasan , phys . lett . , * 59 * , 1424 ( 1987 ) . r. benzi , g. paladin , g. parisi , a. vulpiani , j. phys . a , * 17 * , 3521 ( 1984 ) . d. schertzer and s. lovejoy , _ turbulence and chaotic phenomena in fluids _ , edited by t. tatsumi ( north - holland , amsterdam , 1984 ) . d. schertzer and s. lovejoy , j. geophys . res . , * 92 * , 9693 ( 1987 ) . d. schertzer , s. lovejoy , f. schmitt , y. chigirinskaya and d. marsan , fractals , * 05*,427 ( 1997 ) . c. meneveau and k. r. sreenivasan , phys . lett . , * 59 * , 1424 ( 1987 ) . h. o. peitgen , d. saupe , m. f. barnsley , y. fisher , and m. mcguire , _ the science of fractal images _ ( springer , new york , ny , 1988 ) . m. ausloos and d.h . berman , proceedings of the royal society of london . a. mathematical and physical sciences , * 400 * , 331 ( 1985 ) . h. hamzehpour and m. sahimi , phys . e , * 73 * , 056121 ( 2006 ) . h. a. makse , s. havlin , m. schwartz and h. e. stanley , phys . rev . e,*53 * , 5445 ( 1996 ) . m. sahimi , _ heterogeneous materials ii _ , chap . 5 - 8 ( springer , berlin , 2003 ) . m. sahimi , _ flow and transport in porous media and fractured rock _ , 2nd ed . , chap . 5 ( wiley - vch , weinheim , 2011 ) m. sahimi and s. e. tajer , phys . e , * 71 * , 046301 ( 2005 ) . m. a. knackstedt , a. p. sheppard and w. v. pinczewski , phys . e , * 58 * , r6923 ( 1998 ) ; m. a. knackstedt , a. p. sheppard and m. sahimi , adv . water resour . , * 24 * , 257 ( 2001 ) . s. habenicht , w. bolse , k. p . lieb and k. reimann , phys . * , r2200 ( 1999 ) . u. valbusa , c. boragno and f. b. de mongeot , j. phys . : condens . matter , * 14 * , 8153 ( 2002 ) . f. frost , b. ziberi , a. schindler and b. rauschenbach , app . phys . a : materials science & processing , * 91 * , 4 ( 2008 ) . r. kree , t. yasseri and a. k. hartmann , nucl . ins . meth . in phys . b , * 267 * , 1403 ( 2009 ) . t. yasseri , and r. kree , nucl . ins . meth . in phys . b * 268 * , 2496 ( 2010 ) . a. d. maynard _ et al . _ , nature , * 444 * , 267 ( 2006 ) . r. m. bradley and j. m. e. harper , j. vac . a , * 6 * , 2390 ( 1988 ) . p. sigmund , phys . rev . * 184 * , 383 ( 1969 ) . d. e. newland , _ an introduction to random vibrations , spectral and wavelet analysis _ , third edition , longman scientific technical , ( 1993 ) . h. e. hurst , trans . * 116 * , 770 , ( 1951 ) . c. k. peng , s. buldyrev , a. goldberger , s. havlin , f. sciortino , m. simons and h. e. stanley , nature , * 356 * , 168 ( 1992 ) . c. k. peng , s. v. buldyrev , s. havlin , m. simons , h. e. stanley and a. l. goldberger , phys . e , * 49 * , 1685 ( 1994 ) . j. w. kantelhardt , s. a. zschiegner , e. koscielny - bunde , s. havlin , a. bunde and h. e. stanley , physica a , * 316 * , 87 ( 2002 ) . k. hu , p. c. ivanov , z. chen , p. carpena and h. e. stanley phys . e , * 64 * , 011114 ( 2001 ) . a. arnodo , n. decoster , p. kestener and s. g. roux , advances in imaging and electron physics , * 126 * , 1 ( 2003 ) . s. g. roux , a. arneodo and n. decoster , eur . j. b , * 15 * , 765 ( 2000 ) ; n. decoster , s. g. roux and a. arneodo , _ ibid _ , * 15 * , 739 ( 2000 ) ; a. arnedo , n. decoster and s. g. roux , _ ibid _ , * 15 * , 567 ( 2000 ) . j. f. muzy , e. bacry and a. arneodo , phys . rev . lett . , * 67 * , 3515 ( 1991 ) . zbigniew r. struzik and a. p. j. m. siebes , physica a , * 309 * , 388 , ( 2002 ) . j. arrault , a. arneodo , a. davis and a. marshak , phys . lett . , * 79 * , 75 ( 1997 ) . j. w. kantelhardt , h. e. roman and m. greiner , physica a , * 220 * , 219 ( 1995 ) . h. e. roman , j. w. kantelhardt and m. greiner , europhys . * 35 * , 641 ( 1996 ) . f. shahbazi , s. sobhanian , m. r. rahimi tabar , s. khorram , g. r. frootan and h. zadeh , j. phys . a , * 36 * , 2517 ( 2003 ) . m. vahabi , g. r. jafari and s. m. s. movahed , j. stat . ( 2011 ) p11021 . t. matsubara , apj , 434 , l43 ( 1994 ) . s. bazmi , a. r. vafaei sadr and s. m. s. movahed , in preparing ( 2015 ) .

the _ wilkinson microwave anisotropy probe _ ( _ wmap _ ) mapped the distribution of temperature and polarization over the entire sky in five microwave frequency bands . these full - sky maps were used to obtain measurements of temperature and polarization anisotropy of the cosmic microwave background with the unprecedented accuracy and precision . the analysis of two - point correlation functions of temperature and polarization data gives determinations of the fundamental cosmological parameters such as the age and composition of the universe , as well as the key parameters describing the physics of inflation , which is further constrained by three - point correlation functions . _ wmap _ observations alone reduced the flat cold dark matter (cdm ) cosmological model ( six ) parameter volume by a factor of compared with pre-_wmap _ measurements . the _ wmap _ observations ( sometimes in combination with other astrophysical probes ) convincingly show the existence of non - baryonic dark matter , the cosmic neutrino background , flatness of spatial geometry of the universe , a deviation from a scale - invariant spectrum of initial scalar fluctuations , and that the current universe is undergoing an accelerated expansion . the _ wmap _ observations provide the strongest ever support for inflation ; namely , the structures we see in the universe originate from quantum fluctuations generated during inflation .

1404.5415

the _ wmap _ @xcite spacecraft was designed to measure the full - sky distribution of temperature differences ( anisotropy ) and polarization of the cosmic microwave background ( cmb ) . _ wmap _ is the successor of the legendary _ cosmic background explorer _ ( _ cobe _ ) satellite , whose spectrograph provided a precision - measurement of the cmb blackbody , implying that matter and radiation were in thermal equilibrium , consistent with the expectation of the hot big bang theory of the universe @xcite . the _ cobe _ differential radiometers discovered the primordial ripples in spacetime that existed in the early universe @xcite . with 35 times better angular resolution and 40 times better sensitivity than _ cobe _ , _ wmap _ took the cosmological research with cmb to the next level . _ wmap _ was proposed to nasa as a midex ( medium - class explorers ) mission in 1995 . four of eight co - investigators of _ wmap _ were previously on the _ cobe _ science team . after being selected in 1996 , _ wmap _ launched on june 30 , 2001 , and arrived in its orbit around the second lagrange point ( l2 ) , 1.5 million kilometers from earth , three months later . since then , _ wmap _ operated almost flawlessly for nine years until it left its l2 orbit on september 8 , 2010 , to pass the baton to its successor , the _ planck _ satellite , which arrived at l2 in july 2009 . the _ wmap _ team issued five data releases . the first - year data release ( february 11 , 2003 ) came with 13 papers @xcite and later with one more paper @xcite ; the third - year data release ( march 16 , 2006 ) came with 4 papers @xcite and later with one more paper @xcite ; the five - year data release ( march 5 , 2008 ) came with 7 papers @xcite and later with one more paper @xcite ; the seven - year data release ( january 25 , 2010 ) came with 6 papers @xcite ; and the final , nine - year data release ( december 21 , 2012 ) came with 2 papers @xcite . in addition , detailed descriptions of the mission , data processing , calibration , as well as of the data products for each release are given in the explanatory supplement document @xcite . in this article , we give a brief review on the _ wmap _ experiment , the data analysis , and the main science results from the nine - year observations . _ wmap _ measured the distribution of temperature and polarization of the entire sky in five frequency bands ( 23 , 33 , 41 , 61 , and 94 ghz ) . five frequencies were necessary to separate the cmb from emission of our own galaxy , including synchrotron , free - free , and dust emission ( see section [ sec : fg ] for how to reduce the effect of the galactic emission ) . _ wmap _ has two back - to - back mirrors to focus the incoming electromagnetic waves arriving from two different lines of sight , which are separated by @xmath2 from each other . the size of each mirror is 1.4 m@xmath3 m , providing substantially better angular resolution than _ cobe _ , which did not have mirrors but only horn antennas . the incoming waves collected by two mirrors are received by a pair of feed horns . let us call these two inputs `` a side '' and `` b side . '' each of the inputs from the a and b sides is separated into two orthogonal polarized waves by orthomode transducers ( omts ) , and then sent to a pair of radiometers . _ wmap _ has 20 radiometers , or 10 pairs of radiometers . we call each pair of radiometers a `` differencing assembly '' ( da ) . there are 1 , 1 , 2 , 2 , and 4 das at 23 , 33 , 41 , 61 , and 94 ghz , respectively . each da measures temperature and polarization differences between the a and b sides each of 20 radiometers processes inputs from the a and b sides as follows ( section 2.3 of @xcite ) * the inputs are separated into two linear combinations , @xmath4 and @xmath5 . * these linear combinations are amplified by high electron mobility transistor ( hemt ) amplifiers , yielding @xmath6g_1,\qquad u_2\equiv \left[\frac{a - b}{\sqrt{2}}+n_2\right]g_2 , \end{aligned}\ ] ] where @xmath7 and @xmath8 are the noise added by the amplifiers , and @xmath9 and @xmath10 are the gains of the amplifiers . * the latter combination , @xmath11 , is phased - switched to yield @xmath12 . then @xmath13 and the phase - switched @xmath11 are combined to form @xmath14 and @xmath15 . * the combined signals are detected by the square - law detectors ( diodes ) . the outputs , @xmath16 and @xmath17 , are proportional to @xmath18 and @xmath19 , respectively . * finally , we compute the difference between @xmath16 and @xmath17 , obtaining @xmath20 where @xmath21 is the proportionality constant of the square - law detector . this quantity is thus proportional to the _ difference _ between the powers of light coming from the a and b sides ; i.e. , _ wmap _ measures temperature differences in the sky separated by @xmath2 , and the mean cmb temperature ( 2.725 k ) and many types of undesirable systematic effects cancel out . we need to convert the measured @xmath22 ( in units of voltages ) to a temperature difference in thermodynamic units . we do this by using the dipole anisotropy of the cmb . as earth orbits around sun , the l2 point ( hence _ wmap _ ) also orbits around sun at 30 km / s . this motion creates time - varying dipole anisotropy , creating a sinusoidal signal changing over a year . as we know the mean cmb temperature and the orbital velocity precisely , we know that the amplitude of this signal must be @xmath23 . this fixes the proportionality constant between @xmath22 and @xmath24 , where @xmath25 and @xmath26 are the temperatures toward the a and b sides , respectively . the polarization information is still entangled in @xmath24 measured by each radiometer . to measure polarization , we need to combine the measurements of a pair of radiometers forming each da . let us define @xmath27 and @xmath28 , where @xmath29 and @xmath30 are the outputs of two radiometers . the action of omts gives ( equations 10 and 11 of @xcite ) @xmath31 here , @xmath32 , @xmath33 , and @xmath34 are the stokes parameters describing the incoming waves from the a side . we define the stokes @xmath35 and @xmath36 such that the polarization directions of a pure @xmath35 signal are parallel to either galactic longitudes or latitudes , and the polarization directions of a pure @xmath36 signal are 45 degrees tilted from those of a pure @xmath35 signal . in other words , a pure @xmath35 signal aligns with either the galactic north - south or east - west direction ( see figure [ fig : stokes ] ) . then , @xmath37 is the angle between a meridian through the galactic poles and the projection of the electric field of each output port of the omts on the sky . the sum and difference of the instantaneous outputs of two radiometers thus yield @xmath38 the sum gives the temperature difference ( i.e. , difference of unpolarized intensities ) , while the difference gives a combination of the stokes @xmath35 and @xmath36 . the polarization angles of radiometers were measured on the ground using a polarized source . the uncertainty of the measurement is @xmath39 ( section 2.5 of @xcite ) , and the measurements are within @xmath40 of the design orientation ( section 3 of @xcite ) . in flight , we observe tau a @xcite , and find that the standard deviation of angles measured in five bands is @xmath41 ( see table 15 of @xcite ) . this is consistent with ( and is smaller than ) the scatter of the ground measurements and we conservatively use @xmath42 as an estimate of the systematic error in the polarization angle of _ the instantaneous outputs of two radiometers per da yield the temperature and polarization differences between the a and b sides . the next step is to reconstruct the distribution of temperature ( minus the mean cmb temperature ) and polarization over the entire sky . _ wmap _ scans the full sky in six months . we can then estimate maps of temperature and polarization over the full sky using the six - month data . we first write down the measured time - ordered data ( tod ) as ( section 3.4 of @xcite ) @xmath43 where @xmath44 is the tod of two radiometers , @xmath45 , measured at a given observation time , @xmath46 ; @xmath47 is the actual sky map consisting of @xmath48 at a given sky location ( pixel ) , @xmath49 ; @xmath50 is noise of the tod ; and @xmath51 is the so - called `` mapping matrix '' which projects @xmath47 onto @xmath44 . for an ideal differential experiment , the mapping matrix is a @xmath52 matrix , where @xmath53 is the number of data - recording times and @xmath54 is the number of sky pixels . each row corresponds to one observation , while each column corresponds to a map pixel . each row of @xmath51 has 6 non - zero elements . the non - zero elements are @xmath55 in the columns corresponding to the observed pixels in the @xmath56 map ; and @xmath57 , @xmath58 , @xmath59 , and @xmath60 for the @xmath35 and @xmath36 maps . the plus and minus signs are chosen according to equations ( [ eq : d1 ] ) and ( [ eq : d2 ] ) . each observation is associated with 12 non - zero values of @xmath51 that are distributed in two rows . in reality , there are two dominant nonidealities in radiometers that must be taken into account . one is the `` bandpass mismatch . '' we take the difference between two radiometers to measure polarization . while these radiometers were designed and built to have nearly identical frequency responses ( bandpass ) to the incoming electromagnetic waves , a slight mismatch in the bandpass produces a spurious polarization signal even in the absence of polarization . suppose that the incoming waves are unpolarized and have a spectrum of @xmath61 . due to the bandpass mismatch of two radiometers , they receive the incoming waves at slightly different effective frequencies , @xmath62 and @xmath63 . as a result , the difference between two radiometers does not vanish , producing a spurious polarization , @xmath21 , given by @xmath64 . while the cmb , whose temperature does not depend on frequencies , does not produce a spurious polarization , the other components ( such as galactic emission ) that depend on frequencies do produce a spurious polarization . fortunately , it is relatively straightforward to remove this effect . equations ( [ eq : d1 ] ) and ( [ eq : d2 ] ) show that the real polarization signals are modulated by the angle @xmath65 . on the other hand , a spurious polarization is independent of @xmath65 . therefore , we can separate the real and spurious polarization signals if we have enough coverage in @xmath65 . we modify equations ( [ eq : d1 ] ) and ( [ eq : d2 ] ) as ( equations 17 and 18 of @xcite ) @xmath66 and expand the mapping matrix to a @xmath67 matrix . ( each row has 8 non - zero elements . ) while _ wmap _ s scan pattern allows for a uniform coverage in @xmath65 near the ecliptic poles , it covers only 30% of possible @xmath65 on the ecliptic plane . this produces noisy modes in the reconstructed sky maps , which must be properly de - weighted . ( by comparison , _ planck _ s coverage is @xmath68% on the ecliptic plane . ) the second non - ideality is the `` transmission imbalance , '' which is the difference between the a and b sides ; namely , the a and b sides do not necessarily have equal responses to the incoming waves due to loss ( i.e. , imperfect transmission ) in the system . the spurious polarization is an additive effect , but the transmission imbalance is a multiplicative effect , given by ( equations 19 and 20 of @xcite ) @xmath69\\ \label{eq : d1ss } & & -(1-x_{im})[i_b+q_b\cos 2\gamma_b+u_b\sin 2\gamma_b+s_b],\\ \nonumber d_2&= & ( 1+x_{im})[i_a - q_a\cos 2\gamma_a - u_a\sin 2\gamma_a - s_a]\\ \label{eq : d2ss } & & -(1-x_{im})[i_b - q_b\cos 2\gamma_b - u_b\sin 2\gamma_b - s_b ] , \end{aligned}\ ] ] where @xmath70 is the transmission imbalance factor , which has been measured using the responses of radiometers to the cmb dipole ( table 2 of @xcite ) . we include the transmission imbalance in the mapping matrix by multiplying the a- and b - side elements by @xmath71 and @xmath72 , respectively . the optimal estimator for a sky map , @xmath73 , that is unbiased and has the minimum variance , is given by ( in matrix notation ) @xmath74 where @xmath75 @xmath76 is the noise matrix of the tod . the tod noise of _ wmap _ is stationary , in a sense that the noise matrix is a function of @xmath77 . the noise matrix is given by @xmath78 with ( equation 21 of @xcite ) @xmath79^{-1}d\omega+k\right\ } , & \delta t<\delta t_{\rm max},\\ 0 , & \delta t\ge \delta t_{\rm max } , \end{array } \right.\ ] ] where @xmath80 is in units of samples , and @xmath81 is the time lag at which the tod noise correlation function ( @xmath82 in the integral ) crosses zero , typically @xmath83 seconds . the coefficients , @xmath84 and @xmath85 , are chosen such that @xmath86 is normalized to unity , and that the mean of @xmath87 over @xmath88 vanishes . the tod noise correlation function , @xmath82 , is measured from the data and a functional form is given in equation ( 4 ) of @xcite . given this noise matrix of the tod , we solve the equation , @xmath89 , to find a sky map solution , @xmath90 , using the conjugate gradient method . as described in section [ sec : temp ] , we use the cmb dipole to convert the input signals ( in voltages ) to thermodynamic temperatures . the uncertainty in this conversion ( calibration uncertainty ) per da is 0.2% for the final nine - year temperature and polarization maps , which has unchanged since the five - year analysis ( section 4 of @xcite ) . figure [ fig : wmap9_temp ] shows the nine - year solutions to the full - sky @xmath56 maps ( minus dipole anisotropy ) , while figures [ fig : wmap9_q ] and [ fig : wmap9_u ] show the stokes @xmath35 and @xmath36 maps , respectively , in five frequency bands . maps in five frequency bands measured by _ wmap _ ( adapted from @xcite ) . maps are smoothed to a common gaussian beam of 2 degrees to suppress noise . see figure [ fig : stokes ] for the correspondence between the signs of @xmath35 and polarization directions.,scaledwidth=100.0% ] but for stokes @xmath36.,scaledwidth=100.0% ] figure [ fig : wmap9_temp ] shows that the distribution of the measured temperatures at high frequencies ( 41 , 61 , and 94 ghz ) at high galactic latitudes are quite similar . this means that , at these frequencies , the temperature data at high galactic latitudes are dominated by the cmb which is independent of frequencies . on the other hand , the data at lower frequencies ( 23 and 33 ghz ) are clearly affected by strong galactic emission , and the data near the galactic plane are dominated by the galactic emission at all five frequencies . also , there are many extra - galactic sources ( most of which are synchrotron sources ) all over the sky , which need to be removed from the cosmological analysis . to reduce the effects of the so - called `` foreground '' emission from our galaxy and extra - galactic sources , we combine two methods : one is to simply mask the pixels which are strongly affected by the foreground emission ; and the other is to estimate and remove the foreground emission from the sky maps . three galactic foreground components are known to dominate in the _ wmap _ frequencies ( section 3 of @xcite ) : synchrotron , free - free , and dust . the antenna temperatures of these three components go ( very approximately ) as @xmath91 , @xmath92 , and @xmath93 , respectively . as a result , synchrotron dominates at the lowest frequencies ( 23 and 33 ghz ) , free - free dominates in some regions in the sky at the intermediate frequencies ( 41 and 61 ghz ) , and dust dominates at the highest frequency ( 94 ghz ) . we define the mask for the temperature data as follows ( section 2 of @xcite ) . we first smooth the 23 ghz ( k band ) map to one degree resolution , and remove an estimate of the cmb from this map . the cmb is estimated by the internal linear combination ( ilc ) method ( section 5.2 of @xcite ) . we then mask the pixels brighter than a certain threshold temperature , until 75% or 85% of the sky is left unmasked . we repeat the same procedure for the 41 ghz ( q band ) map . the masks defined by the k- and q - band maps are added to form two masks , `` kq75 '' and `` kq85 '' masks , depending on how much sky was left unmasked in the k- or q - band map . the mask for extra - galactic sources is created using the locations of the known bright radio sources in the literature ( as described in section 7 of @xcite ) , the sources found in the _ wmap _ nine - year data ( section 5.2.2 of @xcite ) , and additional sources found in the _ planck _ early release compact source catalog at 100 ghz @xcite . an exclusion radius of 1.2 degree is used for sources brighter than 5 jy in any _ wmap _ band , and an exclusion radius of 0.6 degree is used for fainter sources . to further reduce foreground emission in unmasked pixels , we estimate the distribution of the diffuse galactic foreground emission over the full sky and remove it from the _ wmap _ maps . we use the difference between the 23 and 33 ghz maps ( which does not contain the cmb ) for synchrotron emission ; a map of h@xmath94 @xcite corrected for extinction and scattering ( section 5.3.1 of @xcite ) for free - free emission ; and a map of dust emission @xcite . these three maps are simultaneously fit to and removed from the 41 , 61 , and 94 ghz maps , yielding the foreground - reduced maps at these frequencies . finally , we slightly enlarge the kq75 and kq85 masks by further masking the regions which have significant excess in the differences between the foreground - reduced 41 and 61 ghz maps , and 61 and 94 ghz maps . the resulting nine - year temperature masks , `` kq75y9 '' and `` kq85y9 , '' retain 68.8% and 74.8% of the sky , respectively . the former is used for testing gaussianity of the temperature data , while the latter is used for the power spectrum analysis . figures [ fig : wmap9_q ] and [ fig : wmap9_u ] show that the polarization maps at low frequencies are dominated by the galactic emission . we thus mask the regions strongly contaminated by foreground emission , and remove an estimate of the foreground emission from unmasked pixels . two polarized foreground components are known to dominate in the _ wmap _ frequencies @xcite : synchrotron and dust . therefore , one might think that reducing foreground in polarization is easier than in temperature , as we have fewer foreground components . in reality , the polarization analysis is more challenging because the cmb signal in polarization is 10 times fainter than in temperature . while the polarized foreground emission is clearly seen in figures [ fig : wmap9_q ] and [ fig : wmap9_u ] , there is no clear evidence for the cmb signals ; thus , we need more sophisticated statistical analysis to extract faint polarization signals of the cmb . we define the mask of polarization maps using the 23 ghz map ( which is dominated by synchrotron ) and a model of dust emission @xcite . we first create lower resolution @xmath35 and @xmath36 maps at the 23 ghz , each containing 3072 pixels . we then compute the total polarization intensity , @xmath95 . as the presence of noise produces a small positive bias in this quantity , we remove it using our estimate of noise in the maps . finally , we mask the pixels that are brighter than a certain threshold polarization intensity . we choose the threshold to be 0.6 times the mean polarization intensity , and call this mask the `` p06 '' mask . as for dust , we use a map of dust emission constructed from the _ wmap _ temperature data using maximum entropy method ( mem ) ( section 5 of @xcite ) . we choose a threshold in this map to be 0.5 times the maximum value found in the polar caps ( @xmath96 degrees ) , and mask the pixels brighter than this threshold value . we then add this dust mask to the p06 mask . we find that extra - galactic sources are minimally polarized in the _ wmap _ frequencies ; thus , we only mask ten bright sources outside of the p06 mask @xcite . the combined p06 mask including synchrotron , dust , and extra - galactic sources retains 73.2% of the sky for the cosmological analysis . to further reduce foreground emission in unmasked pixels , we use the 23 ghz polarization maps to trace synchrotron . as for dust , we use ( equation 15 of @xcite ) @xmath97,\\ u_{\rm dust}(\hat{n})&=&i_{\rm dust}(\hat{n})g_{\rm dust}(\hat{n})\sin[2\gamma_{\rm dust}(\hat{n})],\end{aligned}\ ] ] where @xmath98 is the same dust map that we used for the temperature analysis @xcite . dust emission is polarized because dust grains are not spherical , and are aligned with coherent magnetic fields in our galaxy such that the semi - major axes of grains are perpendicular to the fields . as a result , the polarization directions of dust emission are perpendicular to the field directions . the dust polarization we observe along a particular line of sight is the projection of multiple polarization signals along the line of sight , which is affected by geometry of the fields . we take this projection effect into account by the function @xmath99 , which is calculated using a simple model of fields given in section 4.1 of @xcite . how do we estimate @xmath100 ? as the polarization directions of synchrotron are also perpendicular to the field directions , we could take the polarization directions at 23 ghz as an estimate for @xmath100 ; however , alignment of dust grains with the fields is not necessarily perfect , which yields some differences between @xmath100 and @xmath101 . therefore , we use the polarization directions measured toward stars . while the intrinsic starlight is usually unpolarized ( or polarized very weakly ) , the observed starlight can be polarized due to selective extinction of the starlight when it passes through regions with dust grains . as it gets more extinction along the semi - major axes of dust grains , the observed starlight polarization direction is precisely orthogonal to the polarization direction of dust emission from the same location . we have compiled the existing measurements of starlight polarization in the literature ( section 4.1.2 of @xcite ) , and created a map of the starlight polarization directions , @xmath102 . we then compute the dust polarization direction as @xmath103 . the correlation between @xmath100 computed in this way and @xmath101 shows that these two angles typically agree to 20 degrees . we simultaneously fit the 23 ghz polarization maps and dust polarization maps to reduce polarized foregrounds from the higher frequency maps at 33 , 41 , 61 , and 94 ghz . due to much lower signal - to - noise ratios in the polarization maps , removal of the polarized foreground presents a great challenge to the _ wmap _ analysis . the error in the foreground removal has a non - negligible impact on the inferred value of the optical depth to the thomson scattering , @xmath104 . for example , an alternative foreground removal method presented in @xcite shifts the value of @xmath104 as much as the 1-@xmath105 statistical uncertainty . the steps we have described so far give us foreground - reduced temperature maps at 41 , 61 , and 94 ghz , and foreground - reduced polarization maps at 33 , 41 , 61 , and 94 ghz . the temperature maps come with the kq75 and kq85 masks depending on the purpose of the cosmological analysis , and the polarization maps come with the p06 mask . in this section , we describe how to measure the angular power spectra from these maps . assuming that the distribution of cmb temperatures in the foreground - reduced maps outside the mask is given by a gaussian distribution , which is a good approximation as described in section [ sec : ng ] , a complete description of the measured cmb temperatures is given by the following probability density distribution function ( pdf ) : @xmath106}{\sqrt{\det[2\pi(s+n ) ] } } , \label{eq : pdf}\ ] ] where @xmath107 is the difference between the cmb temperature toward a direction ( or a sky pixel ) @xmath108 and the mean cmb temperature , @xmath109 k , with an extra index @xmath110 denoting a da and an observed period ( e.g. , first year , second year , etc ) . the total covariance matrix consists of the signal matrix , @xmath111 , and the noise matrix , @xmath112 . these matrices are @xmath113 matrices , where @xmath114 is the number of pixels , @xmath115 is the number of das used for the temperature analysis ( 2 at 61 ghz and 4 at 94 ghz ) , and @xmath116 is the number of years . for _ wmap _ , the noise matrix vanishes unless @xmath117 , i.e. , noise is uncorrelated between different das or years . as _ wmap _ is a differential experiment measuring temperature differences between two points separated by 141 degrees , there is a pixel - to - pixel noise correlation at 141 degrees ( figure 11 of @xcite ) . however , this correlation has a negligible influence on the temperature analysis , as the signal covariance due to the cmb totally dominates at large angular scales where this pixel correlation is important . we thus have a simple description of the noise matrix : @xmath118 ^ 2}{n_{{\rm obs},i}^{(a)}}\delta_{ij}\delta_{ab},\ ] ] where @xmath119 sets the overall noise level per da per observation , and @xmath120 gives the corresponding effective number of observations in a sky pixel @xmath108 . we also set the noise matrix to have a large value in the masked pixels , i.e. , we effectively set the noise level to be infinity at the masked pixels . the cmb signal matrix is given by @xmath121 where @xmath122 is the temperature power spectrum of the cmb , and @xmath123 is the so - called `` beam transfer function , '' which is the legendre transform of a symmetrized beam profile of a given da , and @xmath124 is the usual legendre polynomials . an additional smearing due to pixelization may also be included in @xmath123 . the beam transfer functions of all das are determined by the repeated observations of jupiter and the full physical modeling of the optical system of _ wmap _ @xcite . two jupiter observing seasons of @xmath125 days each occur every 395400 days , and we had 17 seasons of the jupiter data over nine years of operation . the jupiter data are used to determine the inner beam profiles directly out to radii at which the antenna gains drop to 2 , 3 , 5 , 6 , and 9 dbi ( or @xmath126 , @xmath127 , @xmath128 , @xmath129 , and @xmath129 db ) at 23 , 33 , 41 , 61 , and 94 ghz , respectively ( section 3 of @xcite ) . then , the physical optics modeling is used from the inner beams out to 7.0 , 5.5 , 5.0 , 4.0 , and 3.5 degrees from the beam center at 23 , 33 , 41 , 61 , and 94 ghz , respectively . the beam response at greater distances from the beam center ( i.e. , far sidelobes ) has been measured on the ground before launch and in - flight using moon @xcite . the accurate determination of the beam transfer functions is crucial for the accurate recovery of the intrinsic cmb power spectrum , @xmath122 , as the errors we make in @xmath123 propagate directly into @xmath122 . we estimate that the 1-@xmath105 uncertainty in the recovered @xmath122 from the nine - year data due to the uncertainty in the beam transfer functions is 0.6% at @xmath130 . this uncertainty is coherent ( correlated ) over a wide range in @xmath131 , and must be included in the cosmological analysis of the cmb power spectrum . how do we infer cosmological parameters from the cmb maps ? we can calculate @xmath122 as a function of cosmological parameters using the linear boltzmann code such as cmbfast @xcite , camb @xcite , and class @xcite . ideally , we wish to evaluate equation ( [ eq : pdf ] ) directly as a function of cosmological parameters given our knowledge of noise , mask , and beam transfer functions ; namely , we interpret equation ( [ eq : pdf ] ) as the likelihood function of temperature data given cosmological parameters , @xmath132 , where @xmath133 denotes a set of cosmological parameters . we then use bayes theorem to obtain the posterior probability of the parameters given the temperature data as @xmath134 . here , @xmath135 is the prior probability of cosmological parameters , and we take it to be uniform within a certain reasonable range of @xmath133 . we then calculate the best - fit values of @xmath133 and the 68% confidence intervals , etc . another approach is to use equation ( [ eq : pdf ] ) as the likelihood function of temperature data given power spectra , @xmath136 , and obtain the posterior probability as @xmath137 with uniform @xmath138 within a certain reasonable range of @xmath122 . we then evaluate @xmath139 for theoretically computed @xmath122 with various values of @xmath133 , and find the best - fit values of @xmath133 and the 68% confidence intervals , etc . in other words , we write the posterior probability of @xmath133 as @xmath140 , where @xmath141 $ ] because we know how to calculate @xmath122 theoretically as a function of @xmath133 . the difference between these two approaches is that the latter approach produces a convenient intermediate product , @xmath139 , which can be made much faster to evaluate than the full likelihood function using the so - called `` blackwell - rao ( br ) estimator '' @xcite . note that the form of @xmath139 is non - gaussian even though @xmath136 is a gaussian ( equation [ eq : pdf ] ) , as @xmath122 is a quadratic function of temperatures . the central limit theorem makes @xmath139 closer to a gaussian distribution for large values of @xmath131 , but it is important to use the full non - gaussian form of @xmath139 at small values of @xmath131 for the cosmological parameter estimation . while it is certainly possible to obtain @xmath139 using the br estimator out to large values of @xmath131 @xcite , the computational cost is still quite substantial . therefore , the _ wmap _ team has adopted a hybrid approach : we use the br estimator to compute the full @xmath139 at @xmath142 from the ilc map with the kq85 mask ( following section 6 of @xcite ) , and use the so - called `` quadratic maximum likelihood ( qml ) estimator '' @xcite for @xmath143 . the qml estimator , @xmath144 , is obtained by taylor - expanding the logarithm of equation ( [ eq : pdf ] ) up to second order in @xmath145 , and maximizing it with respect to @xmath122 , i.e. , @xmath146 . the solution is @xmath147 where @xmath148 is the spherical harmonics coefficient of a map filtered by @xmath149 , @xmath150^{(a)}(\hat{n})$ ] . the matrix @xmath151 is given by @xmath152 where the repeated indices are summed . the qml estimator gives our best estimate for @xmath122 at each @xmath131 if we use the correct @xmath122 in the @xmath153 matrix . we can therefore improve the performance of the estimator by iterating the estimation : assume some reasonable @xmath122 , estimate @xmath122 , use the estimated @xmath122 to recompute the qml estimator , and repeat . if an incorrect @xmath122 is used , the qml estimator does not give the minimum variance , but it is still unbiased . while the qml estimator gives an estimate , we still need to calculate the form of the posterior distribution of @xmath122 . we know that a gaussian approximation is not accurate enough ; thus , we combine a gaussian distribution and a log - normal distribution with appropriate weights to obtain an improved form of the posterior distribution ( following section 2.1 of @xcite ) . we use the nine - year temperature data at 61 and 94 ghz , which have the highest angular resolutions , to compute @xmath122 . figure [ fig : cltt ] shows the nine - year measurements of @xmath122 along with estimates of the 68% cl error bars . the error bars are calculated as follows . given the form of @xmath139 , we calculate the second - order moment ( variance ) of @xmath122 , and parametrize it as @xmath154 ^ 2 \equiv \frac{2(c_\ell+n_\ell)^2}{(2\ell+1)f_{{\rm sky},\ell}^2},\end{aligned}\ ] ] where @xmath155 shows the contribution from instrumental noise and a parameter @xmath156 may be regarded as the effective fraction of sky used for the analysis at each @xmath131 . we know how to calculate @xmath155 from the known properties of noise and beam transfer functions ; thus , for a given value of @xmath122 , the only unknown quantity is @xmath156 . equation ( [ eq : variance ] ) thus provides definition of @xmath156 , which is a slowly - varying function of @xmath131 ( section 2.2.1 of @xcite ) . this equation shows that there is an irreducible uncertainty even in the absence of noise , @xmath157 $ ] . this is the so - called `` cosmic variance '' term , which arises from the fact that @xmath122 is variance of cmb temperatures , and only @xmath158 samples are available for estimating variance at each @xmath131 . at each @xmath131 in @xmath159 , the points with error bars show the binned values of @xmath122 for clarity . the error bars show the standard deviation of @xmath122 from instrumental noise , @xmath160^{1/2}$ ] . the shaded area shows the standard deviation from the cosmic variance term , @xmath161^{1/2}$ ] ( except at very low @xmath131 where the 68% cl from the full non - gaussian posterior probability is shown ) . the solid line shows the theoretical curve of the best - fit @xmath0cdm cosmological model.,scaledwidth=90.0% ] the polarization analysis is similar to the temperature analysis . we begin with a gaussian pdf for temperature and polarization : @xmath162}{\sqrt{\det[2\pi(s+n ) ] } } , \label{eq : polpdf}\ ] ] where @xmath163 , and the signal matrix contains all the power spectrum combinations such as @xmath164 , @xmath165 , @xmath166 , and @xmath167 ( as well as parity - violating combinations , @xmath168 and @xmath169 , if necessary ) . the explicit expressions are given in appendix of @xcite . the te power spectrum does not add much to the parameter constraints but is included in the model fits . the most important information we obtain from the polarization likelihood is the optical depth , @xmath104 , from the ee power spectrum at @xmath170 . we can evaluate the exact likelihood function given by equation ( [ eq : polpdf ] ) for such low multipoles , using the steps described in appendix d of @xcite . more precisely , we use equation ( [ eq : polpdf ] ) to calculate the likelihood using the data at @xmath171 . we use the polarization maps at 33 , 41 , 61 , and 94 ghz , while we use the ilc map for the temperature . figure [ fig : clee ] shows the likelihood of the ee power spectrum , @xmath172 , for @xmath173 through 7 . through 7 obtained from equation ( [ eq : polpdf ] ) ( adapted from @xcite ) . these data fix the optical depth , @xmath104 . the diamonds show the theoretical values of the best - fit @xmath0cdm cosmological model . , scaledwidth=90.0% ] it is still useful to compute the power spectrum of te at high multipoles . for this we use a simplified approach : we do not weight the temperature maps at 61 and 94 ghz , while we weight the polarization maps at 41 , 61 , and 94 ghz by @xmath174 . we then compute @xmath175 , and deconvolve the effects of the mask and weight following appendix a of @xcite . at each @xmath131 in @xmath176 , the points with error bars show the binned values of @xmath165 for clarity . the error bars show the standard deviation of @xmath165 , which include both the instrumental noise and the cosmic variance . the solid line shows the theoretical curve of the best - fit @xmath0cdm cosmological model.,scaledwidth=90.0% ] figure [ fig : clte ] shows the nine - year measurements of @xmath177 along with estimates of the 68% cl error bars . the error bars are calculated as follows : @xmath178 where @xmath179 and @xmath180 are the noise bias spectra of the temperature and e - mode polarization , respectively , and @xmath181 and @xmath182 are the effective sky fractions of the temperature and e - mode polarization data , respectively . as the temperature and e - mode polarization are correlated , we can create images of e - mode polarization around temperature spots by averaging the polarization data around hot and cold spots . figure [ fig : pol ] shows average images of temperature and polarization data . we find that the polarization data around hot and cold spots exhibit radial and tangential polarization patterns , as predicted by simulations . what is the physics behind them ? first of all , the necessary and sufficient conditions for generating non - zero polarization of the cmb are to have thomson scattering and quadrupolar temperature anisotropy around an electron . frequent thomson scatterings between photons and electrons suppress quadrupolar temperature anisotropy around an electron , and thus we need to wait until the photon decoupling epoch ( at which photons and electrons become less strongly coupled ) to produce polarization . how is then quadrupolar anisotropy around an electron created ? it turns out that polarization ( for scalar modes ) traces a velocity gradient field of the plasma around gravitational potentials . suppose that a packet of the plasma is falling into the bottom of the potential well . due to acceleration , a velocity gradient is generated : the front of the packet falls faster than the back of the packet . therefore , an electron at the center of the packet observes redshifted photons from both the front and back of the packet , whereas there is no redshift or blueshift from the sides of the packet . this produces a quadrupolar radiation pattern ( colder along the motion of the packet and hotter in the perpendicular directions ) , and the produced polarization is parallel to the motion of the packet . the polarization pattern around a spherically symmetric gravitational potential well is _ radial _ , and the magnitude of radial polarization is maximal at twice the sound horizon radius at the decoupling epoch ( or 1.2 degrees in the sky ) from the bottom of the potential well @xcite . as the packet approaches the bottom of the potential well , the packet decelerates because of a pressure gradient . in the adiabatic initial condition , the photon density is high at the bottom of the potential well , producing a pressure gradient to decelerate motion of the plasma falling into the potential well . the front of the packet falls slower than the back of the packet . therefore , an electron at the center of the packet observes blueshifted photons from both the front and back of the packet , whereas there is no redshift or blueshift from the sides of the packet . this produces the opposite quadrupolar radiation pattern ( hotter along the motion of the packet and colder in the perpendicular directions ) , and the produced polarization is _ tangential _ to the motion of the packet . the magnitude of tangential polarization is maximal at the sound horizon radius ( or 0.6 degrees in the sky ) from the bottom of the potential well @xcite . by @xmath183 . the lines show the polarization directions , and their lengths are proportional to the magnitude of polarization . the colors of the polarization images are chosen such that blue and red show the tangential and radial polarization patterns , respectively . the data show the predicted tangential and radial polarization patterns ( @xmath184-mode polarization ) , in excellent agreement with the predictions . the maximum of radial polarization around hot spots occurs at 1.2 degrees from the center , whereas the maximum of tangential polarization around hot spots occurs at 0.6 degrees from the center . figure adapted from http://wmap.gsfc.nasa.gov/media/101079/index.html ( credit : nasa / wmap science team ) . , scaledwidth=80.0% ] these predictions have been confirmed by the _ wmap _ polarization data . the bottom panels of figure [ fig : pol ] show the average polarization directions measured around hot and cold temperature spots . on these angular scales ( a few degrees ) , hot and cold spots correspond to potential wells and hills , respectively . ( the high photon energy density at the bottom of the well overcomes the sachs wolfe effect , turning potential wells into hot spots in the sky . ) therefore , we expect each hot spot to come with the radial and tangential polarization patterns at 1.2 and 0.6 degrees from the center , respectively , and each cold spot to come with the opposite patterns . as the magnitude of polarization is small , _ wmap _ can not detect polarization around each spot ; however , by averaging polarization patterns around many spots , we can detect polarization . there are 12387 hot spots and 12628 cold spots outside the galactic mask in the _ wmap _ seven - year temperature map . averaging the polarization data around these spots , the expected polarization patterns ( shown in the `` simulation '' columns in figure [ fig : pol ] ) are clearly detected in the data ( shown in the `` wmap data '' columns ) @xcite . the te cross power spectrum and the average polarization images offer a powerful , precision test of the standard cosmological model . we fix the basic six cosmological parameters by fitting the temperature power spectrum at @xmath185 and the @xmath184-mode polarization power spectrum at low multipoles . we can then predict the cross power spectrum _ without any more additional free parameters_. the prediction matches with the data at the precision shown in figure [ fig : clte ] and [ fig : pol ] . this is a great triumph of the standard cosmological model . the te cross power spectrum offers also a powerful test of one of the generic predictions of cosmic inflation : the presence of `` super - horizon '' fluctuations , whose wavelength is greater than the horizon size at the decoupling time . this test is possible because polarization is generated only when there are free electrons . the reionization of the universe at @xmath186 can generate polarization only on very large angular scales , @xmath170 ; thus , any te signals at high multipoles must be generated at the decoupling epoch . the angle that subtends the radius of the horizon at the decoupling epoch is 1.2 degrees , which corresponds to @xmath187 . therefore , the anti - correlation seen in the te cross power spectrum at @xmath188 provides the direct evidence for the presence of super - horizon fluctuations at the decoupling epoch , i.e. , the key prediction of inflation @xcite . .six cosmological parameters of the standard flat @xmath0cdm model , determined from the cmb data alone . we show the constraints from the _ wmap _ nine - year temperature and polarization power spectra @xcite ; the _ wmap _ data combined with the _ act _ @xcite and _ spt _ @xcite temperature spectra ( `` _ wmap_+_act_+_spt _ '' ) ; and the _ planck _ 15.5-month temperature power spectrum combined with the _ wmap _ nine - year polarization power spectrum ( `` _ planck_+wp '' ) @xcite . [ cols="^,^,^,^",options="header " , ] [ tab : parameters ] the _ wmap _ nine - year temperature and polarization data are consistent with the minimal six - parameter flat @xmath0cdm model @xcite . the high-@xmath131 temperature power spectrum ( @xmath189 ) gives @xmath190 for 1168 degrees of freedom , with the probability to exceed ( pte ) of 25.1% . the high-@xmath131 te spectrum ( @xmath191 ) gives @xmath192 for 777 degrees of freedom , with pte of 16.5% . therefore , the best - fit @xmath0cdm model is a good fit to the high-@xmath131 data . while the low-@xmath131 temperature likelihood ( @xmath193 ) based on the br estimator does not give a @xmath194 value , we find that the best - fit @xmath0cdm model is consistent with the distribution of spectra generated by the br estimator . the low-@xmath131 polarization likelihood ( @xmath195 ) evaluated directly in pixel space gives @xmath196 for 1170 degrees of freedom with pte of 0.13% , which is unusually low . this excess @xmath194 can be interpreted as an additional noise component ( due to , for example , residual foreground emission ) of @xmath197 per @xmath198 pixel ( @xmath199 on a side ) , which is significantly lower than the average standard deviation of @xmath200 . we confirm that this excess noise does not affect the determination of @xmath104 by using differences bewtween frequencies . see section 7.1 of @xcite for details . we assume flat priors on the following six parameters : the amplitude of the primordial power spectrum at @xmath201 , @xmath202 , the tilt of the primordial power spectrum , @xmath203 , the physical baryon density parameter , @xmath204 , the physical cdm density parameter , @xmath205 , the cosmological constant density parameter , @xmath206 , and the optical depth of the reionization , @xmath104 . table [ tab : parameters ] summarizes the constraints of the cosmological parameters from the cmb data alone . adding the smaller - scale cmb data from the _ atacama cosmology telescope _ ( _ act _ ) @xcite and the _ south pole telescope _ ( _ spt _ ) improves the parameter constraints significantly . in particular , statistical significance of a deviation of @xmath203 from unity increases from @xmath207 to @xmath208 . with additional cosmological measurements ( baryon acoustic oscillation @xcite and the local hubble constant @xcite ) this improves to @xmath209 ( 68% cl ) , a @xmath210 deviation from unity . this is a great achievement in cosmology , providing strong evidence for cosmic inflation . the parameters found from the _ planck _ 15.5-month data combined with the _ wmap _ low-@xmath131 polarization data ( `` _ planck_+wp '' ) are consistent with the _ wmap _ and _ wmap_+_act_+_spt _ parameters to within the quoted error bars . with the _ planck_+wp combination , @xmath211 is detected at @xmath212 . this is the first time that @xmath213 is detected with @xmath214 from the cmb data alone . the _ wmap _ measurements also provide definitive evidence for the existence of non - baryonic dark matter with @xmath215 ( 68% cl ) . this measurement comes from a combination of the ratio of the heights of odd and even acoustic peaks giving @xmath216 and the ratio of the heights of the first and other peaks giving the total matter density contributing to gravitational potential well . in other words , _ wmap _ measures the density of matter which interacts with photons , and which does not . ( no matter is left behind . ) the difference between the two provides definitive evidence for non - baryonic dark matter . _ wmap _ has erased lingering doubts about the existence of dark energy . this measurement comes from the peak positions giving the angular diameter distance , @xmath217 , where @xmath218 is the redshift of the photon decoupling . the friedmann equation relates the hubble expansion rate , @xmath219 , to the total energy density in the universe . as the integral is dominated by low redshift contributions , this provides an estimate of the total energy density in a local universe . as we have the complete account of matter density in the universe at any redshifts after @xmath220 , the difference between the total energy density inferred from @xmath221 and the total matter density gives the energy density of some substance which is _ not even matter _ , i.e. , dark energy . strictly speaking , this measurement is possible if we assume flatness of the universe or combine the _ wmap _ data with other cosmological data ( such as the local hubble constant measurements ) . alternatively , we can use the effects of gravitational lensing on the cmb to detect dark energy from the cmb data alone @xcite . the minimal six - parameter model fits all the data we have at the moment ( with a possible exception of the tensor - to - scalar ratio , @xmath222 , which the bicep / keck array collaboration claims to have found recently from the b - mode polarization at degree angular scales @xcite ) . as a result , the _ wmap _ data ( sometimes in combination with other cmb and non - cmb data ) place stringent limits on the parameters beyond the minimal model . spatial geometry of the universe is consistent with flat ( euclidean ) space . by combining the _ wmap_+_act_+_spt _ with the cmb lensing data , we find @xmath223 ( 68% cl ) . when other non - cmb data ( baryon acoustic oscillation and the local hubble constant measurements ) are added , we find a stringent limit of @xmath224 ( 68% cl ) , i.e. , 0.4% measurement . dark energy is consistent with a cosmological constant . the constraints on the equation of state parameter , @xmath225 , are consistent with @xmath226 typically to within 10% ( 95% cl ) , depending on the data combinations . the cosmic neutrino background affects temperature anisotropy of the cmb in four ways : peak locations , early integrated sachs - wolfe effect , anisotropic stress , and enhanced damping tail ( see section 4.3.1 of @xcite for summary ) . using these effects and the _ wmap _ five - year data , we have made the first ( indirect ) detection of the cosmic neutrino background @xcite . the cmb data give the total energy density of neutrinos , @xmath227 , where @xmath228 is the effective number of neutrino species . assuming the standard thermal history of the universe relating the asymptotic neutrino temperature to the cmb temperature as @xmath229 , we use the nine - year data combined with _ act _ and _ spt _ to find @xmath230 ( 68% cl ) , consistent with the standard value of 3.046 to within @xmath231 . the damping tail of the cmb is sensitive to the primordial helium abundance , @xmath232 . the more helium we have , the more electrons are captured by helium nuclei before the decoupling , the fewer electrons are available at the decoupling , the more diffusion damping results . we have made the first detection of this effect by combining the _ wmap _ seven - year data and the small - scale cmb data @xcite . the nine - year data combined with _ act _ and _ spt _ give @xmath233 ( 68% cl ) , consistent with the standard value of 0.25 to within @xmath231 . these measurements of @xmath228 and @xmath232 offer a unique test of the big bang nucleosynthesis @xcite . our measurements are consistent with the standard big bang nucleosynthesis calculations . the _ wmap _ nine - year data alone place a limit on the sum of neutrino masses , @xmath234 ev ( 95% cl ) . adding the baryon acoustic oscillation and the local hubble constant measurements , the limit improves to @xmath235 ev ( 95% cl ) . single - field inflation models predict that the fluctuations in matter and photons trace each other , obeying the adiabatic relation of @xmath236 . we find that this relation holds to better than 7% ( 95% cl ) . this limit plays an important role in constraining the parameter space of axion dark matter models @xcite . the shape of the primordial power spectrum is sensitive to the physics of inflation . the `` running spectral index , '' @xmath237 , is typically predicted to be of order @xmath238 . wmap _ nine - year data alone give @xmath239 , while adding the small - scale cmb data improves the limit to @xmath240 ( 68% cl ) , consistent with a power - law power spectrum to within @xmath231 . finally , inflation generates nearly scale - invariant tensor mode metric perturbations ( gravitational waves ) @xcite , @xmath241 , which also contribute to the observed temperature and polarization anisotropies of the cmb . the amplitude of @xmath241 is parametrized by the `` tensor - to - scalar ratio , '' @xmath222 , defined by @xmath242 . the _ wmap _ data alone give @xmath243 ( 95% cl ) , which improves to @xmath244 by adding the small - scale cmb data . adding the baryon acoustic oscillation measurements improves the limit further to @xmath245 ( 95% cl ) . this limit largely comes from the low multipole temperature data ( see section 3.2.3 of @xcite ) , and thus it is sensitive to our assumption of a power - law power spectrum . including the running index relaxes the limits to @xmath246 ( 95% cl ) with a large running index , @xmath247 , more or less independent of the data sets used . _ wmap _ did not have sufficient sensitivity to detect b - mode polarization . recently the bicep / keck array collaboration claimed to have found b - mode polarization at degree angular scales at 150 ghz . if this signal is cosmological and originates from gravitational waves from inflation , it corresponds to @xmath248 @xcite . as the measurement was done only at one frequency ( the bicep1 data at 100 ghz are too noisy to be useful ) , confirmation of the signal at other frequencies must be made to reject foregrounds . in any case , an independent detection from an independent group is required before interpreting the detected signal as inflationary . inflation predicts that primordial fluctuations originate from quantum fluctuations , and the distribution of primordial fluctuations is nearly a gaussian distribution ( see @xcite for a review ) . sustained inflationary expansion for at least 50 @xmath249-folds requires a field driving inflation to be weakly coupled . the wave function of quantum fluctuations of a scalar field with no interaction in the ground state is precisely a gaussian ; thus , a weakly coupled field is nearly a gaussian field . the linear physics preserves gaussianity , and thus cmb temperature and polarization anisotropies are predicted to obey gaussian statistics with high precision . confirmation of this prediction gives strong evidence for the quantum origin of primordial fluctuations . when the distribution is not a gaussian , the pdf is no longer given by equation ( [ eq : pdf ] ) . when a departure from gaussianity , i.e. , non - gaussianity , is small , we may approximate the pdf by `` taylor - expanding '' around a gaussian distribution . let us do this in harmonic space we obtain an expanded pdf for the spherical harmonics coefficients as @xcite @xmath250\\ & & \times \frac{e^{-\frac12\sum_{\ell m}\sum_{\ell'm'}a^*_{\ell m}(c^{-1})_{\ell m,\ell'm'}a_{\ell'm'}}}{\sqrt{\det(2\pi c)}},\end{aligned}\ ] ] where @xmath251 is the signal plus noise covariance matrix in harmonic space . ( we do not write indices for das or years for simplicity . ) here , the expansion is truncated at the three - point function ( bispectrum ) of @xmath252 , and thus we have assumed that the connected four - point and higher - order correlation functions are negligible compared to the power spectrum and bispectrum . ( this condition is not always satisfied . ) by evaluating the above derivatives , we obtain . ] @xmath253\\ \nonumber & \times & \left\ { 1+\frac16\sum_{{\rm all}~\ell_im_j } \langle a_{\ell_1m_1}a_{\ell_2m_2}a_{\ell_3m_3}\rangle \left [ ( c^{-1}a)_{\ell_1m_1}(c^{-1}a)_{\ell_2m_2}(c^{-1}a)_{\ell_3m_3 } \right.\right.\\ & & \left.\left . -3(c^{-1})_{\ell_1m_1,\ell_2m_2}(c^{-1}a)_{\ell_3m_3 } \right]\right\}. \label{eq : ngpdf}\end{aligned}\ ] ] this formula is useful , as it tells us how to estimate the _ angular bispectrum _ , @xmath254 , optimally from given data by maximizing this pdf . in practice , we usually parametrize the bispectrum using a few parameters ( e.g. , @xmath255 ) , and estimate those parameters from the data by maximizing the pdf with respect to the parameters . in the limit that the contribution of the connected four - point function ( trispectrum ) to the pdf is negligible compared to those of the power spectrum and bispectrum , equation ( [ eq : ngpdf ] ) contains all the information on non - gaussian fluctuations characterized by the covariance matrix , @xmath256 , and the angular bispectrum , @xmath254 . this approach can be extended straightforwardly to the trispectrum if necessary . as the bispectrum has three angular wavenumbers , @xmath257 , @xmath258 and @xmath259 , it can form triangles with various shapes . among all the shapes , the so - called `` local - form bispectrum , '' parametrized by a non - linear parameter @xmath260 @xcite , carries a special significance , as detection of a large local - form bispectrum would rule out all inflation models based on a single energy component with a bunch - davies initial vacuum state and an attractor solution ( see @xcite for the latest discussion on this theorem ) . this triangle has the largest amplitude in the `` squeezed configurations '' in which one of the wavenumbers , say @xmath259 , is much smaller than the other two , i.e. , @xmath261 @xcite . detailed descriptions on what this bispectrum is and what the other shapes are , as well as on how to measure them can be found in @xcite . using the foreground - reduced _ wmap _ nine - year temperature data at 61 and 94 ghz with the kq75 mask , we find @xmath262 ( 68% cl ) , which is consistent with zero to within @xmath231 ; thus , the measurement agrees with the basic prediction of single - field inflation models with a bunch - davies initial vacuum state and an attractor solution . the _ planck _ improves this limit greatly by finding @xmath263 ( 68% cl ) @xcite . one way to generate the local - form bispectrum is to write the primordial curvature perturbation as @xmath264 . this form is called the `` local form '' because both sides are evaluated at the same spatial location , @xmath265 . here , @xmath266 is a gaussian random field , and the curvature perturbation is defined such that the linear sachs wolfe effect gives @xmath267 . using this form and the fact that the variance of @xmath268 is @xmath269 , we find that the 95% upper bound from _ planck _ , @xmath270 , implies that the observed cmb is gaussian to the precision of 0.04% or better . this is a remarkable degree of gaussianity , which provides strong evidence that the observed cmb fluctuations originate from quantum fluctuations generated during single - field inflation models of inflation @xcite make specific , testable predictions . the simplest models based upon a single energy component ( scalar field ) , which slowly rolls down on its potential and drives a sustained quasi - exponential expansion for at least 50 @xmath249-folds , predict that the observable universe is homogeneous and isotropic with flat geometry , and is filled with small fluctuations which are precisely adiabatic and nearly gaussian ( before fluctuations become non - linear ) . both scalar and tensor fluctuations with various wavelengths are generated during inflation . the wavelengths of these fluctuations can exceed the horizon size at the decoupling epoch , and the amplitude of these fluctuations weakly depends on wavelengths . _ all _ of these predictions fit the _ wmap _ data remarkably well : flatness is measured with 0.4% precision ( from _ wmap _ combined with the baryon acoustic oscillation and the local hubble constant measurements ) ; the adiabatic condition holds to better than 7% precision ; a deviation from gaussian fluctuations is restricted to be less than 0.2% ( and 0.04% with the _ planck _ 2013 data ) ; and the presence of super - horizon fluctuations at decoupling is decisively detected in the te cross power spectrum at @xmath188 . the _ wmap _ data combined with the baryon acoustic oscillation and the local hubble constant measurements find convincing evidence for the scale dependence of the scalar initial power spectrum with 4.9@xmath105 significance , with the best - fit value in agreement with the first prediction made in @xcite . while _ wmap _ did not find signatures of tensor fluctuations , the upper bound on the tensor - to - scalar ratio inferred from the temperature data is consistent with many single - field inflation models . figure [ fig : ns - r ] compares the limits on @xmath203 and @xmath222 with a few representative single - field inflation models . , and the tensor - to - scalar ratio , @xmath222 . the red contours show the constraint from the _ wmap _ nine - year data combined with the small - scale cmb temperature data ( _ act _ and _ spt _ ) , the baryon acoustic oscillation data , and the local hubble constant . the blue contours show the constraint from the _ planck _ 15.5-month temperature data combined with the _ wmap _ nine - year polarization data and the baryon acoustic oscillation data . the symbols show the predictions from single - field inflation with monomial potentials , @xmath271 @xcite , with @xmath272 ( black ) , @xmath273 ( white ) , and @xmath274 ( light grey ) , and with a @xmath275 term in the gravitational action ( dark grey ) @xcite.,scaledwidth=90.0% ] the nine years of observations of the _ wmap _ satellite have taught us many things . the current universe is 13.77 billion years old , and consists of 4.6% atoms , 24% cold dark matter , and 71% dark energy @xcite . the nature of dark energy is consistent with that of a cosmological constant . the spatial geometry of the universe is consistent with euclidean geometry . the universe is filled with neutrinos , whose abundance is consistent with the standard model of particle physics . the mass of neutrinos is much less than 1 ev . the measured properties of primordial fluctuations such as adiabaticity , gaussianity , and near scale invariance all point toward a remarkable scenario : the observed fluctuations originate from quantum fluctuations generated during inflation driven by a single energy component . _ wmap _ offered a number of stringent tests of the simplest inflation scenarios : ( 1 ) flat universe , ( 2 ) adiabatic fluctuations , ( 3 ) super - horizon fluctuations , ( 4 ) nearly , but not exactly , scale - invariant initial power spectrum , and ( 5 ) gaussian fluctuations . the simplest scenarios passed all of these tests . the _ planck _ 2013 data have confirmed all of these findings with greater precision . yet , neither the _ wmap _ nor the _ planck _ 2013 data detect the signature of primordial gravitational waves from inflation in cmb . detecting and characterizing the b - mode polarization of the cmb is the next milestone in cosmology . while the bicep / keck array collaboration claims to have found the b - mode polarization from inflationary gravitational waves at 150 ghz , confirmation of the signal at other frequencies and with an independent experiment must be made before we claim a victory in observing all of the inflation predictions . the _ wmap _ mission was made possible by the support of the science mission directorate office at nasa headquarters . this research has made use of nasa s astrophysics data system bibliographic services . all of the scientific products of the _ wmap _ mission are made publicly available at the legacy archive for microwave background data analysis ( lambda ; http://lambda.gsfc.nasa.gov/ ) . support for lambda is provided by nasa headquarters . we acknowledge use of the healpix @xcite , camb @xcite , and cmbfast @xcite packages . ek would like to thank the members of the _ wmap _ science team for letting him join the team and for wonderful collaborations over the last 13 years . ek s work has been supported in part by an alfred p. sloan research fellowship , nasa grants nnx08al43 g and nnx11ad25 g , and nsf grants ast-0807649 and phy-0758153 . clb thanks the entire _ wmap _ science team for the many years of the close , careful , and productive collaboration that advanced cosmology in a significant manner .

we are conducting a search for supermassive black holes ( smbhs ) with masses below by looking for signs of extremely low - level nuclear activity in nearby galaxies that are not known to be agns . our survey has the following characteristics : ( a ) x - ray selection using the _ chandra _ x - ray observatory , since x - rays are a ubiquitous feature of agns ; ( b ) emphasis on late - type spiral and dwarf galaxies , as the galaxies most likely to have low - mass smbhs ; ( c ) use of multiwavelength data to verify the source is an agn ; and ( d ) use of the highest angular resolution available for observations in x - rays and other bands , to separate nuclear from off - nuclear sources and to minimize contamination by host galaxy light . here we show the feasibility of this technique to find agns by applying it to six nearby , face - on spiral galaxies ( ngc 3169 , ngc 3184 , ngc 4102 , ngc 4647 , ngc 4713 , ngc 5457 ) for which data already exist in the _ chandra _ archive . all six show nuclear x - ray sources . the data as they exist at present are ambiguous regarding the nature of the nuclear x - ray sources in ngc 4713 and ngc 4647 . we conclude , in accord with previous studies , that ngc 3169 and ngc 4102 are almost certainly agns . most interestingly , a strong argument can be made that ngc 3184 and ngc 5457 , both of type scd , host agns .

0801.4382

the past decade has seen extraordinary improvement in our understanding of supermassive black holes ( smbhs ) their growth and evolution , and their links with their host galaxies . we now realize that galaxies hosting smbhs at their centers are the rule rather than the exception . the observed correlations of the masses @xmath1 of the smbhs with properties of their host galaxies , for example with the bulge stellar velocity dispersion @xcite , show that there is a close link between the formation and evolution of galaxies and of the smbhs they host . furthermore , a comparison of the smbh mass function required to explain the observed luminosity function of active galactic nuclei ( agns ) with estimates of the local smbh mass function ( e.g. * ? ? ? * ; * ? ? ? * ) shows that not only must smbhs be very common in massive galaxies but that most , if not all , of these black holes are relics of agns active in previous epochs . thus knowledge of the local smbh mass function enables us to put constraints on theories of galaxy and smbh formation and growth , and agn lifetimes . estimates of the local smbh mass function are anchored by resolved stellar or gas dynamical measurements of the masses of @xmath2 smbhs ( see , e.g. , the review by * ? ? ? * ) , and otherwise based on the distribution of host galaxy properties ( luminosity of the bulge or bulge stellar velocity dispersion @xmath3 ) and known scaling relationships between the mass of the smbh and these properties ( most prominently @xmath4 and @xmath5 ; but see @xcite for a more direct estimate ) . most of the measured smbh masses are @xmath6 or greater , however , as the sphere of influence of a less massive smbh is extremely hard to resolve even at moderate distances , even with _ hst_. for example , the sphere of influence of a @xmath7 smbh at 15 mpc is @xmath2 milliarcseconds ( mas ) . as a result , while different estimates of the mass function agree for @xmath8 , the low - mass end ( @xmath9 ) often has discrepancies ( see , e.g. , fig . 7 of * ? ? ? * for a comparison of different authors estimates of the mass function ) . a second source of uncertainty at the low - mass end is the fact that it is unknown how the scaling relationships extrapolate to very late - type spiral galaxies , which have little or no bulge component , and to very low mass galaxies ( de and dsph ) . yet smbhs _ do _ exist in very late - type spirals , e.g. ngc 4395 , a spiral galaxy of type sdm , with @xmath10 @xcite , and in very low mass galaxies , e.g. pox 52 , a dwarf galaxy , with @xmath11 @xcite . questions that naturally arise at this point are : do the scaling relationships break down at low masses ? what determines the mass of an smbh : the mass of the bulge or the mass of the dark matter halo ? is there a lower bound to the local smbh mass function ? a well defined sample of low - mass smbhs is needed to answer these questions . since we can not detect low - mass smbhs by their dynamical signature , looking for them by signs of their accretion activity may be the only viable way of detecting them . this of course limits detection to the subset of low - mass black holes that are active , but this fraction can be expected to be large , for the following reason . we now understand that the `` quasar era '' is a function of luminosity , with the space density of the most luminous quasars peaking at high redshift and that of lower luminosity quasars peaking at progressively lower redshifts @xcite . this is often referred to as the `` downsizing '' of agn activity with cosmic epoch . moreover , at least some models of black hole growth @xcite require anti - hierarchical growth . that is , higher mass smbhs attain most of their mass at high redshift while lower mass smbhs grow at progressively lower redshifts . the trends of agn downsizing and anti - hierarchical growth , extended logically to the smallest mass smbhs , imply that these objects were active in recent times and may still be accreting at the present epoch . the eddington luminosity of a @xmath12 smbh is only @xmath13 erg s@xmath14 ; low - mass smbhs , even if accreting at high rates , will be low - luminosity agns ( llagns ) . the converse is not true ; that is , not all llagns have a low - mass smbh . in addition to a low smbh mass , the low luminosity of an llagn may be caused by a low rate or radiatively inefficient mode of accretion to an smbh of any mass ( * ? ? ? * ; * ? ? ? * present a study of massive smbhs accreting at very low rates ) . finally , obscuration may further lower the observed luminosity . low - ionization nuclear emission - line region ( liner ) nuclei have been studied at multiple wavelengths ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to identify llagns among them , and these studies have demonstrated that these agns are not necessarily the same type of object with the same physical characteristics . since we are interested specifically in the low - mass end of the smbh mass function , it is necessary to identify among the agns those that can be expected to have the smallest black holes . one approach is to use mass estimators based on the luminosity of the agn and the width of the broad component of emission lines in its spectrum . this is the technique that was used by @xcite in constructing their sample of low - mass smbhs . a second approach , and the one we use , is to look for agn activity in galaxies ( late - type spirals , dwarf galaxies ) where the known host galaxy - smbh scaling relationships predict the lowest - mass smbhs would reside . this approach has been used by @xcite to find candidate low - mass smbhs . however , the scaling relationships are only statistical and can not be used to estimate the mass of a particular smbh ; thus the second approach requires an independent estimate of the smbh mass . a search for active low - mass smbhs requires confirmation of the presence of an agn in each candidate nucleus , and measurement of the mass of the smbhs in the confirmed agns . the method used by @xcite has the virtue of effectively combining all of the above into a single step . however , as the luminosity of an agn decreases , the optical spectrum of the galaxy nucleus becomes more and more dominated by host galaxy light , and the signature of the agn becomes difficult to detect . even when the optical spectrum shows no clear evidence of an agn , however , such evidence may still be present in other wavelengths , such as x - ray and radio ( e.g. * ? ? ? * ) , and infrared ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ( see * ? ? ? * for a review of nuclear activity in nearby galaxies ) . for the lowest - luminosity agns , therefore , it is possible that a system based on optical spectra would not classify the nuclei as agns at all . we choose to use x - ray selection to identify candidate agns for the following reasons : first , x - rays can penetrate obscuring material which may be hiding the line emitting regions . second , there are fewer sources of x - rays in a galaxy than there are of optical and uv emission and so dilution of the agn signature by host galaxy light is less of a problem . where the luminosity of the agn is low to begin with , even a modest amount of absorption may result in the signal being below the optical background imposed by the galaxy . third , even if , as expected in some theories @xcite agns that have luminosities or accretion rates below a cut - off value do not have broad - line regions , they should still be detectable in x - rays . x - ray observations have in fact detected agns in what were thought to be `` normal '' galaxies ( e.g. * ? ? ? the disadvantage , as discussed in [ sec : disc ] , is that x - ray observations by themselves can not always distinguish between agns and other x - ray sources , such as x - ray binaries ( xrbs ) and ultraluminous x - ray sources ( ulxs ) . multi - wavelength data are needed to determine the type of source . as the first step towards assembling a sample of low - mass smbhs , we are conducting an x - ray survey of nearby ( within 20 mpc ) , _ quiescent _ spiral galaxies using the _ chandra _ x - ray observatory to look for nuclear x - ray sources , with an emphasis on late - type spirals . the high angular resolution of _ chandra _ is necessary to ensure that any detected source is really at the center and is not an off - nuclear source . the survey is sensitive to an ( unobscured ) smbh of mass @xmath15 radiating at @xmath16 out to the survey limit . we will present details of the survey sample selection and the _ chandra _ observations in a future paper . here , we present the methods used and the feasibility of detecting agns using these methods by applying them to six galaxies that meet selection criteria similar to those used for the survey sample , and for which x - ray data already exist in the _ chandra _ archive . the paper is organized as follows : [ sec : sampsel ] describes the criteria used to select the six galaxies presented here ; [ sec : datan ] describes the observations and data analysis , with individual targets discussed in [ sec : n3169][sec : n5457 ] ; the results are discussed in [ sec : disc ] . starting with third reference catalogue of bright galaxies ( rc3 ; * ? ? ? * ) , we imposed the following main filters . ( 1 ) morphological type : we selected spiral galaxies of type sa through sd ( @xmath17 ) ; ( 2 ) distance : galaxies that had recession velocity @xmath18 km s@xmath14 ; ( 3 ) galactic latitude : @xmath19 to avoid x - ray absorption by gas in our own galaxy ; ( 4 ) inclination : we required @xmath20 , where @xmath21 and @xmath22 are the projected lengths of the semi - major and semi - minor axes , respectively . this was to avoid obscuration by the disk of the host galaxy ; ( 5 ) nuclear inactivity : the galaxy must _ not _ be known to be an agn or starburst . starbursts were excluded since the vigorous star formation also increases the likelihood of the presence of xrbs and ulxs . liners were included as there is still debate about whether the fundamental source of energy in these nuclei is an agn or star formation . we also required that the object have the value of the `` goodposition '' flag set to 1 in the rc3 table in the sloan digital sky survey ( sdss ) database , which indicated that the coordinates of the galaxy were accurate to 0.1 s in ra and @xmath23 in dec . all objects passing these filters were cross - correlated with sdss data release 3 , with a match radius of @xmath24 . this resulted in a list of 76 galaxies . this list was then checked against publicly available data in the _ chandra _ archive . the final sample consists of six galaxies : ngc 3169 ( sa ) , ngc 3184 ( scd ) , ngc 4102 ( sb ) , ngc 4647 ( sc ) , ngc 4713 ( sd ) , and ngc 5457 ( scd ) . details of the observations that were analyzed are given in table [ tab : obs ] . x - ray data were downloaded from the _ chandra _ archive and analyzed using version 3.4 of _ chandra _ interactive analysis of observations ( ciao ) software . in each case the level 2 event list was processed using the observation - specific bad - pixel file and the latest calibration files , and filtered to exclude times when the instrument experienced a background flare . the event lists were then converted to fits images and the ciao wavelet source detection tool _ wavdetect _ was run on them to determine source positions . for each observation , source counts were extracted from a circle centered on the _ wavdetect _ source position , and with a radius equal to the greater of 4.67 pixels ( @xmath25 ) and the 95% encircled - energy radius at 1.5 kev on acis - s at the position of the source . background counts were taken from an annulus with an inner radius of twice , and an outer radius of five times , the source circle radius after excising any sources that happened to fall in the background region . counts were extracted in the 0.3 8.0 kev ( broad ) , 2.5 8.0 kev ( hard ) and 0.3 2.5 kev ( soft ) bands . a hardness ratio ( hr ) was defined as @xmath26 , where @xmath27 and @xmath28 are the counts in the hard and soft bands , respectively . spectral fits were performed with the ciao tool _ sherpa_. uncertainties reported in fit values represent the 90% confidence level for one parameter . in all estimations of the agn eddington ratio we have assumed that the bolometric luminosity is ten times the 210 kev luminosity of the agn . where available , optical and uv data from hst ( wfpc2 ) were downloaded from the archive and analyzed using iraf to extract nuclear fluxes . we note here that we use `` nuclear flux '' to refer to the total background - subtracted flux at the location of the nucleus . we use this flux , and colors if observations in more than one band exist , only as a consistency check , not to obtain photometry of any putative agn . at these extremely low luminosities ( @xmath29@xmath30 erg s@xmath14 ) , the contribution of the agn , even if present , will be a small fraction of the nuclear flux in the optical and uv bands . in most cases , therefore , we expect the fluxes and colors to be consistent with the _ absence _ of an agn . we check for anomalous colors or fluxes , since these may indicate the presence of an additional source of flux , possibly an agn . as such , we perform simple aperture photometry for resolved nuclear sources , with the aperture size matched to the source size . for point sources background - subtracted source counts were determined using ` imexam ` , using a 2-pixel radius aperture for the source , and an annulus with an inner radius of 7 pixels and an outer radius of 12 pixels for the background . these were corrected for cte effects @xcite then extrapolated to a @xmath31 radius aperture using the encircled - energy tables of @xcite . a correction of @xmath32 mag was added to represent the magnitude in an infinite aperture . proprietary acs data for ngc 4713 were made available by martini et al . . the flux of the resolved nuclear source was estimated using aperture photometry on the drizzled image . we use infrared fluxes from the two - micron all sky survey ( 2mass ; * ? ? ? * ) point source catalog ( psc ) for similar consistency checks . individual targets are discussed below . for each target , we first describe the observed properties in the x - ray , and then in other wavebands when such observations exist . we then consider whether the evidence supports the hypothesis that the x - ray source is an agn . this is a galaxy of type sa , at a distance of 19.7 mpc @xcite , and its nucleus is classified as a liner @xcite . the _ chandra_observation of the nucleus of ngc 3169 was analyzed by @xcite and the nucleus identified as a low luminosity agn , but the x - ray data are analyzed and presented here again for completeness . ngc 3169 was observed by _ chandra _ for 2 ks and detected with 159 counts . its hardness ratio @xmath33 makes it the hardest of the six sources discussed here and the only one to have a positive hr . its spectrum , shown in fig . [ fig : n3169s ] , can be reasonably fit by an absorbed power - law , with @xmath34 @xmath35 and @xmath36 . spectral fit parameters are shown in table [ tab : spec ] , with the first row showing a fit using the cash statistic and data binned to five counts per bin , and the second row a fit using the gehrels @xmath37 statistic and data binned by 20 pha channels . best - fit parameter values from both fits are consistent within errors with each other and with the results obtained by @xcite . fluxes used in the analysis are from the cash fit . the unabsorbed broad band flux is @xmath38 erg @xmath35 s@xmath14 , implying a luminosity @xmath39 erg s@xmath14 for the assumed distance . observations with the very large baseline array ( vlba ) show the nucleus is a mas - scale radio source at 5 ghz @xcite . the nuclear flux density is @xmath40 erg @xmath35 s@xmath14 hz@xmath14 , or @xmath41 erg s@xmath14 . the measured brightness temperature exceeds @xmath42 k and rules out starbursts and supernova remnants @xcite . the nucleus was detected by 2mass . the observed magnitudes @xmath43 , @xmath44 , @xmath45 imply luminosities @xmath46 erg s@xmath14 in these bands . the combination of being a sub - parsec scale radio source as well as a luminous hard x - ray source points to the source being an agn . further evidence comes from the high luminosity . the source is obscured , as made clear from the x - ray spectrum . the infrared luminosity therefore is a better indication of the true bolometric luminosity . this luminosity , a few @xmath47 erg s@xmath14 , is higher than is expected from xrbs or nuclear star formation regions . @xcite perform a bulge - disk decomposition of ngc 3169 and use the 2mass k - band luminosity of the galaxy to estimate a central black hole mass @xmath48 . @xcite report the central stellar velocity dispersion @xmath49 km s@xmath14 , which implies @xmath50 . assuming @xmath51 for simplicity , the corresponding @xmath52 erg s@xmath14 . the inferred @xmath53 erg s@xmath14 then implies @xmath54 . this is of type scd , classified as having an nucleus by @xcite , and is at a distance of 8.7 mpc @xcite . it was observed with _ chandra _ twice , one month apart , for 40 ks and 25 ks . both observations used the acis - s3 chip . the nuclear x - ray emission appears to consist of two components ( fig . [ fig : n3184i ] ) , a fainter , point - like source in the north and a brighter , more extended source in the south . the total extent of the nuclear emission is about @xmath55 ( @xmath56 pc ) . both components are soft and have hr @xmath57 . since the northern component has only @xmath2 counts , detailed spectral fitting is not possible . we check for consistency of the data with different models using unbinned data . an absorbed power - law can be fitted to the data , and despite the degeneracy in the parameters , an intrinsic @xmath58 @xmath35 is inconsistent at the 3-@xmath3 level . fixing @xmath59 at the galactic value gives a range of acceptable power - laws with @xmath60 ( 3-@xmath3 limits ) . an unabsorbed blackbody with temperature @xmath61 ( 3-@xmath3 limits ) is also consistent with the data . the observed broadband luminosity of this component , corrected for galactic absorption , corresponds to @xmath62 erg s@xmath14 . the difference in count rates between the two observations is not statistically significant , so this source does not vary on month timescales . there are archival _ hst _ observations in the uv ( wfpc2 , f300w ) and optical ( wfpc2 , f606w ) of this galaxy . the f606w image has the nucleus on the wf4 chip , and clearly shows a central point source . the nucleus is on the pc chip in the f300w image . there is no central point source detected . both images also show diffuse nuclear emission and a prominent spiral arm in the central @xmath23 . other than the nuclear source in the f606w image , there are no obvious counterparts of the x - ray sources in either the uv or the optical . we note here that the coordinates in the _ hst _ images are misaligned with each other and with the x - ray image . the images were therefore aligned by cross - correlating point sources in each individual image with sdss sources that are classified as stars . the aligned _ hst _ images match each other to within @xmath31 , and match the _ chandra _ image to within @xmath63 . these offsets are comparable to the absolute astrometric accuracy of both _ chandra _ and _ the fact that we see x - ray emission in the soft band but almost none in the hard band argues against the emission being the continuum from the agn , whether direct or scattered . the source is not seen in the uv image . both of these points suggest the presence of heavy obscuration . the observed x - rays could instead be soft emission from circumnuclear ionized gas , if the agn is completely obscured and what is visible is mostly re - processed radiation . this is in fact the case in some heavily obscured seyfert 2s ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , where the dominant emission is the soft emission from the narrow - line region . the _ chandra _ and _ hst _ observations , therefore , are inconclusive regarding whether the source is an agn , but do not rule out that possibility either . a stronger argument that the source is an agn derives from infra - red data . this galaxy is part of the _ spitzer _ infrared nearby galaxies survey ( sings ; * ? ? ? @xcite have used the equivalent width of the pah feature at @xmath64 and the fluxes in a mix of high- and low - ionization lines ( [ ] @xmath65 , [ ] @xmath66 , [ ] @xmath67 , [ ] @xmath68 , [ ] @xmath69 , [ ] @xmath70 , [ ] @xmath71 ) to create diagnostic diagrams that distinguish between agn and star - forming galaxies . this nucleus falls into the `` transition '' region between agns and regions . this suggests an agn component to the emission exists that may have been diluted because of the large aperture used ( @xmath72 ) to extract the fluxes . in irac images the nuclear source is resolved ( @xmath73 fwhm compared to the @xmath74 fwhm of the psf ) . nuclear fluxes were extracted using @xmath75 apertures in the irac channels ( d. a. dale , private communication ) . host galaxy emission appears to dominate the mips ( 24 , 70 , and 160 @xmath76 ) fluxes , but the observed irac colors , @xmath77-[4.5 ] = -0.26 $ ] @xmath78 and @xmath79-[8.0 ] = + 0.59 $ ] @xmath78 ( magnitudes in @xmath80 system ) , are redder than more than 80% of normal late - type galaxies @xcite . this is expected if there is an agn , as agn power - law emission falls off more slowly than galactic emission in the nir . thus , the ir line ratios and irac colors strongly argue for the source to be an agn . the nucleus was also detected by 2mass . the @xmath81 , @xmath27 , and @xmath82 magnitudes , from the 2mass point source catalog ( 2mass psc ) , are @xmath83 , @xmath84 , and @xmath85 , which correspond to @xmath86 erg s@xmath14 . such a luminosity can be easily produced by an agn . while all the observations are consistent with there being an agn in ngc 3184 , we can not rule out a nuclear super - starcluster as the source . the x - ray emission could be from one or more xrbs in such a cluster . the x - ray luminosity ( @xmath87 erg s@xmath14 ) is in the range seen from xrbs . the presence of a nuclear star formation region is indicated by the optical line ratios observed by @xcite . in addition , @xcite reports a candidate nuclear star cluster in ngc 3184 . in conclusion , there are two scenarios that are consistent with the observations . in both there is a nuclear star cluster that dominates the optical emission . in the first , there is no agn and the x - rays are produced by one or more xrbs . in the second , there is a low - luminosity agn in the center of the star cluster , so heavily obscured that we do not observe any direct or scattered emission . the x - ray emission arises from photoionized gas immediately surrounding the agn . using the luminosity derived using the bremsstrahlung model above , and assuming that approximately 1% of the agn luminosity is reprocessed into the plasma x - ray emission , the agn has a luminosity @xmath88 erg s@xmath14 . the infrared line ratios and irac colors from _ spitzer _ strongly argue for this scenario . [ fig : n3184s ] this is an sb galaxy , classified as in @xcite and as /liner in the nasa / ipac extragalactic database ( ned ) , at a distance of 17 mpc @xcite . the _ chandra _ observation of this galaxy was presented by @xcite and also by @xcite . there are about 350 counts in the x - ray image of the nucleus . as a hardness ratio map ( fig . [ fig : n4102reg ] ) shows , the hard and soft emission are well segregated into a fairly hard ( hr @xmath89 ) , point - like source with extended , very soft ( hr @xmath90 ) emission to the west of it . _ wavdetect _ , run on hard and soft band images , detects a small hard source and an encompassing soft source whose centers are @xmath23 apart . we therefore fit the spectra of the harder `` core '' and the softer extended emission separately , using counts from the regions shown in fig . [ fig : n4102reg ] . the spectrum of the core has 171 counts , and shows a broad line between 6 and 7 kev , suggestive of reflection . the spectrum was fit after binning to 5 counts per bin , using the cash statistic ( ` cstat ` in _ sherpa _ ) . a simple absorbed power - law model , fit to the energy range @xmath913 kev , shows increasingly positive residuals at energies above 3 kev , again suggestive of a reflection component . the fit quality is poor , with a statistic value of 200 for 27 d - o - f . using a reflection model ( ` xspexrav ` ) reduces the statistic to 60 for 26 d - o - f . the possible line at 6.4 kev is not well constrained because of the lack of counts on the high energy side of the line ( there are 8 counts with @xmath92 kev ) . for the line , we added a gaussian at the fixed position of 6.4 kev and of fixed fwhm @xmath91 kev , allowing only the amplitude to vary . this reduces the statistic further , to 40 for 25 d - o - f . the final parameters are @xmath93 and reflection scaling factor @xmath94 . the formal equivalent width ( ew ) of the line is @xmath95 kev , but the uncertainty in the amplitude of the line is of the order of @xmath96 and in the normalization of the reflected component , @xmath97 . the flux , excluding the line and corrected for galactic absorption , is @xmath98 erg @xmath35 s@xmath14 . a fit using the @xmath37 statistic ( shown in table [ tab : spec ] ) gives best - fit parameter values consistent with those from the cash fit , with the exception of the equivalent width of the line , which is @xmath99 kev in this fit . given the poor determination of the continuum near the line the difference is not significant . the equivalent width of the fe k@xmath100 in type 1 agns is typically of the order of 300 ev , and a large ew implies that the direct continuum is suppressed and the source is reflection dominated . in ngc 4102 the formal ew is an order of magnitude larger ; therefore even after accounting for the large uncertainty we can conclude that the true ew is greater than the typical type 1 value . the extended emission is very soft , with just 5 counts ( out of 115 ) above 2.5 kev . the spectrum was fit after binning to 5 counts per bin and using both the @xmath37 and cash statistics . a bremsstrahlung model provides an acceptable fit ( @xmath101 and cash statistic @xmath102 for 18 d - o - f ) , and the mekal model is less favored ( @xmath103 and cash statistic @xmath104 for 18 d - o - f ) . best - fit parameter values obtained using the two statistics are consistent within the errors . the best - fit temperature @xmath105 kev is higher than what may be expected of 100 pc - scale circumnuclear gas . the plasma surrounding an agn may be photoionized rather than collisionally ionized , and there may be complexity in the spectrum that is hidden because of the poor quality . the flux in the extended emission is @xmath106 erg @xmath35 s@xmath14 . spectral models and best - fit parameter values for both the core and the extended emission are shown in table [ tab : spec ] . the @xmath37 fits for both the core and extended components are shown in fig . [ fig : n4102sc ] . _ hst _ imaging observations in visual and near - infrared ( nir ) bands show that the nucleus is clumpy and has large amounts of dust , and that there is circumnuclear star formation @xcite . high absorption and reddening towards the nucleus is also indicated by the balmer decrement which gives @xmath107 @xcite . the nucleus was detected by 2mass . the 2mass psc gives @xmath108 , @xmath109 , and @xmath110 . for a distance of 17 mpc , these correspond to @xmath111 erg s@xmath14 , @xmath112 erg s@xmath14 , and @xmath113 erg s@xmath14 . the nucleus is highly luminous in the far - infrared ( fir ) . infrared astronomical satellite ( iras ) observations indicate @xmath114 erg s@xmath14 @xcite , where @xmath115 is the luminosity between @xmath116 and @xmath117 , albeit in the arcminute - scale _ iras _ apertures . the nucleus was also detected in radio by the first survey ( 1.4 ghz ) . the stated radio position is within @xmath31 of the _ chandra_position , and therefore within _ chandra _ astrometric accuracy . the radio beam size , however , was @xmath118 . the integrated radio flux density was 223 mjy , with peak flux density of 167 mjy . this is the same integrated 1.4 ghz flux density reported by @xcite , fifteen years prior to the 1997 first observation . @xcite also report a higher resolution 4.9 ghz observation . the 4.9 ghz map shows radio emission extended in the northeast - southwest direction with a total extent of about @xmath119 ( @xmath120 pc ) , and with two peaks separated by @xmath121 ( @xmath122 pc ) and also aligned along a ne - sw axis . the position of the southwest peak coincides with the hard x - ray `` core '' seen in the _ chandra _ observation . the radio spectral index @xmath123 ( @xmath124 ) suggests a synchrotron rather than thermal origin for the radio emission . we first consider the case where there is no agn and all of the observed emission is due to star formation , and in particular consider the radio emitting region , which , as mentioned above , has a size of the order of 400 pc . if this region obeys the radio - fir correlation for non - agn galaxies @xcite , then it has an fir luminosity of @xmath125 erg s@xmath14 , exactly what was measured by _ iras_. therefore the radio and fir observations are consistent with each other and with the star - formation - only hypothesis , requiring only the assumption that this region dominated the fir emission within the _ iras _ aperture . following @xcite , we use the radio emission to estimate a supernova rate @xmath126 and consequently a star formation rate ( sfr ) . the luminosity density measured by first , @xmath127 erg @xmath35 s@xmath14 hz@xmath14 , implies @xmath128 yr@xmath14 , and @xmath129 yr@xmath14 . in the samples of @xcite and @xcite , the galaxies that have this star formation rate have x - ray luminosities ranging from a few @xmath130 erg s@xmath14 to a few @xmath131 erg s@xmath14 . the observed nuclear x - ray luminosity of ngc 4102 , @xmath132 erg s@xmath14 is really a lower limit because the amount of absorption is undetermined , but for the purpose of the argument here may be considered to be consistent with the @xmath133sfr relationship of @xcite and @xcite on the face of it , therefore , all of the observations are consistent with a starburst origin to the emission , but this explanation requires an extraordinarily intense starburst in a non - interacting galaxy . the radio , fir , and x - ray luminosities require @xmath134 yr@xmath14 of massive ( @xmath135 ) star formation within a 400 pc region . for comparison , the massive star formation rate in the interacting starburst galaxy m 82 integrated over the whole galaxy is @xmath136 yr@xmath14 @xcite , and the total sfr ( @xmath135 ) in the merging pair ngc 4038/4039 is estimated to be between 5 and 10 @xmath137 yr@xmath14 @xcite . in favor of the agn hypothesis , ngc 4102 shows an approximately conical region of outflowing gas that has a higher [ ] /h@xmath138 ratio than the surrounding star forming regions @xcite , suggesting exposure to a harder ionizing radiation than stellar continuum emission , and reminiscent of the ionization cones sometimes seen in seyfert 2s ( e.g. * ? ? ? the [ ] line profile in this region is broader than in the surrounding regions and can not be fit with a single gaussian @xcite . @xcite reported a broad component ( fwhm @xmath139 560 km s@xmath14 ) to the [ ] line as well . the broad component is weak , comprising just 57% of the flux in the line in the case of the h@xmath100 and h@xmath138 lines . using this component gives [ ] @xmath140 6583/h@xmath100 = 1.57 , similar to liners and seyferts , and thus @xcite argue for the presence of a very weak seyfert 2 in ngc 4102 . in this respect ngc 4102 is similar to ngc 1042 , where @xcite recently demonstrated the existence of a broad component in [ ] , and that considering only the broad component moves the nucleus into the seyfert / liner regions in line - ratio diagnostic diagrams . finally , the combination of a point - like hard source , soft extended circumnuclear emission , and an fe k@xmath100 line with a large ew , is one often seen in type 2 agns . there is undeniably strong star formation occurring in the nucleus of ngc 4102 , and we can not rule out the extremely large sfr implied if all of the observed emission is imputed to star formation alone . nevertheless , the evidence is strong that there is an agn in ngc 4102 . we conclude that ngc 4102 is another example where an agn and strong star formation co - exist at the nucleus . ngc 4647 is a galaxy of type sc , its nucleus classified as in @xcite , at a distance of 16.8 mpc @xcite . the _ chandra _ observation is of the elliptical galaxy ngc 4649 , and ngc 4647 is on the chip , at an off - axis angle of @xmath141 . two factors complicate the detection of the nucleus of ngc 4647 . first , the nucleus lies within the extended emission from ngc 4649 . second , the nucleus falls on a node boundary of the ccd . _ wavdetect _ at its default filter for source significance ( @xmath142 false source per @xmath143 pixels ) does not detect the nucleus . however , once the elliptical galaxy is modeled and subtracted out of the image , there is a positive residual at the location of the nucleus of ngc 4647 ( see fig . [ fig : n4647xo ] ) . to extract source counts we used a circular region centered on the centroid of the residual and with radius @xmath25 , which is approximately the 95% encircled - energy radius at 1.5 kev at that position . to extract background counts we located another region on the same node boundary that was at the same distance from the center of ngc 4649 as was the source circle . the source region has @xmath144 counts after background subtraction , but this number necessarily has a large uncertainty . we take the radius of the source circle , @xmath25 , as the _ chandra _ positional uncertainty of this source . the nucleus may also have been detected in x - rays by xmm - newton @xcite . while the source was detected with @xmath145 , the positional uncertainty was @xmath75 . the xmm - newton source is soft , similar to the _ chandra _ residual . the xmm - newton and _ chandra _ source positions differ by @xmath119 . there is a nuclear 2mass point source ( @xmath146 ) at a distance of @xmath147 from the _ chandra _ position , but it is embedded in diffuse emission and the flux from the point source is poorly constrained . in the radio , there is a @xmath148 upper limit to the nuclear emission of @xmath149 mjy at 5 ghz @xcite . there is an older report of a radio detection of the nucleus with a flux density of 16 mjy at 1.4 ghz @xcite , but with positional uncertainty @xmath150 . the data are inconclusive at this time as to whether there is an agn in ngc 4647 . the faintness of the source and the positional uncertainties in the existing observations prevent a firm identification and characterization of the source . this galaxy is of type sd , at a distance of 17.9 mpc @xcite . its nucleus was classified as t2 ( transition object with type 2 spectrum ) by @xcite . the nucleus is clearly detected by _ chandra _ , but with only ten counts . it is a very soft source , with nine of the ten counts below @xmath95 kev . this object was also analyzed by @xcite , but since they were looking for hard - band sources , this object was not counted as a detection . the observed count rate corresponds to a flux of ( 510)@xmath151 erg @xmath35 s@xmath14 in the 0.38 kev band , or luminosity ( 35)@xmath152 erg s@xmath14 for our assumed distance . the 210 kev flux depends strongly on the assumed model , from @xmath153 erg @xmath35 s@xmath14 for a thermal bremsstrahlung model with @xmath154 kev , to @xmath155 erg @xmath35 s@xmath14 for a power - law with @xmath156 and galactic absorption . ngc 4713 was observed by _ hst _ in december 2006 . the nucleus is resolved ; thus only a nuclear star cluster or star forming region , and no agn , is detected . the cluster is @xmath157 in diameter in the image , corresponding to a physical size of @xmath158 pc . within this aperture the flux density is @xmath159 erg @xmath35 s@xmath14 @xmath14 , which implies @xmath160 erg s@xmath14 . the nucleus is also in the 2mass psc , with @xmath81 , @xmath27 , @xmath82 luminosities @xmath161 erg s@xmath14 . however , the nucleus is embedded in diffuse extended emission and therefore the reported magnitudes may not be accurate estimates of the nuclear emission . there is an upper limit of @xmath162 mjy to the nuclear radio emission at 15 ghz @xcite , and a similar limit , 1.0 mjy , to the emission at 1.4 ghz from the first survey . as in the case of ngc 4647 , the data are inconclusive regarding the presence of an agn . the data are consistent with the agn hypothesis : the `` transition object '' classification by @xcite implies there may be an agn component in the optical spectrum ; the nucleus is without doubt an x - ray source , though it is not possible to distinguish between emission from circumnuclear gas and the nucleus proper in the current _ chandra _ imaging . the hardness ratio ( with large uncertainty ) is consistent with a @xmath156 power - law . ngc 4713 may be similar to ngc 3184 in having a low - luminosity agn inside a nuclear star cluster . ngc 5457 is a galaxy of type scd at a distance of approximately 7 mpc @xcite . the nucleus was classified as by @xcite . it has been observed multiple times by _ chandra _ for a total observation time of about 1.1 ms . our analysis omits the shorter , and hence low signal - to - noise , exposures . the observations included here are listed in table [ tab : m101 ] . the total usable exposure time is 695 ks . chandra _ clearly resolves two sources in the nuclear region ( fig . [ fig : n5457i ] ) , which we label n and s. the northern source , n , is the nucleus , while the southern , s , is a star cluster @xcite . table [ tab : m101 ] shows the counts and hardness ratios of the two sources in each of the observations . source n ( the nucleus ) varies in brightness by about a factor of 9 over the course of about 8 months ( see table [ tab : m101 ] and fig . [ fig : m101var ] ) . there is no significant change in hardness ratio . in our analysis and discussion below we consider only the nucleus ( source n ) . three nuclear x - ray spectra were extracted : one ( `` merged '' ) from the event list obtained by merging all observations done in 2004 , and fit using the instrument response from obsid 5339 ; one ( `` high state '' ) from merging two observations where the source had a high count rate ( obsids 4736 and 6152 ) and fit using the response from obsid 4736 ; and one ( `` low state '' ) from merging five observations where the count rate was low ( obsids 5300 , 5309 , 4732 , 5322 , 5323 ) , fit using the response from obsid 4732 . the spectra are shown in figs . [ fig : n5457sa ] and [ fig : n5457shl ] ; fit models and parameters are shown in table [ tab : spec ] . the low state spectrum can be fit with an absorbed power - law ( @xmath163 ) but the fit is improved slightly with the addition of a plasma component ( ` xsmekal ` ) at temperature @xmath164 kev ( @xmath165 for two fewer d - o - f ) . the best - fit power - law slope @xmath166 is typical of unabsorbed agn , and best - fit intrinsic absorption is consistent with zero . the unabsorbed 0.38 kev luminosity is @xmath167 erg s@xmath14 . the high state spectrum can also be fit by an absorbed power - law . the best - fit slope is @xmath168 , steeper but consistent with the low state slope within the uncertainties . the differences from the low state spectrum are , first , that the fit requires intrinsic absorption @xmath169 @xmath35 ) , and second , that there is no evidence for the mekal component . if a plasma component exists in the high state its flux falls below that of the power - law component . the unabsorbed 0.38 kev luminosity in the high state is @xmath170 erg s@xmath14 . figure [ fig : n5457sop ] shows the high and low state spectra over - plotted . the spectra are identical below 1 kev , with the entire flux difference arising from the power - law normalization between 1 and 7 kev , and a possible `` shoulder '' between 1 and 2 kev in the high state . a fit where the power - law slope is forced to be identical in the low and high states is of similar statistical significance and produces best - fit values similar to the separate fits . this fit is shown in table [ tab : spec ] in the rows labeled `` sim . low '' and `` sim . the merged spectrum is similar to the high state one in that an absorbed power - law provides a good fit without the need for additional components . we analyzed archival _ hst _ wfpc2 images of ngc 5457 , a 2400 s exposure using the f336w filter ( referred to as u band below ; nucleus on wf3 ) and a 1600 s exposure using the f547 m filter ( referred to as v band below ; nucleus on pc1 ) . the nucleus and the star cluster are detected in both images . in 2-pixel radius apertures , the u band nuclear flux is @xmath171 erg @xmath35 s@xmath14 @xmath14 and the v band flux is @xmath172 erg @xmath35 s@xmath14 @xmath14 , corresponding to @xmath173 and @xmath174 in the stmag system . the 2mass psc contains a source at the position of the nucleus ( x - ray source n ) but none at the location of source s. the given magnitudes @xmath175 , @xmath176 , and @xmath177 all correspond to @xmath178 erg s@xmath14 . the nucleus was not detected by first . @xcite gives an upper limit to the nuclear luminosity at 6 cm corresponding to @xmath179 erg s@xmath14 . @xcite reports @xmath180 km s@xmath14 which implies @xmath181 . the corresponding eddington luminosity is @xmath182 erg s@xmath14 . assuming the source is an agn , and assuming bolometric correction factors of @xmath183 and @xmath184 in the x - ray and ir , respectively , implies that the bolometric luminosity is in the range @xmath185@xmath186 erg s@xmath14 , or that @xmath187@xmath188 . the nuclear source shows several properties typical of an agn . first , variability : the source varies by a factor of nine in eight months in 2004 , and also varied between 2000 and 2004 . second , the x - ray spectrum : the spectrum is an absorbed power - law with slope @xmath189 . in particular , the spectrum can not be fit by a thermal component alone a power - law is necessary . the low state spectrum is reminiscent of highly obscured agn where an unabsorbed but diminished spectrum is seen via scattering . third , x - ray colors : the ratio of 0.32 kev , 25 kev , and 58 kev counts puts this object into the compton - thick agn part of the levenson color - color diagram ( fig . 9 in * ? ? ? * ) . in the context of the latter two points it is worth noting that if the source really is a highly absorbed agn then energy emitted by the agn may show up as re - processed radiation in the infrared . the existence of an infrared source with luminosity @xmath190 erg s@xmath14 is consistent with this picture , though it does not argue for the presence of an agn to exclusion of other types of sources . the x - ray properties of the nucleus are consistent with both the xrb and agn hypotheses , but an agn is not ruled out . the inferred x - ray luminosity ( @xmath191 erg s@xmath14 ) is in the range seen from xrbs , and the observed x - ray colors put this source in the lmxb region of the @xcite color - color diagram . in favor of the xrb scenario is that the variability of the source is similar to the spectral state changes in xrbs . the low state may be interpreted as being the xrb state known as `` hard '' or `` low / hard '' , and the best - fit power law slope , @xmath192 is the value seen in xrbs in this state . the high state would then be the xrb state known as the `` very high '' or `` steep power law '' state . the best - fit power - law slope @xmath193 in this case is less steep than is usually seen in xrbs in this state ( @xmath194 ) but the steeper value is included in the 90% confidence range . however , it must be kept in mind first , that at the quality of the spectra analyzed here the power law slopes are consistent with being identical in the two states , and second , that agns can show the same state behavior ( e.g. 1h 0419 - 577 , * ? ? ? * ) . in xrbs , the steep power - law state is associated with the presence of quasi - periodic oscillations in the x - ray emission , while the hard state is associated with the presence of a radio jet . in principle the presence of these features could provide additional evidence supporting the xrb hypothesis , but it is not currently feasible to detect them in xrbs at the distance of ngc 5457 . thus , there are again two possible scenarios , as in the case of ngc 3184 . the x - ray source could be an hmxb in a super - star cluster . the star cluster would dominate the optical and ir emission . the large variation observed in the x - ray flux rules out the source s being more than one hmxb , as otherwise they would have to be varying in concert . the actual amount of obscuration is important , however , for the plausibility of the hmxb hypothesis since the inferred x - ray luminosity ( @xmath195 erg s@xmath14 in the high state ) is already at the high end of the range of xrb luminosities and there is not much room for a significantly higher absorption - corrected intrinsic luminosity . the alternative scenario is an agn together with a nuclear star cluster . in agns both the intrinsic luminosity and the amount of obscuration are known to vary . the low state x - ray spectrum may be explained as a truly under - luminous ( @xmath196 erg s@xmath14 ) unabsorbed agn , or as an agn where the obscuration is so high that only scattered light , and no direct emission , is seen . the fact that the x - ray colors are similar to those of compton - thick agn @xcite supports the latter view , but we also note the absence of the @xmath197 kev fe k@xmath100 line in the x - ray spectrum that is often present in the reflected component . though both the hmxb and agn hypotheses are possible , on balance the agn appears to be the more plausible one . the motivation for this paper was to evaluate the feasibility of detecting low - mass smbhs in late - type spiral galaxies , which may still be accreting at the current epoch and if so should be detectable in x - rays . the six galaxies studied in this paper are not all late - type , but span the range sa ngc 3169 and ngc 4102 were regarded as low luminosity agns . none of the remaining four nuclei , however , was known to have an accreting smbh , of any mass . ngc 3169 and ngc 4102 are of type sa and sb , which have massive bulges and are expected to have massive smbhs . for galaxies of types scd and sd , on the other hand , the lack of a luminous agn could mean either that there is no smbh or that the mass of the smbh is low . as such , the observations studied here examine both aspects of a search for accreting low - mass smbh : first , are these objects really detectable , given that the accretion rate is expected to be low ? second , are any sources detected in the very latest type spiral galaxies that have small or no bulges ? we first note that of the six galaxies presented here , all six show nuclear x - ray sources . this implies that it is a very common occurrence . in a survey of late - type spiral galaxies such as our ongoing _ chandra _ survey , therefore , the predominant concern is not going to be detection efficiency , but rather identification of the agns among the detected sources . given that the sample presented in this paper consists of only six galaxies , we do not draw statistical conclusions here of the prevalence of very low - luminosity agns in nearby galaxies . but we note that , as shown in 3.13.6 , of the six nuclear x - ray sources , ngc 3169 is almost certainly an agn , and ngc 4102 , ngc 3184 , and ngc 5457 , have very strong , though not conclusive , arguments in favor of their being agns . the two remaining galaxies , ngc 4713 and ngc 4647 , are ambiguous but agns are not ruled out . we discuss below the issues such surveys will face when attempting to identify the nature of the detected sources . the diagnostic tools traditionally used to distinguish agns from non - agns ( e.g. optical line ratios , * ? ? ? * ; * ? ? ? * ) were developed in the course of studying luminous agn . dilution of the agn emission by host galaxy light was not a serious problem and observations with low spatial resolution ( several arcseconds ) sufficed . in the study of agn that are either intrinsically less luminous or are heavily obscured , however , host galaxy light becomes increasingly problematic , and surveys relying on optical spectra ( e.g. * ? ? ? * ; * ? ? ? * ) require careful subtraction of the starlight using galactic spectral templates . in the weakest agns , however , signs of agn emission may not be detected by the usual diagnostics . this problem will persist until optical observations with angular resolution of 110 mas become possible so that host galaxy light can be effectively excluded , though it can be mitigated by using regions of the spectrum where host galaxy emission is negligible , for example very high energy x - rays ( tens to hundreds of kev ) . the detection of a compact radio source unresolved at milliarcsecond - scales , especially if the source is accompanied by jets , would also unambiguously identify the nucleus as an agn . this has been the motivation for radio surveys like that of @xcite . some agns obscured in the optical and uv may be detectable using infrared emission line strengths and ratios ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . while the methods listed above allow the unambiguous identification of agns , observations often do not have the angular resolution or sensitivity to distinguish agn and non - agn flux . other sources of radiation in the vicinity of the agn are , for example , plasma photoionized by the agn itself , or a nuclear star cluster . the targeted agns have very low luminosity and thus even moderate amounts of obscuration may cause a significant decrement in the observed flux . in most cases , therefore , the agn contribution should not be expected to dominate the total observed flux . consequently , identifying these agns requires a different approach than what can be used in the case of the more luminous agns ( seyferts and qsos ) . agns can be identified using x - ray observations ( e.g. with _ chandra _ and _ xmm - newton _ ) solely , but only if they are point sources whose inferred luminosities are greater than @xmath198 erg s@xmath14 . below that value , agns can become indistinguishable from ulxs and xrbs in x - rays . ulxs can have luminosities of a few times @xmath30 erg s@xmath14(e.g . * ) and have x - ray spectra that look similar to agn spectra . it has been suggested @xcite that ulx spectra show a break at @xmath199 kev . agns are not known to show this break . in high quality spectra with a large number of counts it may be possible to exploit this difference to separate ulxs and agns . xrbs have power law spectra with @xmath200 , can show an fe k@xmath100 emission line , and emit hard x - rays , all characteristics of agns as well . additionally , even with _ chandra _ s angular resolution , the physical space probed ranges from the central @xmath201 10 to 100 pc of the galaxy . the existence of one or more xrbs within that region would be unsurprising . however , the fact that the inferred luminosities are as high as @xmath202 erg s@xmath14 severely constrains the expected number of xrbs . for example , for ngc 5457 , one of the four galaxies presented here that are not confirmed agns , @xcite provide a log n - log s relation as well as the surface density of x - ray point sources as a function of radius ( their figs . 3 and 4 ) . approximately 12.5% of the sources have luminosities exceeding @xmath196 erg s@xmath14 . the surface density of sources in the innermost @xmath203 is @xmath204 arcmin@xmath205 . therefore we may expect @xmath206 sources arcmin@xmath205 above the luminosity cut - off in the central @xmath203 , or @xmath207 such sources within the _ chandra_source circle of radius @xmath25 that has been used here . ngc 5457 is of type scd , and can be taken to be representative of the other three galaxies , ngc 4647 , ngc 3184 , and ngc 4713 which are types sc , scd , and sd , respectively . thus invoking xrbs and ulxs alone to explain the x - ray observations leads to physically implausible conditions , such as requiring that most , or all , quiescent , non - starburst spiral galaxies have a ulx or @xmath208 xrbs in the central @xmath209 nevertheless , x - ray observations by themselves will usually be inadequate to distinguish agns from non - agns in any particular instance . information from other wavebands is thus crucial for the identification process . for the six galaxies in this paper multi - wavelength photometry is summarized in tables [ tab : lum ] and [ tab : n31spitz ] . but here too , traditional methods of identifying agns using flux ratios such as @xmath210 , @xmath211 , and @xmath212 , must be used with caution . the low luminosities of the agns mean that the emission in the two bands being compared may not be from the same object . for instance , for an obscured agn surrounded by a nuclear star cluster , the observed x - ray flux may be from the agn but the optical flux may be dominated by the cluster . the existence of an agn must instead be inferred by consistency and plausibility checks considering as much of the spectral energy distribution as possible , and the goal is the rejection of the hypothesis that all of the observed properties can be explained without requiring the presence of an agn . ngc 3184 provides a good example where different modes of observing , imaging and spectroscopy , in two wavebands , x - ray and ir , together make a compelling argument for the presence of an agn where each observation individually is ambiguous . nuclear star clusters are fairly common in spiral galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? a cluster poses two main problems . first , it makes the presence of xrbs more likely . second , if the cluster is young and contains many o and b stars , it may overwhelm agn emission in the uv in addition to the optical ( e.g. ngc 4303 , * ? ? ? * ) . it may be possible in some cases , as for ngc 1042 @xcite and ngc 4102 @xcite , to attempt a separation of the cluster and agn components in the optical spectrum . in addition , stellar spectra , even for late - type stars , fall faster towards the infrared than the power - laws typical of agns . the mid - infrared colors of agns , therefore , tend to be redder than those of stellar populations and this color difference can be used to infer the presence of an agn @xcite . in addition to the spectral energy distribution , source variability can be a discriminant , as agns are known to vary in all wavelengths , whereas a star cluster , say , would not . conversely , if the variation is periodic it would rule out an agn and argue for an xrb instead . a survey of the type discussed here finds agns and strong agn candidates , but does not measure the masses of the smbhs in those agns . measurement of the mass of one of these smbhs will be a difficult endeavor , since the sphere of influence of the black hole can not be resolved with current technology and resources . none of the six objects studied in this paper shows broad optical emission lines whose widths could be used to estimate the bh mass , and this is likely to be typical behavior . spectropolarimetry may uncover broad lines in polarized light in some of the agns . estimates of the smbh mass may be made by using known scaling relationships , with the caveat that the correlations are all based on smbhs two or more orders of magnitude more massive than the ones expected to be found by the survey . the least indirect method is an application of the observed correlation between the x - ray power law slope and eddington ratio @xcite . the eddington ratio and the luminosity in turn provide an estimate of the mass of the smbh . other relationships that can be used are : ( a ) the @xmath1@xmath3 relation @xcite , but this method becomes more and more uncertain as the bulge itself becomes ill - defined in the latest - type spirals ; ( b ) the @xmath1@xmath213 relationship @xcite , which has more scatter and also faces the problem of the definition of the bulge ; ( c ) the @xmath1@xmath214 relation @xcite , which has the advantage that it does not require the presence of a well - defined bulge ; ( d ) the @xmath1@xmath215 relation @xcite , where @xmath215 is the concentration of light . there is also a reported relationship between black hole mass and core radio power @xcite , but this relationship is not as well established as the others , and is based on observations of elliptical galaxies , so its applicability to the spiral galaxies here is uncertain . overall , there is unlikely to be one standard method of measurement that can be applied to these galaxies , but one or more of the above methods may provide useful estimates of or limits to the masses of the smbhs in these agns . mass measurements independent of the scaling relationships are possible if an object turns out to have broad emission lines , like ngc 4395 , in which case line widths or reverberation mapping may be used , or if it has maser emission , like ngc 4258 , in which case gas dynamics can be used . mass measurement in other cases will have to await mas - scale angular resolution in bands other than radio to resolve the sphere of influence of these black holes . of the six galaxies here , ngc 3169 , ngc 4102 , and ngc 5457 have measurements of either the stellar velocity dispersion or the luminosity of the bulge , thus allowing an estimate of their smbh masses ( assuming here that the source in m 101 is an agn ) . the scatter in the @xmath1@xmath3 and @xmath1@xmath213 relations , together with the uncertainty in the observed flux and the bolometric correction , result in uncertainties of about an order of magnitude in the inferred eddington ratio , but all three objects have @xmath216 . this is in the range seen in low - luminosity agns ( e.g. @xmath217 for m 81 ; * ? ? ? * ) , and much higher than the @xmath218 seen in truly quiescent smbhs . these observations indicate that there is indeed a population of accreting smbhs in nearby spiral galaxies that do not show optical signs of activity but can be uncovered by looking for their x - ray emission . such a population will answer the question of whether the bulge is the dominant component that determines the existence , and mass , of a nuclear smbh . the discoveries of agns in the sd galaxies ngc 4395 @xcite and ngc 3621 @xcite , and the strong evidence for agns in the scd galaxies ngc 3184 and ngc 5457 suggest it is not , at least as far as existence is concerned . among the smbhs discovered in the latest - type spirals ( with small or no bulges ) and in the lowest - mass galaxies should be a population of smbhs with masses less than @xmath7 , enabling a systematic study of the low - mass end of the local supermassive black hole mass function . llcccrcccr ngc 3169 & sa & l2 & 10 14 15.0 & + 03 27 58 & 19.7 & 96 & 2001 may 2 & 1614 & 2.0 + ngc 3184 & scd & & 10 18 17.0 & + 41 25 28 & 8.7 & 42 & 2000 jan 8 & 804 & 39.8 + & & & & & & & 2000 feb 3 & 1520 & 21.3 + ngc 4102 & sb & & 12 06 23.1 & + 52 42 39 & 17.0 & 82 & 2003 apr 30 & 4014 & 4.9 + ngc 4647 & sc & & 12 43 32.3 & + 11 34 55 & 16.8 & 81 & 2000 apr 20 & 785 & 36.9 + ngc 4713 & sd & t2 & 12 49 57.9 & + 05 18 41 & 17.9 & 87 & 2003 jan 28 & 4019 & 4.9 + ngc 5457 & scd & & 14 03 12.6 & + 54 20 57 & 7.2 & 35 & & & 750 + cccccccc x - ray & 0.38 kev & 41.7 & 40.2 & 39.0 & 37.3 & 37.538.5 & 38.6 + uv & opt . & f300w & & & & @xmath344 & & + & f336w & & & & & 39.0 & + & f547 m & & & & & 39.5 & + & f606w & & & & 39.5 & & 40.1 + ir & @xmath82 & 42.6 & 42.8 & 38.8 & 40.9 & 41.0 & 41.0 + radio & 15 ghz & & & & & & @xmath345 + & 5 ghz & 37.2 & & & & @xmath346 & + & 1.4 ghz & & 38.0 & 36.9 & @xmath347 & @xmath348 & @xmath349 we thank the referee , and p. salucci and a. seth for helpful comments . we are grateful to d. a. dale for kindly providing _ spitzer _ fluxes for the nucleus of ngc 3184 , and to p. martini , t. bker , and e. schinnerer for providing _ hst _ data for ngc 4713 prior to publication . support for this work was provided by the national aeronautics and space administration through chandra award number go7 - 8111x issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . 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 publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation .

we present detailed integral field unit ( ifu ) observations of the central few kiloparsecs of the ionised nebulae surrounding four active central cluster galaxies ( ccgs ) in cooling flow clusters ( abell 0496 , 0780 , 1644 and 2052 ) . our sample consists of ccgs with h filaments , and have existing data from the x - ray regime available . here , we present the detailed optical emission - line ( and simultaneous absorption line ) data over a broad wavelength range to probe the dominant ionisation processes , excitation sources , morphology and kinematics of the hot gas ( as well as the morphology and kinematics of the stars ) . this , combined with the other multiwavelength data , will form a complete view of the different phases ( hot and cold gas and stars ) and how they interact in the processes of star formation and feedback detected in central galaxies in cooling flow clusters , as well as the influence of the host cluster . we derive the optical dust extinction maps of the four nebulae . we also derive a range of different kinematic properties , given the small sample size . for abell 0496 and 0780 , we find that the stars and gas are kinematically decoupled , and in the case of abell 1644 we find that these components are aligned . for abell 2052 , we find that the gaseous components show rotation even though no rotation is apparent in the stellar components . to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas for all the galaxies . based on optical diagnostic ratios ( [ oiii]/h against [ nii]/h , [ sii],6731/h , and [ oi]/h ) , all galaxies show extended liner emission , but that at least one has significant seyfert emission areas , and at least one other has significant hii like emission line ratios for many pixels . we also show that the hardness of the ionising continuum do not decrease with galactocentric distance within our field - of - view ( with the exception of one galaxy that show a core separation ) as the emission line ratios do not vary significantly with radius . this also indicates that the derived nebular properties are spatially homogeneous . we fit agn and pagb stars photoionisation models as well as shock excitation models to our derived diagnostic ratios . these fits , combined with information from multiwavelength studies reveal that agn photoionisation is the most likely ionisation mechanism in at least two cases even though shocks and pagb stars can not be conclusively eliminated . = 1 [ firstpage ] galaxies : formation galaxies : elliptical and lenticular , cd galaxies : clusters : individual : abell 0496 galaxies : clusters : individual : abell 0780 galaxies : clusters : individual : abell 1644 galaxies : clusters : individual : abell 2052

1303.0884

decades ago , elliptical galaxies were thought to contain very little , if any , gas . studies of galaxy formation , therefore , often focussed on the stellar properties , however we now know that a large fraction of the baryonic mass in massive galaxies is believe to be in diffuse form . thus a complete view of galaxy formation and evolution necessarily incorporates both the stars and hot gas and an understanding of the processes by which these phases interact ( mccarthy et al . 2010 ) . cooling - flow clusters are common in the local universe and massive central cluster galaxies ( ccgs ) are often found at the centres of these systems ( peres et al . if the central cluster density is high enough , intracluster gas can condense and form stars at the bottom of the potential well . since the radiative cooling times for intracluster gas are short enough that gas can cool and settle to the cluster centre ( edge , stewart @xmath4 fabian 1992 ) , it has been suggested that the big envelopes of ccgs may arise from the gradual deposition of this cool gas . more recently , high spectral resolution _ xmm - newton _ observations showed that the x - ray gas in cluster centres does not cool significantly below a threshold temperature of @xmath5 kev ( jordn et al . 2004 , and references therein ) . this initially contradicted the model that these young stars are formed in cooling flows . however , it is possible that star formation is ongoing in cool - core clusters at a much reduced rate ( bildfell et al . 2008 ) . previous studies have reported several examples of ongoing star formation in ccgs , in particular those hosted by cooling - flow clusters ( cardiel , gorgas @xmath4 aragn - salamanca 1998 ; crawford et al . 1999 ; mcnamara et al . 2006 ; edwards et al . 2007 ; odea et al . 2008 ; bildfell et al . 2008 ; pipino et al . 2009 ; loubser et al . however , the origin of the gas fuelling this star formation is not yet known . possible explanations include processes involving cooling flows or cold gas deposited during a merging event ( bildfell et al . these processes will leave different imprints in the dynamical properties , the detailed chemical abundances , and the star formation histories of these galaxies , which can be studied using high - quality spectroscopy ( loubser et al . 2008 ; 2009 ; 2012 ) . observations that support this idea are blue- and ultraviolet - colour ( uv - colour ) excesses observed ( indicative of star formation ) in the central galaxy of abell 1795 by mcnamara et al . ( 1996 ) and molecular gas detected in 10 out of 32 central cluster galaxies by salom @xmath4 combes ( 2003 ) . the observations by cardiel et al . ( 1998 ) were consistent with an evolutionary sequence in which star formation bursts , triggered by radio sources , take place several times during the lifetime of the cooling flow in the centre of the cluster . however , mcnamara @xmath4 oconnell ( 1992 ) found only colour anomalies with small amplitudes , implying star formation rates that account for at most a few percent of the material that is cooling and accreting onto the central galaxy . cooling - flow models for ccg formation also imply the formation of larger numbers of new stars , for which there is no good observational evidence ( athanassoula , garijo @xmath4 garca - gmez 2001 ) . the central cluster galaxies often host radio - loud agn which may account for the necessary heating to counteract radiative cooling ( von der linden et al . 2007 ) . in summary , cooling flow models predict more cooled gas than is observed ( bohringer et al . thus , it is possible that the mass deposited into the molecular clouds is heated by one of several processes - hot young stellar populations , radio - loud agn , x - rays or heat conduction from the intracluster medium itself , shocks and turbulent mixing layers , and cosmic rays . therefore , only a small fraction of the cooled gas is detected ( crawford et al . 2005 ; ferland et al . thus , ccgs lie at the interface where it is crucial to understand the role of feedback and accretion in star formation . within these cooling - flow ccgs , cool molecular clouds , warm ionized hydrogen , and the cooling intracluster medium are related . a complete view of the star formation process incorporates the stars with the gas and an understanding of the processes by which these phases interact , and therefore , requires information from several wavelength regimes . in conclusion , although ccgs are probably not completely formed in cooling flows , the flows play an important role in regulating the rate at which gas cools at the centres of groups and clusters . in the @xmath6cdm cosmology it is now understood that local massive haloes assemble late through the merging of smaller systems . in this picture , cooling flows seem to be the main fuel for galaxy mass growth at high redshift . this source is removed only at low redshifts in group or cluster environments , due to agn feedback ( de lucia @xmath4 blaizot 2007 ; voit et al . indeed , if agn feedback is not properly assumed in hydrodynamical simulations , an apparently bluer bcg is formed as a result of an accelerated late stellar birthrate even after the epoch of quiescent star formation ( romeo et al . 2008 ) . we proposed and obtained ifu observations of the central few kiloparsecs of the ionised nebulae in active ccgs in cooling - flow clusters . these observations will map the morphology , kinematics and ionisation state of the nebulae to gain an understanding of their formation , heating and relationship to the cluster centre . we selected galaxies from the mcdonald et al . ( 2010 ) study , which consisted of an h@xmath0 survey of 23 cooling flow clusters . amongst their conclusions , mcdonald , veilleux @xmath4 rupke ( 2012 ) find a stong correlation between the h@xmath0 luminosity contained in filaments and the x - ray cooling flow rate of the cluster , suggesting that the warm , ionised gas is linked to the cooling flow . we chose objects with confirmed extended h@xmath0 emission , and with near - ir ( 2mass ) , ultraviolet ( galex ) , x - ray data ( chandra ) , and in some cases vla 1.4 ghz fluxes , already available ( mcdonald et al . detailed properties of the host clusters , which are reported to influence the activity in the central galaxy , such as central cooling times and the offset between the cluster x - ray peak and the central galaxy , have been derived . farage et al . ( 2010 ) presented ifu observations of one nearby bcg showing liner - like emission , and brough et al . ( 2011 ) presented ifu data on four ccgs at z @xmath7 to calculate the dynamical masses of ccgs and measure their stellar angular momentum ( although none of their four ccgs contained emission lines ) . the sauron sample includes only one ccg ( m87 ) . hatch , crawford @xmath4 fabian ( 2007 ) and edwards et al . ( 2009 ) analysed data comparable to the data presented here . hatch et al . ( 2007 ) presents ifu observations of six emission - line nebulae in cool - core clusters ( selected from the optical rosat follow - up by crawford et al . 1999 ) with oasis on the wht ( with a limited wavelength range centered around h@xmath0 ) . edwards et al . ( 2009 ) presents ifu observations of nine ccgs in cooling and non - cooling clusters ( also from crawford et al . 1999 ) and within 50 kpc of the x - ray centre , with gmos ifu and oasis on wht ( most galaxies with wavelength ranges also limited to the h@xmath0 region , but three also included h@xmath2 ) . emission line maps of morphology , kinematics , often line ratios , and the stellar continuum have been published for abell 496 by hatch et al . ( 2007 ) , and for abell 2052 by edwards et al . these studies concluded that no proposed heating mechanism reproduces all the emission - line properties within their observed wavelength within a single source . thus , a single dominant mechanism may not apply to all ccg nebulae , and there may be a mixture of heating mechanisms acting within a single nebula ( wilman , edge @xmath4 swinbank 2006 ; hatch et al . 2007 ) . thus to get more information which will enable the dominant mechanism(s ) to be identified , we observed more line ratios over a longer wavelength range ( around h@xmath0 and h@xmath2 which also allowed extinction to be appropriately measured ) . we will be adding the detailed stellar population analysis , and place the derived information from the optical spectra in context with multiwavelength data over the full spectrum in a future paper , to explain the diverse nature of these galaxies . all the data for abell 780 and 1644 is new , and all four of the extinction maps are new . also , the construction of ionisation diagrams and analysis of ionisation processes with shock models and pagb models is also new . our exposure times ( improved from previous studies ) are shown in table [ table : objects ] . this data is also complimentary to the long - slit spectroscopy ( on keck and magellan ) along the h@xmath0 filaments of the four objects ( studied here ) by mcdonald et al . ( 2012 ) . we introduce the sample and detail of the data reductions in sections [ sample ] and [ reduction ] . we then derive the optical extinction as well as the line strengths in section [ extinction ] . we proceed to discuss the four individual cases in section [ figures_nvss ] , and identify the mechanism producing the emission and the ionised gas in these four ccgs in section [ ionisation ] . we summarise the findings of this paper in [ summary ] . we have used the following set of cosmological parameters : @xmath8 = 0.3 , @xmath9 = 0.7 , h@xmath10 = 70 km s@xmath11 mpc@xmath11 . [ fig : thumbnails ] we have chosen our sample of active central cluster galaxies from the h@xmath0 imaging presented in mcdonald et al . ( 2010 ) , who in turn , selected their sample from white , jones @xmath4 forman ( 1997 ) . mcdonald et al . ( 2010 ) enforced the cuts : @xmath12 and @xmath13 , after which they selected 23 clusters to cover the full range of properties , from very rich clusters with high cooling rates to low - density clusters with small cooling flows . their classical cooling rates range from 6.3 431 m@xmath14 yr@xmath11 which means that while covering a large range in properties , their sample consisted of only cooling flow clusters . from their 23 cooling flow clusters , we selected all the clusters with clearly detected h@xmath0 in their centres ( albeit filamentary , extended or nuclear emission ) . in addition , all of these central galaxies have optical imaging , near - ir ( 2mass ) and uv ( galex data ) available . thereafter , we selected all the central galaxies with detailed x - ray ( chandra ) data , as well as vla 1.4 ghz fluxes , available . this resulted in a sub - sample of 10 galaxies . we observed four of these galaxies with the gmos ifu ( as shown in figure [ fig : thumbnails ] ) . we merely chose the objects with the most auxiliary information available . this additional information will be added in the future paper ( where the underlying stellar populations will be analysed in detail ) and might help to constrain the ionisation mechanisms . the rest - wavelength range of the emission lines of interest is 48606731 ( h@xmath2 to [ sii]@xmath16731 ) . the ratio of the forbidden [ nii]@xmath16584 to h@xmath0 line will depend on the metallicity of the gas , the form of the ionising radiation , and the star formation rate . the relative strength of the [ oiii]@xmath15007 and h@xmath2 lines reveals further excitation mechanism and gas metallicity information . the role of agn photoionisation is confined to the central 2 3 " of active , massive nearby elliptical galaxies ( sarzi et al . thus , ifu observations are ideal and will also allow us to study the 2d - distribution of the ionising radiation . in addition to the information from the emission lines , we are able to extract the underlying stellar absorption spectra using the improved gandalf code ( sarzi et al . 2006 ) . thus , the kinematics and morphology of the hot ionised gas and stellar components can be correlated . [ cols="<,^,^,^,^,^,^,^,^ " , ] some ellipticals contain dust in the centre that can be patchy , uniform or filamentary ( laine et al . the long wavelength range of the spectra allows us to constrain the amount of reddening using the observed decrement of the balmer lines , which can be set to have an intrinsic decrement consistent with the recombination theory by treating the lines as a multiplet . the physical constraints on the emission from the higher - order balmer lines also ensures the strength of the corresponding absorption features is correctly estimated . we used the dust models by calzetti et al . ( 2000 ) to calculate the flux attenuation values at the desired wavelength for any given @xmath15 value ( optional see below ) . the balmer decrement assumes a case b recombination for a density of 100 @xmath16 and a temperature of 10@xmath17 k , resulting in the predicted h@xmath0/h@xmath2 ratio of 2.86 ( osterbrock 1989 ) . the code can adopt either a single dust component , affecting both the stellar continuum and the emission - line fluxes , or in addition include a second dust component that affects only the emission - line templates . we did not specify galactic extinction ( from the ned database , schlegel , finkbeiner @xmath4 davis 1998 ) , but we measured a total diffuse component which gave the total extinction ( hence including the foreground galactic extinction ) . this total extinction measured from the balmer decrement are shown in figure [ fig : mcg_extinct ] and averages are noted in table [ table : objects2 ] ( the galactic extinction noted in table [ table : objects2 ] was not subtracted ) . the extinction was smoothed over 0.3 arcseconds , and is only plotted where the velocity dispersion of the h@xmath0 line is less than 500 km s@xmath11 ( to avoid spaxels where the h@xmath0 line could not be separated from the [ nii ] lines ) , and it is only plotted where the amplitude - to - noise ( a / n ) ratio of the h@xmath0 line is higher than 3 ( as defined in sarzi et al . 2006 ) where @xmath18 is the equivalent width of the line , and @xmath19 is the line width . ] . the a / n maps of the h@xmath0 measurements are shown in section [ figures_nvss ] which gives a direct indication of the errors on the extinction . the parameter @xmath15 , i.e. the colour excess between 4350 and 5550 , for the galactic extinction for each of the four galaxies was taken from the ned database ( schlegel et a. 1998 ) , and ranged between 0.037 and 0.140 mag ( table [ table : objects2 ] ) . the parameter r@xmath20 , i.e. the ratio of the absolute extinction at 5550 ( a@xmath20 ) to the colour excess @xmath15 , was taken as 3.1 for the interstellar medium ( cardelli , clayton @xmath4 mathis 1989 ) . the total extinction is given by : @xmath21 the theoretical h@xmath0/h@xmath2 flux ratio of 2.86 may not be the ideal value to use for seyfert - type galaxies , but the actual value is debated . it is often assumed that the h@xmath0 emission in these systems is enhanced due to collisional processes , and several authors use a value @xmath22 of 3.1 ( gaskell @xmath4 ferland 1984 ; osterbrock @xmath4 ferland 2006 ) , although other values have also been determined ( binette et al . 1990 calculate a value of 3.4 ) . the total extinction maps are presented in figure [ fig : mcg_extinct ] , and shows mostly very low extinction , but some morphological features can be seen in mcg-02 - 12 - 039 and ugc09799 . particularly high values of @xmath23 can be seen in ugc09799 . the galactic extinction of pgc044257 is @xmath24 = 0.071 mag , and from long - slit spectra , crawford et al . ( 1999 ) derived the total extinction as 0.46 to 0.63 mag . this agrees with the extinction we derived for the very centre of the galaxy in figure [ fig : mcg_extinct ] , but on average our spatially resolved extinction is slightly lower . the galactic extinction @xmath24 of ugc09799 is 0.037 mag , and crawford et al . ( 1999 ) derived an integrated internal extinction of @xmath23 of 0.22 mag for the centre of this galaxy . this corresponds very well to what we derived and plotted in figure [ fig : mcg_extinct ] , although we find that some regions show much higher internal extinction . the values of extinction determined here may be slightly overestimated due to the choice of intrinsic h@xmath0/h@xmath2 flux used . figure [ fig : thumbnails2 ] show slices with width 1 ( smoothed spatially with gaussian fwhm 1.5 spaxels and using the sauron colourmap ) of pgc026269 showing the wavelength and spatial scale over which the morphology of h@xmath0 changes . the morphology of the nuclear region changes rapidly across the emission lines . this is illustrated in the figure showing a series of monochromatic slices . although the continuum emission is usually smooth , the morphologies of the line emission are not uniform . we are able to measure h@xmath0 line fluxes with 3 a / n accuracy at around 5 @xmath25 10@xmath26 erg @xmath27 s@xmath11 . after correcting for extinction , we are able to measure the following lines within our wavelength range : [ ariv]@xmath14740 ; h@xmath284861 ; [ oiii]@xmath294958,5007 ; [ ni]@xmath295198,5200 ; hei@xmath15876 ; [ oi]@xmath296300,6364 ; [ nii]@xmath296548,6583 ; h@xmath0 at 6563 ; [ sii]@xmath296716,6731 ( see figures [ fig : mcgkinematics2 ] to [ fig : ugckinematics3 ] and [ hbeta ] for the a / n ratios of the individual emission lines ) . h@xmath0 and [ nii]@xmath296548,6583 has already been measured for mcg-02 - 12 - 039 and ugc09799 , and h@xmath2 and [ oiii]@xmath294958,5007 has also been measured for ugc09799 . all the other line measurements are new . .pdf the continuum emission of the galaxies are shown in figure [ fig : thumbnails ] . figure [ fig : mcg_gandalf ] , [ fig : pgc026_gandalf ] , [ fig : pgc044_gandalf ] and [ fig : ugc_gandalf ] shows a random spectrum in the central region of mcg-02 - 12 - 039 , pgc026269 , pgc044257 and ugc09799 respectively . the best - fitting stellar template , gaussians at the emission lines , best - fitting stellar templates with the emission lines subtracted and the relative flux for the measured emission lines are also shown . the absorption - extracted , dereddened maps of the h@xmath0 , [ nii]@xmath16583/h@xmath0 , and [ oiii]@xmath15007/h@xmath2 emission are shown in figure [ haflux ] . for comparison purposes , the kinematics were extracted from the ifu image along the same slit position as the long - slit data in loubser et al . this comparison for mcg-02 - 12 - 039 , pgc026269 and pgc044257 is showed in section [ stellar_kinematics_comparison ] , and the data points compare satisfactory . the absorption - extracted , dereddened maps of the h@xmath0 emission are shown in figure [ haflux ] in units of 10@xmath30 erg @xmath27 s@xmath11 . we also show the useful ratios [ nii]@xmath16583/h@xmath0 , and [ oiii]@xmath15007/h@xmath2 . figures [ fig : mcgkinematics2 ] to [ fig : ugckinematics3 ] shows the a / n ratios of the h@xmath0 , [ nii]@xmath16583 ; [ sii]@xmath296716,6731 ; [ oiii]@xmath15007 ; [ oi]@xmath16300 lines . as mentioned in the introduction , mcg-02 - 12 - 039 and ugc09799 already has previous ifu observations which is compared ( in terms of s / n and lines measurable ) in the beginning of section 3 and the last paragraph of section 4 . the stellar kinematics in figures [ fig : mcgkinematics ] , [ fig : pgc026kinematics ] , [ fig : pgc044kinematics ] and [ fig : ugckinematics ] were measured using all the absorption lines within our wavelength range . for a sample of @xmath3150 elliptical galaxies , sarzi et al . ( 2006 ) find ionised gas velocities ( estimated using the [ oiii]@xmath32 line ) between 250 and 250 km s@xmath11 , and gas velocity dispersion as high as 250 km s@xmath11 . using the same line , we find gas velocities from @xmath33 to @xmath34 km s@xmath11 , and line widths from 200 to 420 km s@xmath11 . using h@xmath0 , we find gas velocities from @xmath35 to @xmath34 km s@xmath11 , and line widths from 200 to 400 km s@xmath11 . the central galaxy ( mcg-02 - 12 - 039 ) of abell 0496 is a fairly weak line emitter ( fabian et al . 1981 ; hu , cowie @xmath4 wang 1985 ) and is host to the compact radio source msh 04 - 112 ( markovic , owen @xmath4 eilek 2004 ) . abell 0496 is a relaxed cluster with a cool core with a central metal abundance enhancement ( tamura et al . 2001 ) . at a redshift of 0.0329 , the galaxy has a linear scale of 0.656 kpc arcsec@xmath11 ( from hatch et al . this cluster has an interesting h@xmath0 morphology ( mcdonald et al . there are at least 5 distinct filaments , with various shapes and directions . the two longest filaments run parallel to each other for @xmath31 12 kpc ( 18.3 arcsec , whereas our observations only cover 3.5 by 5 arcsecs in the centre ) . figure [ haflux ] shows that the peak h@xmath0 emission is also where the dust ( extinction ) is the highest in figure [ fig : mcg_extinct ] . this is similar to the finding of hatch et al . ( 2007 ) , where they found that the peak h@xmath0 + [ nii ] emission corresponds to the area where the three dust lanes meet in the galaxy centre . we confirm the observation by hatch et al . ( 2007 ) , that the line emission follows the general path of the dust ( extinction ) features , but is not as filamentary . ifu observations of this galaxy was also taken by hatch et al . ( 2007 ) , but they only measured the h@xmath0 and [ nii]@xmath296548 , 6583 lines . their spectral resolution is much poorer at 223 - 273 km s@xmath11 than ours at 81 km s@xmath11 . our s / n is also more than five times that of hatch et al . ( 2007 ) , observed with a much bigger instrument . we plot h@xmath0 , [ nii ] , [ sii ] , [ oiii ] and [ oi ] velocities , line width and a / n in figure [ fig : mcgkinematics2 ] and [ fig : mcgkinematics3 ] . the stellar component displays a elongated morphology in the kinematics plotted in figure [ fig : mcgkinematics ] . the gas components also appear elongated ( kinematics plotted in figure [ fig : mcgkinematics2 ] ) , but to a lesser extend . it suggests that the stars and gas are kinematically decoupled . comparison of the gas kinematics ( figures [ fig : mcgkinematics2 ] and [ fig : mcgkinematics3 ] ) with the stellar kinematics presented in figure [ fig : mcgkinematics ] , and figure [ fig : mcg_kinematics ] , as well as by fisher , illingworth @xmath4 franx ( 1995 ) confirms the hatch et al . ( 2007 ) observations that the two components are disconnected . comparison of the plots in figure [ fig : mcgkinematics2 ] showed that , to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas . we find a maximum emission line width of 400 km s@xmath11 as shown in figures [ fig : mcgkinematics2 ] and [ fig : mcgkinematics3 ] . this is again similar to the finding of hatch et al . ( 2007 ) . they found a maximum linewidth of 600 km s@xmath11 in the dust - free central region to the north - east of the galaxy centre , and the rest of the nebula to have a linewidth of 100 250 km s@xmath11 . the stellar line - of - sight velocity reveals that some parts of the nucleas are blueshifted by @xmath31125 km s@xmath11 and some parts are redshifted by the same amount . hatch et al . ( 2007 ) found the southern part of the galaxy to be blueshifted by 200 km s@xmath11 whilst the northern section is marginally redshifted up to + 150 km s@xmath11 . no clear kinematic pattern is associated with the dust structures . the peak- to - peak gas velocity of 250 km s@xmath11 is fairly low compared to the other ccgs , and the stellar component of abell 0496 has a mean rotation of 59 km s@xmath11 ( paper 1 ) . this galaxy coincides with the well - known luminous radio counterpart hydra a. abell 0780 is a poor cluster with an associated cooling flow nebulae . ifu observations of this galaxy has not been presented in the literature before this study . the stellar line - of - sight velocity ( from the ifu data ) is blue - and redshifted by @xmath31175 km s@xmath11 which is slightly higher than that derived from long slit spectroscopy ( 51 @xmath36 20 km s@xmath11 paper 1 ) . we plot the kinematics of the stellar and gaseous components in figures [ fig : pgc026kinematics ] and [ fig : pgc026kinematics2 ] . both the stellar and gaseous components show clear rotation in figures [ fig : pgc026kinematics ] and [ fig : pgc026kinematics2 ] , although it also seems to be decoupled . this system bears resemblance to the bcg ncg 3311 in abell 1060 , studied by edwards et al . ( 2009 ) which also showed striking rotation . the rotation of the warm gas agrees with the long - slit observations by mcdonald et al . ( 2012 ) . we plot h@xmath0 , [ nii ] , [ sii ] , [ oiii ] and [ oi ] velocities , line width and a / n in figure [ fig : pgc026kinematics2 ] and [ fig : pgc026kinematics3 ] . comparison of the plots in figure [ fig : mcgkinematics2 ] showed that , to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas . wise et al . ( 2007 ) present a summary of the x - ray and radio properties of this cluster , showing the excellent correlation between the radio jets and the x - ray cavities . the arcing h@xmath0 filament that mcdonald et al . ( 2010 ) detect north of the ccg ( on a larger scale as this study ) appears to be spatially correlated with the radio jet . our figure [ fig : bpts ] ( d f ) show that our optical observations confirms that the photoionisation is caused by an agn ( the green and red curves ) , and are therefore consistent with the x - ray and radio studies of this system on larger scales . this ccg is one of the rare cases ( occurs in @xmath31 3 per cent of ccgs ; hamer et al . 2012 ) where the ccg is slightly offset from the optical line emission ( johnson et al . the rarity of such offsets points to a large event in cluster evolution , a major cluster merger or possibly a powerful agn outburst . whatever the reason for the separation the gas cooling at the x - ray peak will continue and cooled gas will be deposited away from the ccg ( hamer et al . 2012 ) . ifu observations of this galaxy has not been presented in the literature before this study . the stellar line - of - sight velocity is blue - and redshifted by @xmath31225 km s@xmath11 which is higher than derived from long slit spectroscopy ( 20 @xmath36 16 km s@xmath11 paper 1 ) . we plot the kinematics of the stellar and gaseous components in figures [ fig : pgc044kinematics ] and [ fig : pgc044kinematics2 ] . both the stellar and gaseous components show morphologies ( figures [ fig : pgc044kinematics ] and [ fig : pgc044kinematics2 ] ) that are elongated and aligned . however the recession velocity of the gas is @xmath31 200 km s@xmath11 , which is higher than that of the stars , suggesting that the gaseous and stellar components are decoupled . the rotation of the warm gas agrees with the long - slit observations by mcdonald et al . ( 2012 ) . we plot h@xmath0 , [ nii ] , [ sii ] , [ oiii ] and [ oi ] velocities , line width and a / n in figure [ fig : pgc044kinematics2 ] and [ fig : pgc044kinematics3 ] . to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas . we find the h@xmath0 flux to be uniform in figure [ haflux ] suggesting that all of the gas had the same origin , but the h@xmath0 flux is also quite low . if figure [ haflux ] is compared to figure [ fig : pgc044kinematics2 ] , then it can be seen that some structure is visible where the a / n of h@xmath0 is the highest . the h@xmath0 appear to be quite core - dominated . mcdonald et al . ( 2010 ) found that the direction of the h@xmath0 filaments ( from narrow band imaging ) correlate with the position of nearby galaxies . the filaments in abell 1644 also appear to be star - forming ( mcdonald et al . the star forming filaments that mcdonald et al . ( 2011 ) detected are on a bigger scale than the region we probe in this study . our results show that the very centre of the system display properties of a liner ( see section [ ionisation ] ) , and therefore both mechanisms may be part of this system at different scales . ifu observations of this galaxy ( around h@xmath0 and h@xmath2 ) was previously presented by edwards et al . the observations presented in this study are improved in that it was observed with triple the integration time of the previous observations . the morphology of the continuum emission of this galaxy ( see figure [ fig : thumbnails ] ) is centrally concentrated and condensed , as was also found by edwards et al . ( 2009 ) . just fitting one single gaussian per emission line resulted in very poor fits ( as shown in figure [ fig : ugc_before_after ] ) . we also tried fitting voigt profiles to the [ oiii ] lines , for example , as well as slightly offset blue and red velocity wings . the best fits were consistently achieved using 2 gaussians for each individual line in the [ oiii ] , [ oi ] , [ nii ] and [ sii ] doublets , of which one of the gaussians were broader than the other . the h@xmath2 lines ( and therefore also h@xmath0 ) however required no additional gaussian to achieve a good fit . edwards et al . ( 2009 ) could not detect h@xmath2 emission above the 1@xmath19 level for this galaxy . we do detect h@xmath2 above this level and plot the flux ratios in figure [ haflux ] . figure 16 in edwards et al . ( 2009 ) can be compared with figures [ fig : ugckinematics2 ] and [ haflux ] here both show smooth centrally condensed emission . we plot the kinematics of the stellar and gaseous components in figures [ fig : ugckinematics ] and [ fig : ugckinematics2 ] . we plot h@xmath0 , [ nii ] , [ sii ] , [ oiii ] and [ oi ] velocities , line width and a / n in figure [ fig : ugckinematics2 ] and [ fig : ugckinematics3 ] . to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas . our kinematic analysis can be compared to that of edwards et al . ( 2009 ) , who detected a gradient in the velocity but could not differentiate between a rotation or outflow . our stellar kinematics is shown in figure [ fig : ugckinematics ] , however we do not detect rotational kinematics . edwards et al . ( 2009 ) also found h@xmath0 to have a velocity range of 250 km s@xmath11 to + 150 km s@xmath11 . our h@xmath0 kinematics are shown in [ fig : ugckinematics2 ] and show also show a range ( @xmath31 400 km s@xmath11 ) of velocities for h@xmath0 . the gaseous components show rotation in figure [ fig : ugckinematics2 ] even though no rotation is apparent in the stellar components ( figure [ fig : ugckinematics ] ) . this galaxy has patchy dust in the centre ( laine et al . 2003 , also shown in figure [ fig : mcg_extinct ] ) , and hicks @xmath4 mushotzky ( 2005 ) have deduced star formation from the excess uv ir emission . the fuv / h@xmath0 ratio suggests heating by fast shocks or some other source of hard ionisation ( e.g. , cosmic rays , agn ) . the filaments of abell 2052 have ratios which are consistent with this picture ( mcdonald et al . 2011 ) . the highest extinction ( in figure [ fig : mcg_extinct ] ) does not coincide with the h@xmath0 peak ( appendix b1 ) , and is slightly off centre . this does not necessarily contrast with the results quoted by laine et al . ( 2003 ) that the dust is patchy in the centre , as the centre of the observations might not necessarily be exactly the same as our ifu placement ( see figure 1 ) . figure 1 also shows that our continuum emission is smooth , similar to the conclusion reached by edwards et al . ( 2009 ) . venturi , dallacasa @xmath4 stefanachi ( 2004 ) found a parsec - scale bipolar radio source , and chandra x - ray emission show two bubbles in the icm on a larger scale ( blanton , sarazin @xmath4 mcnamara 2003 ) . our results are consistent with the central emission being that of a liner ( see section [ ionisation ] ) . various mechanisms have been proposed as sources of excitation in the nebulae , including photoionisation by radiation from the agn , cluster x - rays , or hot stars , collisional heating by high - energy particles , shocks or cloud - cloud collisions , and conduction of heat from the x - ray corona ( peterson et al . . however , none so far satisfactorily describes all the characteristic spectra , energetics , and kinematics of the extended emission - line regions ( wilman et al . 2006 ; hatch et al . 2007 ) . we reiterate that the fov of the observations are only 3.5 @xmath25 5 arcsec ( corresponding to the central few parsec see table [ table : objects ] ) . these results only look at the very heart of these sources . therefore , it is really only the emission mechanisms for the gas at the very center of these systems that is being studied here . the forbidden lines such as [ nii]@xmath16583 result from the excitation of n@xmath37 through collisions with electrons liberated through photoionisation . the h@xmath0 emission results from the recombination of the hydrogen ion . [ nii]@xmath16583 flux depends on the n@xmath37 abundance , the strength of the radiation field , and the form of the radiation field : a harder ionising source will produce a greater flux . the h@xmath0 flux also depends on the strength of the radiation field . therefore , the ratio will depend on the metallicity of the gas and the form of the ionising radiation . the form of the ionising radiation and/or the gas metallicity are not uniform but must vary within each galaxy and between the whole sample . different excitation mechanisms may act in different regions ( hatch et al . 2007 , edwards et al . 2009 ) , or the h@xmath0 emission might be disturbed by the presence of companion galaxies ( wilman et al . 2006 ) . a commonly used method to distinguish between the sources of ionisation uses the emission - line diagrams pioneered by baldwin , phillips @xmath4 terlevich ( 1981 , hereafter bpt diagram ) which separate the two major origins of emission : star formation and agn . the diagrams use pairs of emission line ratios , of which the most commonly used is [ oiii]@xmath15007/h@xmath2 against [ nii]@xmath16584/h@xmath0 . extinction - corrected emission - line measurements were used ( even though the bpt diagram is almost insensitive to reddening ) . we plot the bpt diagrams in figures [ fig : mcg_bpt ] and [ fig : ugc_bpt ] using the following criteria : kewley criteria : galaxies above this line are agn . kewley et al . ( 2001 ) used a combination of photoionisation and stellar population synthesis models to place a theoretical upper limit on the location of star forming galaxies on the bpt diagram , @xmath38}{h\beta } = 0.61/{\log([nii]/h\alpha)-0.47}+1.19)$ ] . kauffmann et al . ( 2003 ) criteria : galaxies below this line are purely star forming . between kewley and kauffmann criteria is composite objects . kauffmann et al . ( 2003 ) revised the criteria as follows : a galaxy is defined to be an agn if @xmath38}{h\beta } > 0.61/{\log([nii]/h\alpha)-0.05}+1.3)$ ] . liners seyffert line ( schawinsky et al . 2007 ) : this line distinguishes liners from seyffert galaxies @xmath38}{h\beta } ) = 1.05 \log(\frac{[nii]}{h\alpha})+0.45 $ ] . sdss emission - line galaxies occupy a well - defined region shaped like the wings of a seagull ( stasinska et al . although the exact location of the line dividing the star forming and active galactic nuclei galaxies is still controversial ( kewley et al 2001 ; kauffmann et al 2003 ; stasinska et al 2006 ) . all galaxies show important liner emission , but that at least one has significant seyfert emission areas , and at least one other has significant hii like emission line ratios for many pixels as shown in figures [ fig : mcg_bpt ] and [ fig : ugc_bpt ] . this is in agreement with the long - slit data of these sources plotted on bpt diagrams in figure 3 of mcdonald et al . ( 2012 ) . however , there is a debate about the ionisation mechanism in liners ( low - ionisation nuclear emission - line region ) . the most viable excitation mechanisms are : a low accretion - rate agn ( kewley et al . 2006 ) , photoionisation by old post - asymptotic giant branch ( pagb ) stars ( stasinska et al . 2008 ) , and fast shocks ( dopita @xmath4 sutherland 1995 ) . sarzi et al . ( 2010 ) investigate the ionising sources for the gas in elliptical galaxies based on sauron intergral - field spectroscopy whose spectra are limited to a relatively narrow wavelength range . they conclude that pagb stars are the main source of ionisation . in contrast , annibali et al . ( 2010 ) analyse long - slit spectra of 65 ellipticals and claim that their nuclear line emission can be explained by excitation from the hard ionising continuum from an agn and/or fast shocks . however , they can not completely rule out a contribution from pagb stars at large radii . voit @xmath4 donahue ( 1997 ) suggested that sources of supplementary heating produce the liner - like properties of the spectra , though not necessarily through the same mechanism in all systems . annibali et al . ( 2010 ) found that from the centre outward , galaxies move left and down in the bpt diagram for their study of 65 early - type galaxies . thus , the hardness of the ionising continuum decreases with galactocentric distance ( up to half the half - light radius in the annibali et al . ( 2010 ) sample ) . figures [ fig : mcg_bpt ] and [ fig : ugc_bpt ] also show the flux ratios as function of distance from the galaxy centre . the red circles indicate the central 0.5 @xmath25 0.5 arcsec of the galaxy , the yellow circles 1.0 @xmath25 1.0 arcsec , and the blue circles the full 3.5 @xmath25 5.0 arcsec . the centre of the galaxy was determined as the luminosity peak in the continuum images in figure [ fig : thumbnails ] . in our case , the hardness of the ionising continuum stay mostly uniform with galactocentric distance over out limited spatial extend . pgc044257 show an interesting core separation of the emission in the very centre of the galaxy in figure [ fig : ugc_bpt]a . there is also the debate about which galaxies are liners ( agn ) and which have just liner - like emission ( non - agn ) . the division is usually made by looking at the extend of the liner signature : core dominated mean true liner and diffuse means liner - like . the arising problem is that the scale of the sdss fibers is already too large to make the distinction for most galaxies . for the nearby galaxies , the centroiding of the fibres is not accurate enough to be sure if the measured spectra cover the core of the galaxy . we have core dominated liner emission for at least three out of the four galaxies . additional confirmation comes from the fact that at least one of our galaxies ( pgc026269 ) is a strong radio source ( agn , see [ table : objects2 ] ) . similarly , our findings for ugc09799 agree with edwards et al . ( 2009 ) who found line ratios consistent with seyffert or liner activity in most of their central spaxels for ugc09799 . edwards ( 2009 ) also found that the seyffert signature dominated the central spaxels of the ccg ( ugc09799 ) . the h@xmath0 emission surrounding the bcg in this cluster is coincident with radio - blown bubbles in the central region of the cluster . these bubbles to the north and south of the cluster core are filled with radio emission , which likely originated from the agn within the ccg ( blanton et al . 2003 ) . since the h@xmath0 emission seen by mcdonald et al . ( 2010 ) is primarily along the edges of the northern bubble , they suspect that shocks may be responsible for the heating in this case . however , for pgc044257 , mcdonald et al . ( 2010 ) find very little evidence for an agn in the cluster hosting the galaxy , in terms of the radio power , x - ray morphology and hard x - ray flux . two of the galaxies in the current study have detected x - ray point sources in the ccg ( pgc026269 and ugc09799 ) with large cavities in their x - ray haloes , suggesting that the agn is influencing the surrounding medium . we investigate this further by plotting the other bpt diagrams , on a spaxel - by - spaxel basis , for all four galaxies ( shown in figure [ fig : bpts ] ) : [ oiii]@xmath15007/h@xmath2 vs [ nii]@xmath16584/h@xmath0 , [ oiii]@xmath15007/h@xmath2 vs [ sii]@xmath36717,6731/h@xmath0 , and [ oiii]@xmath15007/h@xmath2 vs [ oi]@xmath16300/h@xmath0 . one bcg pgc026269 show several hii pixels in all three bpt diagrams . figure [ fig : mcg_bpt ] show that these hii pixels occur throughout the centre of the galaxy . on these plots , we compare our observations to the photoionisation models for pagb stars with z = z@xmath14 ( binette et al . these models are consistent with most of our observations . a model with z=1/3 z@xmath14 will be shifted towards lower values on the x - axis of the three bpts . the pagb scenario has recently been revisited by stasnska et al . ( 2008 ) , whose extensive grid of photoionisation models ( see their figure 5 ) cover most of the regions occupied by our spatially resolved measurements . three other grids of ionisation models are overplotted on the bpt diagrams ( figure [ fig : bpts ] ) . the plotted agn photoionisation models ( groves , dopita @xmath4 sutherland 2004 ) have an electron density , n@xmath39 = 100 @xmath16 , metallicities of solar , z = z@xmath14 ( red grids ) , and twice solar ( green grids ) , a range of ionisation parameter ( 3.6 @xmath40 0.0 ) and a power - law ionising spectrum with spectral index @xmath0 = 2 , 1.4 , and 1.2 . a harder ionising continuum , with @xmath0 = 1.2 , boosts [ sii]@xmath416717,6731 and [ oi]@xmath16300 relative to h@xmath0 . we also compared our results with shock models ( allen et al . 2008 , purple grids ) . in figure [ fig : bpts ] , we plot the grids with z = z@xmath14 , preshock densities of 100 @xmath16 , shock velocities of 100 , 500 and 1000 km s@xmath11 , and preshock magnetic fields of b=1 , 5 and 10 @xmath42 g . shock models with a range of magnetic field strengths ( b=1,5 and 10 @xmath42 g ) match our observations . interstellar magnetic fields of b @xmath31 1 10 @xmath42 g are typical of what is observed in elliptical galaxies ( mathews @xmath4 brighenti 1997 ) . overall , shock models reproduce the majority of our data in the three emission - line ratio diagrams . the shock grids with lower metallicity ( e.g. lmc and smc metallicities ) are not consistent with our measurements.brent/itera.html . ] as shown in table [ table : objects2 ] we have very weak as well as strong radio fluxes in our small sample . we therefore believe that we are not particularly prone to biases such as the fact that a priori choice of galaxies with strong radio fluxes will result in finding a sample where high [ nii]/h@xmath0 would be more common than in an optically , or h@xmath0 selected sample only where starforming high h@xmath0/[nii ] sources might be more abundant . we present detailed integral field unit ( ifu ) observations of the central few kiloparsecs of the ionised nebulae surrounding four active ccgs in cooling flow clusters ( abell 0496 , 0780 , 1644 and 2052 ) . our sample consists of ccgs with h@xmath0 filaments . we observed the detailed optical emission - line ( and simultaneous absorption line ) data over a broad wavelength range to probe the dominant ionisation processes , excitation sources , morphology and kinematics of the hot gas ( as well as the morphology and kinematics of the stars ) . two of the four sources have not been observed with ifu data before ( abell 0780 and 1644 ) , and for the other two sources we observed with significantly improved integration times ( and number of lines ) than previous studies ( hatch et al . 2007 , edwards et al . 2009 ) . this will help form a complete view of the different phases ( hot gas and stars ) and how they interact in the processes of star formation and feedback detected in central galaxies in cooling flow clusters , as well as the influence of the host cluster . the total extinction maps are presented in figure [ fig : mcg_extinct ] and shows extinction which agrees well with previously derived long - slit values ( where available ) . from long - slit spectra , crawford et al . ( 1999 ) derived the total extinction of pgc044257 as 0.46 to 0.63 mag . this agrees with the extinction we derived for the very centre of the galaxy in figure [ fig : mcg_extinct ] , but on average our spatially resolved extinction is slightly lower ( 0.195 mag ) . crawford et al . ( 1999 ) derived an integrated internal extinction of @xmath23 of 0.22 mag for the centre of ugc09799 . this corresponds very well to what we derived and plotted in figure [ fig : mcg_extinct ] , although we find that some regions show much higher internal extinction ( on average 0.42 mag ) . we derive a range of different kinematic properties , given the small sample size . for abell 0496 and 0780 , we find that the stars and gas are kinematically decoupled , and in the case of abell 1644 we find that these components are aligned . for abell 2052 , we find that the gaseous components show rotation even though no rotation is apparent in the stellar components . to the degree that our spatial resolution reveals , it appears that all the optical forbidden and hydrogen recombination lines originate in the same gas for all the galaxies . all galaxies show important liner emission , but that at least one has significant seyfert emission areas , and at least one other has significant hii like emission line ratios for many pixels ( consistent with the long - slit observations of mcdonald et al . ( 2012 ) ) . we also show that the hardness of the ionising continuum does not decrease with galactocentric distance ( except for pgc044257 which show an interesting core separation of the emission in the very centre of the galaxy in figure [ fig : ugc_bpt]a ) . the radial profiles of diagnostic line ratios , [ oiii]@xmath15007/h@xmath2 and [ nii]@xmath16584/h@xmath0 , show that they are roughly constant with radius for three of the four galaxies ( all except pgc044257 ) . this indicates that the dominant ionising source is not confined to the nuclear region in the two objects and that the ionised gas properties are homogeneous in the emission - line regions across each galaxy . overall , it remains difficult to dissentangle the dominant photoionisation mechanisms , even with more line measurements . the agn photoionisation models ( with higher metallicity ) are the best able to reproduce our spatially - resolved line ratios the in all of the three bpt diagrams simultaneously of most objects , even though shock models and pagb stars can not be conclusively eliminated . we also do not see extended [ oi]@xmath16300 emission , following the morphology of the strong emission lines ( e.g. farage et al . 2010 ) , therefore , it is unlikely that shock excitation is the dominant ionising source in the galaxies of this limited sample . in addition , multiwavelength observations ( as discussed in the previous section ) favour the agn photoionisation mechanism , especially in the case of pgc026269 and ugc09799 . sil is financially supported by the south african national research foundation . we thank james turner , bryan miller , and michele cappellari for providing helpful scripts , as well as marc sarzi for helpful discussions . based on observations obtained on the gemini south telescope . the gemini observatory is operated by the association of universities for research in astronomy , inc . , under cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( usa ) , the science and technology facilities council ( uk ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) and conicet ( argentina ) . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , california institute of technology . 99 allen m.g . , groves b.a . , dopita m.a . , sutherland r.s . , kewley l.j . , 2008 , apjs , 178 , 20 annibali f. , bressan a. , rampazzo r. , zeilinger w.w . , vega o. , panuzzo p. , 2010 , a@xmath4a , 519 , 40 athanassoula e. , garijo a. , garca - gmez c. , 2001 , mnras , 321 , 353 baldwin j.a . , phillips m.m . , terlevich r. , 1981 , pasp , 93 , 5 bildfell c. , hoekstra h. , babul a. , mahdavi a. , 2008 , mnras , 389 , 1637 binette l. , raga a.c . , calvet n. , canto j. , 1990 , pasp , 102 , 723 binette l. , magris c.g . , stasiska g. , bruzual a.g . , 1994 , a@xmath4a , 292 , 13 blanton e.l . , sarazin c.l . , mcnamara b.r . , 2003 , apj , 585 , 227 bohringer h. , et al . , 2001 , a@xmath4a , 365 , 181 brough s. , tran k .- v . , sharp r.g . , von der linden a. , couch w.j . , 2011 , mnras , 414l , 80 calzetti d. , armus l. , bohlin r.c . , kinney a.l . , koornneef j. , storchi - bergmann t. , 2000 , apj , 533 , 682 cappellari m. , emsellem e. , 2004 , pasp , 116 , 138 cardelli j.a . , clayton g.c . , mathis j.s . , 1989 , apj , 345 , 245 cardiel n. , gorgas j. , aragn - salamanca a. , 1998 , mnras , 298 , 977 crawford c.s . , allen s.w . , ebeling h. , edge a.c . , fabian a.c . , 1999 , mnras , 306 , 857 crawford c.s . , hatch n.a . , fabian a.c . , sanders j.s . , 2005 , mnras , 363 , 216 de lucia g. , blaizot j. , 2007 , mnras , 375 , 2 dopita m.a . , sutherland r.s . , 1995 , apj , 455 , 468 edge a.c . , stewart g.c . , fabian a.c . , 1992 , mnras , 258 , 177 edwards l.o.v . , hudson m.j . , balogh m.l . , smith r.j . , 2007 , mnras , 379 , 100 edwards l.o.v . , robert c. , moll m. , mcgee s.l . , 2009 , mnras , 396 , 1953 fabian a.c . , nulsen p.e.j . , steward g.c . , ku w.h .- m . , malin d.f . , mushotzky r.f . , 1981 , mnras , 196 , 35 farage c.l . , mcgregor p.j . , dopita m.a . , bicknell g.v . , 2010 , apj , 724 , 267 ferland g.j . , fabian a.c . , hatch n.a . , johnstone r.m . , porter r.l . , van hoof p.a.m . , williams r.j.r . , 2009 , mnras , 392 , 1475 fisher d. , illingworth g. , franx m. , 1995 , apj , 438 , 539 gaskell c.m . , ferland g.j . , 1984 , pasp , 96 , 393 gerssen j. , allington - smith j. , miller b.w . , turner j.e.h . , walker a. , 2006 , mnras , 365 , 29 groves b.a . , dopita m.a . , sutherland r.s . , 2004 , apjs , 153 , 9 hamer s.l . , edge a.c . , swinbank a.m. , wilman r.j . , russell h.r . , fabian a.c . , sanders j.s . , salom p. , 2012 , mnras , 421 , 3409 hatch n.a . , crawford c.s . , fabian a.c . , 2007 , mnras , 380 , 33 hicks a.k . , mushotzky r. , 2005 , apj , 635 , 9 hu e.m . , cowie l.l . , wang z. , 1985 , apjs , 59 , 447 johnson r.e . , markevitch m. , wegner g.a . , jones c. , forman w.r . , 2010 , apj , 710 , 1776 jordn a. , ct p. , west m.j . , marzke r.o . , minniti d. , rejkuba m. , 2004 , aj , 127 , 24 kauffmann g. , et al . , 2003 , mnras , 341 , 33 kewley l.j . , dopita m.a . , sutherland r.s . , heisler c.a . , trevena j. , 2001 , apj , 556 , 121 kewley l.j . , groves b. , kauffmann g. , heckman t. , 2006 , mnras , 372 , 961 laine s. , van der marel r.p . , lauer t.r . , postman m. , odea c.p . , owen f.n . , 2003 , aj , 125 , 478 loubser s.i . , sansom a.e . , snchez - blzquez p. , soechting i.k . , bromage g. , 2008 , mnras , 391 , 1009 ( paper 1 ) loubser s.i . , snchez - blzquez p. , sansom a.e . , soechting i.k . , 2009 , mnras , 398 , 133 loubser s.i . , snchez - blzquez p. , 2012 , mnras , 425 , 841 markovi t. , owen f.n . , eilik j.a . , 2004 , proceedings of the riddle of cooling flows in galaxies and clusters of galaxies , eds . t. reiprich , j. kempner , and n. soker . mathews w.g . , brighenti f. , 1997 , in asp conference series `` the nature of elliptical galaxies '' , vol 116 , 396 mccarthy i.g . , et al . , 2010 , mnras , 406 , 822 mcdonald m. , veilleux s. , rupke d.s.n . , mushotzky r. , 2010 , apj , 721 , 1262 mcdonald m. , veilleux s. , rupke d.s.n . , mushotzky r. , reynolds c. , 2011 , apj , 734 , 95 mcdonald m. , veilleux s. , rupke d.s.n . , 2012 , apj , 746 , 153 mcnamara b.r , oconnell r.w . , 1992 , apj , 393 , 579 mcnamara b.r . , wise m. , sarazin c.l . , jannuzi b.t . , elston r. , 1996 , apjl , 466 , 9 mcnamara b.r . et al . , 2006 , apj , 648 , 164 odea c.p . et al . , 2008 , apj , 681 , 1035 osterbrock d.e . , 1989 , astrophysics of gaseous nebulae and active galactic nuclei , university science books , mill valley ca osterbrock d.e . , ferland g.j . , 2006 , astrophysics of gaseous nebulae and active galactic nuclei , university science books , sausalito ca peres c.b . , fabian a.c . , edge a.c . , allen s.w . , johnstone r.m . , white d.a . , 1998 , mnras , 298 , 416 peterson j.r . , kahn s.m . , paerels f.b.s . , kaastra k.s . , tamura t. , bleeker j.a.m . , ferrigno c. , jernigan j.g . , 2003 , apj , 590 , 207 pipino a. , kaviraj s. , bildfell c. , hoekstra h. , babul a. , silk j. , 2009 , mnras , 395 , 462 romeo a.d . , napolitano n.r . , covone g. , sommer - larsen j. , antonuccio - delogu v. , capacciolo m. , 2008 , mnras , 389 , 13 salom p. , combes f. , 2003 , a@xmath4a , 412 , 657 snchez - blzquez p. et al . , 2006 , mnras , 371 , 703 sarzi m. et al . , 2006 , mnras , 366 , 1151 sarzi m. , et al . , 2010 , mnras , 402 , 2187 schawinsky k. , et al . , 2007 , mnras , 382 , 1415 schlegel d.j . , finkbeiner d.p . , davis m. , 1998 , apj , 500 , 525 stasiska g. , cid fernandes r. , mateus a. , sodr l.jr . , asari n.v . , 2006 , mnras , 371 , 972 stasiska g. , vale asari n. , cid fernandes r. , gomez j.m . , schlickmann m. , mateus a. , schoenel w. , sodr l.jr . , 2008 , mnras , 391 , 29 tamura t. , bleeker j.a.m . , kaastra j.s . , ferrigno c. , molendi s. , 2001 , a@xmath4a , 379 , 107 van dokkum p.g . , 2001 , pasp , 113 , 1420 venturi t. , dallacasa d. , stefanachi f. , 2004 , a@xmath4a , 422 , 515 voit g.m . , donahue m. , 1997 , apj , 486 , 242 voit g.m . , cavagnolo k.m . , donahue m. , rafferty d.a . , mcnamara b.r . , nulsen p.e.j . , 2008 , apj , 681 , 5 von der linden a. , best p.n . , kauffmann g. , white s.d.m . , 2007 , mnras , 379 , 867 white d.a . , jones c. , forman w. , 1997 , mnras , 292 , 419 wilman r.j . , , swinbank a.m. , 2006 , mnras , 371 , 93 wise m.w . , mcnamara b.r . , nulsen p.e.j . , houck j.c . , david l.p . , 2007 , apj , 659 , 1153 for comparison purposes , the kinematics were extracted from the ifu image along the same slit position as the long - slit data in paper 1 . this comparison is showed in figures [ fig : mcg_kinematics ] to [ fig : pgc044_kinematics ] , and the data points compare satisfactory . we emphasize that the long - slit data were binned to a constant ( higher ) s / n , where the ifu data are three binned spaxels ( 0.3 ) along the same alignment . [ lastpage ]

when solving linear stochastic differential equations numerically , usually a high order spatial discretisation is used . balanced truncation ( bt ) and singular perturbation approximation ( spa ) are well - known projection techniques in the deterministic framework which reduce the order of a control system and hence reduce computational complexity . this work considers both methods when the control is replaced by a noise term . we provide theoretical tools such as stochastic concepts for reachability and observability , which are necessary for balancing related model order reduction of linear stochastic differential equations with additive lvy noise . moreover , we derive error bounds for both bt and spa and provide numerical results for a specific example which support the theory . [ [ keywords ] ] keywords : + + + + + + + + + model order reduction , balanced truncation , singular perturbation approximation , stochastic systems , lvy process , gramians , lyapunov equations . [ [ ams - subject - classifications ] ] ams subject classifications + + + + + + + + + + + + + + + + + + + + + + + + + + + primary 93a15 , 93b40 , 93e03 , 93e30 , 60j75 . secondary 93a30 , 15a24 .

1611.05494

many mathematical models of real - life processes pose challenges during numerical computations , due to their large size and complexity . model order reduction ( mor ) techniques are methods that reduce the computational complexity of numerical simulations , an overview of mor methods is provided in @xcite . mor techniques such as balanced truncation ( bt ) and singular perturbation approximation ( spa ) are methods which have been introduced in @xcite and @xcite , respectively , for linear deterministic systems @xmath0 here @xmath1 is asymptotically stable , @xmath2 , @xmath3 and @xmath4 , @xmath5 , @xmath6 are state , output and input of the system , respectively . from the gramians @xmath7 and @xmath8 which solve dual lyapunov equations @xmath9 a balancing transformation is found , which is used to project the state space of size @xmath10 to a much smaller dimensional state space ( see , e.g. @xcite ) . recently , the theory for bt and spa has been extended to stochastic linear systems of the form @xmath11 where @xmath12 , @xmath13 and @xmath14 as above , and @xmath15 and @xmath16 ( @xmath17 ) are uncorrelated scalar square integrable lvy processes with mean zero ( often @xmath18 and the special case of wiener processes are considered , see , for example , @xcite ) . in this case bt and spa require the solution of more general lyapunov equations of the form @xmath19 where @xmath20 $ ] for general lvy processes . note that @xmath21 , @xmath22 for the case of a wiener process @xcite . we refer to @xcite for a detailed theoretical and numerical treatment of balancing related mor for ( [ linsysafterdisintr ] ) . in this paper we are going to study balancing related mor for systems of the form @xmath23 where @xmath24 and @xmath25 is the @xmath26 column of @xmath27 . the processes @xmath28 are the components of a square integrable mean zero lvy process @xmath29 that takes values in @xmath30 . consequently , these components are not necessarily uncorrelated . for a general theoretical treatment of sdes with lvy noise we refer to @xcite . the setting in ( [ linsysafterdisintr2 ] ) is of particular interest in many applications . if one is interested in a large number of different realisations of the output @xmath31 ( e.g. to compute moments of the form @xmath32 $ ] ) , then one needs to solve the sde in ( [ linsysafterdisintr2 ] ) a large number of times . for a state space of high dimension this is computationally expensive . reduction of the state space dimension decreases the computational complexity when sampling the solution to ( [ linsysafterdisintr2 ] ) , as the sde can then be solved in much smaller dimensions . hence the computational costs are reduced dramatically . the linear system ( [ linsysafterdisintr2 ] ) is a problem where the control is noise . in this case the standard theory for balancing related mor applied to a deterministic system no longer applies . balanced truncation has been applied to linear systems with white noise before . the discrete time setting was discussed in @xcite . for the continuous time setting , dissipative hamiltonian systems with wiener noise were treated in @xcite , but no error bounds were provided . in this paper we consider both bt and spa model order reduction . as far as we are aware , no theory and in particular error bounds for balancing related mor have been developed for continuous time sdes with lvy noise . using theory for linear stochastic differential equations with additive lvy noise we provide a stochastic concept of reachability . this concept motivates a new formulation of the reachability gramian . we prove bounds for the error between the full and reduced system which provide criteria for truncating , e.g. criteria for a suitable size of the reduced system . we analyse both bt and spa and apply the theory directly to an application arising from a second order damped wave equation . we now consider a particular example which explains why the above setting is of practical interest . [ [ motivational - example ] ] motivational example + + + + + + + + + + + + + + + + + + + + in @xcite the lateral time - dependent displacement @xmath33 of an electricity cable impacted by wind was modeled by the following one - dimensional symbolic second order spde with lvy noise : @xmath34 for @xmath35 $ ] , @xmath36 $ ] and @xmath37 , with boundary and initial conditions @xmath38 for small @xmath39 , the output equation @xmath40 is approximately the position of the middle of the cable . in @xcite , it is shown that transforming this spde in into a first order spde and then discretising it in space , leads to a system of the form ( [ linsysafterdisintr ] ) where @xmath41 . one drawback of the approach above is , that , when the electricity cable is in steady state , the wind has no impact . a more realistic scenario , which models the wind as some form of stochastic input , is the following symbolic equation @xmath42 for @xmath35 $ ] , @xmath36 $ ] and @xmath37 , boundary and initial conditions as in ( [ eq : bcic ] ) , and @xmath16 the components of a square integrable mean zero lvy process @xmath29 that takes values in @xmath30 . in this paper , we consider a framework which covers this model . moreover we modify the output in ( [ introbspout ] ) and let @xmath43 so that both the position and velocity of the middle of the string are observed . transformation and discretisation of this spde leads to a system of the form ( [ linsysafterdisintr2 ] ) where @xmath12 is an asymptotically stable matrix , i.e. @xmath44 . this paper is set up as follows . section [ sec : balancing ] provides the theoretical tools for balancing linear sdes with additive lvy noise . we explain the theoretical concepts of reachability and observability in this setting and show how this motivates mor using bt and spa . moreover we provide theoretical error bounds for both methods . in section [ sec : wave ] we show how a wave equation driven by lvy noise can be transformed into a first order equation and then reduced to a system of the form ( [ linsysafterdisintr2 ] ) by using a spectral galerkin method . numerical results which support our theory are provided in section [ sec : numerics ] . in @xcite balancing related mor was considered for deterministic systems of the form @xmath45 where @xmath46 was assumed to be asymptotically stable , i.e. @xmath44 , @xmath2 , @xmath3 and @xmath47)$ ] for all @xmath48 was a deterministic control . we now turn our attention to a stochastic system @xmath49 which , in section [ subsecspeltralgal ] , represents a spatially discretised version of an spde . the matrices @xmath12 , @xmath13 and @xmath14 are as above and the @xmath30-valued process @xmath50 is a square integrable lvy processes with mean zero . one might interpret system ( [ stochsysnew ] ) as system ( [ detsysant ] ) with @xmath51 but the noise @xmath52 is no control in the classical sense . first of all , stochastic controls were not admissible in the deterministic setting and secondly the classical derivative does not exist . so , if we want to study balancing related mor for the `` particular control '' @xmath53 , we need to make sense of this setting which we do by the ito - type sde in ( [ stochsysnew ] ) . in the deterministic case reachability and observability concepts are introduced to characterise the importance of states . difficult to reach states ( states which require large energy to reach them ) and difficult to observe states ( states which only produce little observation energy ) are seen to be unimportant in the systems dynamics . in balancing related mor the idea is to create a system , where the dominant reachable and observable states are the same . those are then truncated to obtain a reduced order model ( rom ) . applying balancing related mor to ( [ stochsysnew ] ) requires a few modifications compared to the classical deterministic framework . we introduce a stochastic reachability concept in section [ reachsec ] which also leads to a different reachability gramian compared to the deterministic case . for the observation concept we follow the deterministic approach . we then describe the procedure of balancing for systems with additive noise in section [ subsecprocedure ] which is similar to the deterministic case . afterwards , we will discuss two particular techniques which are bt and spa . since it is not a priori clear whether these approaches for system ( [ stochsysnew ] ) perform as well as for deterministic systems , we contribute an error bound for both bt and spa in section [ btandspa ] . these error bounds enable us to point out the cases , where bt and spa work well , and they can be used to find a suitable rom dimension . with suitable reachability and observability concepts we want to analyze which states in system ( [ stochsysnew ] ) are unimportant and hence can be neglected . [ [ reachability ] ] reachability + + + + + + + + + + + + we begin with a stochastic reachability concept , where the particular choice of @xmath50 is taken into account . starting from zero ( @xmath54 ) in @xmath55 we investigate how much the noise can control the state away from zero . we define what is meant be reachability in the stochastic case , where @xmath56 , @xmath57 , denotes the solution to ( [ stochstatenew ] ) with initial condition @xmath58 and noise process @xmath50 . a state @xmath59 is not reachable from zero on the time interval @xmath60 $ ] , @xmath48 , if it is contained in an open set @xmath61 with @xmath62\right\}=0 , \end{aligned}\ ] ] else @xmath63 is reachable . the system is called completely reachable if @xmath64\right\}>0 \end{aligned}\ ] ] for every open set @xmath65 . we refer to @xcite , where weak controllability was analyzed for equations with wiener noise . weak controllability turns out to be similar to condition ( [ completereach ] ) . to characterise the degree of reachability of a state , we introduce finite time reachability gramians @xmath66 $ ] which are the covariance matrices of @xmath67 at fixed times @xmath57 . before we study the meaning of these gramians , we show that @xmath68 is the solution of a matrix differential equation . the matrix - valued function @xmath68 , @xmath57 , is the solution to @xmath69 where @xmath70 $ ] is the covariance matrix of @xmath50 at time @xmath71 . we replace @xmath67 by @xmath72 to shorten the notation in the proof . using ito s formula in corollary [ iotprodformelmatpro ] , we obtain the following for @xmath73 , @xmath57 : @xmath74_t\right)_{i , j=1 , \ldots , n},\end{aligned}\ ] ] where @xmath75 is the @xmath76-th unit vector and we used @xmath54 . inserting the stochastic differential of @xmath72 yields @xmath77 since the ito integrals have mean zero , we have @xmath78=\int_0^t \mathbb e\left[x(s ) x^t(s)\right ] a^t ds+\int_0^t a \mathbb e\left[x(s ) x^t(s)\right ] ds+\left(\mathbb e[e_i^t x , e_j^t x]_t\right)_{i , j=1 , \ldots , n},\end{aligned}\ ] ] where we replaced @xmath79 by @xmath80 . this does not impact the integrals since a cdlg process has at most countably many jumps on a finite time interval ( see ( * ? ? ? * theorem 2.7.1 ) ) . applying corollary [ iotprodformelmatpro ] again , the stochastic differential of @xmath81 is given by : @xmath82_t\right)_{i , j=1 , \ldots , n},\end{aligned}\ ] ] taking the expected value , we have @xmath83b^t=\mathbb e\left([e_i^t bm , e_j^tbm]_t\right)_{i , j=1 , \ldots , n}$ ] . in ( * ? ? ? * theorem 4.44 ) it was shown that the covariance function is linear in @xmath84 , i.e. @xmath85=\mathcal q_m t$ ] . since the @xmath76th component @xmath86 has the same jumps and the same martingale part as @xmath87 , we know by ( [ decomqucov ] ) that @xmath88_t=[e_i^t x , e_j^t x]_t$ ] for @xmath89 . summarizing the results , we have @xmath78=\int_0^t \mathbb e\left[x(s ) x^t(s)\right ] a^t ds+\int_0^t a \mathbb e\left[x(s ) x^t(s)\right ] ds+b\mathcal q_m b^t t,\end{aligned}\ ] ] which concludes the proof . to find a representation for @xmath68 we need the following straightforward result . [ prodeinfachh ] let @xmath90 and @xmath91 , then @xmath92 the product rule yields @xmath93 and integrating gives the result . setting @xmath94 and @xmath95 in proposition [ prodeinfachh ] , we see that @xmath96 solves the differential equation ( [ matrixdiffgl ] ) . since the solution to ( [ matrixdiffgl ] ) is unique , we have @xmath97 consequently , @xmath98 is an increasing function . if @xmath99 , then we obtain the reachability gramian of the deterministic setting ( [ detsysant ] ) , see @xcite . this is also the case if @xmath50 is a standard wiener process . the finite reachability gramian @xmath68 provides information about the reachability of a state which we see from the following identity : @xmath100}\mathbb e \left\langle x(t , 0 , m ) , x \right\rangle^2_{\mathbb r^n}=x^t p(t)x\quad \text{for } \quad x\in\mathbb r^n.\end{aligned}\ ] ] consequently , we know that @xmath101 , @xmath102 $ ] , @xmath103 a.s . if and only if @xmath104 meaning that @xmath67 is orthogonal to @xmath105 . since @xmath106 is symmetric positive semidefinite , we have @xmath107 and hence @xmath108\right\}=1,\end{aligned}\ ] ] we observe from ( [ impreachset ] ) that all the states that are not in @xmath109 are not reachable and thus they do not contribute to the system dynamics . as a first step to reduce the system dimension it is necessary to remove all the states that are not in @xmath109 . we will see in the next proposition , that the finite reachability gramians can be replaced by the infinite gramian @xmath110 since their images coincide . this ( infinite ) gramian exists due to the asymptotic stability of @xmath12 . it is easier to work with @xmath7 since it can be computed as the unique solution to @xmath111 @xmath7 satisfies ( [ lyapeqreach ] ) since @xmath68 satisfies ( [ matrixdiffgl ] ) and @xmath112 if @xmath113 due the asymptotic stability of @xmath12 . for the case @xmath114 this gramian was discussed in ( * ? ? ? * section 4.3 ) in the context of balancing for deterministic systems ( [ detsysant ] ) . [ refsameimbla ] the images of the finite reachability gramians @xmath68 , @xmath115 , and the infinite reachability gramian @xmath7 are the same , that is , @xmath116 since @xmath7 and @xmath68 are symmetric positive semidefinite , it is enough to show that their kernels are equal . let @xmath117 . this implies @xmath118 , since @xmath119 is increasing . hence @xmath120 . on the other hand , if @xmath120 , we have @xmath121 consequently , @xmath122 for all @xmath123 $ ] . since the entries of @xmath124 are analytic functions , the scalar function @xmath125 is analytic , such that @xmath126 on @xmath127 . thus , @xmath128 and the result follows . let us now assume that we already removed all the unreachable states from ( [ stochstatenew ] ) . so , ( [ completereach ] ) holds which implies that @xmath129 . we choose an orthonormal basis of @xmath130 , consisting of eigenvectors @xmath131 of @xmath7 , and the following representation holds : @xmath132 we investigate how much the noise influences @xmath67 in the direction of @xmath133 . if a state remains close to zero , it barely contributes to the system dynamics . those states can be identified with the help of the positive eigenvalues @xmath134 of @xmath7 . using ( [ supkleinpt ] ) and the fact that @xmath106 is increasing , we obtain @xmath135}\mathbb e \left\langle x(t , 0 , m ) , p_k\right\rangle^2_{\mathbb r^n}=p_k^t p(t ) p_k\leq p_k^t p p_k=\lambda_k.\end{aligned}\ ] ] hence , if @xmath136 is small , then the the corresponding coefficient @xmath137 in ( [ eq : expansion ] ) is small ( in the @xmath138 sense ) . this means that the noise hardly steers the state in the direction of @xmath133 . consequently , the states that are difficult to reach are contained in the space spanned by the eigenvectors corresponding to the small eigenvalues of @xmath7 . we continue by reasoning why using the modified reachability gramian @xmath7 is better than using the reachability gramian @xmath139 ( @xmath114 ) of the deterministic system ( [ detsysant ] ) . [ propdworse ] the following properties hold for the ( modified ) reachability gramians @xmath7 and @xmath140 : * in general , we have @xmath141 * if @xmath142 ( positive definite ) , then @xmath143 . * if @xmath144 , then @xmath145 . let @xmath146 , then @xmath147 which is equivalent to @xmath148 on @xmath149 and implies @xmath150 on @xmath149 . equivalently , we have @xmath151 and since @xmath152 if and only if @xmath153 , we have @xmath154 . consequently , we obtain @xmath155 due to @xmath156 and @xmath157 . if @xmath142 , then @xmath158 on @xmath149 implies @xmath148 on @xmath149 . in this case , all the above statements are equivalent . therefore @xmath159 and hence @xmath160 . to prove ( c ) , assume @xmath161 pre- and postmultiplying ( [ lyapeqreach ] ) with @xmath162 and @xmath163 , respectively , yields @xmath164 this implies @xmath165 but if @xmath166 , then there is a @xmath117 such that @xmath167 . we set @xmath168 , @xmath57 , and observe that @xmath169 is an analytic function that is not constantly zero since @xmath170 . consequently , @xmath169 has only countably many zeros such that @xmath171 is a purely positive function up to lebesgue zero sets . hence , @xmath172 such that @xmath173 . having @xmath174 implies ( c ) . by ( [ impreachset ] ) and proposition [ propdworse ] ( a ) , we obtain @xmath175\right\}=1.\end{aligned}\ ] ] one could now think of using @xmath140 instead of @xmath7 but from ( [ pdnaja ] ) not all unreachable states can be identified especially if case ( c ) in proposition [ propdworse ] holds . hence , if we were to use @xmath140 , we would underestimate the set of unreachable states . even if we assume that the system is already completely reachable ( e.g. ( [ completereach ] ) holds ) , inequality ( [ interlambad ] ) can not be obtained with @xmath140 . this means that we can not identify the difficult to reach states with the help of the eigenvalues of @xmath140 . in section [ btandspa ] we will see that @xmath7 , rather than @xmath140 , enters the error bound for the rom . finally we note that the reachability gramian @xmath140 of system ( [ detsysant ] ) does not depend on the input @xmath176 . if a `` noisy control '' is used , this does not apply , since @xmath7 depends on @xmath177 and hence on the lvy process @xmath50 . [ [ observability ] ] observability + + + + + + + + + + + + + we conclude this section by introducing a deterministic observability concept for the output equation @xmath178 corresponding to ( [ stochstatenew ] ) with @xmath179 . we recall known facts from ( * ? ? ? * subsection 4.2.2 ) to characterise the importance of certain initial states in the system dynamics since we are in a situation without noise . we assume to have an unknown initial state @xmath58 in the following observation problem and aim to reconstruct @xmath180 from the observation @xmath181 on the entire time interval @xmath182 . an initial state @xmath180 is not observable if @xmath183 on @xmath182 , i.e. it can not be reconstructed by the observation . otherwise , @xmath180 is called observable . a system a called completely observable if every initial state is observable . in order to determine the observability of a state , we consider the energy that is caused by the observations of @xmath180 : @xmath184 where we used that @xmath185 and set @xmath186 . the observability gramian @xmath8 exists due to the asymptotic stability of @xmath12 and is the unique solution to @xmath187 the above relation is obtained by replacing @xmath12 and @xmath188 in ( [ infinitreachgram ] ) and ( [ lyapeqreach ] ) by @xmath189 and @xmath190 , respectively . from ( [ observenergy ] ) we see that @xmath180 is unobservable if and only if @xmath191 . hence , the system is completely observable if and only if @xmath192 . besides the unobservable states we aim to remove the difficult to observe states from the system in order to obtain an accurate rom . the difficult to observe states are those producing only little observation energy , i.e. the corresponding observations @xmath181 are close to zero in the @xmath193 sense . using ( [ observenergy ] ) again , the difficult to observe states are contained in the eigenspaces spanned by the eigenvectors of @xmath8 corresponding to the small eigenvalues . before considering balanced truncation ( bt ) and singular perturbation approximation ( spa ) we summarise the general theory for balancing and how to find a balancing transformation . states that are difficult to reach have large components in the span of the eigenvectors corresponding to small eigenvalues of the reachability gramian @xmath7 , cf . ( [ interlambad ] ) . similarly , states that are difficult to observe are the ones that have large components in the span of eigenvectors corresponding to small eigenvalues of the observability gramian @xmath8 , see ( [ observenergy ] ) . hence in order to produce accurate roms one eliminates states that are both difficult to reach and difficult to observe . to this end we need to find a basis in which the dominant reachable and observable states are the same , which is done by a simultaneous transformation of the gramians . let @xmath194 be a nonsingular matrix . transforming the states using @xmath195 the system ( [ linsysafterdisintr2 ] ) becomes @xmath196 where @xmath197 , @xmath198 , @xmath199 . the input - output map remains the same , only the state , input and output matrices are transformed . @xmath7 and @xmath8 , the reachability and observability gramians of the original systems which satisfy ( [ lyapeqreach ] ) and ( [ observlyap ] ) can be transformed into reachability and observability gramians of the transformed system @xmath200 and @xmath201 ( by multiplying ( [ lyapeqreach ] ) with @xmath202 from the left and @xmath203 from the right and ( [ observlyap ] ) with @xmath204 from the left and @xmath205 from the right ) . the hankel singular values ( hsvs ) of @xmath206 , where @xmath207 for the original and transformed system are the same . the above transformation is a balancing transformation if the transformed gramians are equal to each other and diagonal . such a transformation always exists if @xmath208 and can be obtained by choosing @xmath209 where @xmath210 are the hsvs . @xmath211 , @xmath212 , @xmath213 and @xmath214 are computed as follows . let @xmath215 , @xmath216 be square root factorisations of @xmath7 and @xmath8 , then an svd of @xmath217 gives the required matrices . with this transformation @xmath218 . below , let @xmath202 be the balancing transformation as stated above , then we partition the coefficients of the balanced realisation as follows:@xmath219},\;\;\ ; t{b } = { \left[\begin{array}{c}}{b}_1\\ { b}_2{\end{array}\right]},\;\;\ ; { ct^{-1 } } = { \left[\begin{array}{cc}}{c}_1 & { c}_2{\end{array}\right ] } , \end{aligned}\ ] ] where @xmath220 etc . furthermore , setting @xmath221}$ ] , where @xmath222 , we obtain the transformed partitioned system @xmath223}&={\left[\begin{array}{cc}}{a}_{11}&{a}_{12}\\ { a}_{21}&{a}_{22}{\end{array}\right]}{\left[\begin{array}{cc}}{x}_1(t)\\ { x}_2(t){\end{array}\right]}dt+ { \left[\begin{array}{c}}{b}_1\\ { b}_2{\end{array}\right]}dm(t),\label{balrelpart}\\ y(t)&= { \left[\begin{array}{cc}}{c}_1 & { c}_2{\end{array}\right]}{\left[\begin{array}{cc}}{x}_1(t)\\ { x}_2(t){\end{array}\right]}.\label{balrelpartout}\end{aligned}\ ] ] in this system , the difficult to reach and observe states are represented by @xmath224 , which correspond to the smallest hsvs @xmath225 , but of course @xmath226 has to be chosen such that the neglected hsvs are small ( @xmath227 ) . we discuss two methods ( bt and spa ) to neglect @xmath224 leading to a reduced system of the form @xmath228 where @xmath229 , @xmath230 and @xmath231 ( @xmath232 ) . [ [ balanced - truncation ] ] balanced truncation + + + + + + + + + + + + + + + + + + + for bt the second row in ( [ balrelpart ] ) is truncated and the remaining @xmath224 components in the first row and in ( [ balrelpartout ] ) are set to zero . this leads to reduced coefficients @xmath233 which is similar to the deterministic case . the next lemma states that bt preserves asymptotic stability , which is known from the deterministic case , see ( * ? ? ? * theorem 7.9 ) . [ btpresstab ] let the gramians @xmath7 and @xmath8 be positive definite and @xmath234 , then @xmath235 for @xmath236 , i.e. @xmath237 and @xmath238 are asymptotically stable . the above lemma is vital for the error bound analysis in section [ btandspa ] . [ [ singular - perturbation - approximation ] ] singular perturbation approximation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + instead of setting @xmath239 , one assumes @xmath240 . this idea originates from the deterministic case , where it can be observed that @xmath224 are the fast variables meaning that they are in a steady state after a short time . in our framework , the classical derivative of @xmath224 does not exist but we proceed with setting @xmath241 in ( [ balrelpart ] ) . this yields an algebraic constraint @xmath242}{\left[\begin{array}{cc}}{x}_1(s)\\ { x}_2(s){\end{array}\right]}ds+{b}_2 m(t)=:r(t),\end{aligned}\ ] ] where we assumed zero initial conditions . applying ito s product formula ( [ profriot ] ) to every summand of @xmath243 ( @xmath244 is the @xmath76th component of @xmath245 ) yields @xmath246_t . \end{aligned}\ ] ] inserting the differential of @xmath245 and exploiting that the expectation of the ito integrals is zero , gives @xmath247=\mathbb e \left[\int_0^t a^t(s ) r(s)ds\right]+\mathbb e\left[\int_0^t r^t(s ) a(s ) ds\right ] + \mathbb e \sum_{i=1}^{n - r}[r_i , r_i]_t , \end{aligned}\ ] ] where we set @xmath248 . setting @xmath249 gives @xmath250_t$ ] . since @xmath244 and @xmath251 have the same martingale parts and the same jumps , their compensator processes coincide by ( [ decomqucov ] ) and hence @xmath252_t=\mathbb e \sum_{i=1}^{n - r}[e_i^t b_2 m , e_i^t b_2 m]_t . \end{aligned}\ ] ] applying ito s product formula to @xmath253 and taking the expectation , we have @xmath254_t=\mathbb e ( b_2 m)^t b_2 m \end{aligned}\ ] ] which implies @xmath255 @xmath103-a.s . using this simplification in ( [ algbraicconst ] ) yields @xmath256 which is well - defined by lemma [ btpresstab ] and which we use in the first row of ( [ balrelpart ] ) and in ( [ balrelpartout ] ) . this leads to reduced order coefficients @xmath257 this reduced model is different to the deterministic case , that requires @xmath258 with an additional term in the output equation which does not depend on the state , see ( * ? ? ? * section 2 ) . in the deterministic case , the rom is balanced @xcite , which is not true here due to the modification . like in the deterministic case , the observability gramian is given by @xmath259 . this property is obtained by multiplying ( [ partobeq ] ) with @xmath260 ( @xmath261 is the matrix of the balanced system ) from the left and with @xmath262 from the right and then evaluating the left upper block of the resulting equation , see also ( [ bzwzw ] ) . the following example shows that the reduced order reachability is @xmath263 is not equal to @xmath264 in general which is different from the deterministic case . let @xmath50 be a standard wiener process , then @xmath114 and set @xmath265 @xmath12 is asymptotically stable and the system is balanced since @xmath266 . we fix the reduced order dimension to @xmath267 and compute the reduced order coefficients by spa in ( [ coefficientspa ] ) . we know that @xmath268 but the reachability gramian is up to the digits shown @xmath269 which we computed numerically . this implies that the hsvs are not a subset of the original ones anymore . here , they are @xmath270 and @xmath271 . we conclude this section by a stability result from @xcite . [ spapresstab ] let the gramians @xmath7 and @xmath8 be positive definite and @xmath234 , then @xmath272 for @xmath273 with @xmath274 . the inverses in lemma [ spapresstab ] exist because of the asymptotic stability of the matrices @xmath237 and @xmath238 , see lemma [ btpresstab ] . before we specify the error bounds for bt and spa , we provide a general error bound comparing the outputs of ( [ stochsysnew ] ) and ( [ generalreducedsys ] ) with asymptotically stable matrices @xmath12 , @xmath275 and initial conditions @xmath54 , @xmath276 . these outputs are then given by ornstein - uhlenbeck processes @xmath277 as mentioned in @xcite , see also @xcite . using these representations and cauchy s inequality as well as ito s isometry ( see @xcite ) , we obtain @xmath278 where @xmath279 is the covariance matrix of @xmath50 . substitution and taking limits yields @xmath280 using the definition of the frobenius norm and the linearity of the trace , we have @xmath281}\mathbb e \left\|y(t)- y_r(t)\right\|_{\mathbb r^p}\leq \left({\operatorname{tr}}\left(c p c^t\right)+{\operatorname{tr}}\left(c_r p_r c_r^t\right)-2{\operatorname{tr}}\left(c p_g c_r^t\right)\right)^{\frac{1}{2}},\end{aligned}\ ] ] where @xmath282 is the reachability gramian of the original system satisfying ( [ lyapeqreach ] ) , @xmath283 the one of the reduced system satisfying @xmath284 and @xmath285 is the solution to @xmath286 equation ( [ gemischtegram ] ) is a consequence of proposition [ prodeinfachh ] with @xmath287 and @xmath288 and the fact that @xmath289 if @xmath290 due to the asymptotic stability of @xmath12 and @xmath275 . the matrices @xmath291 and @xmath292 in the bound in ( [ computeerrorbound ] ) are all well - defined because @xmath12 and @xmath275 are asymptotically stable . the error bound in ( [ computeerrorbound ] ) holds for both bt and spa since both approaches preserve asymptotic stability , see lemmas [ btpresstab ] and [ spapresstab ] . for both bt and spa the representation in ( [ computeerrorbound ] ) can be used for practical computations of the error bound . the gramian @xmath7 is already available since it is required in the balancing procedure . the reduced model gramian @xmath263 is computationally cheap because it is low dimensional assuming that we fix a small rom dimension . the same is true for @xmath292 since it has only a few columns which makes the solution to ( [ gemischtegram ] ) easily accessible . since the error bound ( [ computeerrorbound ] ) is computationally cheap , it can be computed for several rom dimensions and hence be used to find a suitable @xmath226 . in the next two theorems we specify the general error bound in ( [ computeerrorbound ] ) for both bt and spa and represent it in terms of the truncated hsvs @xmath225 of the system . using the balanced realisation ( [ balancedrels ] ) of the original system with @xmath293 and its corresponding partition , we have @xmath294}\hspace{-0.1cm}{\left[\begin{array}{cc}}\sigma_1\hspace{-0.2cm}{}&\\ & \sigma_2{\end{array}\right]}\hspace{-0.05cm}+\hspace{-0.05 cm } { \left[\begin{array}{cc}}\sigma_1\hspace{-0.2cm}{}&\\ & \sigma_2{\end{array}\right]}\hspace{-0.1 cm } { \left[\begin{array}{cc}}a^t_{11}&a^t_{21}\\ a^t_{12 } & a^t_{22}{\end{array}\right]}\hspace{-0.05cm}= & \hspace{-0.05 cm } -{\left[\begin{array}{cc}}\tilde { b}_1\tilde b_1^t&\tilde { b}_1\tilde b_2^t \\ \tilde { b}_2\tilde b_1^t & \tilde { b}_2\tilde b_2^t{\end{array}\right]}\\ \label{partobeq } { \left[\begin{array}{cc}}{a}^t_{11}&{a}^t_{21}\\ { a}^t_{12}&{a}^t_{22}{\end{array}\right]}\hspace{-0.1cm}{\left[\begin{array}{cc}}\sigma_1\hspace{-0.2cm}{}&\\ & \sigma_2{\end{array}\right]}\hspace{-0.05cm}+\hspace{-0.05 cm } { \left[\begin{array}{cc}}\sigma_1\hspace{-0.2cm}{}&\\ & \sigma_2{\end{array}\right]}\hspace{-0.1 cm } { \left[\begin{array}{cc}}a_{11}&a_{12}\\ a_{21 } & a_{22}{\end{array}\right]}\hspace{-0.05cm}= & \hspace{-0.05 cm } -{\left[\begin{array}{cc}}c_1^tc_1&c_1^tc_2 \\ c^t_2c_1 & c^t_2c_2{\end{array}\right]}\end{aligned}\ ] ] where @xmath295}={\left[\begin{array}{c}}{b}_1\mathcal{q}_m^{\frac{1}{2}}\\ b_2 \mathcal{q}_m^{\frac{1}{2 } } { \end{array}\right]}$ ] , @xmath296 and @xmath297 . [ bterrorbound ] let @xmath298 be the output of the reduced order system obtained by bt , then under the assumptions of lemma [ btpresstab ] , we have@xmath299}\mathbb e \left\|y(t)- y_{bt}(t)\right\|_{\mathbb r^p}\leq \left({\operatorname{tr}}(\sigma_2 ( b_2 \mathcal{q}_m b_2^t+2 p_{g , 2 } a_{21}^t))\right)^{\frac{1}{2}},\end{aligned}\ ] ] where @xmath300 are the last @xmath301 rows of @xmath302 with @xmath202 being the balancing transformation . evaluating the left and right upper block of ( [ partobeq ] ) yields@xmath303 from ( [ computeerrorbound ] ) the error bound has the form @xmath304 since @xmath305 . using the balancing transformation @xmath202 and the partition of @xmath306 in ( [ balancedrels ] ) , we obtain @xmath307 . now , the left upper block of ( [ partreacheq ] ) is @xmath308 such that @xmath309 . using the partitions of @xmath310 and @xmath311}$ ] , we obtain @xmath312 . inserting these results into ( [ eq : epserror ] ) gives @xmath313 using @xmath314 and substituting ( [ c1c2eqbt2 ] ) yields @xmath315 multiplying ( [ gemischtegram ] ) with the balancing transformation from the left and using the partitions of @xmath316 and @xmath317 from ( [ balancedrels ] ) yields @xmath318}{\left[\begin{array}{c } } p_{g , 1 } \\ p_{g , 2 } { \end{array}\right]}+{\left[\begin{array}{c } } p_{g , 1 } \\ p_{g , 2 } { \end{array}\right]}a_{11}^t = -{\left[\begin{array}{c}}{b}_1\\ { b}_2{\end{array}\right]}\mathcal q_m b_1^t . \end{aligned}\ ] ] with the first row of this equation , @xmath319 we have@xmath320 and substituting ( [ firstc1c2bt2jk ] ) , we obtain @xmath321 so that @xmath322 inserting this result into ( [ insertforebbt2fh ] ) gives@xmath323 with ( [ firstc1c2bt2jk ] ) and ( [ blaaa ] ) , and the properties of the trace function we obtain @xmath324 similarly @xmath325 can be shown using the right lower blocks of ( [ partreacheq ] ) and ( [ partobeq ] ) . hence , @xmath326 which gives the result . [ spaerrorbound ] let @xmath327 be the output of the reduced order system obtained by spa , then under the assumptions of lemma [ spapresstab ] , we have@xmath299}\mathbb e \left\|y(t)- y_{spa}(t)\right\|_{\mathbb r^p}\leq \left({\operatorname{tr}}(\sigma_2 ( b_2 \mathcal{q}_m b_2^t-2(a_{22}p_{g , 2}+a_{21 } p_{g , 1})(a_{22}^{-1 } a_{21})^t))\right)^{\frac{1}{2}},\end{aligned}\ ] ] where @xmath328 are the first @xmath226 and @xmath300 the last @xmath301 rows of @xmath329 with @xmath202 being the balancing transformation . let @xmath330}$ ] , then , since @xmath331 are invertible by lemma [ btpresstab ] , its inverse is given in block form @xmath332},\end{aligned}\ ] ] where @xmath333 . if we multiply ( [ partobeq ] ) with @xmath260 from the left hand side and select the left and right upper block of this equation , we obtain @xmath334 where @xmath335 and thus @xmath336 furthermore , multiplying ( [ partobeq ] ) with @xmath260 from the left and with @xmath262 from the right , the resulting left upper block of the equation is @xmath337 and thus@xmath338 we define @xmath339 which is the error bound for spa . from the proof of theorem [ bterrorbound ] we know that the following holds@xmath340 by ( [ bzwzw ] ) and the definition of the reachability equation of the rom , we have @xmath341 this leads to @xmath342 we multiply ( [ gemischtegram ] ) with the balancing transformation @xmath202 from the left ( here @xmath343 ) and use the partitions of @xmath316 , @xmath317 from ( [ balancedrels ] ) and the partition of @xmath311}$ ] . thus , @xmath344}{\left[\begin{array}{c } } p_{g , 1 } \\ p_{g , 2 } { \end{array}\right]}+{\left[\begin{array}{c } } p_{g , 1 } \\ p_{g , 2 } { \end{array}\right]}\bar a^t = -{\left[\begin{array}{c}}{b}_1\\ { b}_2{\end{array}\right]}\mathcal q_m b_1^t . \end{aligned}\ ] ] we obtain @xmath345 using the partition of @xmath310 in ( [ balancedrels ] ) . with ( [ barcczwei ] ) we obtain @xmath346 inserting the upper block of ( [ partmixedeqspa ] ) leads to @xmath347 using ( [ barcceins ] ) and the properties of the trace function we have @xmath348 consequently , @xmath349 holds and hence @xmath350 which provides the required result . the error bound representations in theorems [ bterrorbound ] and [ spaerrorbound ] depend on the @xmath301 smallest hsvs @xmath225 . if the corresponding truncated components are unimportant , i.e. they are difficult to reach and observe , then the values @xmath225 are small and consequently the error bound is small . hence , the rom is of good quality . the error bounds in theorems [ bterrorbound ] and [ spaerrorbound ] can be used to find a suitable reduced order dimension @xmath226 . small hsvs @xmath225 for fixed @xmath226 would guarantee a small error . note that , if @xmath114 , as for example in the standard wiener case , then the error bound in theorem [ bterrorbound ] coincides with the @xmath351-error bound in the deterministic case when using a normalised control , see ( * ? ? ? * lemma 7.13 ) . the next section provides a particular sde to which we will apply the theory developed in this section . in this section , we deal with a setting that covers the spde with its output in ( [ inrobspeq])-([introbspoutnew ] ) , a damped wave equation with additive noise which can formally be interpreted as @xmath352 with @xmath353 ( @xmath236 ) , @xmath37 and the @xmath354th component of the control @xmath355 , @xmath356 . here , @xmath357 are the components of an @xmath358-valued lvy processes @xmath50 that is square integrable and has mean zero . this is in contrast to the setting in @xcite where ( [ nichtlinsde ] ) with multiplicative lvy noise was considered , e.g. @xmath359 linear bounded operators and @xmath176 an @xmath360-dimensional stochastic control , @xmath361 and @xmath362 uncorrelated scalar lvy processes . for the stability analysis of the uncontrolled equation ( [ nichtlinsde ] ) with wiener noise ( @xmath363 ) we refer to @xcite . since lvy noise is no feasible control in the framework in @xcite , this setting requires further analysis . we transform damped wave equation with additive noise into a first order spde and define the corresponding solution in section [ setting ] , following the approach in @xcite . in section [ subsecspeltralgal ] , we explain how the resulting first order spde can be approximated by a spectral galerkin scheme . we refer to @xcite , where similar techniques were applied . let @xmath364 be square integrable lvy processes with zero mean that takes values in @xmath30 . moreover , @xmath50 is defined on a complete probability space @xmath365 , is right - continuous and that @xmath366 contains all @xmath103 null sets . ] it is adapted to the filtration @xmath367 and its increments @xmath368 are independent of @xmath369 for @xmath370 . let @xmath371 be a self adjoint and positive definite operator on a separable hilbert space @xmath372 and let @xmath373 be an orthonormal basis of eigenvectors of @xmath374 for @xmath372 , @xmath375 where @xmath376 are the corresponding eigenvalues . we denote the well - defined square root of @xmath374 by @xmath377 . @xmath378 equipped with the inner product @xmath379 represents a separable hilbert space as well . the ( symbolic ) second order spde we consider is given by @xmath380 with initial conditions @xmath381 , @xmath382 , @xmath37 and output @xmath383 we assume @xmath384 and @xmath385 . using the separable hilbert space @xmath386 with the inner product @xmath387 we transform this second order system into a first order system following the approach in @xcite . the system ( [ nichtlinsde2])-([eq : output ] ) can be expressed as : @xmath388 where @xmath389\;\;\;\text{and}\;\;\;\mathcal b_k=\left[\begin{matrix } 0 \\ -\tilde { \mathcal b}_k \end{matrix}\right]\in h.\end{aligned}\ ] ] so far , we only worked with symbolic equation , since the classical derivative of a lvy process does not exist in general . the next lemma from @xcite provides a stability result and is vital to define a cdlg mild solution of ( [ firstordertransf ] ) . [ lemwaveconexsemigr ] for every @xmath37 the linear operator @xmath390 with domain @xmath391 generates an exponential stable contraction semigroup @xmath392 with @xmath393 we use this result to define the solution to ( [ firstordertransf ] ) . an @xmath367-adapted cdlg process @xmath394 is called mild solution to ( [ firstordertransf ] ) if for all @xmath57 @xmath395 we refer to @xcite for the definition of the ito integral in ( [ mildsolalternative ] ) . for the finite dimensional case , the definition of an ito integral with respect to lvy processes can be found in @xcite and ito integrals with respect to martingales are defined in @xcite . if we set @xmath396)$ ] , @xmath397)$ ] , @xmath398 , @xmath399)$ ] and the output operator @xmath400 with @xmath401 @xmath402)$ ] in system ( [ nichtlinsde2])-([eq : output ] ) then we obtain the system in ( [ inrobspeq])-([introbspoutnew ] ) , which is therefore well - defined in the mild sense ( [ mildsolalternative ] ) . we study a spectral galerkin scheme to approximate the mild solution of ( [ firstordertransf ] ) with output ( [ firstordertransfout ] ) , similar to the approach in @xcite ( mainly for spdes with wiener noise ) . this approximation is based on a particular choice of an orthonormal basis @xmath403 of @xmath404 , given by @xmath405 where @xmath373 and @xmath406 are defined in ( [ eigvaleq ] ) , see @xcite . in ( [ inrobspeq])-([introbspoutnew ] ) , we have @xmath407 on @xmath408 $ ] . in this case @xmath409 and @xmath410 for @xmath411 . to approximate the @xmath404-valued process @xmath412 in ( [ firstordertransf ] ) , we construct a sequence @xmath413 of finite dimensional adapted cdlg processes with values in @xmath414 , defined by @xmath415 where we set * @xmath416 for all @xmath417 , * @xmath418 for @xmath419 , * @xmath420 . for the mild solution to ( [ approxsde ] ) , let @xmath421 be a @xmath422-semigroup on @xmath423 given by @xmath424 for all @xmath425 . it is generated by @xmath426 such that the mild solution of equation ( [ approxsde ] ) is @xmath427 since @xmath426 is bounded , the @xmath422-semigroup on @xmath423 is represented by @xmath428 , @xmath57 . we formulate the main result of this section , which uses ideas from @xcite and is proved in appendix [ sec : app2 ] . [ th : mildsol ] the mild solution @xmath429 of equation ( [ approxsde ] ) approximates the mild solution @xmath412 of equation ( [ firstordertransf ] ) , i.e. @xmath430 for @xmath431 and @xmath57 . this implies the convergence of the corresponding outputs @xmath432 . in the following , we make use of the property that the mild and the strong solution of ( [ approxsde ] ) coincide , since we are in finite dimensions . we write the output @xmath433 of the galerkin system as an expression depending on the fourier coefficients of the galerkin solution @xmath429 . the coefficients of @xmath433 are @xmath434 for @xmath435 , where @xmath436 is the @xmath437-th unit vector in @xmath438 . we set @xmath439 and obtain @xmath440 . the components @xmath441 of @xmath72 satisfy @xmath442 using the fourier series representation of @xmath429 , we obtain @xmath443 hence , the vector of fourier coefficients @xmath63 is given by @xmath444 where @xmath445 with @xmath446 ( @xmath447 ) , and @xmath448 the eigenvalues of @xmath449 , and @xmath450 for @xmath451 . we will often make use of the compact form of the sde in ( [ galerkwavenowspec ] ) which is @xmath452 where @xmath29 and @xmath453 $ ] . applying the spectral galerkin method to the system ( [ inrobspeq])-([introbspoutnew ] ) the matrices of the semi - discretised system ( [ galerkwavenowspec ] ) are given by @xmath454 with @xmath455 , @xmath456 with @xmath457)}},\ ] ] and @xmath458 $ ] with @xmath459 & 0 \end{matrix}\right)^t,\\ \mathcal c h_{2 \ell}&=\left(\begin{matrix } 0 & \frac{1}{\sqrt{2\pi}\ell\epsilon}\left [ \cos \left(\ell \left(\frac{\pi}{2 } -\epsilon \right)\right)-\cos\left(\ell\left(\frac { \pi } { 2 } + \epsilon \right)\right)\right]\end{matrix}\right)^t,\end{aligned}\ ] ] where we assume @xmath10 to be even and @xmath460 . in figure [ fig : wavetime ] we plot the numerical solution to the stochastic damped wave equation for @xmath36 $ ] and in the time interval @xmath461 $ ] where we set @xmath462 , @xmath463 and @xmath464 ( e.g. @xmath465 stochastic inputs ) . the weighting functions for the two inputs are @xmath466 and @xmath467 . the noise processes are @xmath468 and @xmath469 is a compound poisson process , where @xmath470 is a poisson process with parameter equal to @xmath71 , @xmath471 are independent uniformly distributed jumps and @xmath472 is a standard wiener process . @xmath473 and @xmath474 are independent . the plot in figure [ fig : wavetime ] shows a particular realisation of the solution to ( [ inrobspeq ] ) at @xmath475 specific times @xmath476 . we see that the string moves up and down as expected due to the nonzero ( stochastic ) input . we observe that the third snapshot is taken after a jump occured in stochastic process . the corresponding output , namely both the position and the velocity in the middle of the string , is shown in figure [ fig : position_velocity ] . in the plot for the velocity the noise generated by the lvy process can be seen . the trajectory of the velocity is impacted by lvy noise with jumps , where the velocity ( e.g. the impact by wind ) is randomly increased or reduced . ) -([introbspoutnew ] ) in the phase plane . ] the trajectory for the position of the cable in figure [ fig : position_velocity ] is smoother as it is the integral of the velocity . finally , figure [ fig : paths ] shows the velocity versus the position of the string , for the same sample path , in a phase portrait . the four jumps are clearly visible . we consider the spectral galerkin discretisation of the second order damped wave equation which we discussed in detail in section [ sec : wave ] , and in particular , the example in ( [ inrobspeq])-([introbspoutnew ] ) with two stochastic inputs and two outputs , namely position and velocity of the middle of the string . we set @xmath477 and choose the weighting functions @xmath478 and the noise processes @xmath28 ( @xmath236 ) as in figure [ fig : wavetime ] . we fix the state dimension to @xmath479 and reduce the galerkin solution by bt and spa . for computing the trajectories of the sde we use the euler - maruyama method ( see , e.g. @xcite ) . and velocity @xmath480 with @xmath481 . ] figures [ fig : btr6 ] and [ fig : btr24 ] show the logarithmic errors for the position @xmath482 and the velocity @xmath480 of the middle of the string , if mor by bt is applied to the wave equation with stochastic inputs when reduced models of dimension @xmath475 and @xmath483 , respectively , are computed . and velocity @xmath480 with @xmath484 . ] the first two plots in each of the figures show the logarithmic mean error for both the position and the velocity . one observation is that the position is generally more accurate than the velocity ( about one order of magnitude here ) , since the trajectories are smoother . moreover , comparing the expected values of the errors of the reduced model of dimension @xmath481 ( first two plots in figure [ fig : btr6 ] ) with the one of dimension @xmath484 ( first two plots in figure [ fig : btr24 ] ) it can be seen that the latter ones are more accurate ( an improvement of about one order of magnitude ) as one would expect . the last two plots in figures [ fig : btr6 ] and [ fig : btr24 ] show the logarithmic errors for position and velocity for one particular trajectory , which is the same as the one for the sample we considered in section [ sec : wave ] . and velocity @xmath480 with @xmath481 . ] figures [ fig : spar6 ] and [ fig : spar24 ] show the logarithmic errors for the position @xmath482 and the velocity @xmath480 of the middle of the string , if mor by spa is applied to the wave equation with stochastic inputs when reduced models of dimension @xmath475 and @xmath483 , respectively , are computed . again , the first two plots show the mean errors while the last two plots show the errors in particular trajectories . and velocity @xmath480 with @xmath484 . ] we observe that the error in the position is smaller than the error in the velocity , and , the error is smaller if a larger dimension of the reduced order model is used . finally , we compare the error bounds for bt ( see theorem [ bterrorbound ] ) and spa ( see theorem [ spaerrorbound ] ) with the worst case mean errors , that is @xmath485 } \mathbb e\left\|y(t)-y_r(t)\right\|_{\mathbb{r}^p } \end{aligned}\ ] ] for both methods in table [ tab : errorbd ] , where @xmath486 is the full output of the original model and @xmath487 the rom output . .error and error bounds for both bt and spa and several dimensions of the reduced order model ( rom ) . [ cols="^,^,^,^,^,^ " , ] first , as expected both mean errors and error bounds are getting smaller the larger the size of the rom . moreover , both error bounds are rather tight and close to the actual error of the rom , e.g. the bounds , which are worst case bounds also provide a good prediction of the true time domain error . we also note that bt performs better than spa , both in actually computed mean errors as well as in terms of the error bounds . we have presented theory for balancing related model order reduction ( mor ) applied to linear stochastic differential equations ( sdes ) with additive lvy noise . in particular we extended the concepts of reachability and observability to stochastic systems and formulated a new reachability gramian . we then showed how balancing related mor which is well known for deterministic systems can be extended to sdes with additive lvy noise , e.g. leads to the solution of a lyapunov equation ( with a slightly different right hand side ) . we proved a general error bound for reduced ( asymptotically stable ) systems in this setting and then gave specific bounds for balanced truncation ( bt ) and singular perturbation approximation ( spa ) which depended on the neglected ( small ) hankel singular values of the linear system . we finally applied our theory to a second order damped wave equation , discretised using a spectral galerkin method , and controlled by lvy noise . the numerical results showed that mor can be applied successfully and that errors for both bt and spa are small , and the error bounds tight . let all stochastic processes appearing in this section be defined on a filtered probability space @xmath488 shall be right continuous and complete . ] . we denote the set of all cdlg square integrable @xmath489-valued martingales with respect to @xmath367 by @xmath490 . let @xmath491 be scalar semimartingales . we set @xmath492 with @xmath493 for @xmath236 . then the ito product formula @xmath494_t\end{aligned}\ ] ] for @xmath57 holds , see @xcite or @xcite for the special case of lvy - type integrals . by ( * theorem 4.52 ) , the compensator process @xmath495 $ ] is given by @xmath496_t=\left\langle m_1^c , m_2^c\right\rangle_t+\sum_{0\leq s\leq t } \delta z_1(s ) \delta z_2(s)\end{aligned}\ ] ] for @xmath57 , where @xmath497 , @xmath498 are the continuous martingale parts of @xmath499 and @xmath500 ( cf . * theorem 4.18 ) ) . the process @xmath501 is a uniquely defined angle bracket process that ensures that @xmath502 is an @xmath367- martingale , see ( * ? ? ? * proposition 17.2 ) . as a simple consequence of ( [ profriot ] ) , we have : [ iotprodformelmatpro ] let @xmath211 be an @xmath503-valued and @xmath212 be an @xmath130-valued semimartingale , then we have@xmath504_t\right)_{{i=1 , \ldots , d}\atop { j=1 , \ldots , n}}\end{aligned}\ ] ] for all @xmath57 . considering the stochastic differential of the @xmath505-th component of the matrix - valued process @xmath506 , @xmath57 , and using ( [ profriot ] ) gives the result , see also @xcite . using @xmath507 for @xmath508 , we obtain @xmath509 ito s isometry ( see e.g. @xcite ) yields that the right hand side can be bounded by @xmath510 $ ] . since @xmath511 is a contraction semigroup , we have@xmath512 by the representation @xmath513 ( @xmath425 ) and lebesgue s theorem , the bound in ( [ inconcon2 ] ) tends to zero for @xmath431 . for @xmath514 we get @xmath515 which tends to zero for @xmath431 and hence @xmath516\rightarrow 0\end{aligned}\ ] ] for @xmath431 by lebesgue s theorem .

in order to understand certain observed features of arc - like giant radio relics such as the rareness , uniform surface brightness , and curved integrated spectra , we explore a diffusive shock acceleration ( dsa ) model for radio relics in which a spherical shock impinges on a magnetized cloud containing fossil relativistic electrons . toward this end , we perform dsa simulations of spherical shocks with the parameters relevant for the sausage radio relic in cluster ciza j2242.8 + 5301 , and calculate the ensuing radio synchrotron emission from re - accelerated electrons . three types of fossil electron populations are considered : a delta - function like population with the shock injection momentum , a power - law distribution , and a power - law with an exponential cutoff . the surface brightness profile of radio - emitting postshock region and the volume - integrated radio spectrum are calculated and compared with observations . we find that the observed width of the sausage relic can be explained reasonably well by shocks with speed and sonic mach number . these shocks produce curved radio spectra that steepen gradually over with break frequency ghz , if the duration of electron acceleration is myr . however , the abrupt increase of spectral index above ghz observed in the sausage relic seems to indicate that additional physical processes , other than radiative losses , operate for electrons with .

1505.04256

radio relics are diffuse radio sources found in the outskirts of galaxy clusters and they are thought to trace synchrotron - emitting cosmic - ray ( cr ) electrons accelerated via diffusive shock acceleration ( dsa ) at cluster shocks ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? so far several dozens of clusters have been observed to have radio relics with a variety of morphologies and most of them are considered to be associated with cluster merger activities ( see for reviews , e.g. , * ? ? ? * ; * ? ? ? for instance , double radio relics , such as the ones in zwcl0008.8 + 5215 , are thought to reveal the bow shocks induced by a binary major merger @xcite . on the other hand , recently it was shown that shocks induced by the infall of the warm - hot intergalactic medium ( whim ) along adjacent filaments into the hot intracluster medium ( icm ) can efficiently accelerate cr electrons , and so they could be responsible for some radio relics in the cluster outskirts ( see , e.g. , * ? ? ? * ) . the radio relic 1253 + 275 in coma cluster observed in both radio @xcite and x - ray @xcite provides an example of such infall shocks . the so - called sausage relic in ciza j2242.8 + 5301 ( @xmath7 ) contains a thin arc - like structure of @xmath8 kpc width and @xmath9 mpc length , which could be represented by a portion of spherical shell with radius @xmath5 mpc @xcite . unique features of this giant radio relic include the nearly uniform surface brightness along the length of the relic and the strong polarization of up to @xmath10 with magnetic field vectors aligned with the relic @xcite . a temperature jump across the relic that corresponds to a @xmath11 shock has been detected in x - ray observations @xcite . this was smaller than @xmath12 estimated from the above radio observation . several examples of mpc - scale radio relics include the toothbrush relic in 1rxs j0603.3 with a peculiar linear morphology @xcite and the relics in a3667 @xcite and a3376 @xcite . the shock mach numbers of radio relics estimated based on x - ray observation are often lower than those inferred from the radio spectral index using the dsa model , for instance , in the toothbrush relic @xcite and in the radio relic in a2256 @xcite . although such giant radio relics are quite rare , the fraction of x - ray luminous clusters hosting some radio relics is estimated to be @xmath13 % or so @xcite . through a number of studies using cosmological hydrodynamical simulations , it has been demonstrated that during the process of hierarchical structure formation , abundant shocks are produced in the large - scale structure of the universe , especially in clusters ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? considering that the characteristic time - scale of cluster dynamics including mergers is @xmath14 gyr , typical cluster shocks are expected to last for about the same period . yet , the number of observed radio relics , which is thought to trace such shocks , is still limited . so it is plausible to conjecture that cluster shocks may ` turn on ' to emit synchrotron radiation only for a fraction of their lifetime . one feasible scenario is that a cluster shock lights up in radio when it sweeps up a fossil cloud , a magnetized icm gas with fossil relativistic electrons left over from either a radio jet from agn or a previous episode of shock / turbulence acceleration ( see the discussions in section 2.5 and figure 1 ) . pre - exiting seed electrons and/or enhanced magnetic fields are the requisite conditions for possible lighting - up of relic shocks . in particular , the elongated shape with _ uniform _ surface brightness and high polarization fraction of radio emission in the sausage relic , may be explained , if a mpc - scale thermal gas cloud , containing fossil relativistic electrons and permeated with regular magnetic field of a few to several @xmath15 g , is adopted . a more detailed description will be given later in section 2.5 . in this picture , fossil electrons are expected to be re - accelerated for less than cloud - crossing time ( @xmath16 ) , which is much shorter than the cluster dynamical time - scale . in addition , only occasional encounters with fossil clouds combined with the short acceleration duration could alleviate the strong constraints on the dsa theory based on non - detection of @xmath17-ray emission from clusters by fermi - lat @xcite . a similar idea has been brought up by @xcite , who reported the discovery of a mpc - scale , elongated relic in the bullet cluster 1e 0657 - 55.8 . they also proposed that the arc - like shape of uniform surface brightness in some radio relics may trace the underlying regions of pre - existing , seed electrons remaining from old radio lobes . on the other hand , @xcite suggested that radio relics could be explained by revival of fossil radio plasma by compression due to a passage of a shock , rather than dsa . in a follow - up study , @xcite showed using mhd simulations that a cocoon of hot radio plasma swept by a shock turns into a filamentary or toroidal structure . although this scenario remains to be a viable explanation for some radio relics , it may not account for the uniform arc - like morphology of the sausage relic . it is now well established , through observations of radio halos / relics and faraday rotation measures of background radio sources , that the icm is permeated with @xmath15g - level magnetic fields ( e.g. * ? ? ? * ; * ? ? ? the observed radial profile of magnetic field strength tends to peak at the center with a few @xmath15 g and decrease outward to @xmath18 g in the cluster outskirts @xcite . a variety of physical processes that could generate and amplify magnetic fields in the icm have been suggested : primordial processes , plasma processes at the recombination epoch , and biermann battery mechanism , combined with turbulence dynamo , in addition to galactic winds and agn jets ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? given the fact that @xmath19 g fields are required to explain the amplitude and width of the observed radio flux profile of the sausage relic ( e.g. * ? ? ? * ; * ? ? ? * ) , the presence of a cloud with enhanced magnetic fields of several @xmath15 g might be preferred to the background fields of @xmath20 g in the cluster periphery . alternatively , @xcite showed that the postshock magnetic fields can be amplified to @xmath21 g level leading to high degrees of polarization , if there exists dynamically significant turbulence in the upstream region of a curved shock . although it is well accepted that magnetic fields can be amplified via various plasma instabilities at collisionless shocks , the dependence on the shock parameters such as the shock sonic and alfvnic mach numbers , and the obliquity of background magnetic fields remains to be further investigated ( see * ? ? ? for example , the acceleration of protons and ensuing magnetic field amplification via resonant and non - resonant streaming instabilities are found to be ineffective at perpendicular shocks @xcite . in several studies using cosmological hydrodynamical simulations , synthetic radio maps of simulated clusters were constructed by identifying shocks and adopting models for dsa of electrons and magnetic field amplification @xcite . in particular , @xcite demonstrated , by generating mock radio maps of simulated cluster samples , that radio emission tends to increase toward the cluster periphery and peak around @xmath22 ( where @xmath23 is the virial radius ) , mainly because the kinetic energy dissipated at shocks peaks around @xmath24 . as a result , radio relics are rarely found in the cluster central regions . re - acceleration of fossil relativistic electrons by cosmological shocks during the large scale structure formation has been explored by @xcite . the radio emitting shocks in these studies look like segments of spherical shocks , moving from the cluster core region into the periphery . we presume that they are generated mostly as a consequence of major mergers or energetic infalls of the whim along adjacent filaments . so it seems necessary to study spherical shocks propagating through the cluster periphery , rather than interpreting the radio spectra by dsa at _ steady planar _ shocks , in order to better understand the nature of radio relics @xcite . according to the dsa theory , in the case of a _ steady planar _ shock with _ constant _ postshock magnetic field , the electron distribution function at the shock location becomes a power - law of @xmath25 , and so the synchrotron emissivity from those electrons becomes a power - law of @xmath26 . the power - low slopes depend only on the shock sonic mach number , @xmath27 , and are given as @xmath28 and @xmath29 for the gasdynamic shock with the adiabatic index @xmath30 @xcite . here we refer @xmath31 as the injection spectral index for a steady planar shock with constant postshock magnetic field . then , the volume - integrated synchrotron spectrum downstream of the shock also becomes a simple power - law of @xmath32 with the spectral index @xmath33 above the break frequency , @xmath34 , since electrons cool via synchrotron and inverse - compton ( ic ) losses behind the shock ( e.g. , * ? ? ? * ; * ? ? ? ' as the spectral index of the flux density , @xmath35 for unresolved sources , so in that case @xmath36 is the same as @xmath37 . here @xmath38 is defined as the spectral index of the _ local _ emissivity , @xmath39 . see equations ( [ alpha])-([anu ] ) . ] such predictions of the dsa theory have been applied to explain the observed properties of radio relics , e.g. , the relation between the injection spectral index and the volume - integrated spectral index , and the gradual steepening of spatially resolved spectrum downstream of the shock . @xcite performed time - dependent , dsa simulations of cr electrons for _ steady planar _ shocks with @xmath40 and constant postshock magnetic fields . several models with thermal leakage injection or pre - existing electrons were considered in order to reproduce the surface brightness and spectral aging profiles of radio relics in ciza j2242.8 + 5301 and zwcl0008.8 + 5215 . adopting the same geometrical structure of radio - emitting volume as described in section 2.5 , they showed that the synchrotron emission from shock accelerated electrons could explain the observed profiles of the radio flux , @xmath41 , of the sausage relic , and the observed profiles of both @xmath41 and @xmath42 of the relic in zwcl0008.8 + 5215 . here @xmath43 is the distance behind the projected shock edge in the plane of the sky . in the case of spherically expanding shocks with varying speeds and/or nonuniform magnetic field profiles , on the other hand , the electron spectrum and the ensuing radio spectrum could deviate from those simple power - law forms , as shown in @xcite . then even the injection slope should vary with the frequency , @xmath44 . here we follow the evolution of a spherical shock expanding outward in the cluster outskirts with a decreasing density profile , which may lead to a curvature in both the injected spectrum and the volume - integrated spectrum . moreover , if the shock is relatively young or the electron acceleration duration is short ( @xmath45 myr ) , then the break frequency falls in @xmath46 ghz and the volume - integrated spectrum of a radio relic would steepen gradually with the spectral index from @xmath31 to @xmath47 over @xmath2 ( e.g. * ? ? ? * ) . in the case of the sausage relic , @xcite and @xcite originally reported observations of @xmath48 and @xmath49 , which imply a shock of @xmath12 . @xcite , however , found a spectral steepening of the volume - integrated spectrum at 16 ghz , which would be inconsistent with the dsa model for a steady planar shock . moreover , @xcite , by performing a spatially - resolved spectral fitting , revised the injection index to a steeper value , @xmath50 . then , the corresponding shock mach number is reduced to @xmath51 . they also suggested that the spectral age , calculated under the assumption of freely - aging electrons downstream of a steady planar shock , might not be compatible with the shock speed estimated from x - ray and radio observations . also @xcite reported that for the relic in a2256 , the volume - integrated index steepens from @xmath52 for @xmath53 mhz to @xmath54 for @xmath55 ghz , which was interpreted as a broken power - law . discoveries of radio relic shocks with @xmath56 in recent years have brought up the need for more accurate understanding of injections of protons and electrons at weak collisionless shocks , especially at high plasma beta ( @xmath57 ) icm plasmas ( e.g. * ? ? ? here @xmath58 is the ratio of the gas to magnetic field pressure . injection of electrons into the fermi 1st - order process has been one of long - standing problems in the dsa theory for astrophysical shocks , because it involves complex plasma kinetic processes that can be studied only through full particle - in - cell ( pic ) simulations ( e.g. * ? ? ? * ; * ? ? ? it is thought that electrons must be pre - accelerated from their thermal momentum to several times the postshock thermal proton momentum to take part in the dsa process , and electron injection is much less efficient than proton injection due to smaller rigidity of electrons . several recent studies using pic simulations have shown that some of incoming protons and electrons gain energies via shock drift acceleration ( sda ) while drifting along the shock surface , and then the particles are reflected toward the upstream region . those reflected particles can be scattered back to the shock by plasma waves excited in the foreshock region , and then undergo multiple cycles of sda , resulting in power - law suprathermal populations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . such ` self pre - acceleration ' of thermal electrons in the foreshock region could be sufficient enough even at weak shocks in high beta icm plasmas to explain the observed flux level of radio relics . in these pic simulations , however , subsequent acceleration of suprathermal electrons into full dsa regime has not been explored yet , because extreme computational resources are required to follow the simulations for a large dynamic range of particle energy . the main reasons that we implement the fossil electron distribution , instead of the shock injection only case , are ( 1 ) the relative scarcity of radio relics compared to the abundance of shocks expected to form in the icm , ( 2 ) the peculiar uniformity of the surface brightness of the sausage relic , and ( 3 ) curved integrated spectra often found in some radio relics , implying the acceleration duration @xmath45 myr , much shorter than the cluster dynamical time . in this paper , we consider a dsa model for radio relics ; a spherical shock moves into a magnetized gas cloud containing fossil relativistic electrons , while propagating through a density gradient in the cluster outskirts . specifically , we perform time - dependent dsa simulations for several spherical shock models with the parameters relevant for the sausage relic . we then calculate the surface brightness profile , @xmath59 , and the volume - integrated radio spectrum , @xmath60 , by adopting a specific geometrical structure of shock surface , and compare them with the observational data of the sausage relic . in section 2 , the dsa simulations and the model parameters are described . the comparison of our results with observations is discussed in section 3 . a brief summary is given in section 4 . we follow the evolution of the cr electron population by solving the following diffusion - convection equation in the one - dimensional ( 1d ) spherical geometry : @xmath61 + p { \partial \over { \partial y } } \left ( { b\over p^2 } g_e \right ) , \label{diffcon}\end{aligned}\ ] ] where @xmath62 is the pitch - angle - averaged phase space distribution function of electrons , @xmath63 is the flow velocity and @xmath64 with the electron mass @xmath65 and the speed of light @xmath66 @xcite . the spatial diffusion coefficient , @xmath67 , is assumed to have a bohm - like dependence on the rigidity , @xmath68 the cooling coefficient @xmath69 accounts for radiative cooling , and the cooling time scale is defined as @xmath70 where @xmath71 is the lorentz factor of electrons . here the ` effective ' magnetic field strength , @xmath72 with @xmath73 , takes account for the ic loss due to the cosmic background radiation as well as synchrotron loss . the redshift of the cluster ciza j2242.8 + 5301 is @xmath7 . assuming that the test - particle limit is applied at weak cluster shocks with @xmath74 several ( see table 1 ) , the usual gasdynamic conservation equations are solved to follow the background flow speed , @xmath63 , using the 1d spherical version of the crash ( cosmic - ray amr shock ) code @xcite . the structure and evolution of @xmath63 are fed into equation ( [ diffcon ] ) , while the gas pressure , @xmath75 , is used in modeling the postshock magnetic field profile ( see section 2.3 ) . we do not consider the acceleration of cr protons in this study , since the synchrotron emission from cr electrons is our main interest and the dynamical feedback from the cr proton pressure can be ignored at weak shocks ( @xmath76 ) in the test - particle regime @xcite . in order to optimize the shock tracking of the crash code , a comoving frame that expands with the instantaneous shock speed is adopted . further details of dsa simulations can be found in @xcite . to set up the initial shock structure , we adopt a sedov self - similar blast wave propagating into a _ uniform static _ medium , which can be specified by two parameters , typically , the explosion energy , @xmath77 , and the background density , @xmath78 @xcite . for our dsa simulations , we choose the initial shock radius and speed , @xmath79 and @xmath80 , respectively , and adopt the self - similar profiles of the gas density @xmath81 , the gas pressure @xmath82 , and the flow speed @xmath83 behind the shock in the upstream rest - frame . for the fiducial case ( * sa1 * model in table 1 ) , for example , the initial shock parameters are @xmath84 , and @xmath85 . for the model parameters for the shock and upstream conditions , refer to table 1 and section 3.1 . we suppose that at the onset of the dsa simulations , this initial shock propagates through the icm with the gas density gradient described by a power law of @xmath86 . typical x - ray brightness profiles of observed clusters can be represented approximately by the so - call beta model for isothermal icms , @xmath87^{-3\beta/2}$ ] with @xmath88 @xcite . in the outskirts of clusters , well outside of the core radius ( @xmath89 ) , it asymptotes as @xmath90 . we take the upstream gas density of @xmath91 as the fiducial case ( * sa1 * ) , but also consider @xmath92 ( * sa3 * ) and @xmath93 constant ( * sa4 * ) for comparison . the shock speed and mach number decrease in time as the spherical shock expands , depending on the upstream density profile . @xcite demonstrated that the shock decelerates approximately as @xmath94 for @xmath93 constant and as @xmath95 for @xmath96 , while the shock speed is almost constant in the case of @xmath92 . as a result , nonlinear deviations from the dsa predictions for steady planar shocks are expected to become the strongest in * sa4 * model , while the weakest in * sa3 * model . in the fiducial * sa1 * model , which is the most realistic among the three models , the effects of the evolving spherical shock are expected to be moderate . the icm temperature is set as @xmath97kev , adopted from @xcite , in most of the models . hereafter , the subscripts `` 1 '' and `` 2 '' are used to indicate the quantities immediately upstream and downstream of the shock , respectively . although the icm temperature is known to decrease slightly in the cluster outskirt , it is assumed to be isothermal , since the shock typically travels only @xmath98 mpc for the duration of our simulations @xmath45 magnetic fields in the downstream region of the shock are the key ingredient that governs the synchrotron cooling and emission of cr electrons in our models . we assume that the fossil cloud is magnetized to @xmath15 g level . as discussed in the introduction , observations indicate that the magnetic field strength decreases from @xmath99 in the core region to @xmath100 in the periphery of clusters ( e.g. , * ? ? ? * ; * ? ? ? this corresponds to the plasma beta of @xmath57 in typical icms ( e.g. , * ? ? ? * ; * ? ? ? on the other hand , it is well established that magnetic fields can be amplified via resonant and non - resonant instabilities induced by cr protons streaming upstream of _ strong _ shocks @xcite . in addition , magnetic fields can be amplified by turbulent motions behind shocks @xcite . recent hybrid plasma simulations have shown that the magnetic field amplification factor due to streaming cr protons scales with the alfvnic mach number , @xmath101 , and the cr proton acceleration efficiency as @xmath102 @xmath103 @xcite . here , @xmath104 is the turbulent magnetic field perpendicular to the mean background magnetic field , @xmath105 is the downstream cr pressure , and @xmath106 is the upstream ram pressure . for typical radio relic shocks , the sonic and alfvnic mach numbers are expected to range @xmath107 and @xmath108 , respectively ( e.g. , * ? ? ? * ) . the magnetic field amplification in both the upstream and downstream of _ weak _ shocks is not yet fully understood , especially in high beta icm plasmas . so we consider simple models for the postshock magnetic fields . for the fiducial case , we assume that the magnetic field strength across the shock transition is increased by compression of the two perpendicular components : @xmath109 where @xmath110 and @xmath111 are the magnetic field strengths immediately upstream and downstream of the shock , respectively , and @xmath112 is the time - varying compression ratio across the shock . for the downstream region ( @xmath113 ) , the magnetic field strength is assumed to scale with the gas pressure : @xmath114^{1/2 } , \label{bd}\ ] ] where @xmath115 is the gas pressure immediately behind the shock . this assumes that the ratio of the magnetic to thermal energy density is constant downstream of the shock . since @xmath82 decreases behind the spherical blast wave , @xmath116 also decreases downstream as illustrated in figure 2 . this fiducial magnetic field model is adopted in most of the models described in table 1 , except * sa2 * and * sa4 * models . the range of @xmath117 is shown for the acceleration duration of @xmath118 myr in table 1 , reflecting the decrease of shock compression ratio during the period , but the change is small . in the second , more simplified ( but somewhat unrealistic ) model , it is assumed that @xmath119 and @xmath120 , and the downstream magnetic field strength is constant , , @xmath121 for @xmath113 . this model was adopted in @xcite and also for * sa2 * and * sa4 * models for comparison in this study . @xcite showed that the postshock synchrotron emission increases downstream away from the shock in the case of a decelerating shock , because the shock is stronger at earlier time . but such nonlinear signatures become less distinct in the model with decreasing downstream magnetic fields , compared with the model with constant downstream magnetic fields , because the contribution of synchrotron emission from further downstream region becomes weaker . as described in the introduction , the new picture of particle injection emerged from recent pic simulations is quite different from the ` classical ' thermal leakage injection model commonly employed in previous studies of dsa ( e.g. , * ? ? ? however , the requirement of @xmath122 for particles to take part in the full fermi 1st - order process , scattering back and forth diffusively across the shock transition zone with thickness , @xmath123 , seems to remain valid @xcite . here , @xmath124 is the most probable momentum of thermal protons with postshock temperature @xmath125 and @xmath126 is the gyroradius of particles . in other words , only suprathermal particles with the gyro - radius greater than the shock thickness are expected to cross the shock transition layer . hence , we adopt the traditional phenomenological model in which only particles above the injection momentum , @xmath127 , are allowed to get injected into the cr populations at the lowest momentum boundary @xcite . this can be translated to the electron lorentz factor , @xmath128 . in the case of expanding shocks considered in this study , @xmath129 decreases as the shock slows down in time . as mentioned in the introduction , one peculiar feature of the sausage relic is the _ uniform _ surface brightness along the mpc - scale arc - like shape , which requires a special geometrical distribution of shock - accelerated electrons @xcite . some of previous studies adopted the ribbon - like curved shock surface and the downstream swept - up volume , viewed edge - on with the viewing extension angle @xmath130 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . we suggest a picture where a spherical shock of radius @xmath131 mpc passes through an elongated cloud with width @xmath132 kpc and length @xmath133 mpc , filled with fossil electrons . then the shock surface penetrated into the cloud becomes a ribbon - like patch , distributed on a sphere with radius @xmath131 mpc with the angle @xmath134 . the downstream volume of radio - emitting , reaccelerated electrons has the width @xmath135 $ ] , as shown in figure 1 of @xcite . hereafter the ` acceleration age ' , @xmath136 , is defined as the duration of electron acceleration since the shock encounters the cloud . this model is expected to produce a uniform surface brightness along the relic length . moreover , if the acceleration age is @xmath137 , the volume - integrated radio spectrum is expected to steepen gradually over 0.1 - 10 ghz @xcite . there are several possible origins for such clouds of relativistic electrons in the icms : ( 1 ) old remnants of radio jets from agns , ( 2 ) electron populations that were accelerated by previous shocks and have cooled down below @xmath138 , and ( 3 ) electron populations that were accelerated by turbulence during merger activities . although an agn jet would have a hollow , cocoon - like shape initially , it may turn into a filled , cylindrical shape through diffusion , turbulent mixing , or contraction of relativistic plasmas , as the electrons cool radiatively . during such evolution relativistic electrons could be mixed with the icm gas within the cloud . we assume that the cloud is composed of the thermal gas of @xmath30 , whose properties ( density and temperature ) are the same as the surrounding icm gas , and an additional population of fossil electrons with dynamically insignificant pressure . in that regard , our fossil electron cloud is different from hot bubbles considered in previous studies of the interaction of a shock with a hot rarefied bubble ( e.g. , * ? ? ? * ) . in the other two cases where electrons were accelerated either by shocks or by turbulence ( see , e.g. , * ? ? ? * ) , it is natural to assume that the cloud medium should contain both thermal gas and fossil electrons . three different spectra for fossil electron populations are considered . in the first fiducial case ( model ) , nonthermal electrons have the momentum around @xmath139 , which corresponds to @xmath140 for the model shock parameters considered here . so in this model , seed electrons with @xmath141 are injected from the fossil population and re - accelerated into radio - emitting cr electrons with @xmath142 . since we compare the surface brightness profiles in arbitrary units here , we do not concern about the normalization of the fossil population . in the second model , the fossil electrons are assumed to have a power - law distribution extending up to @xmath143 , @xmath144 for the modeling of the sausage relic , the value of @xmath145 is chosen as @xmath146 with @xmath147 ( * sa1p * model ) . as mentioned in the introduction , the volume - integrated radio spectrum of the sausage relic seems to steepen at high frequencies , perhaps more strongly than expected from radiative cooling alone @xcite . so in the third model , we consider a power - law population with exponential cutoff as follows : @xmath148 , \label{f1exp}\ ] ] where @xmath149 is the cutoff lorentz factor . this may represent fossil electrons that have cooled down to @xmath150 from the power - law distribution in equation ( [ f1 ] ) . the integrated spectrum of the sausage relic shows a strong curvature above @xmath5 ghz @xcite , which corresponds to the characteristic frequency of synchrotron emission from electrons with @xmath151 when the magnetic field strength ranges @xmath152 ( see equation ( [ nupeak ] ) below ) . so @xmath153 and @xmath154 are chosen for * sc1pex1 * and * sc1pex2 * models , respectively . figure 3 shows the electron distributions in the cloud medium : the thermal distribution for the icm gas with @xmath155 kev and the fossil population . the characteristics of the different models , * sa1 * ( fiducial model ) , * sa1p * ( power - law ) , and * sc1pex1 * ( power - law with an exponential cutoff ) will be described in section 3.1 . note that there could be ` suprathermal ' distribution between the thermal and cr populations . but it does not need to be specified here , since only @xmath156 around @xmath139 controls the spectrum of re - accelerated electrons . we consider several models whose characteristics are summarized in table 1 . for the fiducial model , * sa1 * , the upstream density is assumed to decrease as @xmath157 , while the upstream temperature is taken to be @xmath158 kev . since we do not concern about the absolute radio flux level in this study , @xmath78 needs not to be specified . the preshock magnetic field strength is @xmath159 and the immediate postshock magnetic field strength is @xmath160 during 60 myr , while the downstream field , @xmath116 , is given as in equation ( [ bd ] ) . the initial shock speed is @xmath161 , corresponding to the sonic mach number of @xmath162 at the onset of simulation . in all models with @xmath91 , the shock slows down as @xmath95 , and at the electron acceleration age of 60 myr , @xmath163 with @xmath164 and @xmath165 . the fossil seeds electrons are assumed to have a delta - function - like distribution around @xmath166 . in * sa1b * model , the upstream magnetic field strength is weaker with @xmath167 and @xmath168 during 60 myr . otherwise , it is the same as the fiducial model . so the character ` b ' in the model name denotes ` weaker magnetic field ' , compared to the fiducial model . comparison of * sa1 * and * sa1b * models will be discussed in section 3.4 . in * sa1p * model , the fossil electrons have a power - law population of @xmath169 , while the rest of the parameters are the same as those of * sa1 * model . the character ` p ' in the model name denotes a ` power - law ' fossil population . in * sa2 * model , both the preshock and postshock magnetic field strengths are constant , @xmath170 and @xmath120 , otherwise it is the same as * sa1 * model . in * sa3 * model , the upstream gas density decreases as @xmath171 , so the shock decelerates more slowly , compared to * sa1 * model . considering this , the initial shock speed is set to @xmath172 with @xmath173 . at the acceleration age of 60 myr , the shock speed decreases to @xmath174 corresponding to @xmath175 . in * sa4 * model , the upstream density is constant , so the shock decelerates approximately as @xmath176 , more quickly , compared to * sa1 * model . the upstream and downstream magnetic field strengths are also constant as in * sa2 * model . figure 2 compares the profiles of the flow speed , @xmath63 , and the magnetic field strength , @xmath177 , in * sa1 * and * sa4 * models . note that although the shocks in * sa1 * and * sa4 * models are not very strong with the initial mach number @xmath178 , they decelerate approximately as @xmath95 and @xmath94 , respectively , as in the self - similar solutions of blast waves @xcite . the shock speed decreases by @xmath179 during the electron acceleration age of 60 myr in the two models . in * sb1 * model , the preshock temperature , @xmath180 , is lower by a factor of @xmath181 , and so the initial shock speed @xmath161 corresponds to @xmath182 . the shock speed @xmath183 at the age of 60 myr corresponds to @xmath12 , so the injection spectral index @xmath48 . the ` * sb * ' shock is different from the ` * sa * ' shock in terms of only the sonic mach number . in * sc1pex1 * and * sc1pex2 * models , the fossil electrons have @xmath184 $ ] with the cutoff at @xmath185 and @xmath186 , respectively . the character ` pex ' in the model name denotes a ` power - law with an exponential cutoff ' . a slower initial shock with @xmath187 and @xmath188 is chosen , so at the acceleration age of 80 myr the shock slows down to @xmath189 with @xmath190 . the ` * sc * ' shock differs from the ` * sa * ' shock in terms of the shock speed and the sonic mach number . the integrated spectral index at high frequencies would be steep with @xmath191 , while @xmath192 at low frequencies due to the flat fossil electron population . they are intended to be toy models that could reproduce the integrated spectral indices , @xmath193 for @xmath194 ghz and @xmath195 for @xmath196 ghz , compatible with the observed curved spectrum of the sausage relic @xcite . the local synchrotron emissivity , @xmath197 , is calculated , using the electron distribution function , @xmath198 , and the magnetic field profile , @xmath177 . then , the radio intensity or surface brightness , @xmath59 , is calculated by integrating @xmath199 along lines - of - sight ( loss ) . @xmath200 @xmath43 is the distance behind the projected shock edge in the plane of the sky , as defined in the introduction , and @xmath201 is the path length along loss ; @xmath86 , @xmath43 , and @xmath201 are related as @xmath202 . the extension angle is @xmath203 ( see section 2.5 ) . note that the radio flux density , @xmath204 , can be obtained by convolving @xmath59 with a telescope beam as @xmath205 , if the brightness distribution is broad compared to the beam size of @xmath206 . the volume - integrated synchrotron spectrum , @xmath207 , is calculated by integrating @xmath199 over the entire downstream region with the assumed geometric structure described in section 2.5 . the spectral indices of the local emissivity , @xmath197 , and the integrated spectrum , @xmath60 , are defined as follows : @xmath208 @xmath209 as noted in the introduction , unresolved radio observations usually report the spectral index , ` @xmath36 ' at various frequencies , which is equivalent to @xmath210 here . in the postshock region , the cutoff of the electron spectrum , @xmath211 , decreases downstream from the shock due to radiative losses , so the volume - integrated electron spectrum steepens from @xmath212 to @xmath213 above the break lorentz factor ( see figure 5 ) . at the electron acceleration age @xmath136 , the break lorentz factor can be estimated from the condition @xmath214 @xcite : @xmath215 hereafter , the postshock magnetic field strength , @xmath111 , and @xmath216 are expressed in units of @xmath15 g . since the synchrotron emission from mono - energetic electrons with @xmath217 peaks around @xmath218 the break frequency that corresponds to @xmath219 becomes @xmath220 so the volume - integrated synchrotron spectrum , @xmath60 , has a spectral break , or more precisely a _ gradual _ increase of the spectral index approximately from @xmath31 to @xmath47 around @xmath34 . for electrons with @xmath221 , the radiative cooling time in equation ( [ trad ] ) becomes shorter than the acceleration age if @xmath222 myr . then , the width of the spatial distribution of those high - energy electrons downstream of the shock becomes @xmath223 where @xmath224 is the downstream flow speed . with the characteristic frequency @xmath225 of electrons with @xmath217 in equation ( [ nupeak ] ) , the width of the synchrotron emitting region behind the shock at the observation frequency of @xmath226 would be similar to @xmath227 : @xmath228^{-1/2}. \label{lwidth}\ ] ] here , @xmath229 is a numerical factor of @xmath230 . this factor takes account for the fact that the spatial distribution of synchrotron emission at @xmath225 is somewhat broader than that of electrons with the corresponding @xmath71 , because more abundant , lower energy electrons also make contributions @xcite . one can estimate two possible values of the postshock magnetic field strength , @xmath111 , from this relation , if @xmath231 can be determined from the observed profile of surface brightness and @xmath224 is known . for example , @xcite inferred two possible values of the postshock magnetic field strength of the sausage relic , @xmath232 and @xmath233 , by assuming that the fwhm of surface brightness , @xmath234 , is the same as @xmath235 kpc ( i.e. , @xmath236 ) . on the other hand , @xcite showed that the profile of surface brightness depends strongly on @xmath237 in the case of the assumed geometrical structure . figure 4 illustrates the results of a _ shock with different @xmath237 s . the shock has @xmath238 and @xmath239 . the magnetic field strengths are assumed to be @xmath170 and @xmath120 , upstream and downstream of the shock , respectively . the results shown are at the acceleration age of 80 myr . note that the quantities , @xmath240 , @xmath241 , and @xmath242 , are multiplied by arbitrary powers of @xmath71 and @xmath243 , so they can be shown for different values of @xmath71 and @xmath243 with single linear scales in figure 4 . here , the fwhm of @xmath244 is , for instance , @xmath245 kpc for @xmath246 ( red dotted line in the top panel ) , while the fwhm of @xmath199 is @xmath247 kpc for @xmath248 ghz ( red dotted line in the middle panel ) . note that the values of @xmath71 and @xmath249 are chosen from the sets of discrete simulation bins , so they satisfy only approximately the equation ( [ nupeak ] ) . the bottom panel demonstrates that the fwhm of @xmath59 , @xmath234 , strongly depends on @xmath237 . for instance , @xmath250 kpc for @xmath248 ghz , if @xmath203 ( red dotted line ) . this implies that due to the projection effect , @xmath234 could be substantially different from @xmath251 . so @xmath234 of the observed radio flux profile may not be used to infer possible values of @xmath111 in radio relics , unless the geometrical structure of the shock is known and so the projection effect can be modeled accordingly . note that the quantity @xmath252^{-1 } \cdot({b_2/ 5})^{1/2}$ ] in equation ( [ lwidth ] ) also appears as @xmath253 in equation ( [ fbr ] ) . so the break frequency becomes identical for the two values of @xmath111 , @xmath254 and @xmath255 , that give the same value of @xmath256 . for example , * sa1 * model with @xmath257 and * sa1b * model with @xmath258 ( see table 1 ) produce not only the similar @xmath231 but also the similar spectral break in @xmath60 . but the corresponding values of @xmath219 in equation ( [ pbr ] ) are different for the two models with @xmath254 and @xmath255 . moreover , the amplitude of the integrated spectrum would scale as @xmath259 , if the electron spectrum @xmath260 for @xmath261 is similar for the two models . we compare these two models in detail in the next section . in figure 5 , we first compare the electron spectra and the synchrotron emission spectra in * sa1 * and * sa1b * models . here , the plotted quantities , @xmath262 , @xmath263 , @xmath199 and @xmath60 , are in arbitrary units . the postshock magnetic field strength decreases from @xmath264 to @xmath265 in 110 myr in * sa1 * model , while @xmath266 to @xmath267 in * sa1b * model . in these two models , the values of @xmath268 are similar , so the spectral break in the integrated spectra should be similar as well . the left panels of figure 5 show that the electron spectrum at the shock , @xmath269 , steepens in time as the shock weakens . as expected , the volume - integrated electron spectrum , @xmath270 , steepens by one power of @xmath271 above the break momentum , @xmath272 , which decreases in time due to radiative losses . in both models , however , the slopes , @xmath273 and @xmath274 , deviate from the simple steepening of @xmath275 to @xmath276 above @xmath272 , because of the time - dependence of the shock properties . the right panels of figure 5 show the synchrotron spectrum at the shock , @xmath277 , the volume - integrated synchrotron spectrum , @xmath60 , and their spectral indices , @xmath44 and @xmath210 . note that in both models , the transition of @xmath210 from @xmath31 to @xmath47 is broader than the transition of @xmath278 from @xmath279 to @xmath276 . this is because the synchrotron emission at a given frequency comes from electrons with a somewhat broad range of @xmath71 s . as in the case of @xmath278 , @xmath210 does not follow the simple steepening , but rather shows nonlinear features due to the evolving shock properties . this implies that the simple relation , @xmath280 , which is commonly adopted in order to confirm the dsa origin of synchrotron spectrum , should be applied with some caution in the case of evolving shocks . the highest momentum for @xmath269 , @xmath281^{1/2}$ ] , is higher in * sa1 * model with stronger magnetic field , but the amplitude of @xmath269 near @xmath282 , say at @xmath283 , is greater in * sa1b * model with weaker magnetic field . as a consequence , the ratio of the synchrotron emission of the two models is somewhat less than the ratio of the magnetic energy density , @xmath284 . in * sa1b * model , for example , the amplitudes of @xmath277 and @xmath60 at @xmath285 ghz are reduced by a factor of @xmath286 , compared to the respective values in * sa1 * model . also the cutoff frequencies for both @xmath277 and @xmath60 are lower in * sa1b * model , compared to those in * sa1 * model . as pointed above , @xmath34 is almost the same in the two models , although @xmath272 is different . this confirms that we would get two possible solutions for @xmath111 , if we attempt to estimate the magnetic field strength from the value of @xmath34 in the integrated spectrum . as noted in section 3.3 , the width of the downstream electron distribution behind the shock is determined by the advection length , @xmath287 , for low - energy electrons or by the cooling length , @xmath288 , for high - energy electrons . as a result , at high frequencies the width of the synchrotron emission region , @xmath231 , varies with the downstream magnetic field strength for given @xmath224 and @xmath243 as in equation ( [ lwidth ] ) . in addition , the surface brightness profile , @xmath242 , and its fwhm , @xmath234 , also depend on the extension angle , @xmath237 , as demonstrated in figure 4 . in figures 6 and 7 , the spatial profiles of @xmath242 are shown for eight models listed in table 1 . here , we choose the observation frequencies @xmath289 mhz ; the source frequencies are given as @xmath290 with @xmath7 . while the downstream flow speed ranges @xmath291 in most of the models , @xmath292 in * sc1pex1 * model . so the results are shown at @xmath293 and @xmath294 , except in * sc1pex1 * model for which the results are shown at @xmath295 and @xmath296 . note that the quantity , @xmath297 , is shown in order to be plotted with one linear scale , where @xmath298 is the numerical scale factor specified in each panel . in * sb1 * model with a stronger shock , for example , the electron acceleration is more efficient by a factor of about @xmath299 , compared to other models , so the profiles shown are reduced by similar factors . for a fixed value of @xmath237 , the surface brightness profile is determined by the distribution of @xmath197 behind the shock along the path length @xmath201 , as given in equation ( [ sb ] ) . the intensity increases gradually from the shock position ( @xmath300 ) to the first _ inflection point _ , @xmath301 kpc with @xmath203 , mainly due to the increase of the path length along loss . since the path length starts to decrease beyond @xmath302 , if the emissivity , @xmath199 , is constant or decreases downstream of the shock , then @xmath59 should decrease for @xmath303 . in all the models considered here , however , at low frequencies , @xmath199 increases downstream of the shock , because the model spherical shock is faster and stronger at earlier times . as a result , @xmath59 at 150 mhz is almost constant or decrease very slowly beyond @xmath302 , or even increases downstream away from the shock in some cases ( and * sa4 * models ) . so the downstream profile of the synchrotron radiation at low frequencies emitted by uncooled , low - energy electrons could reveal some information about the shock dynamics , providing that the downstream magnetic field strength is known . the second inflection in @xmath242 occurs roughly at @xmath304 for low frequencies , and at @xmath305 for high frequencies where @xmath306 , with @xmath307 . here , @xmath308 , and @xmath227 is given in equation ( [ lcool ] ) . figures 6 and 7 exhibit that at 30 myr ( @xmath309 ) , the second inflection appears at the same position ( @xmath310 kpc ) for the three frequencies shown . at later times , the position of the second inflection depends on @xmath136 at low frequencies , while it varies with @xmath111 and @xmath243 in addition to @xmath224 . thus , only if @xmath311 , @xmath234 at high frequencies can be used to infer @xmath111 , providing that @xmath237 and @xmath224 are known . the width of the sausage relic , defined as the fwhm of the surface brightness , was estimated to be @xmath312 kpc at 600 mhz @xcite . as shown in figures 6 and 7 , in all the models except * sa4 * , the calculated fwhm of @xmath242 at 600 mhz ( red dotted lines ) at the acceleration age of @xmath313 myr would be compatible with the observed value . in * sa4 * model , the intensity profile at 600 mhz seems a bit too broad to fit the observed profile . figures 8 and 9 show the volume - integrated synchrotron spectrum , @xmath60 , and its slope , @xmath210 , for the same models at the same ages as in figures 6 and 7 . here @xmath60 is integrated over the curved , ribbon - like downstream volume with @xmath203 , as described in section 3.2 . in the figures , @xmath60 is plotted in arbitrary units . the filled circles represent the data points for the integrated flux of the sausage relic , which were taken from table 1 of @xcite and re - scaled roughly to fit by eye the spectrum of * sa1 * model at @xmath314 ( the red dotted line in the top - left panel of figure 8) . * observational errors given by @xcite are about 10 % , so the error bars are in fact smaller than the size of the filled circles in the figures , except the one at 16 ghz with 25 % . * at first it looks that in most of the models , the calculated @xmath60 reasonably fits the observed data including the point at 16 ghz . but careful inspections indicate that our models fail to reproduce the rather abrupt increase in the spectral curvature at @xmath5 ghz . the calculated @xmath60 is either too steep at low frequencies @xmath315 ghz , or too flat at high frequencies @xmath316 ghz . for instance , @xmath60 of the fiducial model , * sa1 * , is too steep at @xmath317 ghz . on the other hand , * sa1p * model with a flatter fossil population and * sb1 * model with a flatter injection spectrum ( due to higher @xmath27 ) seem to have the downstream electron spectra too flat to explain the observed data for @xmath318 ghz . as mentioned before , the transition of the integrated spectral index from @xmath31 to @xmath319 occurs gradually over a broad range of frequency , @xmath2 . if we estimate @xmath34 as the frequency at which the gradient , @xmath320 , has the largest value , it is @xmath321 , 1 , 0.3 ghz at @xmath322 , respectively , for all the models except * sc1pex1 * model ( for which @xmath210 is shown at different epochs ) . since @xmath34 is determined mainly by the magnetic field strength and the acceleration age , it does not sensitively depend on other details of the models . in * sc1pex1 * model , the power - law portion ( @xmath323 ) gives a flatter spectrum with @xmath324 at low frequencies , while the newly injected population at the shock with @xmath189 results in a steeper spectrum with @xmath325 at high frequencies . note that the spectral index estimated with the observed flux in @xcite between 2.3 and 16 ghz is @xmath326 , implying @xmath189 . in figure 10 , @xmath60 s from all the models considered here are compared at @xmath327 myr ( * sa * and * sb * shocks ) or @xmath328 myr ( * sc * shocks ) . the observed data points of the sausage relic are also shown . in most of the models ( * sa * models ) , the shocks have @xmath329 at 60 myr , so the predicted @xmath60 between 2.3 and 16 ghz is a bit flatter with @xmath330 than the observed spectrum with @xmath326 at the same frequency range . as mentioned above , * sa1p * and * sb1 * models produce @xmath60 , which is significantly flatter at high frequencies than the observed spectrum . the toy model , * sc1pex1 * , seems to produce the best fit to the observed spectrum , as noted before . in short , the volume - integrated synchrotron spectra calculated here steepen gradually over @xmath2 with the break frequency @xmath3 ghz , if @xmath331 myr . however , all the models considered here seem to have some difficulties fitting the sharp curvature around @xmath5 ghz in the observed integrated spectrum . this implies that the shock dynamics and/or the downstream magnetic field profile could be different from what we consider here . perhaps some additional physics that can introduce a feature in the electron energy spectrum , stronger than the ` @xmath276 ' steepening due to radiative cooling , might be necessary . in figure 11 , we present for the three models , * sa1 * , * sa1b * , and * sc1pex1 * , the mean intensity spectrum in the downstream regions of @xmath332 $ ] , @xmath333 , where @xmath334 and @xmath335 runs from 1 to 6 . this is designed to be compared with the ` spatially resolved ' spectrum behind the shock in radio observations ( e.g. , * ? ? ? the figure shows how the downstream spectrum cuts off at progressively lower frequencies due to radiative cooling as the observed position moves further away from the shock . we propose a model that may explain some characteristics of giant radio relics : the relative rareness , uniform surface brightness along the length of thin arc - like radio structure , and spectral curvature in the integrated radio spectrum over @xmath336 ghz . in the model , a spherical shock encounters an elongated cloud of the icm thermal gas that is permeated by enhanced magnetic fields and an additional population of fossil relativistic electrons . as a result of the shock passage , the fossil electrons are re - accelerated to radio - emitting energies ( @xmath337 ) , resulting in a birth of a giant radio relic . in order to explore this scenario , we have performed time - dependent , dsa simulations of spherical shocks with the parameters relevant for the sausage radio relic in cluster ciza j2242.8 + 5301 . in the fiducial model , the shock decelerates from @xmath338 ( @xmath339 ) to @xmath340 ( @xmath164 ) at the acceleration age of 60 myr . the seed , fossil electrons with @xmath341 are assumed to be injected into the cr population , which is subsequently re - accelerated to higher energies . such shocks are expected to produce the electron energy spectrum , @xmath342 , resulting in the synchrotron radiation spectrum with the injection index , @xmath343 , and the integrated index , @xmath330 , at high frequencies ( @xmath344 ghz ) . we consider various models with a range of shock parameters , different upstream gas density profiles , different downstream magnetic field profiles , and three types of fossil electron populations , as summarized in table 1 . adopting a ribbon - like curved shock surface and the associated downstream volume , which are constrained by the extension angle ( or viewing depth ) of @xmath203 as detailed in section 2.5 ( e.g * ? ? ? * ; * ? ? ? * ) , the radio surface brightness profile , @xmath242 , and the volume - integrated spectrum , @xmath60 , are calculated . \1 ) two observables , the break frequency in the integrated synchrotron spectrum , @xmath34 , and the width of the synchrotron emission region behind the shock , @xmath231 , can have identical values for two values of postshock magnetic field strength ( see equations [ [ fbr ] ] and [ [ lwidth ] ] ) . \2 ) the observed width of the surface brightness projected onto the sky plane , @xmath234 , strongly depends on the assumed value of @xmath237 ( see figure 4 ) . so @xmath234 may not be used to estimate the postshock magnetic field strength , unless the projection effects can be modeled properly . \3 ) the integrated synchrotron spectrum is expected to have a spectral curvature that runs over a broad range of frequency , typically for @xmath345 . for a shock of @xmath346 with the postshock magnetic field strength , @xmath347 or @xmath348 , the integrated spectral index increases gradually from @xmath343 to @xmath349 over @xmath285 ghz , if the duration of the shock acceleration is @xmath286 myr \4 ) assuming that the upstream sound speed is @xmath350 ( @xmath351 kev ) as inferred from x - ray observation , a shock of @xmath346 and @xmath352 ( e.g. , * sa1 * model ) can reasonably explain the observed width , @xmath353 kpc @xcite , and the curved integrated spectrum of the sausage relic @xcite . * sb1 * model with a shock of @xmath354 , however , produces the integrated spectrum that seems too flat to explain the observed spectrum above @xmath355 ghz . \5 ) we also consider two toy models with power - law electron populations with exponential cutoffs at @xmath356 , @xmath357 $ ] ( * sc1pex1 * and * sc1pex2 * models ) . they may represent the electron populations that were produced earlier and then have cooled down to @xmath356 . * sc1pex1 * model with a weaker shock ( @xmath189 ) reproduces better the characteristics of the observed integrated spectrum . but the steepening of the integrated spectrum due to radiative cooling alone may not explain the strong spectral curvature above 1.5 ghz toward 16 ghz . \6 ) this strong curvature at @xmath5 ghz may imply that the downstream electron energy spectrum is influenced by some additional physical processes other than radiative losses , because the integrated spectrum of radiatively cooled electrons steepens with the frequency only gradually . this conclusion is likely to remain unchanged even in the case where the observed spectrum consists of the synchrotron emission from multiple shocks with different mach numbers , as long as the postshock electrons experience only simple radiative cooling . the authors thank the anonymous referee for his / her thorough review and constructive suggestions that lead to a significant improvement of the paper . hk was supported by basic science research program through the national research foundation of korea ( nrf ) funded by the ministry of education ( 2014r1a1a2057940 ) . dr was supported by the national research foundation of korea through grant nrf-2014m1a7a1a03029872 and nrf-2012k1a3a7a03049606 . lccccccc & @xmath358 & 2.5 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363 + * sa1b * & @xmath358 & 0.25 & @xmath364 & @xmath360 & @xmath361 & @xmath362 & @xmath363 + * sa1p * & @xmath358 & 2.5 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath365 + * sa2*@xmath366 & @xmath358 & 2.0 & @xmath367 & @xmath360 & @xmath361 & @xmath362 & @xmath363 + * sa3 * & @xmath368 & 2.5 & @xmath359 & @xmath360 & @xmath369 & @xmath370 & @xmath363 + * sa4*@xmath366 & @xmath78 & 2.0 & @xmath367 & @xmath360 & @xmath371 & @xmath372 & @xmath363 + * sb1 * & @xmath358 & 2.5 & @xmath373 & @xmath374 & @xmath361 & @xmath375 & @xmath363 + * sc1pex1*@xmath376 & @xmath358 & 2.5 & @xmath377 & @xmath360 & @xmath378 & @xmath379 & @xmath380 + * sc1pex2*@xmath381 & @xmath358 & 2.5 & @xmath377 & @xmath360 & @xmath378 & @xmath379 & @xmath382 + -0.8 cm , and the magnetic field , @xmath177 , of the spherical shock in * sa1 * ( top two panels ) and * sa4 * ( bottom two panels ) models at the acceleration age , @xmath383 , 60 , 110 myr ( black solid , red dotted , and blue dashed lines , respectively).,title="fig : " ] 1.5 cm kev ( red solid line ) , fossil electrons with @xmath384 in * sa1 * model ( black solid ) , a power - law of @xmath385 in * sa1p * model ( blue solid ) , and a power - law with an exponential cutoff in * sc1pex1 * model ( green solid ) . see table 1 for the different model parameters . the vertical dashed line demarcates the injection momentum , @xmath139 . an unspecified suprathermal distribution is shown in the red dotted line , but the electron distribution below @xmath139 is not relevant for dsa . , title="fig : " ] , @xmath241 , and @xmath242 for a steady _ planar _ shock with @xmath238 and @xmath239 at the acceleration age of 80 myr . here , @xmath386 is the downstream distance away from the shock , while @xmath43 is the distant behind the shock projected in the sky plane . top : @xmath387 is plotted for the electron lorentz factor , @xmath388 ( black solid line ) , @xmath389 ( red dotted ) , @xmath390 ( blue dashed ) , @xmath391 ( green long dashed ) , @xmath392 ( magenta dot - dashed ) , and @xmath393 ( cyan dot - long dashed ) . middle : @xmath394 is plotted for the observation frequency , @xmath395 ghz ( black solid line ) , 0.594 ghz ( red dotted ) , 1.33 ghz ( blue dashed ) , 2.36 ghz ( green long dashed ) , 16.7 ghz ( magenta dot - dashed ) , and 29.8 ghz ( cyan dot - long dashed ) . bottom : @xmath396 is plotted for the extension angle @xmath397 , and @xmath398 ( solid , dotted , and dashed lines , respectively ) . the red ( blue ) curves are for @xmath399 ( 1.33 ghz ) . the redshift of the host cluster is assumed to be @xmath7 here . ] -0.2 cm ( upper four panels ) and * sa1b * model with @xmath167 ( lower four panels ) at @xmath383 , 60 , 110 myr ( solid , dotted , and dashed lines , respectively ) . left : electron distribution function at the shock position , @xmath400 ( black lines ) , volume - integrated electron distribution function , @xmath401 ( red lines ) , and slopes of electron distribution functions , @xmath402 ( black lines ) and @xmath403 ( red lines ) . right : synchrotron spectrum at the shock position , @xmath404 ( black lines ) , volume - integrated synchrotron spectrum , @xmath405 ( red lines ) , and synchrotron spectral indices , @xmath406 ( black lines ) and @xmath407 ( red lines).,title="fig : " ] -0.3 cm , 600 , and 1400 mhz ( black solid , red dotted , blue dashed lines ) in * sa1 * , * sa1b * , * sa1p * , and * sc1pex * models ( from top to bottom panels ) . see table 1 for the model parameters . the results are shown at the acceleration age , @xmath136 , specified in each panel . for the extension angle , @xmath408 is adopted . the quantity @xmath409 is plotted with a scale factor @xmath298 to present it in the same ordinate scale for all the models.,title="fig : " ] -0.3 cm , 600 , and 1400 mhz ( black solid , red dotted , blue dashed lines ) in * sa2 * , * sa3 * , * sa4 * , and * sb1 * models ( from top to bottom panels ) . see table 1 for the model parameters . the results are shown at @xmath410 30 , 60 , and 110 myr ( from left to right panels ) . for the extension angle , @xmath408 is adopted . the quantity @xmath409 is plotted with a scale factor @xmath298 to present it in the same ordinate scale for all the models.,title="fig : " ] -0.3 cm , and its spectral index , @xmath210 , at @xmath411 , 60 , and 110 myr ( black solid , red dotted , and blue dashed lines , respectively ) for * sa1 * , * sa1b * , and * sa1p * models ( top three panels ) , and at @xmath411 , 80 , and 126 myr ( black solid , red dotted , and blue dashed lines , respectively ) for * sc1pex1 * model ( bottom panel ) . filled circles show the data from @xcite , scaled to fit by eye the red dotted line ( @xmath412 ) for * sa1 * model . observational errors are small , about @xmath413 , except for the data at 16 ghz with @xmath414 ( shown in a vertical bar).,title="fig : " ] -0.3 cm , and its spectral index , @xmath210 , at @xmath411 , 60 , and 110 myr ( black solid , red dotted , and blue dashed lines , respectively ) for * sa2 * , * sa3 * , * sa4 * , and * sb1 * models . filled circles show the data from @xcite , scaled as in figure 8 . , title="fig : " ] -1.2 cm , for all the models in table 1 . in the left panel , * sa1 * ( black solid line ) , * sa1b * ( red dotted ) , * sa1p * ( blue dashed ) , * sc1pex1 * ( green long dashed ) , and * sc1pex2 * ( magenta dot - dashed ) models are shown . in the right panel , * sa2 * ( black solid line ) , * sa3 * ( red dotted ) , * sa4 * ( blue dashed ) , and * sb1 * ( green long dashed ) models are shown . the results are shown at 60 myr , except for * sc1pex1 * and * sc1pex2 * models for which the results are shown at 80 myr . filled circles show the data from @xcite , scaled as in figure 8 . observational errors ( vertical bars ) are small , about @xmath413 , except for the data at 16 ghz with @xmath414.,title="fig : " ] -0.3 cm , averaged over @xmath415$]behind the shock , where @xmath334 , @xmath416 1 , 2 , 3 , 4 , 5 , 6 ( black solid , red dotted , blue dashed , green long dashed , magenta dot - dashed , cyan dot - long dashed lines , respectively ) for * sa1 * , * sa1b * and * sc1pex1 * models . the results are shown at the acceleration age , @xmath136 , specified in each panel.,title="fig : " ]

inspired by the results of recent experimental discoveries for charm and bottom baryons , the masses and magnetic moments of the heavy baryons with containing a single heavy quark are studied within light cone qcd sum rules method . our results on the masses of heavy baryons are in good agreement with predictions of other approaches , as well as with the existing experimental values . our predictions on the masses of the states , which are not yet discovered in the experiments , can be tested in the future experiments . a comparison of our results on the magnetic moments of these baryons and the hyper central model predictions is presented . pacs : 11.55.hx , 13.40.em , 14.20.lq , 14.20.mr

0807.3481

recently , considerable experimental progress has been made in the spectroscopy of baryons containing a single heavy quark . the cdf collaboration has observed four bottom baryons @xmath1 and @xmath2 @xcite . the do @xcite and cdf @xcite collaborations have seen the @xmath3 . the babar collaboration discovered the @xmath4 state @xcite . the cdf sensitivity appears adequate to observe new heavy baryons . study of the electromagnetic properties of baryons can give noteworthy information on their internal structure . one of the main static electromagnetic parameters of the baryons is their magnetic moments . magnetic moments of the heavy baryons in the framework of different approaches are widely discussed in the literature . in @xcite the magnetic moments of heavy baryons containing c - quark have been calculated using the naive quark model . in @xcite , the magnetic moments of charm and bottom baryons are computed in quark model and in @xcite heavy baryon magnetic moments are investigated within soliton type approaches . calculation of the magnetic moments of heavy baryons in the framework of relativistic quark model is done in @xcite . in @xcite the magnetic moments of @xmath5 and @xmath6 baryons are estimated in qcd sum rules with external electromagnetic field . the magnetic moments of the @xmath7 , @xmath8 and @xmath9 transition magnetic moments in light cone qcd sum rules are calculated in @xcite . the magnetic moments of @xmath10 in the same framework is studied in @xcite ( a detailed description of this method can be found in @xcite and references theirin ) . in the present work , we study the magnetic moments and masses of the ground state baryons with total angular momentum 3/2 and containing one heavy quark within light cone qcd sum rules . the paper is organized as follows . in section 2 , the light cone qcd sum rules for mass and magnetic moments of heavy baryons are calculated . section 3 is devoted to the numerical analysis of the mass and magnetic moment sum rules and discussion . before giving detailed calculation of the magnetic moments and masses of the heavy baryons , few words about their classifications are in order . the baryons with single heavy quarks belongs to either antisymmetric @xmath11 or symmetric @xmath12 flavor representation . for the s - wave heavy baryons , total flavor -spin wave function of the two light quarks must be symmetric , since their color wave function is antisymmetric . hence the total spin of the two light quarks is either @xmath13 for @xmath12 or @xmath14 for @xmath11 . the total angular momentum and parity of the s - wave heavy baryons are @xmath15 or @xmath16 for @xmath12 and @xmath15 for @xmath11 . the antitriplet @xmath17 of baryons contain isosinglet @xmath7 with quark content @xmath18 $ ] and the isodoublet @xmath19 with quark content @xmath20 $ ] and @xmath21 $ ] . the sextet @xmath12 of baryons involves isotriplet @xmath22 with quark content @xmath23 , @xmath24 and @xmath25 and the isodoublet @xmath26 with quark content @xmath27 and @xmath28 . here @xmath29 $ ] and @xmath30 denote the antisymmetric and symmetric flavor wave functions . after these preliminary remarks let us start calculation of the magnetic moments of the heavy flavored hadrons . the basic objects in lcsr method is the correlation function where hadrons are represented by the interpolating quark currents . for this aim , we consider the correlator @xmath31 where @xmath32 is the interpolating current of the heavy baryon and @xmath33 means the electromagnetic field . in qcd sum rules method , this correlation function is calculated in two different approaches : on the quark level , it describes a hadron as quarks and gluons interacting in qcd vacuum . in the phenomenological side , it is saturated by a tower of hadrons with the same flavor quantum numbers . the magnetic moments are determined by matching two different representations of the correlation function , i.e. , theoretical and phenomenological forms , using the dispersion relations . from eq . ( [ t ] ) , it follows that to calculate the correlation function from qcd side , we need the explicit expressions of the interpolating currents of heavy baryons with the angular momentum @xmath0 . the main condition for constructing the interpolating currents from quark field is that they should have the same quantum numbers of the baryons under consideration . for the heavy baryons with @xmath0 , the interpolating current is chosen in the following general form @xmath34 where c is the charge conjugation operator and a , b and c are color indices . the value of a and quark fields @xmath35 and @xmath36 for each heavy baryon is given in table 1 . @xmath37&@xmath38 & u&u + @xmath39&@xmath40&u&d + @xmath41&@xmath38&d&d + @xmath42&@xmath40&s&u + @xmath43&@xmath40&s&d + @xmath44 & @xmath38&s & s + the phenomenological part of the correlation function can be obtained by inserting the complete set of states between the interpolating currents in ( [ t ] ) with quantum numbers of heavy baryons . @xmath45 where @xmath46 , @xmath47 and q is the photon momentum . the vacuum to baryon matrix element of the interpolating current is defined as @xmath48 where @xmath49 is the residue and @xmath50 is the rarita - schwinger spinor . the matrix element @xmath51 entering eq . ( [ t2 ] ) can be parameterized in the following way @xmath52 \nonumber\\&-&\frac{q_{\mu}q_{\nu}}{(2m_{b})^{2}}\left[\gamma_{\rho}(g_{1}+g_{2})+ \frac{(p_{1}+p_{2})_{\rho}}{2m_{b}}g_{2}+q_{\rho}g_{3}\right]\left.\vphantom{\int_0^{x_2}}\right\}\bar u_{\nu}(p_{1}),\nonumber\\\end{aligned}\ ] ] where @xmath53 is the photon polarization vector , @xmath54 and @xmath55 are the form factors and they are functions of @xmath56 . in order to obtain the explicit expressions of the correlation function , we perform summation over spins of the spin 3/2 particles using @xmath57 using eqs . ( [ t2]-[raritabela ] ) in principle one can write down the phenomenological part of the correlator . but , the following two drawbacks appear : a ) all lorentz structures are not independent , b ) not only spin 3/2 , but spin 1/2 states also contribute to the correlation function . indeed the matrix element of the current @xmath32 between vacuum and spin 1/2 states is nonzero and is determined as @xmath58 where the condition @xmath59 is imposed . there are two different ways to remove the unwanted spin 1/2 contribution and retain only independent structures in the correlation function : 1 ) introduce projection operators for the spin 3/2 states , which kill the spin 1/2 contribution , 2 ) ordering dirac matrices in a specific order and eliminate the structures that receive contributions from spin @xmath60 states . in this work , we will follow the second method and choose the ordering for dirac matrices as @xmath61 . with this ordering for the correlator , we get @xmath62,\end{aligned}\ ] ] where @xmath63 and at @xmath64 , @xmath65 is the magnetic moment of the baryon in units of its natural magneton , @xmath66 . the factor 3 is due the fact that in the non - relativistic limit the interaction hamiltonian with magnetic field is equal to @xmath67 . on qcd side , the correlation function ( [ t ] ) can be evaluated using operator product expansion . after contracting out the quark pairs in eq . ( [ t ] ) using the wick s theorem , we get the following expression for the correlation function in terms of quark propagators @xmath68\mid\{s_{q}^{ca ' } \gamma_{\nu}s'^{bb'}_{q_{2}}\gamma_{\mu}s_{q_{1}}^{ac'}\nonumber\\&+&s_{q}^{cb ' } \gamma_{\nu}s'^{aa'}_{q_{1}}\gamma_{\mu}s_{q_{2}}^{bc'}+ s_{q_{2}}^{ca ' } \gamma_{\nu}s'^{bb'}_{q_{1}}\gamma_{\mu}s_{q}^{ac'}+s_{q_{2}}^{cb ' } \gamma_{\nu}s'^{aa'}_{q}\gamma_{\mu}s_{q_{1}}^{bc'}\nonumber\\&+ & s_{q_{1}}^{cb ' } \gamma_{\nu}s'^{aa'}_{q_{2}}\gamma_{\mu}s_{q}^{bc'}+s_{q_{1}}^{ca ' } \gamma_{\nu}s'^{bb'}_{q}\gamma_{\mu}s_{q_{2}}^{ac ' } + tr(\gamma_{\mu}s_{q_{1}}^{ab'}\gamma_{\nu}s'^{ba'}_{q_{2}})s^{cc'}_{q}\nonumber\\&+&tr(\gamma_{\mu}s_{q}^{ab'}\gamma_{\nu}s'^{ba'}_{q_{1}})s^{cc'}_{q_{2}}+tr(\gamma_{\mu}s_{q_{2}}^{ab'}\gamma_{\nu}s'^{ba'}_{q})s^{cc'}_{q_{1}}\}\mid 0\rangle,\end{aligned}\ ] ] where @xmath69 and @xmath70 is the full heavy ( light ) quark propagator . in calculation of the correlation function from qcd side , we take into account terms linear in @xmath71 and neglect quadratic terms.the correlator contains three different contributions : 1 ) perturbative contributions , 2 ) mixed contributions , i.e. , the photon is radiated from freely propagating quarks at short distance and at least one of quark pairs interact with qcd vacuum non - perturbatively . the last interaction is parameterized in terms of quark condensates . 3 ) non - perturbative contributions , i.e. , when photon is radiated at long distances . in order to get expressions of the contributions when the photon is radiated at short distance , it is enough to replace the propagator of the quark that emits the photon by @xmath72 where the fock - schwinger gauge , @xmath73 has been used . the expressions of the free light and heavy quark propagators in x representation are : @xmath74 where @xmath75 are bessel functions . the non - perturbative contributions to the correlation function can be obtained from eq . ( [ tree expresion.m ] ) by replacing one of the light quark propagators ( the quark that emits the photon ) by @xmath76 where @xmath77 is the full set of dirac matrices @xmath78 and sum over index @xmath79 is implied . under this procedure , two remaining quark propagators are full propagators involving perturbative as well as non - perturbative contributions . therefore , for the calculation of the correlation function from qcd side , the expressions of the heavy and light quark propagators in the presence of external field are needed . the light cone expansion of the quark propagator in external field is done in @xcite . the propagator receives contributions from @xmath80 , @xmath81 and @xmath82 , where @xmath83 is the gluon field strength tensor . in present work , we neglect terms with two gluons as well as four quarks operators due to the fact that their contributions are small @xcite . in this approximation the heavy and light quark propagators have the following expressions : @xmath84 , \nonumber \\ s_q(x ) & = & s_q^{free } ( x ) - \frac{m_q}{4 \pi^2 x^2 } - \frac{\langle \bar q q \rangle}{12 } \left(1 - i \frac{m_q}{4 } \not\!x \right ) - \frac{x^2}{192 } m_0 ^ 2 \langle \bar q q \rangle \left ( 1 - i \frac{m_q}{6}\not\!x \right ) \nonumber \\ & & - i g_s \int_0 ^ 1 du \left[\frac{\not\!x}{16 \pi^2 x^2 } g_{\mu \nu } ( ux ) \sigma_{\mu \nu } - u x^\mu g_{\mu \nu } ( ux ) \gamma^\nu \frac{i}{4 \pi^2 x^2 } \right . \nonumber \\ & & \left . - i \frac{m_q}{32 \pi^2 } g_{\mu \nu } \sigma^{\mu \nu } \left ( \ln \left ( \frac{-x^2 \lambda^2}{4 } \right ) + 2 \gamma_e \right ) \right ] . \end{aligned}\ ] ] here , we would like to make the following remark about the parameter @xmath85 . in order to achieve a factorization of large and small scales in the ope , all infrared logarithms should be removed from coefficient functions and absorbed in the matrix elements of operators . in our case , this means that the @xmath86 must be included in the condensates of different operator or distribution amplitude . a more detailed discussion on this point can be found in @xcite . for this reason , we will choose the scale parameter @xmath85 as a factorization scale , i.e. , @xmath87 . in order to calculate the contributions of the photon emission from large distances , the matrix elements of nonlocal operators @xmath88 between the photon and vacuum states are needed , @xmath89 . these matrix elements are determined in terms of the photon distribution amplitudes ( da s ) as follows @xcite . @xmath90 \int_0 ^ 1 du e^{i \bar u q x } h_\gamma(u ) \nonumber \\ & & \langle \gamma(q ) \vert \bar q(x ) \gamma_\mu q(0 ) \vert 0 \rangle = e_q f_{3 \gamma } \left(\varepsilon_\mu - q_\mu \frac{\varepsilon x}{q x } \right ) \int_0 ^ 1 du e^{i \bar u q x } \psi^v(u ) \nonumber \\ & & \langle \gamma(q ) \vert \bar q(x ) \gamma_\mu \gamma_5 q(0 ) \vert 0 \rangle = - \frac{1}{4 } e_q f_{3 \gamma } \epsilon_{\mu \nu \alpha \beta } \varepsilon^\nu q^\alpha x^\beta \int_0 ^ 1 du e^{i \bar u q x } \psi^a(u ) \nonumber \\ & & \langle \gamma(q ) | \bar q(x ) g_s g_{\mu \nu } ( v x ) q(0 ) \vert 0 \rangle = -i e_q \qq \left(\varepsilon_\mu q_\nu - \varepsilon_\nu q_\mu \right ) \int { \cal d}\alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) q x } { \cal s}(\alpha_i ) \nonumber \\ & & \langle \gamma(q ) | \bar q(x ) g_s \tilde g_{\mu \nu } i \gamma_5 ( v x ) q(0 ) \vert 0 \rangle = -i e_q \qq \left(\varepsilon_\mu q_\nu - \varepsilon_\nu q_\mu \right ) \int { \cal d}\alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) q x } \tilde { \cal s}(\alpha_i ) \nonumber \\ & & \langle \gamma(q ) \vert \bar q(x ) g_s \tilde g_{\mu \nu}(v x ) \gamma_\alpha \gamma_5 q(0 ) \vert 0 \rangle = e_q f_{3 \gamma } q_\alpha ( \varepsilon_\mu q_\nu - \varepsilon_\nu q_\mu ) \int { \cal d}\alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) q x } { \cal a}(\alpha_i ) \nonumber \\ & & \langle \gamma(q ) \vert \bar q(x ) g_s g_{\mu \nu}(v x ) i \gamma_\alpha q(0 ) \vert 0 \rangle = e_q f_{3 \gamma } q_\alpha ( \varepsilon_\mu q_\nu - \varepsilon_\nu q_\mu ) \int { \cal d}\alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) q x } { \cal v}(\alpha_i ) \nonumber \\ & & \langle \gamma(q ) \vert \bar q(x ) \sigma_{\alpha \beta } g_s g_{\mu \nu}(v x ) q(0 ) \vert 0 \rangle = e_q \qq \left\ { \left[\left(\varepsilon_\mu - q_\mu \frac{\varepsilon x}{q x}\right)\left(g_{\alpha \nu } - \frac{1}{qx } ( q_\alpha x_\nu + q_\nu x_\alpha)\right ) \right . q_\beta \nonumber \\ & & - \left(\varepsilon_\mu - q_\mu \frac{\varepsilon x}{q x}\right)\left(g_{\beta \nu } - \frac{1}{qx } ( q_\beta x_\nu + q_\nu x_\beta)\right ) q_\alpha \nonumber \\ & & - \left(\varepsilon_\nu - q_\nu \frac{\varepsilon x}{q x}\right)\left(g_{\alpha \mu } - \frac{1}{qx } ( q_\alpha x_\mu + q_\mu x_\alpha)\right ) q_\beta \nonumber \\ & & + \left . \left(\varepsilon_\nu - q_\nu \frac{\varepsilon x}{q.x}\right)\left ( g_{\beta \mu } - \frac{1}{qx } ( q_\beta x_\mu + q_\mu x_\beta)\right ) q_\alpha \right ] \int { \cal d}\alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) qx } { \cal t}_1(\alpha_i ) \nonumber \\ & & + \left[\left(\varepsilon_\alpha - q_\alpha \frac{\varepsilon x}{qx}\right ) \left(g_{\mu \beta } - \frac{1}{qx}(q_\mu x_\beta + q_\beta x_\mu)\right ) \right . q_\nu \nonumber \\ & & - \left(\varepsilon_\alpha - q_\alpha \frac{\varepsilon x}{qx}\right ) \left(g_{\nu \beta } - \frac{1}{qx}(q_\nu x_\beta + q_\beta x_\nu)\right ) q_\mu \nonumber \\ & & - \left(\varepsilon_\beta - q_\beta \frac{\varepsilon x}{qx}\right ) \left(g_{\mu \alpha } - \frac{1}{qx}(q_\mu x_\alpha + q_\alpha x_\mu)\right ) q_\nu \nonumber \\ & & + \left . \left(\varepsilon_\beta - q_\beta \frac{\varepsilon x}{qx}\right ) \left(g_{\nu \alpha } - \frac{1}{qx}(q_\nu x_\alpha + q_\alpha x_\nu ) \right ) q_\mu \right ] \int { \cal d } \alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) qx } { \cal t}_2(\alpha_i ) \nonumber \\ & & + \frac{1}{qx } ( q_\mu x_\nu - q_\nu x_\mu ) ( \varepsilon_\alpha q_\beta - \varepsilon_\beta q_\alpha ) \int { \cal d } \alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) qx } { \cal t}_3(\alpha_i ) \nonumber \\ & & + \left . ( q_\alpha x_\beta - q_\beta x_\alpha ) ( \varepsilon_\mu q_\nu - \varepsilon_\nu q_\mu ) \int { \cal d } \alpha_i e^{i ( \alpha_{\bar q } + v \alpha_g ) qx } { \cal t}_4(\alpha_i ) \right\},\end{aligned}\ ] ] where @xmath91 is the magnetic susceptibility of the quarks , @xmath92 is the leading twist 2 , @xmath93 , @xmath94 , @xmath95 and @xmath96 are the twist 3 and @xmath97 , @xmath98 , @xmath99 ( @xmath100 ) are the twist 4 photon da s , respectively . the photon da s is calculated in @xcite and for completeness their explicit expressions are presented in the numerical analysis section . the measure @xmath101 is defined as @xmath102 note that , in definition of photon da s with leading twist there are terms proportional to the strange quark mass . in our calculations , we neglect these terms . using the expressions of the light and heavy full propagators and the photon da s and separating the coefficient of the structure @xmath103 , the expression of the correlation function from qcd side is obtained . separating the coefficient of the same structure from phenomenological part and equating these representations of the correlator , sum rules for the magnetic moments of the @xmath0 heavy baryons is obtained . in order to suppress the contribution of higher states and continuum , borel transformation with respect to the variables @xmath104 and @xmath105 is applied . before presenting the sum rules for the magnetic moments , few words about the relations between the correlation functions are in order . note that , the coefficient of any structure in the correlation function can be written in the form @xmath106 where @xmath107 corresponds to the emission of the photon from the quark @xmath108 , @xmath109 to emission from @xmath110 and @xmath111 to emission from the heavy quark @xmath112 . note that the functions @xmath113 ( @xmath114 ) do not depend on the quark charges in the limit where we neglect electromagnetic corrections . from the explicit form of the interpolating current it follows that it is symmetric under the exchange of the light quarks , i.e. @xmath115 , and for this reason @xmath116 . due to the symmetries between the interpolating currents , any interpolating current can be obtained from the interpolating current for @xmath117 . this also leads to relations between their correlation functions . in terms of the defined @xmath118 and @xmath119 , all the correlation functions can be written as : @xmath120 and the relations for the charmed baryons can simply be obtained from eq . ( [ relations ] ) , by the replacement @xmath121 . note that in the @xmath122 limit , the correlation functions can be expressed as : @xmath123 where , for simplicity , we have used the short hand notation @xmath124 ( @xmath114 ) since in the assumed @xmath122 limit , all the light quarks have the same mass , condensates , etc . from eq . ( [ relations2 ] ) follows the well known @xmath122 relations between the correlation functions , and hence the magnetic moments ( part of these relations were derived in @xcite ) : @xmath125 and the corresponding expressions for the charmed baryons . any deviation from these relations in the magnetic moments is a sign of @xmath122 violation . note that the first relation in eq . ( [ relations3 ] ) is a direct consequence of the isospin subgroup of @xmath122 . since in this work , we set @xmath126 and @xmath127 , i.e. we assume isospin symmetry , our results satisfy this relation exactly . and hence , in the following , we will not show numerical results form the magnetic moments of @xmath117 . once the explicit forms of the functions @xmath128(@xmath114 ) are known , the sum rules for the magnetic moments can be obtained using eq . ( [ relations ] ) and @xmath129 the functions @xmath128 can be written as : @xmath130 where @xmath131 \nonumber\\&+&m_{0}^{2}<q_{2}q_{2}>\left[\{-\frac{m_q^2}{16\pi^{2}m^{4 } } -\frac{19}{288\pi^{2}m^{2}}\}m_{q_{1}}ln(\frac{s - m_q^{2}}{\lambda^{2}})\right . \nonumber\\&+&\frac{1}{288\pi^{2}m_q^2}\{8(1+\psi_{11})m_q-[-9 + 4\psi_{02}+5\psi_{10}+8\psi_{12 } -9\psi_{21}+8\psi_{22}\nonumber\\&+&4\gamma_{e}(23+\psi_{02}+2\psi_{12 } + \psi_{22})+4\psi_{32}]m_{q_{1}}-6(-1 + 2\psi_{12}+\psi_{22})m_{q_{2 } } \nonumber\\&+&6(16 + 2\psi_{02}-18\psi_{03}-50\psi_{10}+9\psi_{21}+93\psi_{23}-4\psi_{32}+120\psi_{33}+45\psi_{43 } ) \nonumber\\&\times&m_{q_{1}}ln(\frac{s - m_q^{2}}{\lambda^{2}})\}\left.\vphantom{\int_0^{x_2}}\right ] \nonumber\\&+&<q_{2}q_{2}>\frac{1}{16\pi^{2}}\left[\vphantom{\int_0^{x_2}}4(\psi_{10}-\psi_{21})m_{q}-\{-1 + 4\psi_{01}-\psi_{02}+2\psi_{11}\right . + \psi_{22}\nonumber\\&+&2\gamma_{e}(1+\psi_{02}+2\psi_{12}+\psi_{22})\}m_{q_{1}}-(1+\psi_{02})m_{q_{2 } } + 2m_{q_{1}}\{ln(\frac{\lambda^{2}}{m_q^{2}})\nonumber\\&+&(2-\psi_{02}-2\psi_{10 } + 3\psi_{22}+2\psi_{32})ln(\frac{s - m_q^{2}}{\lambda^{2 } } ) + ln(\frac{m_q^{2}(s - m_q^{2})}{\lambda^{2}s})\}\left.\vphantom{\int_0^{x_2}}\right ] \nonumber\\&+&<q_{1}q_{1}>\frac{1}{96\pi^{2}}\left[\vphantom{\int_0^{x_2}}\right . -3\{(\psi_{10}-\psi_{21})m_q+m_{q_{2}}\}\mathbb{a}(u_{0})\nonumber\\&+&2\left(\vphantom{\int_0^{x_2}}\right . \{-1 + 4\psi_{01}-\psi_{02}+2\psi_{11}+\psi_{22}+2\gamma_{e}(1+\psi_{02}+2\psi_{12}+\psi_{22})\}\nonumber\\&\times&m_{q_{2 } } ( \eta_{1}-3\eta_{2})+2(\psi_{21}-\psi_{10})\{m_q(\eta_{1}+3\eta_{2})+2m_{q_{2}}\eta_{1}-(m_q+m_{q_{2}})(\eta_{4}-\eta_{3})\}\nonumber\\&-&2(\eta_{1}-3\eta_{2})\left\{\vphantom{\int_0^{x_2}}\right.m_qln(\frac{s}{m_q^{2}})+m_{q_{2}}\{ln(\frac{\lambda^{2}}{m_q^{2}})+(2-\psi_{02}-2\psi_{10}+3\psi_{22}+2\psi_{32})\nonumber\\&\times&ln(\frac{s - m_q^{2}}{\lambda^{2}})+ln(\frac{m_q^{2}(s - m_q^{2})}{\lambda^{2}s})\}\left.\vphantom{\int_0^{x_2}}\right\}\nonumber\\&+&3m_q^{2}\chi_{i}\{2m_q\psi_{10}-(m_q - m_{q_{2 } } ) ( \psi_{20}-\psi_{31})+2m_qln(\frac{m_q^{2}}{s})\}\left.\varphi_{\gamma}(u_{0})\vphantom{\int_0^{x_2}}\right)\left.\vphantom{\int_0^{x_2}}\right]+\nonumber\\&+&\frac{m_q^{3}}{256\pi^{4}}\left[\vphantom{\int_0^{x_2}}\right.m_q(12\psi_{10 } -6\psi_{20}+4\psi_{30}-5\psi_{42}-4\psi_{52}+m_{q_{1}}(18 - 18\psi_{01}+114\psi_{10}\nonumber\\&-&66\psi_{20}+12\psi_{31 } ) -24m_{q_{2}}(2\psi_{10}-\psi_{20}+\psi_{31})-12\{6+\psi_{10}+12ln(\frac{\lambda^{2}}{m_q^{2}})\}\nonumber\\&\times&m_{q_{1}}ln(\frac{\lambda^{2}}{m_q^{2}})-12m_qln(\frac{s}{m_q^{2}})+12\left(\vphantom{\int_0^{x_2}}\right . \{2+\psi_{01}-ln(\frac{\lambda^{2}}{s})\}m_{q_{1}}ln(\frac{\lambda^{2}}{s})\nonumber\\&+&ln(\frac{m_q^{2}}{s } ) \{2(1 - 4ln(\frac{s - m_q^{2}}{\lambda^{2}}))m_{q_{1}}-4m_{q_{2}}\}+m_{q_{1}}\{(1+\psi_{01}-7\psi_{10}+3\psi_{20 } \nonumber\\&+&2ln(\frac{s}{m_q^{2}}))ln(\frac{s - m_q^{2}}{\lambda^{2}})-(3+\psi_{10})ln(\frac{m_q^{2}(s - m_q^{2})}{\lambda^{2}s})+\psi_{01}ln(\frac{s(s - m_q^{2})}{\lambda^{2}m_q^{2}})\}\left.\vphantom{\int_0^{x_2}}\right)\nonumber\\&- & 24\{4li_{2}(1-\frac{m_q^{2}}{s})-li_{2}(1-\frac{s}{m_q^{2}})\}m_{q_{1}}\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+ & \frac{f_{3\gamma}m_q^{2}}{48\pi^{2}}\left[\vphantom{\int_0^{x_2}}-2\eta'\{\psi_{10}+ln(\frac{m_q^{2}}{s})\}+(3\psi_{31 } + 4\psi_{32}+3\psi_{42})\psi^a(u_{0})\vphantom{\int_0^{x_2}}\right],\end{aligned}\ ] ] @xmath132\nonumber\\&+ & m_{0}^{2}<q_{2}q_{2}>\left[\vphantom{\int_0^{x_2}}\right.-\frac{5(1+\psi_{11})}{288\pi^{2}m_q } + \frac{m_{q_{1}}}{18\pi^{2}m^{2}}ln(\frac{s - m_q^{2}}{\lambda^{2}})-\frac{m_{q_{1}}}{36\pi^{2}m_q^{2}}\{\psi_{02}-\psi_{10 } \nonumber\\&+&2(\psi_{12}+\psi_{22})+\gamma_{e}(-4+\psi_{02}+2\psi_{12}+\psi_{22})+\psi_{32}+3(1-\psi_{02}-2\psi_{10}\nonumber\\&+ & 3\psi_{22}+2\psi_{32})ln(\frac{s - m_q^{2}}{\lambda^{2}})\}\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+ & < q_{1}q_{1}>\frac{1}{16\pi^{2}}\left[\vphantom{\int_0^{x_2}}\right.(\psi_{02}-1)m_{q_{1}}-2(\psi_{10}-\psi_{21 } ) ( m_q-2m_{q_{2}})+2m_qln(\frac{s}{m_q^{2}})\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+ & < q_{2}q_{2}>\frac{1}{16\pi^{2}}\left[\vphantom{\int_0^{x_2}}\right.(\psi_{02}-1)m_{q_{2}}-2(\psi_{10}-\psi_{21 } ) ( m_q-2m_{q_{1}})+2m_qln(\frac{s}{m_q^{2}})\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+&\frac{m_q^{3}}{768\pi^{4 } } \left[\vphantom{\int_0^{x_2}}\right.(-4\psi_{30}+5\psi_{42}+4\psi_{52})m_q+18\psi_{31}(m_{q_{1}}+m_{q_{2 } } ) \nonumber\\&-&12\psi_{10}\{7m_q-9(m_{q_{1}}+m_{q_{2}})\}-6\psi_{20}\{m_q+3(m_{q_{1}}+m_{q_{2}})\}\nonumber\\&-&24\{2(1+\psi_{10})m_q-3(m_{q_{1}}+m_{q_{2}})\}ln(\frac{m_q^{2}}{s})\nonumber\\&+&36\{m_q-(1+\psi_{10})(m_{q_{1}}+m_{q_{2}})\}ln(\frac{s}{m_q^{2}})\left.\vphantom{\int_0^{x_2}}\right],\end{aligned}\ ] ] @xmath133 \nonumber\\&+&<q_{1}q_{1}><q_{2}q_{2}>\left[-\frac{m_q^3m_{q_{2}}\mathbb{a}(u_{0})}{48m^{4 } } + \frac{m_q}{72m^{2}}\{2m_{q_{2}}(-\eta_{1}+3\eta_{2})+3m_q\mathbb{a}(u_{0})\}\right.\nonumber\\&+&\frac{1}{72}\{2(4\eta_{1}+2\eta_{3}-4\eta_{4})+3\mathbb{a}(u_{0})+6m_qm_{q_{2}}\chi_{i}\varphi_{\gamma}(u_{0})\}\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+&m_{0}^{2}<q_{2}q_{2}>\left[\frac{\gamma_{e}m_{q_{1}}m^{2}}{3\pi^{2}m_q^{2 } } + \frac{m_q^{2}}{96\pi^{2}m^{2}}\{2m_{q_{2}}-3m_{q_{1}}ln(\frac{\lambda^{2}}{m_q^{2}})\}\right.\nonumber\\&-&\frac{1}{288\pi^{2}}\{18m_q + 9m_{q_{1}}+14m_{q_{1}}(2\gamma_{e}+ln(\frac{\lambda^{2}}{m_q^{2}}))+4m_{q_{2}}\ } \nonumber\\&+&\frac{f_{3\gamma}m_q^{2}m_{q_{2}}(\eta'+4\psi^a(u_{0}))}{216m^4}\left.\vphantom{\int_0^{x_2}}\right]\nonumber\\&+&<q_{2}q_{2}>\left[\frac{\gamma_{e}m^2m_{q_{1}}}{4\pi^{2}}-\frac{1}{36}f_{3\gamma}m_{q_{2}}(\eta'+3\psi^a(u_{0}))\right]\nonumber\\&-&<q_{1}q_{1}>\left[\frac{\gamma_{e}m^2m_{q_{2}}}{12\pi^{2}}(\eta_{1}-3\eta_{2})\right],\end{aligned}\ ] ] @xmath134\nonumber\\ & + & < q_{1}q_{1}><q_{2}q_{2}>\left[-\frac{1}{3}+\frac{m_q(m_{q_{1}}+m_{q_{2}})}{12m^{2}}\vphantom{\int_0^{x_2}}\right]\nonumber\\&+&m_{0}^{2}<q_{1}q_{1}>\left[-\frac{\gamma_{e}m_{q_{2}}m^{2}}{12\pi^{2}m_q^{2 } } + \frac{1}{144\pi^{2}}\{(-9 + 8\gamma_{e}+4ln(\frac{\lambda^{2}}{m_q^{2}}))m_{q_{2}}+m_{q_{1}}\ } \vphantom{\int_0^{x_2}}\right]\nonumber\\&+&m_{0}^{2}<q_{2}q_{2}>\left[-\frac{\gamma_{e}m_{q_{1}}m^{2}}{12\pi^{2}m_q^{2 } } + \frac{1}{144\pi^{2}}\{(-9 + 8\gamma_{e}+4ln(\frac{\lambda^{2}}{m_q^{2}}))m_{q_{1}}+m_{q_{2}}\ } \vphantom{\int_0^{x_2}}\right],\nonumber\\\end{aligned}\ ] ] where @xmath135 is the dilogarithm function . the functions also entering eqs . ( [ rho1]-[gamma2 ] ) are given as @xmath136 and @xmath137 , @xmath138 , @xmath139 and @xmath140 are the photon distribution amplitudes . note that in the above equations , the borel parameter @xmath141 is defined as @xmath142 and @xmath143 . since the masses of the initial and final baryons are the same , we will set @xmath144 and @xmath145 . the contributions of the terms @xmath146 are also calculated , but their numerical values are very small and therefore for customary in the expressions these terms are omitted . for calculation of the magnetic moments of the considered baryons , their residues @xmath49 as well as their masses are needed ( see eq . ( [ magneticmoment1 ] ) ) . note that many of the considered baryons are not discovered yet in the experiments . the residue is determined from two point sum rules . for the interpolating current given in eq . ( [ currentguy ] ) , we obtain the following result for @xmath147 : @xmath148,\end{aligned}\ ] ] where @xmath149\right.\nonumber\\&-&<q_{1}q_{1}>\left[\frac{2(\psi_{02}+2\psi_{10}-2\psi_{21}-1)m_{q}+(\psi_{02}-1)(3m_{q_{1}}-2m_{q_{2}})}{32\pi^{2}}\right]\nonumber\\&-&\frac{m_{q}^{3}}{512\pi^{4}}\left[-8(3\psi_{31 } + 2\psi_{32})m_{q_{1}}+3\{(2\psi_{30}-4\psi_{41}+\frac{5}{2}\psi_{42}+2\psi_{52})m_{q}-4\psi_{42}m_{q_{1}}\}\right.\nonumber\\&+&\left.\left.6(2\psi_{10}-\psi_{20})(\frac{3}{2}m_{q}-2m_{q_{1}})-6(4m_{q_{1}}-3m_{q})ln(\frac{m_{q}^{2}}{s})\right]\right\}\nonumber\\&+&e^{-m_{q}^{2}/m^2}\left\{m_{0}^{2}<q_{1}q_{1}><q_{2}q_{2}>\left[\frac{-5m_{q}^{3}m_{q_{1}}}{144m^{6}}+\frac{m_{q}(m_{q}+5m_{q_{1}})}{24m^{4}}-\frac{5}{24m^{2}}\right]\right.\nonumber\\&+&<q_{1}q_{1}><q_{2}q_{2}>\left[\frac{1}{12}-\frac{m_{q}m_{q_{1}}}{12m^{2}}\right]+m_{0}^{2}<q_{1}q_{1}>\left[\frac{6m_{q_{2}}-7m_{q_{1}}}{192\pi^{2}}\right]\left.\vphantom{\int_0^{x_2}}\right\}.\end{aligned}\ ] ] the masses of the considered baryons can be determined from the sum rules . for this aim , one can get the derivative from both side of eq . ( [ resediueguy1 ] ) with respect to @xmath150 and divide the obtained result to the eq . ( [ resediueguy1 ] ) , i.e. , @xmath151}{\left[\vphantom{\int_0^{x_2}}\pi'+\pi'(q_{1}\longleftrightarrow q_{2})\right]}.\end{aligned}\ ] ] in this section , we perform numerical analysis for the mass and magnetic moments of the heavy flavored baryons . firstly , we present the input parameters used in the analysis of the sum rules : @xmath152 , @xmath153 , @xmath154 @xcite , @xmath155 and @xmath156 @xcite . the value of the magnetic susceptibility was obtained in various papers as @xmath157 @xcite , @xmath158 @xcite and @xmath159@xcite . before proceeding to the results for the magnetic moments , we calculate the masses of heavy flavored baryons predicted from mass sum rule . obviously , the masses should not depend on the borel mass parameter @xmath141 in a complete theory . however , in sum rules method the operator product expansion ( ope ) is truncated and as a result the dependency of the predictions of physical quantities on the auxiliary parameter @xmath141 appears . for this reason one should look for a region of @xmath141 such that the predictions for the physical quantities do not vary with respect to the borel mass parameter . this region is the so called the working region and within this region the truncation is reasonable and meaningful . the upper limit of @xmath141 is determined from condition that the continuum and higher states contributions should be small than the total dispersion integral . the lower limit is determined by demanding that in the truncated ope the condensate term with highest dimension remains small than sum of all terms , i.e. , convergence of ope should be under control . the aforementioned conditions for bottom ( charmed ) baryons are satisfied when @xmath141 varies in the interval @xmath160 ( @xmath161 ) . in figs . 1 - 6 , we presented the dependence of the mass of the heavy flavored baryons on @xmath141 . from these figures , we see very good stability with respect to @xmath141 . the sum rule predictions of the mass of the heavy flavored baryons are presented in table 2 in comparison with some theoretical predictions and experimental results . note that the masses of the heavy flavored baryons are calculated in the framework of heavy quark effective theory ( hqet ) using the qcd sum rules method in @xcite . @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 + this work&@xmath167&@xmath168 & @xmath169&@xmath170&@xmath171&@xmath172 + @xcite&@xmath173&@xmath174 & @xmath175&@xmath176&@xmath177&@xmath178 + @xcite&6.088 & 2.768 & 5.834&2.518&5.963&2.654 + @xcite&-&-&5.805&2.495&-&- + @xcite&6.090&2.770&5.850&2.520&5.980&2.650 + @xcite&-&2.768&-&2.518&-&- + @xcite&6.083&2.760&5.840&-&5.966&- + @xcite&6.060&2.752&5.871&2.5388&5.959&2.680 + exp@xcite&-&2.770&5.836&2.520&-&2.645 + after determination of the mass as well as residue of the heavy flavored baryons our next task is the calculation of the numerical values of their magnetic moments . for this aim , from sum rules for the magnetic moments it follows that the photon da s are needed @xcite : @xmath179 , \nonumber \\ { \cal t}_1(\alpha_i ) & = & -120 ( 3 \zeta_2 + \zeta_2^+)(\alpha_{\bar q } - \alpha_q ) \alpha_{\bar q } \alpha_q \alpha_g , \nonumber \\ { \cal t}_2(\alpha_i ) & = & 30 \alpha_g^2 ( \alpha_{\bar q } - \alpha_q ) \left((\kappa - \kappa^+ ) + ( \zeta_1 - \zeta_1^+)(1 - 2\alpha_g ) + \zeta_2 ( 3 - 4 \alpha_g)\right ) , \nonumber \\ { \cal t}_3(\alpha_i ) & = & - 120 ( 3 \zeta_2 - \zeta_2^+)(\alpha_{\bar q } -\alpha_q ) \alpha_{\bar q } \alpha_q \alpha_g , \nonumber \\ { \cal t}_4(\alpha_i ) & = & 30 \alpha_g^2 ( \alpha_{\bar q } - \alpha_q ) \left((\kappa + \kappa^+ ) + ( \zeta_1 + \zeta_1^+)(1 - 2\alpha_g ) + \zeta_2 ( 3 - 4 \alpha_g)\right),\nonumber \\ { \cal s}(\alpha_i ) & = & 30\alpha_g^2\{(\kappa + \kappa^+)(1-\alpha_g)+(\zeta_1 + \zeta_1^+)(1 - \alpha_g)(1 - 2\alpha_g)\nonumber \\&+&\zeta_2 [ 3 ( \alpha_{\bar q } - \alpha_q)^2-\alpha_g(1 - \alpha_g)]\},\nonumber \\ \tilde { \cal s}(\alpha_i ) & = & -30\alpha_g^2\{(\kappa - \kappa^+)(1-\alpha_g)+(\zeta_1 - \zeta_1^+)(1 - \alpha_g)(1 - 2\alpha_g)\nonumber \\&+&\zeta_2 [ 3 ( \alpha_{\bar q } - \alpha_q)^2-\alpha_g(1 - \alpha_g)]\}.\end{aligned}\ ] ] the constants appearing in the wave functions are given as @xcite @xmath180 , @xmath181 , @xmath182 , @xmath183 , @xmath184 , @xmath185 , @xmath186 , @xmath187 and @xmath188 . the sum rules for magnetic moments also contain the auxiliary parameters : borel parameter @xmath141 and continuum threshold @xmath189 . similar to mass sum rules , the magnetic moments should also be independent of these parameters . in the general case , the working region of @xmath141 and @xmath189 for the mass and magnetic moments should be different . to find the working region for @xmath141 , we proceed as follows . the upper bound is obtained requiring that the contribution of higher states and continuum should be less than the ground state contribution . the lower bound of @xmath141 is determined from condition that the highest power of @xmath190 be less than say @xmath191 of the highest power of @xmath192 . these two conditions are both satisfied in the region @xmath193 and @xmath194 for baryons containing b and c - quark , respectively . the working region for the borel parameter for mass and magnetic moments practically coincide , but again we should stress that , this circumstance is accidental . in figs . 7 - 16 , we present the dependence of the magnetic moment of heavy flavored baryons on @xmath141 at two fixed values of continuum threshold @xmath189 . from these figures , we see that the magnetic moments weakly depend on @xmath189 . the maximal change of results is about @xmath195 with variation of @xmath189 . the magnetic moments also are practically insensitive to the variation of borel mass parameter when it varies in the working region . we should also stress that our results practically do nt change considering three values of @xmath91 which we presented at the beginning of this section . our final results on the magnetic moments of heavy flavored baryons are presented in table 3 . for comparison , the predictions of hyper central model @xcite are also presented . the quoted errors in table 3 are due to the uncertainties in @xmath196 , variation of @xmath189 and @xmath141 as well as errors in the determination of the input parameters . @xmath197 & @xmath198&@xmath199 + @xmath200&@xmath201 & @xmath202 + @xmath203&@xmath204&@xmath205 + @xmath206&@xmath207&@xmath208 + @xmath209&@xmath210&@xmath211 + @xmath212&@xmath213&@xmath214 + @xmath215&@xmath216&@xmath217 + @xmath218&@xmath219&@xmath220 + @xmath221&@xmath222&@xmath223 + @xmath224&@xmath207&@xmath225 + @xmath226&@xmath227&@xmath228 + @xmath229&@xmath230&@xmath231 + although the @xmath122 breaking effects have been taken into account through a nonzero @xmath232-quark mass and different strange quark condensate , we predict that @xmath122 symmetry violation in the magnetic moments is very small , except the relations @xmath233 and @xmath234 , where the @xmath122 symmetry violation is large . for the values of the magnetic moments , our results are consistent with the results of @xcite except for the @xmath197 , @xmath221 and especially for the @xmath215 , @xmath224 which we see a big discrepancy between two predictions . in summary , inspired by recent experimental discovery of the heavy and flavored baryons @xcite , the mass and magnetic moments of these baryons with @xmath0 are calculated within the qcd sum rules . our results on the masses are consistent with the experimental data as well as predictions of other approaches . our results on the masses of the @xmath235 , and @xmath236 can be tested in experiments which will be held in the near future . the predictions on the magnetic moments also can verified in the future experiments . two of the authors ( k. a. and a. o. ) , would like to thank tubitak , turkish scientific and research council , for their partial financial support through the project number 106t333 and also through the phd scholarship program ( k. a. ) . 99 t. aaltonen et . al , cdf collaboration , phys . * 99 * ( 2007 ) 202001 . v. abazov et . al , do collaboration , phys . rev . lett . * 99 * ( 2007 ) 052001 . t. aaltonen et . al , cdf collaboration , phys . * 99 * ( 2007 ) 052002 . b. aubert et . al , babar collaboration , phys . * 97 * ( 2006 ) 232001 . a. l. choudhury , v. joshi , phys . d * 13 * ( 1976 ) 3115 . d. b. lichtenberg , phys . d * 15 * ( 1977 ) 345 . r. j. johnson , m. shah - jahan , phys . d * 15 * ( 1977 ) 1400 . s. k. bose , l. p. singh , phys . d * 22 * ( 1980 ) 773 . l. y. glozman , d. o. riska , nucl . phys . a * 603 * ( 1996 ) 326 . b. julia - diaz , d. o. riska , nucl . phys . a * 739 * ( 2004 ) 69 . s. scholl , h. weigel , nucl . phys . a * 735 * ( 2004 ) 163 . y. s. oh , b. y. park , mod . phys . lett . a * 11 * ( 1996 ) 653 . a. faessler et . al , phys . d * 73 * ( 2006 ) 094013 . s. l. zhu , w. y. hwang , z. s. yang , phys . d * 56 * ( 1997 ) 7273 . t. m. aliev , a. ozpineci , m. savci , phys . d * 65 * ( 2002 ) 096004 . t. m. aliev , a. ozpineci , m. savci , phys . d * 65 * ( 2002 ) 056008 . t. m. aliev , k. azizi , a. ozpineci , phys . d * 77 * ( 2008 ) 114006 . p. colangelo , a. khodjamirian , in at the frontier of particle physics / handbook of qcd , edited by m. shifman ( world scientific , singapore , 2001 ) , vol . 3 , p. 1495 ; v. m. braun prep . hep - ph/9801122 ( 1998 ) ; i. i. balitsky , v. m. braun , a. v. kolesnichenko , nucl . b * 312 * ( 1989 ) 509 . i. i. balitsky , v. m. braun , nucl . phys . b * 311 * ( 1989 ) 541 . v. m. braun , i. e. filyanov , z. phys . c * 48 * ( 1990 ) 239 . k. g. chetyrkin , a. khodjamirian , a. a. pivovarov , phys . b * 661 * ( 2008)250 ; i. i. balitsky , v. m. braun , a. v. kolesnichenko , nucl . phys . b * 312 * ( 1989 ) 509 . p. ball , v. m. braun , n. kivel , nucl . b * 649 * ( 2003 ) 263 . m. c. banuls , i. scimemi , j. bernabeu , v. gimenez , a. pich , phys . d * 61 * ( 2000 ) 074007 . v. m. belyaev , b. l. ioffe , jetp * 56 * ( 1982 ) 493 . j. rohrwild , jhep * 0709 * ( 2007 ) 073 . v. m. belyaev , i. i. kogan , yad . fiz . * 40 * ( 1984 ) 1035 . x. liu , h. x. chen , y. r. liu , a. hosaka , s. l. zhu , arxiv : 07100123 ( hep - ph ) . d. ebert , r. n. faustov , v. o. galkin , phys . d * 72 * ( 2005 ) 034026 . s. capstick , n. isgur , phys . d * 34 * ( 1986 ) 2809 . r. ronaglia , d. b. lichtenberg , e. predazzi , phys . d * 52 * ( 1995 ) 1722 . m. j. savage , phys . b * 359 * ( 1995 ) 189 . e. jenkins , phys . d * 54 * ( 1996 ) 4515 ; ibid * 55 * ( 1997 ) r10 . n. mathur , r. lewis , r. m. woloshyn , phys . d * 66 * ( 2002 ) 014502 . c. amsler et al . ( particle data group ) , phys . b * 667 * ( 2008 ) 1 . b. patel , a. k. rai , p. c. vinodkumar , j. phys . g * 35 * ( 2008 ) 065001 .

tangle machines are a topologically inspired diagrammatic formalism to describe information flow in networks . this paper begins with an expository account of tangle machines motivated by the problem of describing ` covariance intersection ' fusion of gaussian estimators in networks . it then gives two examples in which tangle machines tell stories of adiabatic quantum computations , and discusses learning tangle machines from data .

1511.04919

the kantian explanation for ` the unreasonable effectiveness of mathematics in the natural sciences' is that the patterns which we seek out in nature are those patterns with which we are already familiar as innate categories of perception @xcite . when we encounter a new phenomenon , perhaps we first seek to describe this phenomenon in the category of perception most familiar to us . taken one step further , perhaps human choice of research topics is itself influenced by the scientific language of the day . the conventional graphical language in computer science is the language of flowcharts ( _ e.g. _ @xcite ) . a flowchart is a labeled directed graph in which edges represent transitions that can be concatenated . thus , an ordered sequence of edges @xmath0 in which the head of @xmath1 is the tail of @xmath2 for all @xmath3 represents a transition from the tail of @xmath4 to the head of @xmath5 . edges can be prepended or appended to such sequences , and the transition represented by prepending @xmath4 to @xmath6 is guaranteed to coincide with the transition represented by appending @xmath5 to @xmath7 . this property is called _ associativity_. thus , the language of flowcharts natively describes _ sequential _ processes , although this paradigm may be expanded via a _ concurrency symbol_. in previous work we proposed a different diagrammatic formalism called _ tangle machines_. a tangle machine looks visually similar to a coloured tangle diagram in low dimensional topology @xcite . it consists of arcs which we think of as analogous to computer registers and of crossings that we call _ interactions _ which we think of as analogous to logic gates . the interaction below involves three arcs called an _ agent _ , an _ input patient _ , and an _ output patient_. the interaction is labeled by a binary operation @xmath8 , and for labels @xmath9 for the input patient and @xmath10 for the agent , the label of the output patient is @xmath11 . @xmath12{\small $ x$ } \psfrag{y}[c]{\small $ y$ } \psfrag{x}[l]{\small $ x { \triangleright}y$ } \includegraphics[width=0.1\textwidth]{cross.eps}\ ] ] we call operation @xmath8 a _ fusion _ if it satisfies three properties : 1 . @xmath13 for all @xmath9 . thus , fusing information with itself should be the identity operation . an input label can uniquely be recovered from its corresponding output label together with the agent label . thus , fusion should ` not forget ' . 3 . fusing @xmath14 with @xmath15 should equal @xmath16 . thus , fusion should count the redundant appearance of @xmath17 only once towards the final result . the language of tangle machines natively describes networks of fusions ( strengthening the requirement that the fusion ` not forget ' strengthened to @xmath18 being a bijection ) . there is nothing inherently sequential about fusions , so tangle machines and flowcharts are quite different . what is the expressive power of tangle machines ? more prosaically , what tales can a tangle machine tell ? let us adopt an abstract view of a computation as a scheme to apply the information content of a computer program to the information content of input data in order to generate output data . some computations ( in this generalized sense ) are fusions . it turns out that we can describe any turing machine by a tangle machine under this paradigm . if we limit the number of interactions then we can describe any computation in the complexity class @xmath19 @xcite . parenthetically , we do nt know how much more we can describe by tangle machines for example , can tangle machines with bounded interactions describe any computation in complexity class @xmath20 ? what more can be described by tangle machines ? archetypal mathematical examples of fusions are convex combinations and group conjugations , but we suggest that an archetypal real - world example of a fusion operation is fusion of estimates of a random variable . if the random variable follows a bernoulli distribution then estimator fusion and convex combination essentially coincide ( an example of this is in @xcite ) . we consider fusion of normally distributed estimates . the kalman filter is a linear optimal measure to fuse gaussian estimates , but it requires knowledge of how the estimate errors depend on one another . this information is not contained in the estimates themselves and may sometimes be entirely unknown . _ covariance intersection _ is a method to fuse gaussian estimates whose error cross - correlations are unknown . section [ s : ci ] introduces tangle machines as a diagrammatic language for information flow in networks in which information is fused by covariance intersection . it contains a geometric interpretation of covariance intersection as a choice of a point on a geodesic on a _ statistical manifold _ whose points parameterize normal distributions . section [ s : ldt ] explains the connection of tangle machines to low dimensional topology . section [ ss : knots ] recalls a diagrammatic algebra from low dimensional topology . section [ ss : generalize ] generalizes and discusses quandles , after which section [ ss : info ] defined tangle machines and shows how they may describe information fusion networks . the above sections represent an expository account of previous published work of the authors . as discussed in @xcite , tangle machines ( with colours suppressed ) topologically arise as diagrams of embedded networks of intervals and spheres in standard euclidean @xmath21space . section [ s : quantum ] takes tangle machines into the realm of quantum information , presenting several adiabatic quantum computations ( aqc s ) with different performance features . section [ ss : timeentanglement ] presents a computation in which a product state evolves into an entangles state , and section [ ss:2sat ] presents a boolean @xmath22sat problem over four qubits . a further example may be found in ( * ? ? ? * section 5 ) . section [ s : utility ] discusses potential utility of the tangle machine description . inherited from low dimensional topology , tangle machines have topological invariants that are characteristic quantities giving high - level information about what is happening in the machine @xcite . in particular , there are meaningful notions of _ capacity _ and of _ complexity _ for tangle machines . in addition , tangle machines provide a flexible description , and this flexibility may perhaps be useful for fault tolerance @xcite . section [ s : learn ] concludes by discussing the problem of adapting existing causality - detection algorithms to detect tangle machines . we illustrate with an example in which we detect a single crossing in real - world google search data . [ r]covariance intersection [ r]coloured tangle diagrams [ r ] ' '' '' [ c]tangle machines [ l]tms for qi [ l]quantum info fusion a coloured tangle representing this paper . at each crossing , the output topic is an application of the input topic to the agent ( overcrossing ) topic , perhaps in a different sense for each crossing . this figure suggests how a coloured tangle can tell a tale and has no rigourous mathematical meaning.,title="fig:",scaledwidth=60.0% ] the father of the use of tangle diagrams as a diagrammatic language for computation is louis kauffman , who represented automata , nonstandard set theory , and lambda calculus with tangle diagrams @xcite . buliga has suggested to represent computations using a different calculus of coloured tangles @xcite . in another direction , a different diagrammatic calculus , originating in higher category theory , has been used in the theory of quantum information see _ e.g. _ @xcite . in all of these approaches an element in a topologically inspired diagrammatic algebra represents a computation and equivalent diagrams represent bisimilar computations . tangle machines are distinguished by the role of colours , by locality of orientations , and by binding together of operations in ` interactions ' , as discussed in @xcite . was partially supported by the helmsley charitable trust through the agricultural , biological and cognitive robotics initiative of ben - gurion university of the negev . information geometry applies the methods of differential geometry to probability theory @xcite . its main role so far has been to provide geometric coordinate - free elucidation to known facts . for example , the cramr rao inequality over @xmath23 which bounds variance from below by the reciprocal of fisher information becomes a statement about the statistical manifold changing more in the vicinity of points of high curvature than in the vicinity of points of low curvature . a _ statistical manifold _ corresponding to an @xmath24parameter family of distributions has a point for each possible value of the parameters . in particular , the statistical manifold for all normal distributions over @xmath23 is a surface whose coordinates are @xmath25 , the order and the variance of the distribution . a statistical manifold has a riemannian metric called the _ fisher information metric_. in the case of gaussian random variables the metric is given by the inverse of the covariance matrix ( over @xmath23 by the reciprocal of the variance as in the cramr rao bound ) . we will be considering the problem of estimate fusion in an information geometric setting . the estimates will be points on the statistical manifold . our main observation will be that that a fusion corresponds to a choice of a point on a geodesic . in this section we ask the question : how can we fuse estimates of an unknown quantity whose interdependence is unknown ? consider a pair of sensors which provide noisy observations about an unknown state vector @xmath26 . each sensor runs its own filtering algorithm to yield an estimate @xmath27 of @xmath26 and an estimate @xmath28 of the error covariance matrix of @xmath26 . @xmath29 { \stackrel{\textup{\tiny def}}{=}}e\left[(x-\hat{x}_i)(x-\hat{x}_i)^t \right ] \\ - e\left[x-\hat{x}_i\right ] e\left[x-\hat{x}_i\right]^t \quad \text{$i=0,1,2$}\enspace .\end{gathered}\ ] ] we wish to optimally integrate these estimates @xmath30 and @xmath31 . it is important that the fused estimate be _ consistent _ ( or _ conservative _ ) , meaning that : @xmath32 \enspace , \ ] ] _ i.e. _ that the matrix difference @xmath33 $ ] is positive semi - definite . we would like all estimators to be consistent because inconsistent estimators may diverge and cause errors . if the correlations between the estimate errors are known , a linear optimal fusion in the sense of minimum mean estimation error ( mmse ) is given by the _ kalman filter_. but such error cross - correlations are typically unknown and may be unmeasurable in real - world settings . ignoring error cross - correlations can cause the kalman filter estimates to be non - conservative and perhaps to diverge @xcite . the standard approach in applications is to increase the system noise artificially . but proper use of this heuristic requires substantial empirical analysis and compromises the integrity of the kalman filter framework @xcite . _ covariance intersection _ provides a method to fuse estimates whose error cross - correlations are unknown in a way that guarantees that the resulting estimate is consistent @xcite . the covariance intersection method to fuse estimators @xmath30 with @xmath31 requires a choice of _ weight _ @xmath34 . the bottleneck for practical covariance intersection is the computation of the optimal value for the weight @xmath35 with respect to some ( typically nonlinear ) cost function such as logdet or trace @xcite . in the present paper we treat @xmath35 as a formal parameter or as an unknown constant for each pair of estimates to be fused . we denote the covariance intersection of @xmath30 with @xmath31 with respect to @xmath34 as follows : @xmath36 we construct @xmath37 as follows : [ eq : ci ] @xmath38 @xmath39 the working principle of covariance intersection is that it fuses two conservative estimates into a third conservative estimate . the reason , for gaussian estimators , is that the _ covariance ellipse _ of the fused estimator includes the intersection of the covariance ellipse of the estimators being fused ( figure [ fig : cvinter ] ) . covariance ellipse _ of a covariance matrix @xmath40 is the locus of vectors @xmath41 such that @xmath42 where @xmath43 is some arbitrary ( but fixed ) constant . for gaussian estimators , @xmath44 represents the fisher information . for details see @xcite . [ c]@xmath45 [ c]@xmath46 [ l]@xmath47 the covariance ellipse of the fused estimator is a minimal ellipse that includes within it the intersection of the covariance ellipses of the estimates being fused.,title="fig:",scaledwidth=60.0% ] we prove the following proposition in the appendix . [ p : geodesic ] the @xmath48parameter family of gaussian covariance intersections @xmath49}$ ] parameterizes the geodesic from @xmath30 to @xmath31 . [ c]@xmath50 [ l]@xmath51 [ c]@xmath52 [ c]@xmath53 [ c]@xmath9 [ c]@xmath10 [ l]@xmath54 [ c]@xmath55 [ c]@xmath56 [ c]_interaction _ [ c]@xmath57 a basic information fusion operation ( left ) and its geometric representation ( right ) . here @xmath50 denotes a gaussian distribution with mean @xmath58 and covariance matrix @xmath40 , which represents a point in the manifold.,title="fig:",scaledwidth=90.0% ] covariance intersection has the following properties , which are evident from the geometric interpretation and can also be derived directly from : idempotence : : @xmath59 reversibility : : the function @xmath60 from the statistical manifold to itself is a bijection . thus , an estimate @xmath43 can uniquely be recovered from @xmath61 , @xmath62 , and @xmath35 . no double counting : : for any estimators @xmath43 , @xmath61 , and @xmath63 and for any weights @xmath64 , we have : @xmath65 thus , the redundant appearance of @xmath63 in @xmath66 and in @xmath67 counts only once towards the final result . this is the key property of covariance intersection that makes it insensitive to the dependence of different estimators . this section provides a new strand in our tale . thus , we shall put aside all that came before , and we shall begin afresh . the fundamental problem in knot theory is to distinguish knots , which are smooth embeddings @xmath68 of a directed circle into the @xmath69sphere , considered up to an equivalence relation called _ ambient isotopy_. identifying @xmath70 and choosing a projection of the knot onto a generic plane in @xmath71 disjoint from the knot , we may represent @xmath72 as a _ knot diagram _ such as one of those drawn in figure [ f : knot ] . _ reidemeister s theorem _ tells us that two knot diagrams represent ambient isotopic knots if and only if they are related by a finite sequence of the three _ reidemeister moves _ of figure [ f : reidemeistermoves ] . .32 [ c ] the unknot , the trefoil , and the figure - eight knot.,title="fig:",scaledwidth=67.0% ] .32 [ c ] [ c ] [ c ] the unknot , the trefoil , and the figure - eight knot.,title="fig:",scaledwidth=80.0% ] .32 [ c ] [ c ] [ c ] [ c ] the unknot , the trefoil , and the figure - eight knot.,title="fig:",scaledwidth=80.0% ] [ r]_r2 _ [ r]_r3 _ [ r]_r1 _ the reidemeister moves . orientations are arbitrary . to execute a reidemeister move , cut out a disc inside a tangle diagram containing one of the patterns above , and replace it with another disc containing the pattern on the other side of the reidemeister move.,title="fig:",scaledwidth=70.0% ] how are we to distinguish two non - ambient - isotopic knots ? for example , how are we to distinguish the _ trefoil knot _ from the _ unknot _ in figure [ f : knot ] ? the original idea of tietze @xcite , as explained later by fox @xcite , is to colour each arc in the knot diagram in one of three colours @xmath73 , represented numerically as @xmath74 , @xmath48 , and @xmath22 , subject to the constraints that at least two different colours are used , and if two adjacent colours to a crossing are @xmath75 and @xmath76 then the third adjacent colour is @xmath77 . @xmath12{\small $ x$ } \psfrag{y}[c]{\small $ y$ } \psfrag{x}[l]{\small $ x { \triangleright}y$ } \includegraphics[width=0.1\textwidth]{cross.eps}\ ] ] the trefoil can be coloured according to these rules ( figure [ f : colknot ] ) , but the unknot and the figure - eight knot can not . this property of being _ @xmath69colourable _ is preserved under reidemeister moves ( figure [ f : reidemeistermoves1 ] will later show a more general statement ) , and thus we have shown that the trefoil is knotted and is not ambient isotopic to the figure - eight knot . but how to show that the figure - eight knot is knotted ? take five colours @xmath74 , @xmath48 , @xmath22 , @xmath69 , and @xmath21 , and colour according to the same procedure . the figure - eight knot is @xmath78colourable , but the trefoil and the unknot are not . generalizing , we arrive at the idea of defining an algebraic structure by which a knot can be coloured . its elements are a set @xmath79 and it comes equipped with an operation @xmath80 satisfying the following axioms : idempotence : : @xmath81 automorphism : : the function : @xmath82 is an automorphism of @xmath83 for all @xmath84 . in particular it is a bijection of sets and : @xmath85 a structure @xmath86 is called a _ quandle _ ( _ e.g. _ @xcite ) . several examples will be given in section [ ss : info ] . .32 [ c]@xmath74 a trivial colouring of the unknot , a @xmath69colouring of the trefoil , and a @xmath78colouring of the figure - eight knot . the quandle operation is @xmath87.,title="fig:",scaledwidth=67.0% ] .32 [ c]@xmath74 [ c]@xmath48 [ c]@xmath22 a trivial colouring of the unknot , a @xmath69colouring of the trefoil , and a @xmath78colouring of the figure - eight knot . the quandle operation is @xmath87.,title="fig:",scaledwidth=80.0% ] .32 [ c]@xmath74 [ c]@xmath48 [ c]@xmath22 [ c]@xmath69 a trivial colouring of the unknot , a @xmath69colouring of the trefoil , and a @xmath78colouring of the figure - eight knot . the quandle operation is @xmath87.,title="fig:",scaledwidth=80.0% ] we did not use all of the properties of a knot when defining a quandle colouring . as far as colourings are concerned , several properties of a knot were irrelevant . let us relax these , with a view to motivating tangle machines in section [ ss : info ] . * we did not care about the number of connected components or whether they were closed . quandle colourings are defined for _ tangles _ , that are _ concatenations _ of crossings , _ i.e. _ shapes composed by scattering crossings in the plane and connecting up endpoints via disjoint line segments ( figure [ f : comb ] shows how a knot diagram can be constructed this way).two tangles are equivalent if they are related by reidemeister moves . see figure [ f : tangle ] . * we did not care about planarity . in terms of colourings , nothing is lost by allowing the connecting segments to intersect transversely . relaxing the planarity requirement gives rise to _ virtual tangles_. * the orientations of the arcs were only important to give directionality to the quandle operation to specify which the ` input underarc ' is , versus the ` output underarc ' . there is no reason for directions of overcrossing arcs to match up ; a global direction on an arc plays no role . restricting orientations to overcrossing arcs and making them local introduces ` disorientations ' into our virtual tangles ( _ e.g. _ @xcite ) a knot formed by concatenating crossings . orientations ignored.,scaledwidth=70.0% ] .41 two equivalent tangles.,title="fig : " ] .41 two equivalent tangles.,title="fig : " ] we will also want to consider quandles in a more general sense , in which @xmath79 comes equipped with not one , but an entire family @xmath88 of binary operations , subject to the following axioms : idempotence : : @xmath89 automorphism : : the function : @xmath82 is an automorphism of @xmath90 for all @xmath84 and for all @xmath91 . in particular each such function is a bijection of sets , and : @xmath92 such structures , perhaps subject to extra axioms which hold in our cases of interest , are variously called _ distributive @xmath93idempotent right quasigroups _ @xcite , _ @xmath94family of quandles _ @xcite , and _ multiquandles _ @xcite . we continue to call them quandles . we list some examples of quandles : colours might be elements of a group @xmath93 , and the operation might be conjugation : @xmath95 the pair @xmath96 is called a _ conjugation quandle_. if the group @xmath93 is a dihedral group and we colour by reflections , we recover the notion of an @xmath24colouring of a knot . _ e.g. _ @xcite . [ e : linearquandle ] colours might be elements of a real vector space @xmath79 and the operations might be convex combinations : @xmath97 the pair @xmath98 is called a _ linear quandle_. [ e : loglinear ] in the same setting as example [ e : linearquandle ] , consider the operations : @xmath99 the pair @xmath100 is called a _ loglinear quandle_. the attentive reader will have noticed that the properties of covariance intersection are similar to the quandle axioms . as we have seen , the set of points of a statistical manifold of gaussian distributions satisfies the quandle axioms with respect to covariance intersection ` choose a point on a geodesic ' operations . in this section , we define tangle machines and colour them by estimators . quandle operations will be covariance intersections . the basic building block of tangle machine is called an _ interaction _ ( figure [ f : kebab ] ) . it involves one strand coloured @xmath10 which we call a _ agent _ and strands coloured @xmath101 which we call _ patients_. we draw @xmath10 as a thick horizontal line , and we label line segments above @xmath10 by @xmath101 correspondingly . an interaction has a weight @xmath102 associated to it , and the strands below @xmath10 are labeled by @xmath103 correspondingly . we may orient the strand labeled @xmath10 or all strands labeled @xmath9 ( these two conventions are interchangeable ) . sometimes we may be sloppy and orient both , in which case the orientations of the patients should be ignored . .35 [ c]@xmath10 [ c]@xmath104 [ r]@xmath105 [ c]@xmath106 [ l]@xmath107 [ c]@xmath108 [ c]@xmath109 [ c]@xmath10 [ c]@xmath110 an interaction . we usually draw agents as thick lines.,title="fig : " ] .35 an interaction . we usually draw agents as thick lines.,title="fig : " ] interactions may be concatenated as shown in figure [ f : concat ] . when labels are estimators , the concatenation of interactions represents a pair of related choices of points on geodesics , as shown in figure [ fig : manifold1 ] . one reason that we allow multiple input patients for interactions is that the covariance intersection parameter @xmath35 is an estimated quantity . in general it will be different for different estimator fusions . when we know that it is the same , this information is important for us and we would like to record it . we typically represent agents with multiple agents by thickening the overarc . endpoints can be concatenated if they share the same colours ( colours are suppressed in the above figure ) . at the third stage , note that in our formalism , only agents are oriented , and no compatibility requirement is imposed . the concatenation line chosen is arbitrary , and in particular it may intersect other concatenation lines . ] [ c]@xmath50 [ l]@xmath51 [ c]@xmath111 [ c]@xmath17 [ c]@xmath110 [ l]@xmath112 [ c]@xmath113 [ c]@xmath9 [ c]@xmath10 [ c ] @xmath114 [ c]@xmath115 [ c]@xmath116 [ c]@xmath55 [ c]@xmath56 [ c]_tangle machine _ an information fusion network a tangle machine ( left ) and its geometric representation on a statistical manifold ( right).,title="fig:",scaledwidth=90.0% ] the lines used for concatenation do not matter , and the equivalence relation on diagrams generated by different choices of concatenating lines , and changes of local orientation , and adding or removing agents where ` nothing happens ' , is illustrated in figure [ f : local_moves_machines ] . the first two moves allow us to ignore orientations of agents when they do nt matter . reidemeister moves for machines are illustrated in figure [ f : reidemeistermoves1 ] , where the general form of r3 is obtained in figure [ f : r3-box ] where we first replace a box by a thickened line for convenience . we follow this convention for the remainder of the paper . .4 [ c]@xmath110 [ c]@xmath9 [ c ] [ c ] .5 [ c]@xmath10[c]@xmath9[c]@xmath9[r][ld]@xmath117[c ] [ c]@xmath110 + [ c]*(r3 ) * [ c]@xmath8 [ c]@xmath118 [ c]_i1 _ [ c]@xmath9 [ c]_vr1_[c]_vr2_[c]_vr3 _ [ c]_sv_[c]_i2_[c]_fm1_[c]_fm2_[c]_i3_[c]_st _ [ c]_st _ cosmetic moves for machines . where directions are not indicated , the meaning is that the move is valid for any directions , and the same for colourings . here , @xmath118 denotes the ( implicitly defined ) inverse operation to @xmath8.,title="fig:",scaledwidth=85.0% ] one notion of tangle machine computation is that it is a choice of arcs as _ input registers _ and another choice of arcs as _ output registers_. local moves may not involve these registers . if the colours of the input registers uniquely determine the colours of the output registers , then we consider this determination as a computation . we may think of interactions as being analogues to that of logic gates . but because interaction concatenation is non - sequential ( an interaction may even be concatenated to itself ) , this notion of computation is also non - sequential , and a tangle machine computation may be thought of as taking place ` all at once ' via an oracle . when interactions represent choice of point on a geodesic between the input patient and the agent on a statistical manifold , machines represent a network of these . quantum information is based on quantum probability theory , in which _ density matrices _ take the place of classical probability distributions . a set of density matrices may be given the structure of a statistical manifold . a tangent vector to this manifold is a _ hamiltonian matrix _ , which physically describes the dynamics of a quantum system . one research direction would be to colour tangle machines by density matrices and to fuse them using points on the geodesic between them on the statistical manifold . instead , we colour our machines by hamiltonians and we consider the linear quandle with operations indexed by a time parameter @xmath119 @xcite . adiabatic quantum computation ( aqc ) considers evolving hamiltonians of the form @xmath120 to perform quantum computation @xcite . analogously to in which the statistical moments of two estimators are fused , the interaction with output : @xmath121 may be viewed as a fusion of two statistical ensembles underlying two quantum systems . as before the parameter @xmath122 determines the contribution of each component to the new ensemble , @xmath123 . the quantum mechanical counterpart of independence of estimators is that the hamiltonians commute , @xmath124 = 0 $ ] , in which case the underlying density matrices satisfy : @xmath125 this is analogous to the formula for a geodesic on a statistical manifold in section [ s : ci ] . the idea of aqc is to evolve a hamiltonian @xmath126 representing a problem whose solution ( the ground state ) is easy , into a different and perhaps more complicated hamiltonian @xmath127 whose ground state is the solution to the computational task of interest . the computation initializes the system in its ground state , the ground state of @xmath126 , and then slowly evolves its hamiltonian to @xmath127 . this process is called _ quantum annealing_. by the adiabatic theorem , the system remains in its ground state throughout the evolution process , and the computation concludes at the ground state of @xmath127 , that is the sought - after solution . the _ computing time _ of an aqc is inversely proportional to the square of the minimal energy gap between the ground state and the rest of the spectrum , namely to the square of @xmath128 , where @xmath129 are the underlying energy eigenvalues of the hamiltonian . our objective is thus to maximize the minimal @xmath130 along the evolution path or to minimize the computation time . as shown in a previous paper @xcite and as we will again illustrate below , a tangle machine may be used to represent a piecewise - defined evolution path of a hamiltonian . it is well - known that nonlinear evolution paths speed up quantum computation @xcite . the topology of the machine itself is only of interest if we can measure its colours at intermediate times and at intermediate arcs . the technological barrier to implementation is the _ no - cloning theorem _ which tells us that we can not copy a state , and therefore we have lost it the moment we have measured it . from a future technological perspective we might envision a solution in which concatenation of interactions occurs by means of _ quantum teleportation _ of an entire hamiltonian from one interaction to the next . in such a case the tangle machine representation of an aqc would have all of the same features as a tangle machine representation of a classical information fusion network . without such technology , the tangle machines below merely tell us a tale of hamiltonian evolution . we define one additional piece of structure . choose a _ time deformation parameter _ @xmath131 and define the normalized time variables : @xmath132 with @xmath133 . a time parameter @xmath134 is associated with the _ @xmath75-th time - frame _ , and by convention all interactions within this ` time - frame ' ( drawn as a region in the plane ) are labeled @xmath134 . for small @xmath135 , the computations in the @xmath136-th time - frame nearly terminate when the computations in the @xmath137 frame begin . in the limit @xmath138 we recover sequential computation . our diagrams are drawn along a time axis showing the time - frames . interactions appearing in a particular time - frame , say the @xmath136-th one , are understood to be executed using the respective time parameter , _ i.e. _ @xmath139 . in what follows we use the shorthand @xmath140 for @xmath139 . in this section we describe a computation involving a single pair of qubits in which a product state evolves into an entangled state . as we shall see , the computation time is proportional to the amount of entanglement . we begin with a standard adiabatic computation @xmath141 where the initial and problem hamiltonians are given by : @xmath142 @xmath143 where @xmath144 are fixed parameters . the _ projectors _ are defined as : @xmath145 where @xmath26 is the pauli x matrix . the initial ground state is the product @xmath146 whereas the final ground state is a state ( written above in the computational basis z ) whose entanglement entropy is @xmath147 a maximally entangled state is obtained for @xmath148 . figure [ fig : time_ent ] graphs computation time against an increasing @xmath149 for two different values of @xmath43 . for two values of @xmath43 : ( blue ) @xmath150 , and ( red ) @xmath151.,scaledwidth=65.0% ] let s solve the same problem using a tangle machine and time deformation . we choose time variables as : @xmath152 let us solve the same computational problem in two different ways , represented by the tangle machines in figure [ fig : entang ] : @xmath153 { \triangleright}_0 h_1 \enspace ; \ ] ] @xmath154 { \triangleright}_0 \left[(1 \otimes p_x^1 ) { \triangleright}_1 h_1 \right]\enspace .\ ] ] the computation time is graphed with respect to @xmath149 in figure [ fig : time_ent1 ] . the performance of the standard aqc from figure [ fig : time_ent ] is also shown for comparison . here @xmath155 . [ r]@xmath156 [ r]@xmath157 [ c]@xmath127 [ c]@xmath158 [ c]@xmath159 [ c]@xmath160 [ c]@xmath161 [ c]@xmath162 [ c]@xmath163 [ c]@xmath164 [ c]@xmath165 for @xmath150 . showing time - deformed aqc ( with @xmath155 ) @xmath166 ( thick red ) , @xmath167 ( thick green ) , and standard aqc @xmath168 ( thin blue ) . the black line corresponds to no time deformation , i.e. @xmath169 { \triangleright}_0 h_1 $ ] , where @xmath170 is executed using a time variable @xmath171.,scaledwidth=65.0% ] the difference between the two time - deformed computations , @xmath167 and @xmath166 , is evident . figure [ fig : time_ent11 ] illustrates this further by showing the gap between the computation time of either schemes with respect to @xmath149 . we see that the discrepancy between the two computations reaches a peak for a maximally entangled final ground state . and @xmath155.,scaledwidth=65.0% ] the performance of the time - deformed algorithms for @xmath151 and @xmath172 are shown in figure [ fig : time_ent12 ] . this time the computation @xmath166 has a clear advantage over @xmath167 and the standard aqc as the amount of entanglement reaches its peak , around @xmath173 . for @xmath151 . showing time - deformed aqc ( with @xmath172 ) @xmath174 ( thick red ) , @xmath167 ( thick green ) , and standard aqc ( thin blue ) . the black line corresponds to no time deformation , _ i.e. _ @xmath169 { \triangleright}_0 h_1 $ ] , where @xmath170 is executed using a time parameter @xmath171.,scaledwidth=65.0% ] in this example we consider a boolean 2-sat problem over four variables ( qubits ) . we wish to find a satisfiable assignment for the following disjunctive normal form ( dnf ) expression @xmath175 there are four such assignments , namely , @xmath176 , @xmath177 , @xmath178 , @xmath179 . the final hamiltonian corresponding to this problem is composed of 2-local hamiltonians . to see this we define the projector @xmath180 where @xmath181 is the pauli z matrix . we see that : @xmath182 an encoding of the above 2-sat is then given by : @xmath183 \otimes \frac{1}{2 } \left [ p_z^2 + ( 1-p_z)^2 \right ] \\ = 1 - \frac{1}{4 } \left [ p_z^4 + p_z^2 ( 1-p_z)^2 + ( 1-p_z)^2p_z^2 + ( 1-p_z)^4 \right]\enspace , \end{gathered}\ ] ] where we have used shorthands @xmath184 and @xmath185 for @xmath186 and for @xmath187 respectively . the problem hamiltonian has a degenerate ground state whose basis vectors are satisfying assignments for our 2-sat . suppose there is an oracle with a probability of @xmath188 of yielding any one of the four satisfying assignment . this oracle is represented by a black - box hamiltonian whose ground state is a superposition of the four assignments . hence , @xmath189 the initial hamiltonian has a ground state @xmath190 , _ i.e. _ a superposition of @xmath191 possible assignments . it is given by @xmath192 we describe a number of computations which solve the 2-sat problem using the oracle s knowledge . these computations differ both in the way in which the oracle s `` answers '' are processed and in their evaluation of the sat clauses . the details are as follows . in the first round of experiments we assume that @xmath127 is given to us . in this case we employ two different time - deformed tangle machines , drawn in figure [ fig : qudit ] , which represent the following computations : [ eq : defaqc ] @xmath193 @xmath194 [ c]@xmath126 [ c]@xmath195 [ c]@xmath127 [ c]@xmath158 [ c]@xmath159 [ c]@xmath160 [ c]@xmath161 [ c]@xmath162 [ c]@xmath163 [ c]@xmath164 [ c]@xmath165 in another round of experiments we embed @xmath127 into the tangle machine . hence , instead of using the expression for @xmath127 in , we consider @xmath127 to be computed during the evolution of the system . in other words , we let : @xmath196 { \triangleright}_{\star } \left[(1-(1-p_z)^2p_z^2 ) { \triangleright}_{\star } ( 1-(1-p_z)^4)\right]\enspace , \ ] ] where @xmath197 . note that indeed @xmath198 substituting @xmath127 with its evolving version @xmath199 in yields [ eq : defaqc1 ] @xmath200 @xmath201 the diagrams describing the two computations are shown in figure [ fig : qudit1 ] . the dashed region represents the logic underlying @xmath199 . note that the number of inputs / outputs of this _ submachine _ corresponds to the number of conjunctive clauses in the original dnf . [ c]@xmath126 [ c]@xmath195 [ r]@xmath202 [ r]@xmath203 [ r]@xmath204 [ r]@xmath205 [ c]@xmath158 [ c]@xmath159 [ c]@xmath160 [ c]@xmath161 [ c]@xmath162 [ c]@xmath163 [ c]@xmath164 [ c]@xmath165 [ c]@xmath199 figure [ fig : qudit2 ] shows the energy gap over time for the various modes of computation in this example . here we use @xmath155 . ( thin blue ) , standard aqc using @xmath199 ( green ) , time - deformed aqc using @xmath127 ( purple ) , time - deformed aqc @xmath167 using @xmath199 ( red ) , time - deformed aqc @xmath166 using @xmath199 ( black).,scaledwidth=65.0% ] in this paper we have introduced tangle machines and they have told us tales of information flow in networks in both classical and quantum settings . in what way can this distributive structure be of use to information theorists and to computer scientists ? we present two advantages that such a topological description provides following @xcite and @xcite . knot diagrams are planar projections of knots in @xmath71 equipped with over / under information . knots in @xmath69space are ambient isotopic if and only if any two diagrams that represent them are related by a finite sequence of reidemeister moves . is there an analogous topological interpretation for tangle machines ? we have shown that tangle machines arise as planar projections of objects we call _ inca foams _ @xcite , that are necklaces of spheres glued together along disjoint discs , all jointly embedded in @xmath206 . two inca foams are ambient isotopic if and only if any two tangle machines which represent them are related by a finite sequence of reidemeister moves . this topological interpretation provides us tools to construct and to study _ topological invariants _ of tangle machines , that are characteristic quantities which coincide when tangle machines are equivalent . these will ideally have clear and compelling interpretations in terms of information . we mention two invariants . an inca foam and a tangle machine representing it.,scaledwidth=80.0% ] shannon capacity : : there is a well - defined notion of _ shannon capacity _ for a graph which measures the shannon capacity of an communications channel defined by the graph ( _ e.g. _ @xcite ) . this notion extends to tangle machines @xcite . + define an assignment of colours to arcs of a tangle machine @xmath57 to be _ confusable _ is there is one of these colours that is uniquely determined by the others . let @xmath207 denote the number of distinct assigments of @xmath136 colours to arcs of @xmath57 which @xmath57 admits . now define : + @xmath208{\mathrm{cap}_k(m)}\ ] ] complexity : : knots admit a binary operation called _ connect sum _ under which they form a commutative monoid : + @xmath209{trefoilcol}\quad \raisebox{20pt}{{\begin{minipage}{8pt}\includegraphics[width=8pt]{hash}\end{minipage } } } \quad \includegraphics[height=45pt]{trefoilcol}\quad \raisebox{20pt}{$=$ } \quad \includegraphics[height=45pt]{trefoil - connsum1}\ ] ] + 0.42 [ c ] [ c ] [ c ] [ c ] + 0.42 [ c ] [ c ] [ c ] [ c ] + coloured knots also form a commutative monoid under the connect - sum operation @xcite . although many related diagrammatic algebraic structures do not @xcite , tangle machines also form a commutative monoid under the connect - sum operation @xcite . more generally , we may define two submachines as being _ independent _ if they are connected along a set of arcs all of which share the same colour @xcite . the maximal number of nontrivial connect - summands is a topological invariant of tangle machines , and is a ( rough ) measure of tangle machine complexity . see figure [ f : complexity ] . a second advantage of the tangle machine formalism is that , unlike directed graphs , it is flexible . reidemeister moves give different machines representing different information flows which globally achieve the same task . in section [ s : quantum ] we will give examples in which the performance of one machine is significantly better than the performance of an equivalent machine although both achieve the same task . further examples are in @xcite . here , following @xcite , we shall present an idea for an application of the tangle machine description to fault tolerance of sensor networks which perform fusion using covariance intersection . we may imagine a realization of a tangle machine using with modular robots called ` cubelets ' as shown in figure [ f : cubelets ] . [ c]@xmath10 [ c]@xmath9 [ l]@xmath17 the r3 move realized using cubelets @xcite . both these networks have the same outputs and thus represent different realizations of essentially the same modular robot . the cubelets themselves do not move , and what changes is only the configuration of which cubelet sends its output to which other cubelet.,title="fig:",scaledwidth=90.0% ] three transmitting devices continuously stream data @xmath210 , @xmath211 , and @xmath212 . our task is to combine these data streams , eliminating redundancy ( _ e.g. _ because of the problem of double counting @xcite ) . two schemes to combine the data streams at a fixed time @xmath213 are represented by the left and the right hand side of the r3 move in figure [ f : reidemeistermoves1 ] . in the left machine , combine @xmath210 and @xmath211 with @xmath212 with parameter @xmath214 to obtain fused data streams @xmath215 and @xmath216 . assume that all estimators are gaussian and that the fusion is performed using covariance intersection . we read the fusion from bottom to top because the overcrossing arc is directed from left to right . if it were oriented from right to left , we would read from top to bottom and we would be filtering out the data stream @xmath212 from @xmath215 and from @xmath216 . on the right , @xmath215 is combined with @xmath216 using a weight @xmath217 . the left and the right machine describe equivalent data stream fusion schemes , which is visually indicated by the fact that the diagrams are related by sliding one overcrossing arc over another . indeed , the result of the data stream fusion in the right machine is @xmath218 , which is the same combined data stream as in the left machine because redundant double appearance of @xmath212 in @xmath215 and @xmath216 is eliminated by @xmath110 by assumption . however , there is an important difference between these two schemes on the left and right of the r3 move . imagine at some time @xmath219 that stream @xmath10 becomes faulty . in this case the left machine is superior because it contains the intermediate data stream @xmath215 which might be useful even when @xmath220 is junk . conversely , if @xmath17 becomes faulty at some time @xmath221 then the right machine would be preferred . the top overcrossing arc might slide back and forth at different times . thus the machines involved in the r3 move might be describing the underlying logic of a simple fault - tolerant data stream fusion network . a more general version of the same scheme is described in figure [ fig : topo ] . consider the network illustrated in the upper left corner of figure [ fig : topo ] . this network has a set of outputs outside the bounding disk ( not shown ) . some set of intermediate edges lies inside the subnetwork designated by @xmath222 and represented as an empty circle . in the course of network operation , erroneous streams of data cause one or more of the edges @xmath74,@xmath48 , and @xmath22 , to carry faulty pieces of information , _ e.g. _ biased , inconsistent and otherwise unreliable estimates . this is detected inside the network . to inhibit the influence of this contamination on edges within @xmath222 , the network performs reidemeister moves on itself , transforming itself into an equivalent network . this is achieved by ` sliding ' the faulty edges all the way over @xmath222 , by repeated application of the second and third reidemeister moves . in the resulting topology , the faulty edges have no effect on @xmath222 , and so local costs are improved . [ c]@xmath222 [ c]_fault _ @xmath223 [ c]_fault _ @xmath224 [ c]_fault _ @xmath225 [ c]_fault _ @xmath226 [ c]_fault _ @xmath227 [ c]_fault _ @xmath228 [ c]@xmath74 [ c]@xmath48 [ c]@xmath22 in the above example , things were as easy as sliding faulty edges forward , but this is only because the networks we considered are topologically uninteresting . in the general case an optimal configuration may be difficult to identify . let s now try a different way to connect tangle machines to the real world . given a data set we may ask whether there exists a tangle machine which adequately the flow of information such a set . can a tangle machine represent a data - generating process for real world - data ? in this section we identify an interaction within real world data . it may be possible to expand our method and to adapt existing causality - detection algorithms to construct tangle machines from data . one autoregressive system @xmath229 _ granger causes _ another autoregressive system @xmath230 if lagged values @xmath231 contain information about @xmath232 beyond the information contained in past values of @xmath233 itself for some fixed @xmath234 and for all @xmath235 . magnitude of granger causality is measured a quantity called the _ transfer entropy _ from @xmath26 to @xmath233 . a bayesian network is a directed acyclic graph whose nodes represent random variables and whose edges represent conditional dependencies . generalizing , nodes of bayesian networks may represent autoregressive systems and edges may represent transfer entropies from the tail to the head of each edge @xcite . thus the tail of an edge granger causes its head . there is a well - developed family of algorithms to learn transfer entropies for a given bayesian network from time - series data . we consider an interaction with a single input patient as representing a triangular bayesian network . the agent corresponds to the source in the network , and the two patients correspond to the remaining two nodes . thus , our idea is that the agent granger causes both patients . see figure [ fig : bayesian ] . our model is : @xmath236 where all variables are assumed to be normally distributed , @xmath102 terms are constant , and @xmath237 terms are noise . under the above assumptions , a short computation shows that the expression for the transfer entropies is maximized when the @xmath102 parameters are maximal @xcite . thus to maximize the transfer entropy for a given data set , our goal becomes to solve the following least - square problem to estimate the @xmath102 parameters : @xmath238 we then choose the configuration that corresponds to the greater value among each pair of parameters @xmath239 , @xmath240 , and @xmath241 . this process fits a single input - patient interaction to a triangular bayesian network . by extension , an interaction with agent @xmath233 , input patients @xmath242 , and output patients @xmath243 with @xmath244 corresponds to a bayesian network with nodes @xmath245 in which the @xmath246 arise as mixtures of autoregressive processes such that @xmath233 only acts on one of these and is independent of the others . the extension of our method to these and to more complicated bayesian networks will be considered in future work . [ c]@xmath233[c]@xmath26[c]@xmath181 [ r]@xmath26 [ c]@xmath233 [ l]@xmath181 we present an example in which we have learnt an interaction from real - life search data . around 2006 , the field of signal processing was revolutionized by the appearance of the concept of _ compressed sensing_. compressed sensing recovers signals using far fewer observations than required by the shannon nyquist sampling theorem . several people were involved , one of whom was emmanuel candes . this story may be described by an interaction of the form : e. candes _ causes _ a shift of state from _ nyquist sampling _ to _ compressed sensing_. we used _ google trends _ to generate time - series data of hits per year from 2000 to 2010 for the three search terms : e. candes , shannon nyquist sampling , and compressed sensing . we then used the filtering - based detection algorithm described in @xcite to recover the underlying bayesian network . as shown in figure [ fig : detection ] , the actual relationship between the time - series could be recovered up to a time order of shannon nyquist sampling and compressed sensing . there was an ambiguity between two possible interactions both of which identify e. candes as a cause of the paradigm shift . a refined and expanded version of this method will be needed to detect machines with more than one interaction . [ c]_time series data _ 99 [ 1]#1 urlstyle [ 1]doi:#1 abramsky s. & coecke , b. 2009 categorical quantum mechanics . in _ handbook of quantum logic and quantum structures _ , vol . 2 , 261323 . aharonov , d. & van dam , w. & kempe , j. & landau , z. & lloyd , s. & regev , o. 2008 adiabatic quantum computation is equivalent to standard quantum computation . _ siam review _ * 50*(4 ) , 755787 . amari , s. & nagaoka , h. 2007 methods of information geometry . translations of mathematical monographs ( vol . 191 ) , american mathematical soc . baez , j. & stay , m. 2011 physics , topology , logic and computation : a rosetta stone . in _ new structures for physics _ , lecture notes in phys . * 813 * , 95172 . bohl , m. & rynn , m. 2007 tools for structured and object - oriented design . prentice hall , upper saddle river , nj , usa . buliga , m. 2011 braided spaces with dilations and sub - riemannian symmetric spaces . in _ geometry . exploratory workshop on differential geometry and its applications _ , ( d. andrica & s. moroianu ed . ) , cluj - napoca 2135 . buliga , m. 2011 computing with space : a tangle formalism for chora and difference . preprint . buliga , m. & kauffman , l. 2014 chemlambda , universality and self - multiplication . in _ alife 14 : proceedings of the fourteenth international conference on the synthesis and simulation of living systems _ , 490497 . carmi , a. 2013 compressive system identification : sequential methods and entropy bounds . _ digit . signal process . _ * 23*(3 ) , 751770 . carmi , a.y . & moskovich , d. 2014 low dimensional topology for information fusion . in _ bict14 : proceedings of the 8th international conference on bio - inspired information and communications technologies _ , acm / eai , 251258 . carmi , a.y . & moskovich , d. 2014 tangle machines * ii * : invariants . carmi , a.y . & moskovich , d. 2015 computing with coloured tangles . _ symmetry _ * 7*(3 ) , 12891332 . carmi , a.y . & moskovich , d. 2015 tangle machines . a _ * 471 * , 20150111 . conference version : calude , c.s . , dinneen , m.j . ucnc 2015 , lncs , vol . 9252 , 277 289 . springer , heidelberg . carmi , a.y . & moskovich , d. 2015 inca foams . chang , k.c . , chong , c .- y . & mori , s. 2010 analytical and computational evaluation of scalable distributed fusion algorithms . _ ieee transactions on aerospace and electronic systems _ * 46*(4 ) , 20222034 . chen , l. & arambel , p.o . & mehra , r.k . 2002 estimation under unknown correlation : covariance intersection revisited . _ ieee t. automat . contr . _ * 47*(11 ) , 18791882 . clark , d. & morrison , s. & walker , k. 2009 fixing the functoriality of khovanov homology . _ _ * 13*(3 ) , 14991582 . cubelets robot construction for kids | modular robotics . retrieved from http://www.modrobotics.com/cubelets/ erickson , m.j . 2014 introduction to combinatorics . discrete mathematics and optimization * 78 * ( 2nd ed . ) . john wiley & sons . farhi , e. & goldstone , j. & gutmann , s. & sipser , m. 2000 quantum computation by adiabatic evolution . farhi , e. & goldstone , j. & gutmann , s. 2002 quantum adiabatic evolution algorithms with different paths . farhi , e. & goldstone , j. & gosset , d. & gutmann , s. & meyer , h.b . & shor , p. 2011 quantum adiabatic algorithms , small gaps , and different paths . _ quantum information & computation _ * 11*(34 ) , 181214 . fox , r.h . 1961 a quick trip through knot theory . in : _ topology of 3-manifolds and related topics _ , m.k . fort ( ed . ) , prentice - hall , nj , 120167 . franken , d. & hupper , a. 2005 improved fast covariance intersection for distributed data fusion . in _ proceedings of the 8th international conference on information fusion_. ishii , a. , iwakiri , m. , jang , y. & oshiro , k. 2013 a @xmath94family of quandles and handlebody - knots . _ illinois j. math . _ , * 57 * 817838 . jazwinski , a.h . 1970 stochastic processes and filtering theory . new york : academic press . joyce , d. 1982 a classifying invariant of knots : the knot quandle . _ j. pure appl . algebra _ * 23 * , 3765 . julier , s. uhlmann , j.k . 2001 general decentralized data fusion with covariance intersection . in _ handbook of multisensor data fusion _ , crc press , boca raton , fl . kauffman , l.h . 1994 knot automata . in _ twenty - fourth international symposium on multiple - valued logic , conference proceedings _ , may 25 - 27 1994 , boston , massachusetts , ieee computer society press , 328333 . kauffman , l.h . 1995 knot logic . in _ knots and applications _ ( ed l. kauffman ) , series of knots and everything * 6 * , world scientific publications , 1110 . kim , virtual knot groups and their peripheral structure , _ j. knot theory ramifications _ * 9*(6 ) ( 2000 ) 797812 . t. kishino and s. satoh , a note on non - classical virtual knots , _ j. knot theory ramifications _ * 13*(7 ) ( 2004 ) 845856 . knuth , k.h . 2015 the deeper roles of mathematics in physical laws . third place , fqxi essay contest : `` 2015 trick or truth : the mysterious connection between physics and mathematics . '' retrieved from http://fqxi.org/data/essay-contest-files/knuth_knuth---fqxi-essay---.pdf moskovich , d. 2006 surgery untying of coloured knots . _ * 6 * , 673697 . ` arxiv : math.gt/0506541 ` niehsen , w. 2002 information fusion based on fast covariance intersection filtering . in _ proceedings of the fifth international conference on information fusion , 2002 _ * 2*(2 ) , 901904 . pearl , j. 2009 _ causality : models , reasoning , and inference . _ cambridge university press . przytycki , j.h . 2011 distributivity versus associativity in the homology theory of algebraic structures . _ demonstration mathematica _ * 44*(4 ) , 823869 . tietze , h.f.f . 1908 ber die topologische invarianten mehrdimensionaler mannigfaltigkeiten . _ monatshefte fr mathematik und physik _ * 19 * 1118 . vicary , j. 2012 higher semantics for quantum protocols . in _ proceedings of the 27th annual acm / ieee symposium on logic in computer science _ , 606615 . the proof of proposition [ p : geodesic ] is the combination of the following two lemmas which are surely well - known . let @xmath251 be unnormalized gaussian densities parameterized by their mean and covariance , @xmath252 and @xmath253 , respectively . the pdf @xmath250 is given by the expression : @xmath254\right)\enspace .\ ] ] let @xmath265 be a space of probability density functions ( pdfs ) over some domain . consider the functional @xmath266 defined by @xmath267 where @xmath268 , @xmath248 , and @xmath269 are pdfs in @xmath265 and @xmath270 is the kullback leibler divergence . then the following holds for any @xmath271 $ ] and @xmath272 : @xmath273 writing down the kullback leibler divergence explicitly , we have : @xmath274 = \\ = \min_{g(x ) } \int_x g(x ) \log \frac{g(x)}{p(x)^{1-\omega } q(x)^\omega } \mathop{dx}= \\ = \min_{g(x ) } \left[\int_x g(x ) \log \frac{g(x)}{p(x)^{1-\omega } q(x)^\omega } \mathop{dx } + \int_x p(x)^{1-\omega } q(x)^\omega \mathop{dx } - \int_x p(x)^{1-\omega } q(x)^\omega \mathop{dx } \right ] = \\= \min_{g(x ) } \left[\mathrm{kl}\left ( g(x ) \bigg \| \frac{p(x)^{1-\omega } q(x)^s}{\int_x p(x)^{1-\omega } q(x)^s \mathop{dx } } \right ) \right ] - \int_x p(x)^{1-\omega } q(x)^\omega \mathop{dx } \enspace .\end{gathered}\ ] ] but @xmath275 vanishes if and only if its arguments coincide , and the result follows .

scientists have long sought to understand how vascular networks supply blood and oxygen to cells throughout the body . recent work focuses on principles that constrain how vessel size changes through branching generations from the aorta to capillaries and uses scaling exponents to quantify these changes . prominent scaling theories predict that combinations of these exponents explain how metabolic , growth , and other biological rates vary with body size . nevertheless , direct measurements of individual vessel segments have been limited because existing techniques for measuring vasculature are invasive , time consuming , and technically difficult . we developed software that extracts the length , radius , and connectivity of _ in vivo _ vessels from contrast - enhanced 3d magnetic resonance angiography . using data from 20 human subjects , we calculated scaling exponents by four methods two derived from local properties of branching junctions and two from whole - network properties . although these methods are often used interchangeably in the literature , we do not find general agreement between these methods , particularly for vessel lengths . measurements for length of vessels also diverge from theoretical values , but those for radius show stronger agreement . our results demonstrate that vascular network models can not ignore certain complexities of real vascular systems and indicate the need to discover new principles regarding vessel lengths . department of biomathematics , david geffen school of medicine , university of california , los angeles , los angeles , ca , usa + * 2 * department of radiological sciences , biomedical physics , and bioengineering , university of california , los angeles , los angeles , ca , usa + * 3 * department of ecology and evolutionary biology , university of california , los angeles , los angeles , ca , usa + * 4 * santa fe institute , santa fe , nm , usa current address : university of pennsylvania , department of biology , 433 s university ave , philadelphia , pa 19104 , usa + * [email protected]

1509.02824

networks that supply resources are essential to the maintenance and growth of many natural and engineered systems . these resource - distribution networks are pervasive throughout biology . examples include the tracheal system in insects , xylem networks in plants@xcite , foraging trails of ant colonies@xcite , and cardiovascular systems in animals@xcite . because the cardiovascular system delivers resources and energy to the body , its structure at least partly determines rates of growth and metabolism @xcite . moreover , the cardiovascular system plays a role in many diseases such as heart disease , stroke , inflammation , and malignant tumor growth@xcite . the pervasiveness and importance of material transport in biology motivates the search for basic principles that help shape resource - distribution networks in general and the cardiovascular system in particular . the theory of the structure of vascular networks has roots in the early 20th century@xcite . more recent theories predict properties of the entire vascular network by assuming a hierarchical structure and include those of west , brown and enquist@xcite ( henceforth the wbe model ) , banavar et al.@xcite , dodds@xcite , huo and kassab@xcite , and others@xcite . these theories assume or predict how vascular structure , such as the radius of vessels , changes as the network branches from aorta to capillaries . while early theories focused on vessel radius , recent theories also incorporate how vessel length changes across the network@xcite . many also assume symmetric branching child vessels have identical properties@xcite . in symmetric , hierarchical models , knowledge of both radius and length enables derivations of how metabolic rate varies with body size across species@xcite and how organismal and tumor growth rate change with body size@xcite . several theories predict vascular structures will be found to be self - similar some aspect of the network can be viewed as a rescaled copy of the whole@xcite across specific ranges of spatial scale@xcite . self similarity necessarily leads to relationships and distributions that are characterized by power laws , whose exponents we call the scaling exponents@xcite . self similarity can either be a strict or statistical property . each new chamber of the nautilus is a larger exact copy of the previous chamber , while along coastlines shorter segments exhibit rescaled statistical properties of the longer segments@xcite . self similarity in nature suggests scale - free principles that constrain structure . an organism may need to maintain a certain shape at all stages of growth , use a common developmental program at all scales , or cope with physical processes with no preferred scale such as energy minimization or turbulence in fluids . self similarity greatly simplifies many calculations , with applications from cardiac physiology@xcite to allometric scaling@xcite . real vasculature is known to deviate from hierarchical , symmetric models in many ways , leading to criticism and debate about leading models@xcite . without reliable data , it is impossible to determine whether these deviations can be ignored , so that existing theories can accurately predict newly observed features such as curvature in scaling relationships@xcite . price et al.@xcite recently decried the lack of data for individual vessel segments that are needed for tests of scaling theory . direct measurements and tests may reject existing principles while laying the groundwork for the discovery of new patterns and principles in vascular architecture . in addition , measurements can parameterize existing equations to obtain more exact predictions for metabolic rates and growth curves , and can be used to examine the natural variation in the parameters across species , tissues , and tumor types . vessel segment data preserves the asymmetry , reticulation , tortuosity , and other features of real vasculature , and quantitative data about these features may inform future models . furthermore , extreme values or distinct patterns of variation may be signatures of pathologies that could eventually be used as diagnostics . recent work involving direct measurements of plant architecture has begun to realize this potential@xcite . comprehensively characterizing vascular structure and obtaining reliable estimates for vascular scaling exponents requires large numbers of measurements across orders - of - magnitude in spatial scale ranging from 0.004 mm to 15 mm for vessel radii in humans . measurement is complicated because vascular systems are intertwined with tissues throughout the body at a wide range of spatial scales@xcite . even gross morphological measurements have historically taken impressive and time - consuming efforts and required invasive methods such as casting@xcite . zamir@xcite and kassab@xcite constructed explicit descriptions of small regions of vascular systems such as the coronary artery by perfusing fixed specimens with a silicone or acrylic polymer , dissolving tissue away and examining each vessel segment under a microscope . these approaches enabled the measurement of tens of thousands of vessels from fixed specimens that were then used to test and develop vascular system models@xcite . however , the process of casting may enlarge or damage vessels , and little of this raw data is publicly available for analysis . in addition , many more measurements across the range of scales are needed to identify the principles that shape vascular architecture . different physical principles may dominate at different scales , and mapping out different regimes will require large amounts of data at each scale . in the wbe model@xcite , the dominant mechanism of energy loss for blood flow in the arterioles ( radius @xmath0 1 mm ) is viscous dissipation , but near the heart ( vessel radius @xmath1 1 mm ) pulsatile flow and reflection of pressure waves along vessel walls dominates@xcite . the transitions between these regimes are neither well - understood theoretically nor described empirically@xcite . the pulsatile regime the focus of our measurements has greater variation in the number of branching orders and size of vessels across species and is thus the primary determinant of scaling of metabolic and other vital rates with body size@xcite . large amounts of vascular data across all relevant spatial scales are contained within existing angiographic images ( e.g. , mri and x - ray ) . these images are obtained non - invasively , thus avoiding problems of damage to vasculature and allowing for the possibility of longitudinal studies . the latent data within these images represents a tremendous opportunity . all that is needed is a reliable and automatic method for extracting vascular data from angiographic images . we have developed novel software that imports 3d images , creates a topologically and spatially explicit map of the blood vessel network , and measures the radius , length , and volume of all visible vessels . we have applied this software to 20 magnetic resonance angiograms of living humans to obtain 3015 data points that range in radius from 0.6 - 6.8 mm , representing an order of magnitude . this range corresponds to large vessels in the pulsatile flow regime relevant to allometric scaling@xcite . we then analyzed these data based on the four distinct methods ( eqs [ eq : murray_gen]-[eq : regres ] ) for measuring vascular scaling exponents described below . vascular scaling exponents encapsulate how radius and length of vessels change across the network . virtually all scaling relationships for local or global properties can be expressed in terms of these vascular scaling exponents . consequently , we view these scaling exponents as forming the foundation of most modern biological scaling theory and make them the primary focus of our analysis . we here describe four distinct methods for calculating vascular scaling exponents . because the radius of the vessel plays a primary role in determining both the flow rate and resistance to blood flow through the vessel , theories for the vascular scaling exponent for vessel radius often focus on the power to pump blood from the heart to the capillaries . it is argued@xcite that this power will have been minimized by natural selection to allow as much power as possible to be available for foraging , growth , and reproduction . one classical approach minimizes power loss of blood flow due to viscous dissipation and due to cost of blood volume in order to derive that flow rate , @xmath2 , depends on the cube of the vessel radius , @xmath3 . combining this result with conservation of fluid flow at a branching junction yields murray s law , @xmath4 , where @xmath5 is the radius of the parent vessel segment and @xmath6 is the radius of the @xmath7th child ( distal ) segment . another classical approach for the cardiovascular system is to minimize wave reflections in pulsatile flow , as womersley and west et al . have done@xcite . this approach leads to area - preserving branching ( or da vinci s rule ) , so that the sum of the cross - sectional areas ( @xmath8 ) of child vessels equals the cross - sectional area of a parent vessel at a branching junction . in the wbe model , reflections dominate for large vessels while dissipation dominates for small vessels ( @xmath0 1 mm ) . moreover , in the wbe model , a volume - servicing argument@xcite is used to derive an analogous relationship for vessel lengths , @xmath9 , while huo and kassab assume the same relationship but allow the exponent to vary from length - preserving ( exponent of 1 ) to volume - servicing ( exponent of 3 ) . optimization has been a common approach to developing vascular models throughout the past century , but it has been highly debated as to which properties are optimized and what are the tradeoffs among them@xcite . for instance , banavar et al.@xcite optimize for efficient transport within three - dimensional bodies and dodds@xcite minimizes network volume as required to continually supply metabolites within a body . indeed , different principles and assumptions lead to a variety of relationships between the flow rate and vessel radius . consequently , we express a generalized form of murray s law @xmath10 in which we define the vascular scaling exponent , @xmath11 , for vessel radius to be consistent with the notation of price et al.@xcite . the analogous generalization for vessel length is @xmath12 where @xmath13 is the vascular scaling exponent for vessel length . to ease computation , many models further assume that vascular networks are strictly self - similar and symmetrically branching child vessels all have identical properties within a branching level . in this case , scale factors and associated scaling exponents can be defined for each branching level , @xmath14 , which represents the number of branching junctions from the heart to that vessel . following the notation of the wbe model , the scale factors are @xmath15 and @xmath16 . for dichotomous branching , we can solve for @xmath17 and @xmath18 using eqs [ eq : murray_gen ] and [ eq : volserv_gen ] , to find @xmath19 furthermore , for these idealized networks , the frequency distributions of radius and length follow power laws with scaling exponents @xmath20 and @xmath21 . because there are @xmath22 vessels of radius @xmath23 , eqs [ eq : bgqs ] give the two power - law relationships @xmath24 similarly , for any vessel , its radius and length are related to the number of downstream endpoints ( e.g. capillaries ) , @xmath25 , by @xmath26 eqs [ eq : murray_gen]-[eq : regres ] constitute four methods of calculating vascular scaling exponents . each method relies on different types and levels of information . first , for each branching junction , the generalizations of murray s law ( eq [ eq : murray_gen ] ) and volume servicing ( eq [ eq : volserv_gen ] ) can be numerically solved for the exponents @xmath11 and @xmath13 by newton s method . the solution provides a direct local measurement of the scaling exponents at every junction that we call _ conservation - based _ scaling exponents . the value is undefined if a child vessel has radius or length greater than its parent . second , for each parent - child pair of vessel segments , the ratio of vessel radius and length can be calculated . by eq [ eq : bgqs ] , these scale factors can be used to compute a second local measure our _ ratio - based _ scaling exponents . third , across all vessels and junctions , empirical distributions of radii and lengths can be fitted to power laws to produce what we term the _ distribution - based _ scaling exponents , as in eq [ eq : powlaw ] . fourth , across all vessel segments , log - log regressions of radii and lengths versus the number of downstream endpoints can be performed to derive _ regression - based _ scaling exponents , following eq [ eq : regres ] . these latter two methods each provide single values for the vascular scaling exponents , @xmath11 and @xmath13 , across the whole network , and they do not rely on information about the geometry of individual branching junctions . in the literature , these four methods for measuring scaling exponents are often used interchangeably@xcite . however , these are only proven to be identical for symmetrically - branching , strictly self - similar networks . furthermore , it is unknown whether measurements of vascular scaling exponents using these four methods ( eqs [ eq : murray_gen]-[eq : regres ] ) will produce values that are approximately similar or significantly different . if they differ , this raises questions about which of these four measures of vascular scaling exponents , if any , best corresponds to the scaling relationships predicted by ideal networks or observed empirically for metabolic and growth rates . to begin to answer these questions , we now report results obtained by applying our new software , angicart , to 20 contrast - enhanced 3d magnetic resonance image volumes for human head and torso . we collected 3015 segments across 1473 branching junctions . of the junctions , 1422 were recorded as dichotomous and 51 as trichotomous . the number of branches on all paths ( subjects pooled ) between the aorta and the smallest observable vessels was typically between 3 and 10 ( middle 68% ) . each 3d image volume , corresponding to 151 segments on average , took about 2.7 minutes of cpu time on a single 2.3ghz intel processor . as described in the methods , we used an automated circularity criterion as an indicator of possible errors in order to omit vessel identification errors . our results are based on radius , length , and volume measurements from only the 1240 segments for which we saw no indication of potential error . the complete , raw output of our software before any filtering or analysis is available as . the software source code and original imagery enabling the exact reproduction of this dataset are available online as a repository in the git revision control system at ` https://github.com/mnewberry/angicart ` . we attempt to solve eq [ eq : murray_gen ] numerically at each branching junction . we label the parent at each branching junction by traversing the vascular network starting at the vessel segment of greatest radius , which we assume is a segment in the aorta . if all children are smaller than the parent , the equation is guaranteed to have exactly one solution , and numerical convergence to the solution is fast . otherwise , the equation may have zero , one , or two solutions . such cases may represent real anatomical variation , or misidentification of the parent vessel due to errors or ambiguous topology such as the circle of willis . we consider only the simplest case , in which children are smaller than their parent , because otherwise the solutions are difficult to interpret in the context of existing vascular network models . all child radii are smaller than the parent in 82% ( 222 ) of junctions , and all child lengths are smaller in only 35% ( 94 ) of junctions . we show distributions of these measures of the vascular scaling exponents , @xmath11 and @xmath13 , across branching junctions in fig [ fig : qdist ] . the arithmetic mean and associated 95% confidence intervals for these conservation - based measurements of @xmath11 and @xmath13 are presented in table [ tab : all ] . l c c c c c c * radius exponent measures * & @xmath27 & @xmath11 & 95% ci & @xmath28 & @xmath29 + conservation ( node ) & 222 & 0.49 & @xmath30 0.03 & 0.23 & + ratio ( node ) & 703 & 0.43 & @xmath30 0.03 & 0.39 & + distribution ( network ) & 657 & 0.30 & @xmath30 0.04 & & 0.90 + regression ( network ) & 1240 & 0.41 & @xmath30 0.02 & & 0.66 + * theory * & & & & & + wbe small vessels & & 0.33 & & & + wbe large vessels & & 0.50 & & & + banavar et al.@xcite & & 0.50 & & & + huo and kassab@xcite & & 0.330.50 & & & + murray s law & & 0.33 & & & + + * length exponent measures * & @xmath27 & @xmath13 & 95% ci & @xmath28 & @xmath29 + conservation ( node ) & 94 & 1.40 & @xmath30 0.20 & 0.98 & + ratio ( node ) & 703 & 0.17 & @xmath30 0.12 & 1.57 & + distribution ( network ) & 518 & 0.73 & @xmath30 0.10 & & 0.68 + regression ( network ) & 1240 & 0.94 & @xmath30 0.05 & & 0.19 + * theory * & & & & & + wbe all vessels & & 0.33 & & & + banavar et al.@xcite & & 0.50 & & & + huo and kassab@xcite & & 0.331.00 & & & + for vessel radius from solutions to eq . ( [ eq : murray_gen ] ) using empirical measurements of vessel radii extracted from magnetic resonance angiography using our software , and ( b ) the analogous frequency distribution of vascular scaling exponent @xmath13 for vessel segment length . ] topological information allows us to compute @xmath17 and @xmath18 directly for each parent - child pair of vessel segments . our dataset contains 703 pairs of parent and child segments with dichotomous branching . a small proportion of the @xmath17 and @xmath18 values may be over - estimated due to misidentification of the parent - child relationship ( see methods ) . this bias would produce underestimates in @xmath11 and @xmath13 , but because misidentification is very infrequent , the magnitude of this bias is expected to be within the measurement error . the distribution of @xmath17 and @xmath18 is displayed in fig [ fig : betadist ] , and the arithmetic mean and associated 95% confidence intervals of the ratio - based vascular scaling exponents calculated using eq . ( [ eq : bgqs ] ) are shown in table [ tab : all ] . , the ratio of child to parent radii , and ( b ) the analogous distribution of scale factor @xmath18 , the ratio of child to parent lengths . ] for symmetrically branching , self - similar networks , the frequency of radius and length measurements follow power - law distributions . we did a linear fit to the log - log transformed histograms of radius and length measurements ( the log of eq . ( [ eq : powlaw ] ) ) using sma regression . we derived empirical 95% confidence intervals by resampling with different bin sizes and cutoff values for the tail ( see ) . the fits are shown in fig [ fig : rdist ] . the scaling exponents obtained from our fits are given in table [ tab : all ] . , and ( b ) the log of length , @xmath31 . fit lines and slope values are shown . the correlation coefficient @xmath32 for the log relative frequency versus the logs of radius and length respectively are 0.90 and 0.68 . ] by taking the logarithm of eq . ( [ eq : regres ] ) , we can estimate the vascular scaling exponents , @xmath11 and @xmath13 , by performing regressions of the logarithm of the number of downstream tips @xmath33 against the logarithm of radius , @xmath34 , and the logarithm of length , @xmath31 , respectively @xcite . regression lines are shown in fig [ fig : betanreg ] . the measured slopes , which are the estimates of the vascular scaling exponents , and associated 95% ci are shown in table [ tab : all ] . , and ( b ) the log of length , @xmath31 , against the log of the number of downstream tips , @xmath33 . fit lines and slope values are shown . correlation coefficients @xmath32 are 0.66 and 0.19 , respectively ( @xmath35 for each ) . ] as in all analyses based on experimental data , measurement errors affect uncertainties in quantities calculated from the raw data . consequently , we investigate the sensitivity of our calculated scaling exponents and entire analysis to the choice of threshold intensity in our algorithm as well as to gaussian noise in the image quality . an intensity threshold is used in our algorithm to select the voxels from the image that describe the shape of the vessel lumen . lower thresholds reveal more vessels , so for our analysis above , we used the minimum threshold that produced reliable segments , as described in and . because the threshold affects the boundaries of the vessel lumen , the vessel radius also depends on the threshold . for each image there is a minimum acceptable threshold that we used above , and for our sensitivity analysis , we also chose a maximum threshold to be the largest value for which at least 30 vessel segments are visible . for our plots in fig [ fig : thresh_sens ] , we normalized the threshold to range from 0 and 1 . that is , 0 is the threshold used for the results above , and 1 is the maximum threshold for which at least 30 vessel segments are visible . for each normalized threshold increment of 0.05 between 0.00 and 1.00 , we ran our entire analysis and calculated the four scaling exponents for radius and length ( fig [ fig : thresh_sens ] ) . the values of the scaling exponents at a normalized threshold of 0.00 recapitulate table [ tab : all ] . our results remain qualitatively similar as we increase the threshold and include fewer segments . however , at higher normalized thresholds , we can no longer resolve differences between some exponents that are resolvable at normalized threshold 0.00 , as expected from the law of large numbers . column of table [ tab : all ] . the top right panel shows the number of vessel segments used to calculate the four scaling exponents for vessel length versus the normalized threshold value , as in the @xmath27 column of table [ tab : all ] . for both of these top two panels , the number of data points decreases with threshold value as it must . the bottom left panel depicts how the four calculated scaling exponents for vessel radius vary with normalized threshold value . the bottom right panel depicts how the four calculated scaling exponents for vessel length vary with normalized threshold value . horizontal lines indicate points of comparison to theory as described in table [ tab : all ] . ] noise in images may also cause errors in the identification of vessels or in estimates of the radius or length of vessels . we conducted a sensitivity analysis similar to the above by adding gaussian noise that varied in magnitude from 0.47 ( the measured baseline level of noise in the foreground of one image ) to 4.7% ( 10 times the baseline noise level ) of the maximum voxel intensity . results ( ) show no significant changes in our results with higher levels of noise . as another measure of uncertainty , we located vessels with radius estimates that differed with threshold despite high reliability in vessel identification ( vessel endpoints were similar across at least 5 threshold values ) . we located 12 vessel segments from 9 different patients that matched this criterion . across these 12 vessels , the mean radius estimate varied from 0.9 to 6.7 mm , but the coefficient of variation ( = standard deviation / mean across thresholds ) ranged only from 0.02 to 0.08 . moreover , the coefficient of variation was uncorrelated with radius ( pearson correlation = 0.07 ) , suggesting that individual vessel radius measurements are precise to roughly @xmath36 ( 2 times the coefficient of variation ) regardless of vessel size . for a specific threshold , we calculate how the measured vessel radius differs from the mean of the vessel radius across all thresholds . at each threshold , this difference tended to be in the same direction ( mostly positive or mostly negative ) across the 12 vessels we measured . moreover , the magnitude of the difference was proportional to vessel radius ( most @xmath37 ) . consequently , the ratio of vessel radii at any specific threshold is roughly equal to the ratio of the means of the vessel radii across thresholds and also equal to the ratio of vessel radii at any other specific threshold . that is , within the plausible range of thresholds that we explore here , the calculated scaling ratios are largely independent of the choice of threshold value , implying that the choice of threshold does not create any bias in the calculated scaling exponents or ratios . of course , our sensitivity analysis can not exclude all possible sources of error or bias . in the methods sections , we discuss how our results might be affected by other possible sources of error , such as tree topology identification , the patient population , small vessel censoring , vessel lumen misidentification , skeleton line selection , centerline quantization , and vessel segment misattribution . the positional accuracy of mri is high , so errors arise due to classification or interpretation of voxels . because the threshold parameter and noise in the image primarily control how we classify voxels the first step of analysis these are the major determinants of subsequent errors . in our analysis , as presented in figs [ fig : thresh_sens ] and , no systematic biases are observable , so we conclude that our results are highly robust to the largest and most notable sources of uncertainty . the estimated values of vascular scaling exponents obtained using our four different methods are all presented in table [ tab : all ] . all pairs of measures are statistically significantly different ( welch s t - test , @xmath35 ) except for the ratio - based and regression - based @xmath11 ( @xmath38 ) . values of @xmath11 based on conservation rules at branching junctions , scale factors for parent - child pairs , and regression of @xmath34 versus @xmath39 are all between @xmath40 ( the wbe prediction for large vessels ) and @xmath41 ( murray s law , the banavar et al . prediction@xcite , and the wbe prediction for small vessels ) , and the remaining distribution - based @xmath11 includes @xmath42 in its 95% ci . the conservation - based exponent @xmath11 is not statistically - significantly different from @xmath40 . different measures for @xmath13 range from @xmath43 to @xmath44 and all are statistically significantly different ( @xmath35 ) from each other and from the volume - servicing and area - servicing values of @xmath45 and @xmath46 respectively . it is notable , however , that the difference between regression - based and distribution - based measurements of @xmath13 is no longer resolvable at normalized thresholds higher than 0 . the distribution - based and regression - based exponents lie between area - servicing and length - preserving . these discrepancies between measures suggest that vessel segment lengths are poorly modeled by strictly self - similar and symmetrically branching networks . our software acquired direct measurements of a large number of connected vessel segments from _ in vivo _ angiography . we calculated vascular scaling exponents in these data using four methods to directly compare values from real vascular networks with each other and with theoretical values from the wbe model , murray s law , banavar et al.@xcite , and huo and kassab@xcite . intriguingly , our results lead to contrasting conclusions for the changes in vessel radius and length across scale . for the vascular scaling exponent @xmath11 that quantifies changes in the radius , the conservation - based and ratio - based estimates are closer to @xmath47 than @xmath42 . these estimates support work on large vessels by west et al . , zamir and banavar@xcite , in contrast to murray s law , which does not distinguish large and small vessels . west et al . derive that the dominant source of power loss for large vessels ( estimated to be @xmath48 1 mm ) is the reflection of pressure waves at branching junctions , while for small vessels , power loss is dominated by viscous dissipation between blood and the vessel walls . because of the resolution of our mri volumes , we are able to extract data mostly for vessels with a radius greater than 1 mm , corresponding to the large vessel regime in the wbe model . consequently , our results for the scaling of radii are supportive of the area - preserving branching of large vessels , corroborate recent findings for plants@xcite , and reject the possibility that murray s law might apply , either generally or on average , to junctions of large vessels . this result demonstrates that minimizing energy - dissipation due to blood flow does not capture the guiding principles that shape the vascular system across all scales . future studies using higher resolution angiography ( e.g. , micro - ct ) to obtain data for small vessels are needed to test murray s law and the wbe prediction for small vessels and to determine if minimizing energy dissipation is a relevant principle at any scale . for the vascular scaling exponent , @xmath13 , for vessel lengths , the discrepancies between predicted and estimated values are more difficult to reconcile and interpret . none of the four measures of @xmath13 agree with each other or provide support for volume- or area - servicing branching , while only the regression - based method provides support for length preservation . within the wbe model , @xmath13 is predicted to be @xmath49 based on an argument that the vascular network must be volume - servicing for the entire body@xcite . this volume - servicing argument has been questioned on theoretical grounds@xcite , and here we provide empirical evidence that volume - servicing or any other conservation law for length does not hold locally at branching junctions . indeed , for the conservation - based exponents , 65% of branching junctions violate the model so severely that exponent values are undefined . the volume - servicing argument is supposed to apply across at least the vast majority of scales and is a key element of the wbe explanation for the @xmath50 allometric scaling relationship between metabolic rate and body size . the breakdown between this argument and the real vascular networks we measured may occur because , contrary to the wbe argument , the length of a vessel segment is not a reliable indicator of the volume it services . the correlations between length and number of downstream endpoints , a proxy for volume serviced , is very low compared to the same correlation for radius ( 0.2 versus 0.7 ) . considering only the largest vessels , this is not surprising . the ascending aorta is only a few centimeters in length and services most of the body , while the carotid artery is much longer ( at least 10 cm in length ) and services only half the head . our results imply that either modification of the volume - servicing argument is needed or some new principle yet to be discovered guides the distribution of vessel lengths as the vascular network branches throughout the body . these new developments could lead to corrections to the power - law predictions of the original wbe theory that may agree better with recent findings of `` curvature '' in the allometric relationship@xcite . beyond the differences discussed thus far , vessel lengths and radii also differ in their distributions for vascular scaling exponents and scale factors . measurements of @xmath11 and @xmath17 exhibit a strong central tendency ( figs [ fig : qdist ] and [ fig : betadist ] ) , while the scale factor for length , @xmath18 , has a highly skewed distribution with typical values that are not well - described by the mean . thus , a derivation implicitly based on a mean - value approximation may be successful for predicting vessel radii but fail to predict scaling relationships involving vessel lengths . thus , while hierarchical symmetric models may fail outright to adequately describe vessel lengths , the discrepancies between vessel radii in real networks and idealized models , such as the wbe model , may only result in minor corrections to model predictions . this may help explain the success of the wbe model in predicting a wide range of phenomena . differences in results for vessel radius and length could be tied to different strengths of the constraints on vessel geometry . radii and length distributions have previously been observed to differ in the external branching of plants and leaves@xcite . one explanation for this is that viscous power loss depends much more sensitively on vessel radius ( as a @xmath51th power , @xmath52 ) than on vessel length ( linearly , @xmath53 ) . thus , the strength of selection for optimal vessel radius is much stronger than for optimal vessel length , implying evolution has more often sacrificed vessel length when negotiating tradeoffs in anatomy . another potential explanation is that vessel radii are self - similar due to a local constraint at each branching junction , whereas vessel lengths may be constrained only at larger scales organs and organisms that more accurately capture how the vascular network needs to span and feed a spatially inhomogeneous body . disagreement about the value of the length scaling exponent between our four methods indicates that assumptions of the simplest model must be violated so strongly as not to hold even approximately . that is , strict self similarity , symmetric branching or both must be strongly violated for the real vascular networks we measured . our data reveal pervasive asymmetry in both radius and length between child vessels . how far the results for symmetric networks generalize to asymmetric networks has been explored very little@xcite . the differences we observe between different measures of vascular scaling exponents could be explained by the inability of existing theories to account for the asymmetry of real vascular networks . developing a theory to account for asymmetric branching may be challenging . for instance , accounting for asymmetry would require at least two scale factors for radii ( e.g. , @xmath54 and @xmath55 ) and two for lengths ( e.g. , @xmath56 and @xmath57 ) . these additional scale factors and associated scaling exponents would necessarily change our analysis and our estimates for the ratio - based scaling exponent , and would potentially change our interpretation of the distributions of @xmath17 and @xmath18 . rather than thinking of distributions of @xmath17 and @xmath18 , we would think of joint distributions of @xmath54 and @xmath55 , for example . for similar reasons , our interpretation and analysis of the frequency distribution of radius and length could be altered , thus affecting the estimates of the distribution - based exponents as well . angicart outputs @xmath17 and @xmath18 values for each vessel pair , and can provide the detailed information required for future studies of multiple scale factors for length or radius and asymmetric branching . all of the models discussed in this paper ignore any reticulation or loops in the vessel topology , in contrast to recent work on leaf venation networks@xcite . our analysis also follows this assumption . however , loops are known to occur anatomically in healthy ( circle of willis ) and diseased ( tumors , arteriovenous malformations ) tissue . extensions to our software and to theory could address this issue . such an extension could be used to investigate abnormal tumor vasculature@xcite , or allow new theory to be developed to explain the normal anatomical function of reticulation . there are also other spatial aspects that have received theoretical attention , such as branching angle , that our software is already capable of recording . many more tests could be performed with data on microvasculature . for instance , huo and kassab@xcite have published scaling relationships for how crown volume and length change with stem radius . testing these requires knowledge of the full crown , down to the microvasculature . similarly , dodds@xcite makes predictions for virtual vessels that coincide with real vessels only at the smallest scales . we developed new software and applied it to mri of human head and torso to obtain one of the most detailed datasets for examining branching architecture in vascular networks . in addition , we conducted a comprehensive data analysis that uses both local and global methods to measure scaling exponents . together , this new software , data , and analysis provides valuable information for answering fundamental questions about vascular system morphology . the public release of our imaging software , angicart , should enable researchers to ground future vascular network theories in empirical data . the software facilitates comparison across spatial scales and between studies by operating uniformly on all tomographic imaging methods . because imaging can be done non - invasively , our method affords the opportunity to record all spatial information of _ in vivo _ vasculature through time or across development . we explain four different methods to estimate vascular scaling exponents from spatially - explicit data . although researchers use and sometimes interchange these four methods , we found that all four methods can lead to different results , and that for scaling exponents for vessel lengths , these differences can be dramatic . this result is in stark contrast to theoretical calculations for idealized , symmetric networks that predict all methods will give identical values . we advise caution when interpreting different methods and estimates as the same scaling exponent because this could lead to misperceptions and disagreements among studies . for instance , regression - based estimates are the most common across levels of biological organization while distribution - based estimates are used for forests@xcite , so comparing these estimates to each other must be done with care . the differences we observe call for a new understanding of the relationships between the local geometry of vessels and the global properties of vascular systems . new theory should be developed to accommodate the anatomical variation and asymmetric branching we observe in real networks . after local institutional review board approval and written informed consent had been obtained , 20 consecutive adult patients with clinically suspected supraaortic arterial occlusive disease were prospectively enrolled to evaluate new mri methods for the study of carotid atherosclerosis@xcite . although other data was recorded in that study , we use only the images . in that study less than 0.5% of observed vessel segments had notable luminal narrowing , so we conclude patient selection and enrollment did not affect our results . we acquired contrast - enhanced magnetic resonance angiograms ( cemra ) of the upper torso , neck , and head in the 20 human subjects ( n=20 ) using a 3 tesla siemens trio scanner ( siemens medical solutions , erlangen , germany ) . the data acquisition details have been previously described@xcite . in brief , the cemra images were acquired after an antecubittal vein injection of gadolinium based contrast agent ( gd - dtpa , magnevist , bayer shering pharma ag , berlin , germany ) . the image volumes have dimensions that are typically close to @xmath58 voxels , with each voxel nearly isotropic and between @xmath59 m and @xmath60 m. the resolution and imaged volume are typical of high - quality 3 t mri . the point - spread - function for mri is known to be precise and equivalent to the programmed pixel size@xcite . in practice , the geometric accuracy is known to be sub - millimeter@xcite . the vessel networks in each image are clearly visible due to the sharp image contrast provided by the presence of the contrast agent , which makes the blood appear bright relative to dark non - blood tissues . we averaged each @xmath61-voxel cube of adjacent voxels into a single 1.4 - 1.8 mm voxel to remove noise , reduce processing time , and match conventionally - acquired resolutions . this reduced noise - induced errors without substantially changing the number of vessel segments represented . in two of our 20 image volumes , segmentation failed because bright , non - blood tissues were present very close to the blood volume , so that no threshold value excluded all non - blood objects as described in . we did not record any vessel segments from these image volumes . we saw no relationship between failure of segmentation and vascular system geometry . we created a free , open source software package angicart to read tomographic images of vascular networks , to automatically decompose a vessel lumen into vessel segments , and to measure the geometry and topology of the segments ( fig [ fig : overview ] ) . the software and data used in this study are available on the internet ( ` https://github.com/mnewberry/angicart/ ` ) under a gnu public license . our software starts by classifying voxels in a 3d image as part of the vessel lumen if they are within the largest connected group of voxels that exceed an intensity threshold@xcite , as with other level set or thresholding methods . we use a manual binary search to select the threshold value that best matched visual identification of vessels . the result is a 3d binary image , called the _ network mask_. next , we use spatial criteria to find the endpoints of the vessel network , where the vessels become too small to detect . given these endpoints , we find the centerline and branch points of vessels by skeletonization@xcite ( see ) . we implement skeletonization using an erosion technique successively removing voxels until no more are removable without disconnecting the endpoints in the mask@xcite . the voxels that remain after erosion lie within approximately 1 voxel - width of the true centerline . this information allows us to partition the network mask into segments by attributing each voxel to the segment whose centerline is closest to it . we record how vessels are connected and measure the length of each segment s centerline and the volume of each segment . following a geometric argument ( see ) , quantization error leads us to overestimate length . we compute radius as @xmath62 . radii are therefore underestimated on average due to the quantization error in length measurements . this bias affects the accuracy of individual length and radius measurements , but does not bias estimates of scaling exponents ( relative measures ) as long as the percent error does not change systematically with vessel size , which it does not . as a final filter of possibly misclassified vessels , we omit vessels in which more than 20% of the voxels lie further than @xmath63 from the centerline , or whose total volume is less than 4 voxels . further details of each step are presented in the . we determined the conservation - based node scaling exponents by solving eq . ( [ eq : murray_gen ] ) numerically using newton s method implemented in ocaml@xcite and iterated until the sum of powers was within 0.00001 of 1 . we estimated the regression - based scaling exponents using standard major axis ( sma ) regression of the natural log of radius ( length ) against the natural log of the number of downstream endpoints@xcite . we used sma regression because the variability and uncertainty in the y - axis ( vessel radius or length ) is as large as the variability and uncertainty in the x - axis ( number of downstream endpoints ) . furthermore , sma is appropriate because our goal is to obtain the best estimate of the scaling exponents ( slopes ) and not the best prediction of @xmath64 given @xmath65 . we estimated distribution - based scaling exponents by fitting the tail of the probability distribution of radius and length to a power law . that is , we binned log - transformed data and determined the slope of the log of probability density versus the log of radius and length using sma regression@xcite . we used 20 bins and discarded 5 and 7 initial bins of radius and length respectively . blood vessels near the resolution limit of mri may not be visible . although dimensions measured from observed small vessels are used in our other methods , counts of small vessels are unreliable due to censoring . thus , we discarded initial bins in order to exclude vessel sizes where non - uniform censoring of values might occur . we computed standard errors by varying bin size and the number of initial bins discarded ( up to @xmath66 each ) and using the middle 95th percentile of these values . by binning our data and fitting the power law using sma regression , we avoid problems that can arise when using maximum - likelihood estimators to fit our power - law distributions . specifically , the maximum - likelihood estimators are derived with specific assumptions and support choices such as smooth and continuous or discrete and integer , as in clauset et al.@xcite . in contrast , our distributions may be somewhere in between : continuous with an increased likelihood to take values near certain points , such as powers of @xmath17 times the aorta radius . our sma regression on bins of simulated vessel data produced stable estimates with relatively little bias in comparison to fits based on published maximum - likelihood estimators . we thank j. paul finn and derek lohan for providing the mri image data used in this study . we thank elizabeth sweedyk , robert zinkov and thomas bushnell , bsg for helpful discussions of the algorithms , datastructures and programming of angicart . we thank joshua plotkin and david mccandlish for helpful conversations . we are grateful to cormac mccarthy for comments on the manuscript . 10 price ca , enquist bj , savage vm . a general model for allometric covariation in botanical form and function . proc natl acad sci u s a. 2007;104(32):13204 . latty t , ramsch k , ito k , nakagaki t , sumpter dj , middendorf m , et al . structure and formation of ant transportation networks . journal of the royal society interface . 2011;8(62):12981306 . jun j , pepper jw , savage vm , gillooly jf , brown jh . allometric scaling of ant foraging trail networks . evolutionary ecology research . 2003;5(2):297303 . west gb , brown jh , enquist bj . . 1997;276(5309):122 . murray cd . proc natl acad sci u s a. 1926;12(3):207 . womersley jr . method for the calculation of velocity , rate of flow and viscous drag in arteries when the pressure gradient is known . the journal of physiology . 1955;127(3):553563 . thompson daw . . cambridge press , cambridge ; 1917 . savage vm , deeds ej , fontana w. . plos comput biol . 2008;4(9 ) . jain rk . . 2005;307(5706):58 . herman ab , savage vm , west gb . a quantitative theory of solid tumor growth , metabolic rate and vascularization . plos one . 2011;6(9):e22973 . savage vm , herman ab , west gb , leu k. using fractal geometry and universal growth curves as diagnostics for comparing tumor vasculature and metabolic rate with healthy tissue and for predicting responses to drug therapies . discrete and continuous dynamical systems - series b. 2013;18(4):10771108 . womersley j. oscillatory flow in arteries : effect of radial variation in viscosity on rate of flow . the journal of physiology . 1955;127(2):38 . banavar jr , maritan a , rinaldo a. . 1999;399(6732):130131 . banavar jr , cooke tj , rinaldo a , maritan a. form , function , and evolution of living organisms . proc natl acad sci u s a. 2014;p . 201401336 . optimal form of branching supply and collection networks . phys rev lett . 2010 jan;104:048702 . huo y , kassab gs . intraspecific scaling laws of vascular trees . journal of the royal society interface . 2012;9(66):190200 . savage v , bentley l , enquist b , sperry j , smith d , reich p , et al . hydraulic trade - offs and space filling enable better predictions of vascular structure and function in plants . proc natl acad sci u s a.2010;107(52):2272222727 . brown jh , gillooly jf , allen ap , savage vm , west gb . toward a metabolic theory of ecology . . 2004;85(7):17711789 . kolokotrones t , savage vm , deeds ej , fontana w. . 2010;464(7289):753756 . sperry js , smith dd , savage vm , enquist bj , mcculloh ka , reich pb , et al . a species - level model for metabolic scaling in trees i. exploring boundaries to scaling space within and across species . functional ecology . 2012;26(5):10541065 . allmen ei , sperry js , smith dd , savage vm , enquist bj , reich pb , et al . a species - level model for metabolic scaling of trees ii . testing in a ring - and diffuse - porous species . functional ecology . 2012;26(5):10661076 . mandelbrot bb . fractals : form , chance , and dimension . wh freeman san francisco ; 1977 . zamir m. on fractal properties of arterial trees . journal of theoretical biology . 1999;197(4):517526 . schroeder mr . fractals , chaos , power laws : minutes from an infinite paradise . courier dover publications ; 2012 . how long is the coast of britain ? statistical self - similarity and fractional dimension . science . 1967;156(3775):636 . zamir m. the physics of coronary blood flow . springer verlag ; 2005 . dodds ps , rothman dh , weitz js . re - examination of the `` 3/4-law '' of metabolism . journal of theoretical biology . 2001;209(1):927 . kozowski j , konarzewski m. is west , brown and enquist s model of allometric scaling mathematically correct and biologically relevant ? functional ecology . 2004;18(2):283289 . price ca , weitz js , savage vm , stegen j , clarke a , coomes da , et al . testing the metabolic theory of ecology . ecology letters . 2012;15(12):14651474 . price ca , symonova o , mileyko y , hilley t , weitz js . leaf extraction and analysis framework graphical user interface : segmenting and analyzing the structure of leaf veins and areoles . plant physiology . 2011;155(1):236 . mckown ad , cochard h , sack l. decoding leaf hydraulics with a spatially explicit model : principles of venation architecture and implications for its evolution . the american naturalist . 2010;175(4):447460 . biomechanics : circulation . springer ; 1997 . huang w , yen r , mclaurine m , bledsoe g. . journal of applied physiology . 1996;81(5):2123 . kaimovitz b , huo y , lanir y , kassab gs . . american journal of physiology - heart and circulatory physiology . 2008;294(2):h714 . kassab gs , rider ca , tang nj , fung yc . . american journal of physiology - heart and circulatory physiology . 1993;265(1):h350 . zamir m , wrigley s , langille b. the journal of general physiology . 1983;81(3):325 . apol mef , etienne rs , olff h. revisiting the evolutionary origin of allometric metabolic scaling in biology . functional ecology . 2008;22(6):10701080 . sperry js , smith dd , savage vm , enquist bj , mcculloh ka , reich pb , et al . a species - level model for metabolic scaling in trees i. exploring boundaries to scaling space within and across species . functional ecology . 2012;26(5):10541065 . savage vm , bentley lp , enquist bj , sperry js , smith dd , reich pb , et al . hydraulic trade - offs and space filling enable better predictions of vascular structure and function in plants . proc natl acad sci u s a. 2010;107(52):2272222727 . banavar jr , moses me , brown jh , damuth j , rinaldo a , sibly rm , et al . a general basis for quarter - power scaling in animals . proc natl acad sci u s a. 2010;107(36):1581615820 . bentley lp , stegen jc , savage vm , smith dd , allmen ei , sperry js , et al . an empirical assessment of tree branching networks and implications for plant allometric scaling models . ecology letters . 2013;16(8):10691078 . west gb , enquist bj , brown jh . a general quantitative theory of forest structure and dynamics . proc natl acad sci u s a. 2009;106(17):70407045 . zamir m. . in : scaling in biology . oxford university press , usa ; 2000 . p. 129143 . price ca , wing s , weitz js . scaling and structure of dicotyledonous leaf venation networks . ecology letters . 2012;15(2):8795 . turcotte d , pelletier j , newman w. networks with side branching in biology . journal of theoretical biology . 1998;193(4):577592 . mileyko y , edelsbrunner h , price ca , weitz js . hierarchical ordering of reticular networks . 2012;7(6):e36715 . blonder b , violle c , bentley lp , enquist bj . venation networks and the origin of the leaf economics spectrum . ecology letters . 2011;14(2):91100 . katifori e , szllsi gj , magnasco mo . damage and fluctuations induce loops in optimal transport networks . physical review letters . 2010;104(4):048704 . corson f. fluctuations and redundancy in optimal transport networks . physical review letters . 2010;104(4):048703 . lohan dg , barkhordarian f , saleh r , krishnam m , salamon n , ruehm sg , et al . mr angiography at 3 t for assessment of the external carotid artery system . american journal of roentgenology . 2007;189(5):1088 . liang zp , lauterbur pc . principles of magnetic resonance imaging . spie optical engineering press ; 2000 . scheib s , gianolini s , lomax n , mack a. high precision radiosurgery and technical standards . in : wowra b , editor . gamma knife radiosurgery . springer ; 2004 . p. 923 . tsai ps , kaufhold jp , blinder p , friedman b , drew pj , karten hj , et al . correlations of neuronal and microvascular densities in murine cortex revealed by direct counting and colocalization of nuclei and vessels . the journal of neuroscience . 2009;29(46):14553 . kirbas c , quek f. . acm computing surveys . 2004;36(2):81121 . zhou y , toga aw . efficient skeletonization of volumetric objects . visualization and computer graphics , ieee transactions on . 1999;5(3):196209 . fisher rb , dawson - howe k , fitzgibbon a , robertson c , trucco e. dictionary of computer vision and image processing . wiley chichester ; 2005 . leroy x , doligez d , frisch a , garrigue j , rmy d , vouillon j. the ocaml system release 4.00 . 2012;. warton di , wright ij , falster ds , westoby m. bivariate line - fitting methods for allometry . biological reviews . 2006;81(2):259291 . clauset a , shalizi cr , newman m. power - law distributions in empirical data . siam review . 2009;51(4):661703 . aylward sr , bullitt e. . medical imaging , ieee transactions on . 2002;21(2):6175 . jiang y , zhuang z , sinusas aj , papademetris x. . in : computer vision and pattern recognition workshops ( cvprw ) , 2010 ieee computer society conference on . ieee ; 2010 . p. 178185 . sniedovich m. dijkstra s algorithm revisited : the dynamic programming connexion . control and 2006;35(3):599 . antiga l , steinman da . robust and objective decomposition and mapping of bifurcating vessels . medical imaging , ieee transactions on . 2004;23(6):704713 . kline tl , zamir m , ritman el . . annals of biomedical engineering . 2010;p . the first step in our process is to obtain a binary mask ( or set of voxels ) that describes the shape of the blood vessel network . separating a structure of interest from the background is known in the computer vision literature as segmentation@xcite . the literature on image segmentation is voluminous and includes studies specifically examining segmentation of tomographic images of blood vessels . there are many approaches to segmentation of blood vessels , including region - growing , ridge - tracing , watershed models , spectral approaches , and deformable contours@xcite . we chose a threshold - based approach for its conceptual simplicity , estimability of errors , and lack of implicit assumptions about blood vessel geometry . there are more sophisticated approaches , but they rely on circularity@xcite , volume - servicing@xcite , or other assumptions to improve their visual quality . we avoid these assumptions as they might bias the output in favor of models our measurements are designed to test . by using simple thresholds rather than bayesian classification , we remain agnostic to the image capture method by ignoring the physics of the imaging system . to select a mask representing the blood vessel network , we calculated the set of voxels that passed an image - specific intensity threshold , then chose the largest connected group of such voxels . to identify the largest connected group , and at many other steps of the analysis , we consider a set of voxels as an adjacency graph . in an adjacency graph , the nodes are the voxels , and the edges are connections between neighboring nodes . we considered two voxels to be connected if they were adjacent in any sense , that is , the voxels touch at at least one corner . the distance along the edge is the distance between the center points of neighboring voxels . in the language of graph theory , the largest connected group of voxels that pass a given threshold is the largest connected component of the adjacency graph . a typical network mask is shown in panel 2 of fig . ( [ fig : overview ] ) . we use only the largest connected component because smaller groups of voxels are either noise , patient motion artifacts , fatty tissue , or isolated vessels whose relationship to the rest of the network can not be determined . finding the largest connected component is the rate - limiting process at this stage , but its time is still @xmath67 for an @xmath27-voxel image . the threshold parameter affects how much of the network is visible . the value affects the volume measurement of each vessel segment to some degree , but the magnitude is expected to be small@xcite . at lower ( more permissive ) thresholds , more of the network is visible , but dimmer objects that are not blood are more likely to be part of the largest connected component . fortunately , these misidentified objects are readily identifiable to a human observer , allowing us to easily choose a threshold at which no such objects appear . the optimal threshold is the lowest threshold that does not misidentify any objects . we chose an optimal threshold for each image using a manual binary search . the skeletonization process removes voxels from the network mask until only those required for maintaining a single connected component remain . without some initial set of non - removable nodes , this process would remove all nodes . therefore , it is important to reliably determine a set of non - removable nodes network endpoints that represent the most distal visible part of each vessel branch . given the network endpoints , skeletonization will reduce the network mask to centerlines , but skeletonization itself can not identify the endpoints , because they may be removed without disconnecting the graph . failure to identify endpoints before skeletonization results in the loss of vessel segments from detection ( see methods ) . we identify endpoints as the local maxima of a distance transform starting from an interior voxel . distance transforms are discussed in the next section . this transform assigns higher values to voxels that are more distal in the network , measured along the contours of vessels . voxels at a local maximum of distance are the network endpoints . we consider a local maximum any voxel whose distance value is greater than all of its neighbors . since this sometimes leads to multiple endpoints in a single terminal vessel , we collapse endpoints that are within the largest vessel radius ( 7 mm ) of one another and joined by a straight line through the network mask . we use skeletonization to identify vessel centerlines and branch points . a skeleton is an irreducible set of voxels that connect a set of endpoints . that is , the removal of any non - endpoint voxel in the skeleton breaks its adjacency graph into at least two connected components . thus , we compute our skeleton by recursively removing ( eroding ) voxels from the network mask , provided that they do not break the new mask into two components . there typically does not exist a unique skeleton for a given network mask and set of endpoints . we chose a particular skeleton in which the remaining voxels conform to the blood vessel centerlines by preferentially removing nodes from the outside ( surface ) of the mask first . to remove outermost voxels first , we use a distance transform to rank the voxels according to a measure of how close they are to the outside of the mask . we chose a measure that is lower for voxels closer to the exterior of the mask , and lower when a voxel has more neighboring voxels outside the mask . by removing voxels with more outside neighbors first , we remove voxels at convex points on the surface , smoothing the surface during erosion . a distance transform @xmath68 of a voxel @xmath69 has the property that @xmath70 where @xmath71 is the set of @xmath69 s 26 neighbors and @xmath72 is the euclidean distance from @xmath69 to a neighbor @xmath73 , which depends on whether the neighbors share one , two , or four corners ( ie , a face ) . this property is valid except when @xmath69 is the origin , or part of an initial boundary condition . given a value @xmath68 at any set of voxels , this property ( [ eq : dt ] ) can be used to extrapolate the distance transform to any connected set of voxels using a greedy algorithm . we use an analog of dijkstra s algorithm@xcite . we supply the values @xmath68 at points on the surface the boundary condition . we choose the boundary to be the set @xmath74 such that @xmath75 of voxels with any neighbors outside the network mask @xmath76 , and define @xmath68 on the boundary to be @xmath77 this inverse sum of inverses assigns lower scores to voxels with more neighbors outside the network mask . erosion requires checking each voxel for whether its removal disconnects the remaining voxels . calculating the number of connected components of the remaining voxels is computationally expensive ( @xmath78 ) , making the erosion process notoriously slow@xcite . however , we greatly speed our algorithm by exploiting the fact that removal of a voxel can not disconnect the whole graph unless it disconnects its neighbors . checking the last condition is fast ( @xmath79 ) a few passes of removing such voxels eliminate most of the voxels in the network , creating a close approximation to the skeleton . to erode the last few removable voxels , we use the whole - network algorithm , calculating connected components for each remaining voxel . this novel algorithm produces a skeleton in a small fraction of the time . because the skeleton is not unique , different erosion algorithms result in different skeletons . for our 20 images , our erosion leads to adequate representations of vessel centerlines . we expect any erosion algorithm that preferentially removes voxels roughly according to the above criteria to well - represent the centerlines and give similar results . using the skeleton , we can detect which voxels correspond to the branch points of the blood vessel network . the skeleton itself is a set of voxels , and hence its adjacency graph is not a tree because small cycles occur near branch points . thus , we first compute a minimal spanning tree of the skeleton . in this tree , nodes with more than two neighbors exactly correspond to branch points , and nodes with only one neighbor are vessel endpoints . branch points and endpoints demarcate vessel segments . we consider the length of a vessel segment to be the distance along the tree between the segment s ends . this distance is the sum of distances between the centers of adjacent voxels , which are either @xmath80 , @xmath81 or @xmath82 times the voxel width . this path length along the voxel grid is greater than or equal to the euclidean distance between two points , and thus quantization of the vessel centerline will cause lengths to be overestimated in comparison to other studies which do not require vessel contours to conform to the grid@xcite . much as with the manhattan metric ( discrete @xmath83 norm ) , the ratio between the grid distance and euclidean distance is independent of the scale of the grid . that is , at any given orientation relative to the grid , the bias is a fixed ratio independent of line length . this error is correctable with smoothing , but because our use of length measurements is to determine scaling exponents ( relative measures ) , our results are not affected by biases that increase length of vessels by a constant factor . consequently , we choose to ignore biases of this type . given the centerlines of vessel segments , we can attribute each voxel in the network mask to the vessel segment whose centerline lies closest to it . to accomplish this , we use dijkstra s algorithm to generate a shortest path tree in which any path along the center lines is zero distance . removing all branch points then breaks the shortest path tree into one connected component per vessel segment . the connected component of each segment contains all information about the segment size , shape , and spatial position . the volume of a segment is the number of voxels multiplied by the volume of each voxel . we compute the radius of each segment from the length and volume as @xmath84 . this is effectively equivalent to averaging multiple measurements of radius along the vessel , resulting in a low error in radius . we also label vessel segments and record their topology to compute the number of downstream endpoints , scaling exponents @xmath11 and @xmath13 , and scaling ratios @xmath17 and @xmath18 . segment decomposition leads to erroneous vessel segments when : 1 . a closely - spaced bundle of vessels is identified as a single vessel ; 2 . endpoint identification misses an endpoint ; 3 . the network mask contains a loop ; or 4 . patient motion artifacts cause large volumes of blood to appear as patches which are skeletonized as individual blood vessel segments . case 1 introduces segments with erroneously large radii into the distributions , but it occurs rarely , and usually in conjunction with a loop . cases 2 and 3 cause one segment to be missed , and its voxels attributed to adjacent segments , but these malformed segments are detectable . most of the voxels of tubular vessel segments are within a distance from the centerline less than the average vessel radius . we eliminate any vessel in which more than 20% of the voxels are further than @xmath63 from the centerline , or whose total volume is less than 4 voxels . this criterion also eliminates most of the vessels erroneously introduced in case 4 . * sensitivity of the measured scaling exponents to added image noise ranging from 1 to 10 times the baseline rate ( 0.47% ) . * the top left panel shows the number of vessel segments used to calculate the four scaling exponents for vessel radius versus the magnitude of the added noise . the top right panel shows the number of vessel segments used to calculate the four scaling exponents for vessel length versus the magnitude of added noise . the bottom left panel depicts how the four calculated scaling exponents for vessel radius vary with the magnitude of added noise . the horizontal lines indicate the wbe predictions for large and small vessels . the bottom right panel depicts how the four calculated scaling exponents for vessel length vary with the magnitude of image noise . the horizontal line indicates the wbe prediction . * vessel segment dataset . * the complete , raw output of angicart on the 18 images used in the study , in tab - separated values ( tsv ) format is available at + ` https://github.com/mnewberry/angicart/raw/example/example.dicom_small.all.tsv `

we reformulate the adiabatic inflow - outflow ( adios ) model for radiatively inefficient accretion flows , treating the inflow and outflow zones on an equal footing . for purely adiabatic flows ( i.e. , with no radiative losses ) , we show that the mass flux in each zone must satisfy with , in contrast to previous work in which is a free parameter but in rough agreement with numerical simulations . we also demonstrate that the resulting two - zone adios models are not dynamically self - consistent without the introduction of an energy source close in to the central regions of the flow ; we identify this with the energy liberated by accretion . we explore the parameter space of non - radiative models and show that both powerful winds and gentle breezes are possible . when small radiative losses ( with fixed efficiency ) are included , any centrally injected energy flux is radiated away and the system reverts to a power - law behavior with , where falls in a small range determined by the fractional level of radiative losses . we also present an adios model for hypercritical ( super - eddington ) disk accretion , in which the radiative losses are closely related to the flow geometry . we suggest that hyperaccretion can lead to either winds or breezes . accretion , accretion discs black hole physics hydrodynamics

1110.5356

radiatively inefficient accretion onto black holes can occur in the limits of both low and high accretion rates . although black holes can hypothetically absorb the gravitational energy liberated during accretion ( begelman 1979 ) , in the realistic case of a rotating accretion flow the accretion tends to be strongly inhibited , with much less gas reaching the black hole than is supplied at large distances . the reason for this is that the outward transport of angular momentum also involves the transport of energy , which causes the gas far from the black hole to become unbound ( narayan & yi 1994 , 1995 ) . there is ample observational evidence that black holes can be fussy eaters . " the galactic center black hole appears to be accreting at a rate orders of magnitude below the bondi rate ( melia & falcke 2001 , and references therein ) , as are the black holes in some elliptical galaxies with weak x - ray sources ( di matteo et al . 2000 ; mushotzky et al . 2000 ; baganoff et al . these systems are believed to be radiatively inefficient because their densities are low enough to inhibit cooling . at the other extreme , cooling is suppressed in hypercritical ( super - eddington ) accretion flows because radiation is trapped by the large optical depth . hyperaccreting sources such as the galactic binary ss433 ( king , taam & begelman 2000 ; begelman , king & pringle 2006 ) , galactic microquasars during outburst ( mirabel & rodriguez 1999 ) , and possibly ultraluminous x - ray sources ( king 2008 ; gladstone , roberts & done 2009 ) all appear to produce powerful outflows . numerical simulations of nonradiative accretion onto black holes also show this inhibition , either through the development of a zonal structure with inflowing gas near the rotational equator , sandwiched between outflowing zones at high latitudes , or through convective motions internal to the flow ( e.g. , stone , pringle & begelman 1999 ; igumenschev & abramowicz 1999 , 2000 ; stone & pringle 2001 ; hawley & balbus 2002 ; mckinney & gammie 2002 ; igumenshchev , narayan & abramowicz 2003 ; de villiers , hawley & krolik 2003 ; yuan & bu 2010 ) . blandford & begelman ( 1999 , 2004 ; hereafter bb99 and bb04 ) proposed a simple model for radiatively inefficient accretion flows , dubbed the adiabatic inflow - outflow solution " ( adios ) . in their steady state , self - similar formulation , there are inflowing and outflowing zones with equal and opposite mass fluxes that vary with radius according to @xmath0 with @xmath5 . in addition , there is presumably a small constant mass flux that reaches the black hole and is accreted but this need not be modeled in detail . in the one - dimensional , one - zone models developed in bb99 ( and presented using a different parameterization in bb04 ) , the vertically averaged conservation laws and dynamical constraints applied only to the inflowing zone impose well - defined relationships among the parameters describing the model , such as the specific angular momentum of the gas , its effective binding energy ( characterized by the bernoulli function ) , and the mass flux index @xmath4 . interestingly , @xmath4 is not determined by these relations , but can take on any value between 0 and 1 . in addition to analyzing some of the two - dimensional characteristics of the inflow zone ( and showing that the 1d models provide an excellent first approximation ) , bb04 considered the outflow zone in some detail . in particular , they showed that outflowing streamlines can be self - consistently patched onto the inflow zone at each radius . by assuming that the streamlines are adiabatic , laminar , and inviscid , they produced a family of self - similar wind models to represent the outflow zone . while the bb04 models provide a proof of concept " for adios , they do not necessarily provide an accurate depiction of even the qualitative features of a realistic zonal flow . in particular , the assumption of a wind with adiabatic , inviscid streamlines leads to very different kinds of flows in the inflow and outflow zones . while the inflow models depict a highly turbulent , well - mixed flow , the outflow models consist of a well - ordered flow pattern , with streamlines at higher latitudes reflecting conditions much closer to the black hole than streamlines close to top of the inflow zone . for example , since the terminal velocity along each streamline scales with the escape speed at its footpoint , the outflow models of bb04 exhibit large shears in the vertical direction , with the speed approaching infinity " near the evacuated cone that defines the wind s centrifugal barrier ( see , e.g. , fig . 6 of bb04 ) . numerical simulations that show the generic characteristics of an adios ( e.g. , stone et al . 1999 ; hawley & balbus 2002 ) often seem to exhibit much messier outflows than the bb04 models . the outflow region appears to be as chaotic as the inflow , with the typical speeds at each radius reflecting the local keplerian velocity . moreover , in adiabatic simulations one does not see a range of values for @xmath4 , but rather ( where this is measurable ) a tendency for the effective value of @xmath4 to be close to 1 . in this paper , we present a new formulation of adios in which the inflow and outflow zones are treated on an equal footing . given the success of the 1d inflow model as demonstrated in bb04 , we adopt a one - dimensional model for each zone and mainly consider self - similar flow . we show that the introduction of a relatively simple two - zone model has a dramatic effect on the character of the solutions : instead of obtaining solutions for any value of @xmath4 in the range @xmath6 , we find that non - radiative solutions must have @xmath1 . such flows are regulated by a conserved , outward flux of energy through the flow ignored in the bb99 and bb04 models and in fact require the presence of such a flux for self - consistency . the source of this flux must be the energy liberated by the residual accretion onto the black hole . the inflows associated with these solutions can range from marginally bound to nearly keplerian , depending on how large the net energy flux is and how it is partitioned between the inflow and outflow zones . outflow solutions come in two varieties : powerful winds with speeds of order the keplerian speed and gentle , subsonic breezes " driven by viscous stresses . flows with self - similar radiative losses have mass flux indices @xmath7 that depend on the rate of energy loss . these flows also depend on the existence of a central energy flux , but the net flux in this case is gradually eroded by radiative losses . taking the limit as the radiative loss rate goes to zero , we find that these solutions also approach @xmath1 . the inflows in these cases are always relatively tightly bound . finally , we present two - zone adios models for hyperaccretion , i.e. , super - eddington accretion , in which the radiative losses are tied closely to the geometrical and dynamical structure of the flow . the special form of the energy loss rate destroys self - similarity on large scales , but at sufficiently small radii we can model the radiative effects as small corrections to nonradiative flows . by extrapolating these effects to scales where they become nonlinear , we develop an approximate picture for the flow patterns , luminosities and accretion rates of hyperaccreting systems . we suggest that there may be two modes of hyperaccretion , corresponding to the wind and breeze solutions mentioned above . the plan of the paper is as follows . in section 2 we review the one - zone adios model of bb99 and bb04 , introduce the equations for the two - zone model , and show that two - zone models without a net energy flux are not self - consistent . in section 3 , we explore the parameter space for perfectly non - radiative two - zone models and in section 4 we consider the effects of ( self - similar ) radiative losses . in section 5 we examine the constraints associated with dissipative heating . we present our model for hyperaccretion in section 6 . finally , we discuss our results and summarize our conclusions in section 7 . we first review the one - zone " adios model proposed by bb99 and bb04 , referring readers to section 2.1 of bb04 for further details . adopting the conventions that the local accretion rate @xmath8 and outward viscous couple @xmath9 are positive ( where @xmath10 is used in bb04 to represent the viscous couple per unit mass ) , we obtain the following equations expressing the conservation of mass , angular momentum and energy in a steady state : [ angmom1 ] ddr(g - m l ) = - ( 1 + ) l dmdr [ energy1 ] ddr(g - m b ) = ( -1 ) b dmdr , where @xmath11 is the specific angular momentum , @xmath12 is the angular speed , and [ bernouilli ] b = l^22r^2 - 1r + h is the bernoulli function when @xmath13 ( where we use a roman typeface for the gravitational constant to avoid confusion with the viscous couple ) and @xmath14 is the specific enthalpy ( @xmath15 for an ideal gas with adiabatic index @xmath16 ) . the terms on the right - hand side of equations ( [ angmom1 ] ) and ( [ energy1 ] ) represent the rates at which angular momentum and energy are lost to the outflow , per unit radius . for power - law solutions with @xmath0 , @xmath17 , @xmath18 , these forms for the angular momentum and energy loss are fully general with the appropriate choice of the constants @xmath19 and @xmath20 . in addition to the conservation laws , the inflow is subject to a dynamical equilibrium condition based on the radial component of the momentum equation . given the self - similar scalings for pressure and density , @xmath21 and @xmath22 ( bb99 ) , and assuming that the inflow is very subsonic , we obtain @xmath23 eliminating @xmath24 in favor of @xmath25 , one finally obtains a three - parameter family of adios models . the independent parameters may be taken to be @xmath26 , @xmath27 and @xmath28 , or @xmath4 , @xmath19 and @xmath20 , or any combination of these . bb04 show that solutions can be found for any value of @xmath4 in the interval @xmath5 . in this interval , the constants of integration for the conservation equations may be neglected and one obtains simple relations among the various quantities . in particular , one can show that the bernoulli function satisfies [ bg ] b = glm r^2 1-n 1 - n . since all quantities on the right - hand side of eq . ( [ bg ] ) are positive with the possible exception of @xmath29 , we must have @xmath30 in order for @xmath25 to be negative and the inflow to be bound . but one can also show that @xmath31 as @xmath32 . the sign of @xmath25 therefore appears to be undefined in this limit . indeed , if one integrates the energy conservation equation in the limit @xmath32 , one finds that the left - hand side must integrate to a constant , whereas the right - hand side diverges logarithmically if @xmath33 . we conclude that the constant of integration which physically represents a conserved radial energy flux through the disk can not be neglected in this limit . let us now generalize the one - dimensional adios formulation to include the outflow as well as the inflow . in bb04 , the outflow was treated in a fundamentally different way from the inflow . by patching an adiabatic , inviscid streamline onto the inflow at every radius , bb04 showed how the outflowing gas could carry away the required amounts of angular momentum and energy . however , the resulting flow patterns exhibit enormous shear in the poloidal ( @xmath34 ) direction and large angular gradients of temperature and density . this is because each streamline reflects the conditions prevailing at the radius where it was launched : streamlines launched near a given radius will lie close to the inflow - outflow boundary and have speeds close to the escape speed at that radius , whereas streamlines closer to the axis will have come from smaller radii and have correspondingly larger speeds and higher specific entropies . simulations , however , suggest that this view of independent , inviscid streamlines may not be accurate . they suggest that the outflow is as chaotic as the inflow and is much better described as being well - mixed . " this suggests that it might be better to treat the mean properties of the outflow at each radius , rather than trying to keep the properties of the individual streamlines separate . in this spirit , we treat the outflow using the same parametrization as the inflow . denoting the inflowing ( outflowing ) zones by subscripts 1 ( 2 ) , and keeping the sign conventions from section 2.1 ( and bb99 , bb04 ) we write down a set of coupled conservation equations : [ angmom2 ] ddr(g_1 - m_1 l_1 ) = - g_12 , [ angmom3 ] ddr(g_2 + m_2 l_2 ) = g_12 , [ energy2 ] ddr(g_1 _ 1 - m_1 b_1 ) = - q_12 - _ 1 , [ energy3 ] ddr(g_2 _ 2 + m_2 b_2 ) = q_12 - _ 2 , where @xmath35 and @xmath36 represent the fluxes of angular momentum and energy per unit radius , respectively , transferred from zone 1 to zone 2 , and @xmath37 and @xmath38 represent net radiative losses from each zone . by mass conservation , the mass fluxes in zones 1 and 2 must be related according to @xmath39 , where @xmath40 is a constant representing the flux of gas that reaches the central object . we can therefore drop the subscript on @xmath41 and write @xmath42 . the angular momentum and energy fluxes transferred between zones cancel when the pairs of conservation equations are summed . we can integrate the summed pairs of equations , to obtain [ fl ] m ( l_1 - l_2)= g_1 + g_2 + f_l - m_0 l_1 , [ fe ] m ( b_1 - b_2)= g_1 _ 1 + g_2 _ 2 - f_e + ^r_r_0 ( _ 1 + _ 2 ) dr - m_0 b_1 , where the constants of integration @xmath43 and @xmath44 represent outward fluxes of angular momentum and energy , respectively , and we assume that the viscous stresses vanish at the innermost radius @xmath45 . since we are interested in radii far outside the inner boundary , where @xmath46 , we can neglect the terms containing @xmath40 in both equations . since @xmath47 is an increasing function of @xmath48 , we can neglect @xmath49 as well . since the viscous stress is assumed to be directed outward , eq . ( [ fl ] ) then implies that @xmath50 , i.e. , that the inflow is spinning more rapidly than the outflow at each radius . in the one - zone adios model , bb99 and bb04 neglected @xmath44 as well , which is justifiable for self - similar solutions with @xmath7 . in the two - zone case , however , this assumption yields the curious result that @xmath51 , i.e. , that the inflowing gas is _ less _ bound than the outflowing gas . since the motivation for considering adios models is the tendency of accreting gas to become unbound if it has no way to give up energy or angular momentum , this result seems suspect . indeed , we shall show below that it is also incompatible with the condition of dynamical equilibrium . to assess the self - consistency of the two - zone models , we introduce a generalized version of the dynamical equilibrium condition . we need to include inertial forces in the outflow zone , where the speed ( @xmath52 ) may be close to sonic . we continue to assume that the inflows are driven by viscous stresses at ( presumably ) very subsonic speeds . when mass is being transferred from the inflowing zone to the outflowing zone , the introduction of slowly moving gas emulates a drag force in the outflow region . for a power - law model with @xmath0 , the resulting deceleration is given roughly by [ inertial ] - v_2 2r _ 2 mr = -n v_2 ^ 2r , where we ignored the speed of the inflowing gas . the radial momentum equation for zone 2 then takes the form @xmath53 where we have defined an isothermal sound speed @xmath54 . the radial momentum equation for zone 1 takes the form @xmath55 similar except for the neglect of dynamical terms . as in bb04 , we ignore the angular correction terms associated with the finite scale height of each zone , effectively assuming that each zone is geometrically thin . this will lead to errors of order @xmath56 , where @xmath14 is the scale height . by comparing the one - dimensional models to two - dimensional ( gyrentropic ) models , bb04 showed that the 1d models provide an excellent approximation , at least for the inflow zones . since the outflow zones sandwich the inflow , we are also implicitly assuming that the @xmath57 is substantially larger than @xmath58 . as in the one - zone models , we can eliminate one of the variables in each zone in favor of the bernoulli function . we modify the bernoulli function for zone 2 to include the radial kinetic energy : [ ber2 ] b_2 = v_2 ^ 2 2 + l_2 ^ 22 r^2 - 1r + -1 a_2 ^ 2 , with a similar expression ( absent the dynamical term ) for zone 1 . if we subtract eq . ( [ mom2 ] ) from eq . ( [ mom1 ] ) and eliminate @xmath59 , we obtain @xmath60 since the coefficient multiplying @xmath61 is always negative , eq . ( [ delb1 ] ) shows that @xmath62 and @xmath63 must have _ opposite _ signs if dynamical terms are negligible . this contradicts the conservation laws , equations ( [ fl ] ) and ( [ fe ] ) , which demand that these quantities have the same sign in the zero - flux ( @xmath64 ) limit . in order to change the sign of @xmath62 , the term proportional to @xmath65 must be sufficiently positive and large . a necessary condition is [ nlimit ] n < 5 - 32 ( + 1 ) , which restricts @xmath4 to values between 0 ( for @xmath66 ) and @xmath67 ( for @xmath68 ) . to check whether consistency is possible even for such low values of @xmath4 , we use the momentum equation to eliminate @xmath65 in favor of @xmath59 . we obtain @xmath69 since @xmath70 , the coefficient of @xmath61 is still negative , implying that the other term must be positive . but in the coefficient of @xmath59 both the numerator and denominator are positive , implying that @xmath71 . this would mean that the outflow must be colder " than the inflow , which seems unlikely given that it lies on top of the inflow , is in vertical pressure balance with it , and must have a higher specific entropy since it consists of adiabatic gas that has already passed through the inflow and has suffered dissipative heating in the process . thus , it appears that the constraints of _ dynamical _ equilibrium in a two - zone adios are incompatible with the constraints of energy and angular momentum conservation , if the net energy flux passing through the system is negligible . this bolsters our original argument , based on the implausibility of @xmath62 being positive on energetic grounds . how is this paradox resolved ? the only reasonable conclusion seems to be that @xmath72 and that a two - zone adios model requires a substantial flux of energy passing through the system from small radii . if radiative losses are negligible , this flux is conserved . but in the presence of radiative losses , this flux must be large enough to both change the sign of @xmath62 at small radii and compensate for the integrated radiative losses further out . in the limit of negligible radiative losses , this implies that @xmath73 is independent of radius and @xmath4 is not a free parameter , but must instead be @xmath74 . we first consider the case in which radiative losses can be neglected , and therefore set @xmath75 . compared to the one - zone adios model ( bb04 ) , the two - zone model would appear to be more restrictive since @xmath1 and @xmath76 . however , two - zone adios models still form a three - parameter family because the energy flux @xmath44 is split into a portion that travels through the inflow and a portion that travels through the outflow . defining [ g12 ] g_12 = ( 1 + ) l_1 d m d r , we can take @xmath19 to be a third parameter representing the lever arm " associated with angular momentum transfer , with a permissible range @xmath77 . on physical grounds , we expect values of @xmath78 to apply if organized magnetic fields mediate the transfer of angular momentum ( bb04 ) . denote the dimensionless energy flux by [ mudef ] ( where we recall that @xmath79 ) and let @xmath80 be the fraction of the flux that propagates through the outflow . if we define the dimensionless variables [ dimensionless ] g ; = r^3/2 ; br b , with subscripts 1 or 2 as appropriate , then the integrated conservation equations ( [ angmom2])([energy3 ] ) become @xmath81 we therefore have four conservation equations relating the 6 dimensionless variables @xmath82 , and @xmath83 note that @xmath8 is not an independent variable , since solutions are invariant under transformations that keep @xmath84 and @xmath85 fixed . in addition , we have two dynamical equilibrium conditions , which we can cast in terms of the bernoulli function by eliminating the sound speed @xmath86 and introducing a dimensionless radial speed @xmath87 : @xmath88 \\ \label{dimlesseqn2 } b_2 & = & { 3 - \gamma \over 3 ( \gamma -1 ) } \left [ 1 - { 3 + \gamma \over 2 ( 3 - \gamma ) } \ell_2 ^ 2 \right ] + { 5\gamma - 3 \over 6 ( \gamma - 1 ) } u_2 ^ 2 . \end{aligned}\ ] ] we thus have 6 equations for the 7 variables ( including @xmath89 ) , which formally define a line through the @xmath90 parameter space . however , we can make the assumption that the viscous stress , in either zone , is capable of driving only very subsonic radial velocities . therefore , if @xmath91 is non - negligible , then @xmath89 is negligible ; conversely , a dynamically significant outflow speed implies that @xmath92 , for our purposes . we can therefore consider two distinct cases , for each of which we have 6 variables and 6 conditions . we refer to these two cases as wind " solutions ( @xmath93 ) and breeze " solutions ( @xmath94 ) , respectively . given a complete set of parameters @xmath95 , the two - zone adios model should be completely determined in each limit . as the wind solution is slightly simpler algebraically , we start by analyzing it . setting @xmath96 in the dimensionless equations , we immediately obtain [ b2wind ] b_2 = ; _ 2 = 23 ( 1 + ) _ 1 . since @xmath97 , the outflow is always unbound in the wind limit . the equations for the inflow region are self - contained , i.e. , they do not depend on quantities from zone 2 , implying that the same solution applies for both the wind and breeze cases . the bernoulli function in zone 1 is given by eq . ( [ dimlesseqn1 ] ) , but an alternate expression can be derived by eliminating @xmath98 between the energy and angular momentum equations : [ b1wind ] b_1 = 13 ( 1 - 2)_1 ^ 2 - ( 1- ) . comparing the two expressions , we obtain a simple expression for @xmath99 in terms of the input parameters : [ ell1 ] _ 1 ^ 2 = 2 3-+ 3 ( -1)(1-)3 + + 2(-1)(1 - 2 ) . given @xmath99 , we can substitute from eq . ( [ b2wind ] ) into eq . ( [ dimlesseqn2 ] ) and solve for @xmath100 . we are now able to map out the parameter space for wind solutions . the three constraints on physical solutions are : * @xmath101 , to ensure that the rotation is sub - keplerian and the pressure ( sound speed ) is positive ; * @xmath102 , to ensure that the inflowing gas is bound ; and * @xmath103 , a self - consistency condition on the wind solutions . these constraints are plotted in figures [ fig : wind5343 ] and [ fig : windeta43 ] . plane for @xmath104 and @xmath105 , with the angular momentum - transfer parameter @xmath19 set to zero . here and elsewhere , @xmath106 is the outward energy flux normalized to @xmath107 and @xmath80 is the fraction of that flux passing through the outflow zone . for each shaded region , the left boundary corresponds to an inflow zone with keplerian rotation ( zero - pressure ) , while the right boundary corresponds to marginally bound inflow ( @xmath108 ) . the lower boundary of each region corresponds to the limit of vanishing outflow speed ( @xmath94 ) . below this boundary , the outflow must be driven by viscous stress and the system is in the breeze " limit.,width=311,height=315 ] figure [ fig : wind5343 ] shows the parameter space available to wind solutions in the @xmath109 plane , for two values of @xmath16 . the angular momentum - transfer parameter @xmath19 is set to zero , corresponding to the purely local transfer of angular momentum from inflowing to outflowing gas , i.e. , there is no extra moment arm in the transfer due , for example , to an organized magnetic field ( bb04 ) . the permitted regions are bounded on the left and right by the conditions @xmath110 and @xmath108 , respectively . this ordering is somewhat counterintuitive , because @xmath80 represents the fraction of net energy flux that passes through the _ outflow_. for a fixed total energy flux @xmath106 , a larger value of @xmath80 means less energy flowing through zone 1 , yet the trend is for the binding energy of zone 1 to _ decrease _ with increasing @xmath80 . equally counterintuitive , we note that for fixed @xmath80 , an increase of total energy flux @xmath106 drives zone 1 from unbound flow toward a thin , cold state . the lower limit of each permitted region corresponds to zero outflow speed @xmath89 , clearly a constraint on the validity of the wind model . the fact that the allowed parameter space for @xmath111 is more restictive than for @xmath66 indicates that it requires more energy flowing through the system ( and a larger fraction of it flowing through the outflow zone ) to drive an outflow in @xmath104 gas . this is not surprising , given that @xmath104 is the softer equation of state . and different values of the angular momentum - transfer parameter @xmath112 . the more efficiently angular momentum is transferred from inflowing to outflowing gas , the lower the central energy flux that is required to drive a wind.,width=307,height=309 ] figure [ fig : windeta43 ] shows the effect of varying @xmath19 , with @xmath104 . as the efficiency of angular momentum transfer increases , less energy flux ( and a smaller fraction flowing through the outflow ) is required to drive a wind . in the limiting case @xmath113 , winds can be driven without any net energy flux through the system . to study the breeze solutions , we express @xmath114 in terms of @xmath115 by setting @xmath116 in eq . ( [ dimlesseqn2 ] ) . eliminating @xmath91 between the angular momentum and energy equations for zone 2 , we obtain a second equation for @xmath114 , in terms of @xmath115 and @xmath117 . eliminating @xmath114 between these equations , we obtain a quadratic equation for @xmath115 : [ ell2quad ] 7 - 3 6(-1 ) _ 2 ^ 2 - 23 ( 1 + ) _ 1_2 + - 3 - 3(-1 ) = 0 , the solution of which is @xmath118.\end{aligned}\ ] ] since @xmath117 is still given by eq . ( [ ell1 ] ) , we can solve for all the properties of the breeze solutions in terms of the input parameters . breeze solutions are subject to four constraints : * @xmath101 as before , to ensure that the rotation of the inflowing gas is sub - keplerian and the pressure ( sound speed ) is positive ; * @xmath102 , also as before , to ensure that the inflowing gas is bound ; * @xmath119 , the assumption that the stress points outward in the outflow , also ensuring that @xmath120 ; and * @xmath115 has real solutions according to eq . ( [ ell2quad2 ] ) , which serves as a self - consistency condition for the breeze solutions . from eq . ( [ ell2quad2 ] ) , one can easily see that the condition @xmath119 is well - satisfied when the discriminant in the expression for @xmath115 vanishes . one can also show that this condition continues to be satisfied at all values of the parameters for which a breeze solution exists . therefore , the third condition is superfluous and the parameter space of viable breeze solution is determined by the conditions on @xmath117 , @xmath121 , and the existence of a real solution for @xmath115 . these constraints are plotted in figures [ fig : breeze5343 ] and [ fig : breezeeta43 ] . all breeze solutions have unbound outflow , @xmath97 . plane for @xmath104 and @xmath105 , with the angular momentum - transfer parameter @xmath19 set to zero . for each shaded region , the left boundary corresponds to an inflow zone with keplerian rotation ( zero - pressure ) , while the right boundary corresponds to marginally bound inflow ( @xmath108 ) as in the wind solutions . the upper boundary of each region corresponds to the limit of vanishing discriminant in eq . ( [ ell2quad2 ] ) . above this boundary , no breeze solution exists.,width=307,height=277 ] as before , each solution space occupies a strip between the keplerian limit at the left and the limit of marginally bound inflow at the right . the upper boundary marks the locus on which the discriminant of the quadratic equation ( [ ell2quad ] ) vanishes . no physical breeze solution exist for parameters above this boundary . by comparing fig . [ fig : breeze5343 ] with fig . [ fig : wind5343 ] , one can also note that the parameter space for breeze solutions overlaps with the parameter space for winds ; thus , in principle one could have either a breeze or a wind model for the same set of parameters @xmath122 . figure [ fig : breezeeta43 ] shows the effect of varying @xmath19 , with @xmath104 . as in the wind case , a higher efficiency of angular momentum transfer decreases the energy flux required to drive a wind . and different values of the angular momentum - transfer parameter @xmath112 . the overlapping bands show the same trend as for the wind models , i.e. , that a higher angular momentum transfer efficiency reduces the energy flux needed to drive a breeze . note that the upper - right threshold for the parameter space is very insensitive to the value of @xmath19.,width=307,height=260 ] we are now in a position to combine our results for the breeze and wind models , to show the different possible types of solution and how they relate to the input parameters . in figures [ fig : windbreeze43 ] and [ fig : windbreeze53 ] we show the combined solution space for @xmath111 and @xmath105 , respectively . the first thing to notice is that wind and breeze regions overlap substantially . moreover , it is possible that both rapidly rotating ( plus sign in eq . [ [ ell2quad2 ] ] ) and slowly rotating breeze solutions may be possible , although the slowly rotating solutions are permitted only in a very narrow band of parameter space . ( this constraint is based on the assumption that the outflow rotates in the same sense as the inflow . ) these results provide a complete picture of the two - zone adios models when losses are neglected . we will now consider models that include self - similar radiative losses . parameter space of solutions for @xmath111 , with @xmath123 . both wind and breeze solutions may exist in the dark shaded area . slowly rotating breeze solutions ( i.e. , with the minus sign in eq . [ [ ell2quad2 ] ] ) may exist only in the narrow strip between the wind - only area and the wind or breeze " region ; elsewhere only rapidly rotating breeze solutions are allowed.,width=307,height=307 ] , but for @xmath105.,width=307,height=307 ] radiative losses are included on the right - hand side of equations ( [ energy2 ] ) and ( [ energy3 ] ) by specifying the form of @xmath37 and @xmath38 . to guarantee self - similar scaling for these losses , we follow narayan & yi ( 1994 ) and define a parameter @xmath124 , with @xmath125 , such that @xmath126 equals the net heating rate per unit radius . since the total viscous heating rate per unit radius is @xmath127 , we have @xmath128 , with a similar expression for zone 2 . note that @xmath129 corresponds to the nonradiative case we have been considering , while @xmath130 implies that all locally dissipated energy is radiated away . we also specify @xmath131 , as in bb04 . we define the dimensionless constants @xmath132 and @xmath133 as in eq . ( [ dimensionless ] ) , but now define @xmath134 and @xmath135 such that @xmath136 and @xmath137 are functions of @xmath48 . the dimensionless form of the cooling rate then has the form @xmath138 in zone 1 , with a similar form for zone 2 . under these assumptions , the conservation equations admit power - law solutions of the form [ radsolns ] \{m , g_1 , g_2 } = \ { a , c_1 , c_2 } r^-q , where @xmath139 , @xmath140 , @xmath141 , and @xmath142 are constants . we may choose the overall normalization arbitrarily , so we set @xmath143 . substituting eq . ( [ radsolns ] ) into the conservation equations , we can obtain solutions for arbitrary choices of @xmath19 , @xmath144 and @xmath145 , but for the purpose of this section we can simplify the algebra considerably by considering the case @xmath146 with @xmath147 . we then have [ c1 ] c_1 = _ 1 3 - 2 q and [ c2 ] c_2 = _ 1 - _ 2 - c_1 defining @xmath148 , we obtain [ yeq ] ( y - 3f)(_1 - _ 2)(_1 + y _ 2 ) + y ( y-3)(b_2 - b_1 ) = 0 , which appears to be a quadratic equation for @xmath149 ( or @xmath150 ) but is actually cubic because @xmath121 and @xmath114 depend linearly on @xmath150 . to see this , we modify the momentum equations ( [ mom2 ] ) and ( [ mom1 ] ) , replacing @xmath4 by @xmath151 . we then have @xmath152 \\ \label{b2q } \lefteqn{b_2 = { 1 \over ( \gamma -1)(3 + 2q ) } \left [ 3 - \gamma - 2q(\gamma-1 ) \right . \nonumber } \\ & & \left . - { 3 + \gamma -2q ( \gamma-1)\over 2 } \ell_2 ^ 2 \right]+{5\gamma - 3 -2q(\gamma -1 ) \over 2 ( \gamma - 1)(3 + 2q ) } u_2 ^ 2 . \end{aligned}\ ] ] given @xmath117 and @xmath150 , we can calculate @xmath20 from the equation [ betaq ] 1- = 11-q , which derives from eq . ( [ energy2 ] ) . these formal solutions to the equations suffer from the self - consistency problem described in section 2.3 unless the system carries a central energy flux @xmath153 to partially compensate for radiative losses . from eq . ( [ fe ] ) , however , we see that the presence of the energy flux would spoil the assumed power - law dependence unless [ ferad ] f_e = ^_r_0 ( _ 1 + _ 2 ) dr . in such a _ critically cooled _ solution , the system forgets " about the energy flux injected at the center and instead carries an @xmath48-dependent effective energy flux [ ferad2 ] ^_r ( _ 1 + _ 2 ) dr , which allows self - consistency to be maintained . we further note that if @xmath44 were smaller than the energy integral in eq . ( [ ferad ] ) , then no solution would be possible cooling would exceed the total energy supply . presumably , any radiative system will adjust so that the accretion rate yields a central energy flux at least large enough to balance cooling . conversely , if @xmath44 exceeded the energy integral , then radiative losses would have a minor effect and the nonradiative ( @xmath1 ) behavior would be preserved . below we consider critically cooled models in the breeze and wind limits . to extract the breeze solutions , we set @xmath116 in eq . ( [ b2q ] ) but make no assumption about @xmath154 . substituting for @xmath155 in eq . ( [ yeq ] ) and canceling a factor of @xmath156 , we obtain the cubic equation [ yeq2 ] ( y - 3f)(_1 + y _ 2 ) + y ( y-3)3 + - 2q(-1)2 ( -1)(3 + 2q ) ( _ 1 + _ 2 ) = 0 , where we recall that @xmath157 . breeze solution must satisfy the constraints : * @xmath158 ; * @xmath159 ; and * @xmath160 . the last condition implies @xmath161 . since @xmath162 , the second term on the left - hand side of eq . ( [ yeq2 ] ) is negative and we must have @xmath163 , which translates to @xmath164 . if @xmath102 as well , then eq . ( [ betaq ] ) implies that [ betaqmin ] 1 - > q 1-q . the allowed range of @xmath165 is determined by the second and third constraints . the actual value of @xmath117 ( or @xmath121 ) is irrelevant , since from eq . ( [ yeq2 ] ) we see that @xmath150 is a function of @xmath166 , not of either angular momentum separately . the second and third constraints imply @xmath167 ; we map this region in fig . [ fig : q ] . ( @xmath168 ) , normalized to the radiative fraction @xmath169 , as a function of @xmath169 for @xmath104 and @xmath66 , . for each case , the lower bound corresponds to vanishing angular momentum in the outflow , @xmath170 , while the upper bound corresponds to vanishing viscous stress , @xmath171 . note that @xmath150 vanishes @xmath172 in the limit of negligible radiative losses , thus recovering the case @xmath1 ( @xmath173).,width=318,height=304 ] one can derive these constraints analytically in the limit of negligible radiative losses , @xmath174 : [ qzero ] 3(- 1)3 + 1 < q1-f < 27 ( -1)23 - 3 . as the radiative fraction @xmath169 goes to zero , @xmath150 does as well and the solution approaches the case @xmath1 . for the wind case , we take @xmath175 , implying @xmath176 . combining equations ( [ c1 ] ) and ( [ c2 ] ) , we obtain [ ell2q ] _ 2 = 2_1 ( 1-q ) 3 - 2 q , which generalizes eq . ( [ b2wind ] ) for @xmath115 in the case @xmath146 . equation ( [ yeq ] ) can then be written in the form [ ell2q2 ] b_2 - b_1 = 3(1-f ) - 2q 2q ( 3 - 2q ) _ 1 ^ 2 = b_1q , where the second version of the equation comes from eq . ( [ energy3 ] ) with @xmath177 . these equations are identical to eq . ( [ betaq ] ) . wind solutions must satisfy the constraints : * @xmath158 ; and * @xmath103 . the latter condition places an upper limit on @xmath165 that is identical to the upper boundary for the breeze zone in fig . [ fig : q ] , i.e. , corresponding to @xmath171 . the allowed zone for wind solutions extends down to @xmath178 for all @xmath124 . the parameter space for breezes therefore completely overlaps with that for winds , in the @xmath179 plane . our two - zone models of adios flows have been based on the conservation laws of angular momentum and energy , supplemented by dynamical equilibrium conditions in both the inflowing and outflowing zones . in addition , these zones must obey the laws of thermodynamics , which describe how energy is distributed among various forms . consideration of the thermodynamics of dissipative heating can impose additional constraints on the possible solutions . since we have already found that models are completely determined by the choice of three input parameters ( @xmath180 ) plus , under certain conditions , an assumption about whether the outflow is a wind or a breeze with either fast or slow rotation it is clear that any further restrictions may lead to constraints on the possible values of these parameters . for example , it may not be possible for the constant energy flux through the flow , or its partition between inflowing and outflowing zones , to take on any value the allowed values may instead be dictated by thermodynamics . we are basically concerned with estimating the rate of entropy generation within each zone of the flow . under self - similar conditions with @xmath1 , the pressure scales with radius as @xmath181 and the density scales as @xmath182 implying that the specific entropy function scales as @xmath183 . the entropy thus increases with decreasing radius , as expected for inflowing gas that undergoes viscous dissipation ( but does not radiate ) as it drifts inward . however , the self - similar nature of our two - zone model implies that the same radial entropy scaling must also apply to the outflowing gas in other words , the specific entropy of the outflowing gas must decrease as the gas flows outward . this entropy decrease in the outflowing zone indicates that the entrainment of low - entropy gas from zone 1 must outweigh the entropy increase due to local dissipation . these arguments serve as a warning of the care required in attempting to estimate dissipative effects . setting @xmath184 , we write the vertically integrated entropy equation for zone 1 in the approximate form [ entropy ] -a_1 ^ 2 -1 m s_1 r + 2r a_1 ^ 2 -1 _ 1 v_z s z dz = -g_1 d_1 dr - d qdr , where the integral is over zone 1 and we have glossed over such subtleties as the difference between the vertical average of the sound speed used here ( recall that @xmath185 ) and those used in the dynamical equilibrium conditions . a similar equation applies in zone 2 , except that the first term on the left - hand side has the opposite sign . for now we ignore the @xmath186 term , which represents energy lost per unit radius due to conduction or radiation . we set @xmath187 , and assume that the second term on the left - hand side can be approximated by [ entropy2 ] a_1 ^ 2 - 1 dm dr s , where @xmath188 is the difference in specific entropy between the top and bottom of the zone under consideration . presumably @xmath189 in both zones 1 and 2 . writing @xmath190 in both zones , redefining @xmath86 so it is dimensionless ( @xmath191 ) , and using the dimensionless variables defined earlier , we obtain [ entropy3 ] s_1 = 3 ( -1 ) 2g_1 _ 1a_1 ^ 2 - 3 - 2 [ entropy4 ] s_2 = 3 ( -1 ) 2g_2 _ 2a_2 ^ 2 + 3 - 2 . thus , we see that in order to preserve the self - similar scaling , there must be a large entropy jump across zone 2 , but there need not be one across zone 1 . for the sake of making a crude estimate , we will assume that the entropy jump across zone 1 is negligible , and that all the entropy change due to mass transfer between zones occurs in zone 2 . in this limit , we have [ entropy5 ] g_1 _ 1a_1 ^ 2 = ( 1 - 2)_1 ^ 2 2(1 - _ 1 ^ 2 ) = 3 - 3 ( -1 ) , where we have used the equations of energy conservation and dynamical equilibrium in zone 1 to express the left - hand side in terms of @xmath117 . the entropy jump across zone 2 , which must be approximately equal to @xmath192 if we assume that @xmath193 , is likely to be even larger than the estimate given in eq . ( [ entropy4 ] ) , which only includes the @xmath194component of the stress . in addition to the possibility of a significant @xmath195 stress due to the vertical shear in angular velocity , there may be large @xmath196 stress . this will certainly be a factor for the wind solutions , where we have already had to take into account the drag on the outflow due to entrainment of gas with little radial velocity ( cf . [ [ mom2 ] ] ) . solving eq . ( [ entropy5 ] ) for @xmath99 , we obtain [ entropy6 ] _ 1 ^ 2 = 2 ( 3-)2 ( 3- ) + 3 ( -1)(1 - 2 ) . in the limit @xmath197 , angular momentum loss carries off all the accretion energy and we are left with a thin , keplerian disk . for all other cases , viscous heating is important and @xmath101 , as expected . using eq . ( [ dimlesseqn1 ] ) to express the inflow bernoulli function in terms of @xmath99 , we obtain [ entropy7 ] b_1 = - 2 ( 3-)2 ( 3- ) + 3 ( -1)(1 - 2 ) . for the case @xmath198 , @xmath199 and the disk is marginally bound . in this limit , the locus of permitted values of @xmath106 and @xmath80 would track the right - hand boundary of the solution space in figures [ fig : windbreeze43 ] and [ fig : windbreeze53 ] . for @xmath200 the disk is bound and the locus lies inside the parameter space . for @xmath201 , the locus hugs the left - hand boundary of the solution space and the bernoulli function has the keplerian value @xmath202 . it proves interesting to generalize our dissipative model to the case where @xmath203 is positive , e.g. , due to mixing with the hotter gas in the outflow zone . in that case , eq . ( [ entropy5 ] ) is replaced by the inequality [ entropy8 ] ( 1 - 2)_1 ^ 2 2(1 - _ 1 ^ 2 ) > 3 - 3 ( -1 ) . since the left - hand side is a monotonically increasing function of @xmath99 , this means that additional entropy generation leads to a colder , more rapidly rotating inflow , and to a more negative binding energy . this surprising result could explain why simulations of radiatively inefficient flows tend to exhibit inflow zones with clearly negative values of the bernoulli function , rather than values hovering close to zero ( e.g. , stone et al . 1999 ) . the effect of radiative losses is to replace the dimensionless stress @xmath204 in equation ( [ entropy3 ] ) or ( [ entropy4 ] ) by @xmath205 , according to the self - similar model of radiative losses adopted in section 4 . for the inflow zone , we obtain [ heating1 ] _ 1 ^ 2 = 2 ( 3-)2 ( 3- ) + 3 f(-1)(1 - 2 ) , confirming that increasing the losses ( decreasing @xmath124 ) makes the inflow colder and the rotation faster . the bernoulli function is [ entropy7 ] b_1 = - ( 3-)2 ( 3- ) + 3 f(-1)(1 - 2 ) . thus , as expected , an increase in radiative losses makes the inflow more bound . on the @xmath206 parameter space , this moves the possible solutions to the left of the zero - energy curve . if @xmath106 is fixed , this means that @xmath80 is smaller , corresponding to a smaller fraction of the energy flux passing through the outflow . to apply these ideas in a well - defined context , we consider the case of hypercritical accretion , in which matter is supplied at a rate that far exceeds the eddington limit . such a flow is radiatively inefficient because of the large optical depth ; indeed , the radiation becomes effectively trapped " in the flow ( advected faster than it can escape ) at roughly the radius within which the liberated power approaches the eddington limit ( begelman 1979 ) . radiative losses are determined using the vertical ( @xmath207 ) component of the momentum equation . since radiation from zone 1 has to pass through zone 2 , we set @xmath208 in eq . ( [ energy2 ] ) and absorb the radiative losses from zone 1 into the overall energy loss term @xmath36 . net radiative losses are thus given by @xmath209 . in considering the vertical dynamical equilibrium of zone 2 we must take account of the outflow speed , which can be appreciable . modeling the vertical velocity profile by @xmath210 , we find that the dimensionless vertical acceleration is @xmath211 . there is also a vertical drag force , associated with lifting material out of zone 1 and mixing it with the outflowing material in zone 2 ( cf . [ [ inertial ] ] ) , that gives a deceleration @xmath212 . inserting these terms into the momentum equation , we obtain [ edot1 ] _ 2 = 4 m cr ( 1 + u_2 ^ 22)h_2 r where @xmath57 is the height of the outflow layer , @xmath213 is the opacity , and we include the @xmath214 factor for clarity ( elsewhere we have taken @xmath13 ) . we have doubled the losses from one layer to take account of the fact that two radiating outflow layers sandwich the inflow zone . to estimate @xmath57 , we integrate the vertical momentum equation assuming a uniform density . this gives @xmath215 , where @xmath216 is the dimensionless isothermal sound speed on the equator . we therefore have [ edot2 ] _ 2 = 4 m cr a_2 ( 1 + u_2 ^ 22)^1/2 . we define a dimensionless accretion rate @xmath137 by @xmath217 and assume that matter is supplied at some large radius @xmath218 at a rate @xmath219 . if @xmath218 is taken to be the radius at which radiation is marginally trapped by the inflow , then @xmath220 . if the inner radius @xmath221 , then it is possible to develop a series solution in powers of a small parameter @xmath222 at radii @xmath223 . importantly , the dimensionless coefficients we have used to characterize the flows remain constant to first order in @xmath222 they vary with radius only at @xmath224 . in contrast , the mass flux and viscous stress develop a logarithmic radial behavior at first order in @xmath222 , which truncates the adios at @xmath218 . to see how this behavior arises , we develop solutions to first order for zone 1 , anticipating the radial dependence of @xmath137 ( which is governed by the form of @xmath209 ) by writing @xmath225 \\ g_1 & = & \dot m_a r^{3/2}\left[g_{1a } - g'_{1a } \delta \ln\left ( { r\over r_{\rm in } } \right ) \right ] , \end{aligned}\ ] ] where quantities denoted by a subscript @xmath86 " are regarded as zeroth order and quantities with a subscript @xmath132 " are first order . we correspondingly define @xmath226 and @xmath227 . substituting these forms into the conservation equations , we recover equations ( [ dimlesseqns ] ) and ( 23 ) as zeroth - order relations . to first order in the angular momentum equation , the terms proportional to @xmath228 give @xmath229 , implying that @xmath230 . the remaining first - order terms give the correction to the specific angular momentum , [ ell1b ] _ 1b = -23 _ 1a . identifying @xmath231 as in the nonradiative case , we obtain [ hyerbeta ] ( 1 - ) b_1a = ( 1- ) . in the presence of radiative losses , the energy transfer rate @xmath232 is no longer zero , but behaves as a first - order quantity . to proceed further , we need to use the perturbed version of the radial momentum equation . as before , we have @xmath233 , so that @xmath234 with @xmath235 to first order . since @xmath236 with @xmath237 to first order , the density must contain the same first - order corrections as @xmath137 , and we have @xmath238 . the equilibrium condition is then @xmath239 which we solve to first order after expanding @xmath99 and taking @xmath240 . specializing to @xmath104 in the bernoulli function , we obtain @xmath241 a similar procedure can be carried out for zone 2 , but here we quote only the important result obtained by comparing terms in the integrated , summed energy equation ( [ fe ] ) . noting that @xmath242 and that @xmath243 , where @xmath244 is the eddington luminosity , we find that [ mudef ] = a_2 m_a ( 1 + u_2 ^ 22)^1/2 . the correction term in eq . ( [ mdota ] ) is negative , implying that the adios is quenched at large radii by radiative losses . in particular , if we extrapolate the logarithmic correction out to radii at which the correction is of order unity ( recognizing that the approximation is a very crude one in this limit ) , we find that @xmath8 has a maximum at a radius which we identify as @xmath218 , implying [ rout ] ^-1 = 1 + ( r_out r_in ) = m_a m_out . for exponentially large values of @xmath245 , as we expect in hyperaccreting systems such as ss433 ( where this ratio may be @xmath246 ) , @xmath222 may fall in the range @xmath247 , justifying our approximation scheme . to put this on a more quantitative footing , we consider solutions for @xmath104 , @xmath146 ( as seems reasonable for winds driven primarily by the effects of radiation pressure ) . figure [ fig : hyperwindbreeze ] shows solution curves for eq . ( [ rout ] ) in the @xmath109 plane for different values of @xmath248 , superimposed on the combined parameter space for winds and breezes from fig . [ fig : windbreeze43 ] . all three breeze solutions ( solid curves ) are self - consistent where they cross the lower shaded ( breeze ) region . wind solutions with @xmath249 can also be self - consistent , since these cross the two upper shaded regions where winds are permitted . however , according to our simple model there are no wind solutions for @xmath250 . a striking feature of the wind curves is their sensitivity to the value of @xmath248 . as noted earlier , we generally expect @xmath220 , since this is the condition that radiation trapping be marginally effective at the outer edge of the adios . but it makes a qualitative difference whether the value of @xmath248 is 1/2 or 1 the former case permits either breezes or winds , whereas only breezes seem to be compatible with the latter . our results suggest that there may be two distinct modes of hyperaccretion breezes or winds and opens the possibility of a system switching between them . we will investigate the implications of this possible bimodal behavior in a separate paper . plane , superimposed on the combined parameter space for @xmath104 winds and breezes from fig . [ fig : windbreeze43 ] . solid curves show breeze solutions for ( top to bottom ) @xmath251 , while dashed curves show wind solutions for the same values of @xmath248 . self - consistent solutions are possible only where a wind or breeze curve crosses the corresponding shaded region . thus , winds are possible only for @xmath252 , whereas breezes are possible for all three values of @xmath248 . the type of solution chosen may depend on the outer boundary conditions of the flow.,width=311,height=300 ] we have presented a new version of the adiabatic inflow - outflow ( adios ) model for radiatively inefficient accretion , treating the outflow on a similar footing to the inflow . instead of matching a laminar , adiabatic wind solution to a single - zone model for the inflowing region , we treat the outflow as a well - mixed " region characterized by vertically integrated fluxes of mass , energy and angular momentum . this two - zone formulation is simpler than the original adios models presented in bb99 and bb04 , and reflects the global conservation laws in a more transparent way . more importantly , it avoids the large velocity shears and entropy gradients predicted as a function of height in the original adios model . if the outflows are highly turbulent , then we argue that a vertically averaged description of the outflow may be more realistic . this simple modification of the adios model has fundamental implications for the parameter space of physically realizable systems . for perfectly adiabatic ( nonradiating ) systems , it implies that the dependence of the mass flow on radius is always described by @xmath1 , i.e. , @xmath253 , in contrast with the original formulation in which @xmath4 is an undetermined parameter with a value between 0 and 1 . moreover , a two - zone adios model with a well - mixed outflow requires a conserved energy flux to propagate through _ both _ the inflowing and outflowing zones . absent this external energy supply , there is not enough power to propel the outflow , while the inflowing gas is unable to relinquish enough energy to become bound ( in the sense that its bernoulli function is negative ) . in the original adios model , where this energy flux is set to zero , the inflow is able to remain bound only because it interacts with the local streamlines of the outflow , which carry much less energy per unit mass than the streamlines originating much closer to the central object . but in the two - zone model , large amounts of mass injected into the outflow at large @xmath48 mix with large quantities of energy liberated much closer in . we identify the conserved energy flux with the energy released by the flux of matter ( @xmath40 ) that actually reaches the black hole . we characterize this energy flux by two parameters , an overall energy efficiency @xmath106 defined by the ratio of the energy flux to @xmath254 ( which is independent of radius ) , and a parameter @xmath80 , describing the fraction of @xmath106 that propagates through the outflow . physical models exist only for a fixed range within the interval @xmath255 that narrows with increasing @xmath106 . if @xmath106 could reach values @xmath256 , the required value of @xmath80 would approach 1 , implying that nearly all the energy must propagate through the outflow . we stress that @xmath106 is not an accretion efficiency in the usual sense , because it compares the energy liberated by the accreted flux @xmath40 to the binding energy of distant matter that never reaches the black hole . if the actual accretion efficiency is @xmath257 , so that the energy flux is @xmath258 , then the accretion rate is [ mdotzero ] m_0 = 1- . a larger value of @xmath106 merely means that equality between @xmath173 and @xmath259 const . occurs farther from the black hole . in this case , there would be a wide radial zone in which @xmath40 dominates the mass flow . such a region would resemble a unidirectional accretion flow ( similar to the early advection - dominated accretion flow [ adaf ] models of narayan & yu [ 1994 , 1995 ] ) , rather than an inflow - outflow system , and would therefore develop a positive bernoulli function ( bb99 ) . for this reason , we expect that two - zone adios models are restricted to values of @xmath260 or 3 . because our model is self - similar , we can not treat boundary conditions in detail . however , we suspect that the value of @xmath106 ( and therefore the accretion rate into the black hole ) and possibly the value of @xmath80 can adjust in response to the energetic demands of the system . a couple of our results suggest this . first , when weak radiative losses are introduced into the model in a self - similar way , the mass flux index @xmath4 drops below 1 by an amount that depends on the fractional energy loss per decade of radius . the effect of this is that @xmath40 determines the energy flux through the flow at small radii , but that the flow gradually forgets " about the inner boundary with increasing @xmath48 , i.e. , as the centrally injected energy is radiated away . the case of hypercritical accretion provides a more dramatic example , since solutions with a large ratio of outer - to - inner radius must have @xmath261 . apparently , matter supplied to a black hole at a highly super - eddington rate will also be accreted at a highly super - eddington rate , in order to supply enough energy to match radiative losses in addition to powering the flow . nevertheless , both @xmath106 and @xmath80 must fall into fairly narrow ranges in order for the flows to be self - consistent . ( the parameter @xmath20 , which describes energy transfer from inflow to outflow , must also self - regulate to @xmath74 for @xmath1 ) . despite the highly constrained parameter space , our solutions display a striking range of behaviors from gentle breezes , governed by viscous stress , to transsonic winds . such bimodal behavior extends to hypercritical accretion , where the dependence of radiative losses on flow geometry and dynamics breaks self - similarity close to the trapping radius . we note that shakura & sunyaev ( 1973 ) foresaw the possibility of both powerful winds and gentle outflows in their pioneering discussion of hypercritical accretion . both breezes and winds are unbound , in the sense that the bernoulli function is positive , but any significant radiative losses near the outer radii of the adios could quench the weaker outflows . it is possible that breeze solutions represent large - scale circulations rather than genuine mass loss , in which case the total mass of the system would grow with time if matter continues to be introduced from outside . the strength of the adios approach is that it relies on simple conservation laws and dynamical equilibrium conditions , which impose clear constraints on the space of possible flows . thus it can provide a useful framework against which numerical models or observations can be analyzed . beyond the application of conservation laws and dynamical relations , the adios idea relies on the assumption that some physical mechanism exists to bifurcate the available energy and to launch an outflow that is spatially distinct from the inflowing gas . in the case of the two - zone models presented here , it also assumes that the inflow and outflow zones are both well - mixed . these assumptions may not be valid , or may apply only in certain types of systems ( e.g. , depending on boundary conditions , magnetic field structures [ see , e.g. , beckwith , hawley & krolik 2008 , 2009 ] , or the relative importance of magnetic and radiative stresses ) . but the two - zone conservation laws and dynamical relations are very general and can be adapted to treat alternative assumptions about the flow pattern . for example , one can devise a two - fluid model that might capture the physics of a radiatively inefficient flow dominated by nonlinear convection , with the inflow and outflow zones interpenetrating one another . such a model could represent a convection - dominated accretion flow " ( cdaf : narayan , igumenshchev & abramowicz 2000 ; quataert & gruzinov 2000 ) . we will present such a model , and compare its properties to those of the two - zone adios model , in a forthcoming paper . i thank phil armitage , kris beckwith and roger blandford for helpful discussions . this work was supported in part by nsf grant ast-0907872 .

this paper investigates the excitation of various natural modes in a magnetized bi - ion or dusty plasma . the excitation is provided by parametrically pumping the magnetic field . here two ion - like species are allowed to be fully mobile . this generalizes our previous work where the second heavy species was taken to be stationary . their collection of charge from the background neutral plasma modifies the dispersion properties of the pump and excited waves . the introduction of an extra mobile species adds extra modes to both these types of waves . we firstly investigate the pump wave in detail , in the case where the background magnetic field is perpendicular to the direction of propagation of the pump wave . then we derive the dispersion equation relating the pump to the excited wave for modes propagating parallel to the background magnetic field . it is found that there are a total of twelve resonant interactions allowed , whose various growth rates are calculated and discussed .

physics0601193

the basic natural modes of magnetized plasmas such as those that occur in molecular clouds , cometary plasmas and stellar atmospheres are of great interest . when the frequencies are low then a class of linear waves , referred to collectively as alfvn waves , are known to exist and are of importance to the understanding of many basic plasma phenomena @xcite . the linear approximation to these waves breaks down at large amplitudes where nonlinear effects become important in their propagation . one such large amplitude wave is a magnetoacoustic wave , which modifies the background magnetic field in an oscillatory fashion , and so can be considered as a pump wave that drives other waves nonlinearly . such large - amplitude pump waves may occur in conditions such as seen in solar and space plasmas . for example , solar shock waves can set up large amplitude standing magnetoacoustic waves in coronal loops or magnetic flux tubes @xcite . by considering perturbations of this large amplitude pump , we can investigate the possibility of exciting natural modes of the system , such as alfvn waves , due to a resonant interaction between the pump wave and the natural waves . in the single ion species case this basic phenomenon was predicted by @xcite , and subsequently pursued by several authors @xcite . for a pump magnetic field parallel to the background magnetic field in a single ion species plasma , the excited waves are alfvn waves travelling in opposite directions along the magnetic field @xcite . if ion - cyclotron effects are included , the excited waves are the fast and slow ( ion cyclotron ) alfvn waves travelling in opposite directions @xcite . more general behavior is allowed in the case where the excited waves are permitted to travel obliquely to the magnetic field @xcite . the parametric excitation of waves in a dusty magnetized plasma has been investigated in ref.@xcite , but in the approximation where the dust is taken to be immobile . in that case the oppositely travelling pair of waves are modified , with the presence of a cutoff frequency in the fast alfvn wave . furthermore , an interaction between a pair of fast waves or a pair of slow waves was found to exist . the corresponding growth rates of the slow - fast and fast - fast pairs were maximized as a function of the dust concentration . many space and laboratory plasmas are multicomponent , i.e. , contain multiple ion species . it is therefore of great interest to investigate the effects of the additional species on the linear and nonlinear properties of the waves in the plasma . for instance , it is often the case that in the presence of an additional ion component an extra mode is excited , or forbidden regions of frequency are introduced . one area of study of multi - ion plasmas is the bi - ion plasma . the inclusion of one extra ion species often captures the basic information of several extra species . the bi - ion plasma has particular importance in plasma fusion , laboratory plasmas and in astrophysical environments , with the secondary ion usually positively charged . a dusty plasma adds another level of interest to the topic of multicomponent plasmas , due to the dust properties . dust is an additional impurity of large mass and often of negative charge . under the simplest approximation , all the dust grains may be considered to have the same mass and equilibrium charge , and so are equivalent to a second ion species . this is the case considered in this paper . it is known that the inclusion of dust in a plasma may introduce cutoff frequencies into the basic alfvn waves , and introduce a low frequency mode , whose nature is different from that in a bi - ion plasma due to the dust grain s extremely high mass @xcite . qualitative differences to the bi - ion case arise when charge perturbations of the grains are included , leading to an additional damping mechanism @xcite , or when a spectrum of dust grain sizes and charges is allowed for @xcite . dust is found , in varying amounts , in many astrophysical and space environments such as molecular clouds and the rings of saturn . dusty plasmas have also been studied closely experimentally , since the heavy grain mass introduces low frequency effects that may be studied in real time . an immediate consequence of the presence of an additional ion species is its modification of the background free - electron number density . this influences the propagation of plasma and electromagnetic waves @xcite . if in the case of a dusty plasma we fix the grain charge , then the grain becomes entirely defined in terms of its mass and charge , and as such acts just like another ion in the plasma ( albeit of negative charge ) . the state of charge neutrality may be written as : @xmath0 here @xmath1 is the number density of plasma electrons ( with the charge @xmath2 ) , and the two ion species ( of signed charge @xmath3 and @xmath4 ) , respectively . for laboratory dusty plasmas , the grain charge is negative ( i.e. , @xmath5 ) and large ( @xmath6 ) , so that an appreciable proportion of the negative charge in the plasma may reside on the dust particles . for astrophysical dusty plasmas , @xmath7 may only be of the order of unity , which is often the case for a canonical bi - ion plasma . in environments such as the interstellar medium , where the dust grain is in an electron - proton plasma with little ultra - violet radiation present , the dust grains acquire a negative charge . on the other hand , exposure to ultraviolet light from nearby stars can cause ionization of the grains , leaving a residual positive charge . in this paper , we investigate the propagation of plane hydromagnetic waves ( alfvn waves and magnetoacoustic waves modified by the presence of a secondary ion or dust species in a bi - ion plasma ) , propagating parallel to the pumped magnetic field of a large amplitude magnetoacoustic wave . we generalize ref.@xcite where the second heavy space was assumed immobile , to the case where both ions are fully mobile . a further generalization is the inclusion of pressure . the background magnetic field and plasma density are taken to be uniform , at frequencies well below the electron plasma and cyclotron frequencies . first , we find the dispersion equation of the pump wave , and then concentrate on pump waves of large wavelength . we then obtain a coupled pair of equations of motion governing the perturbed plasma . resonant interactions are sought , and growth rates of the waves parametrically excited by the pump are calculated and discussed . the most general set of equations used to describe two mobile ion - like species , plus electrons , is a three - fluid model . in this picture we employ three momentum equations for the electrons and the two ion species , where we include both the ions inertia terms , while ignoring the electron inertia . this is valid if the frequencies of interest are well below the electron cyclotron frequency . in addition we use maxwell s equations and two mass continuity equations for each of the ion species , ignoring maxwell s displacement current . the primary species ( assumed positively charged ) shall be denoted with a 1 " subscript , and the secondary species ( either positively or negatively charged ) shall be denoted with a 2 " subscript . in this notation we may write the current density as @xmath8 where the two ion species and electrons have velocities denoted by @xmath9 , @xmath10 , and @xmath11 , respectively . we employ the parameters @xmath12 and @xmath13 , which measure the distribution of charge in the plasma amongst the ions . employing the total charge neutrality condition , given in eq.([movequil ] ) , we may write this as @xmath14 in terms of these parameters . note that in the limit of a single ( primary ) ion species , we have @xmath15 and @xmath16 . ignoring collisions , but including the effects of pressure , the starting equations for the velocities , electric and magnetic fields and each number density are the momentum equations , the two ion continuity equations , and ampere s law neglecting the displacement current : @xmath17 here , @xmath18 are the two heavy species and electron pressures , and @xmath19 are the heavy masses . the magnetic field @xmath20 includes the background magnetic field @xmath21 . finally , by assuming either an isothermal or an adiabatic equation of state , we have @xmath22 for each species , where @xmath23 are the individual sound speeds . though it is not imperative , we shall assume isothermal changes , which permits us to write down @xmath23 in terms of the plasma temperatures , i.e. , @xmath24 is the square of each thermal speed , where @xmath25 is each temperature and @xmath26 is boltzmann s constant . at this point we choose to eliminate the electron variables from eqs.([movi])([movmax2 ] ) , and use faraday s law to eliminate the electric field @xmath27 . there are then two choices for the way to proceed . we can add and subtract the momentum equations and deal with a total fluid velocity and the current density @xmath28 . instead we shall employ the most direct method , which is to deal with the ion velocities @xmath9 and @xmath10 separately , since our equations will exhibit the most symmetry this way . by employing the charge neutrality condition to all orders , the following system of nonlinear partial differential equations is found : @xmath29 in eqs.([movcont])([moveuld ] ) @xmath30 are the densities of each massive component of the plasma and @xmath31 are the corresponding ( signed ) cyclotron frequencies , with @xmath32 . also @xmath33 and @xmath34 are pseudo - squared thermal speeds associated with the electron pressure ( if @xmath35 we allow @xmath36 to be imaginary ) . the presence of both @xmath37 and @xmath38 in both equations of motion , is a consequence of the elimination of the electron variables from the equations . to be more explicit , the electron density fluctuations generate fluctuations in both the ion densities through the charge neutrality condition , which produces this coupling in the momentum equations . the second term on the right hand side of eq.([movmag1 ] ) is the hall term ; it is important when the wave frequency is comparable to either of the cyclotron frequencies . also , we see that the two species are strongly coupled through the momentum equations , via the third term on the right hand side in each . this is an additional hall - type term associated with the relative motion of the two species . note eqs.([movcont])([moveul ] ) reduce , in the case of a single species plasma ( @xmath39 where @xmath40 is the wave frequency and @xmath41 are the alfvn speeds associated with each ion . the charge neutrality condition implies that the wave frequencies are restricted to the regime @xmath42 , which may be restated as @xmath43 there exists a wide range of physical environments for which all of the above mentioned conditions are met . suppose there is a constant background magnetic field in the @xmath44-direction , given by @xmath45 . we now periodically pump the field , with a periodic modulation : @xmath46\hat{{\bf z } } , \label{movinitb}\ ] ] where @xmath47 is the pump frequency . here @xmath48 is a constant dimensionless quantity which determines the amplitude of the pump . more precisely , @xmath48 is an expansion parameter , which permits us to keep track of our terms , by matching powers of @xmath48 . the parameter @xmath49 on the other hand is a dimensionless quantity that we include to capture any necessary frequency information , i.e. , @xmath50 , such that the average of @xmath49 over @xmath47 is @xmath51 . by specifying a particular choice of normalization condition , we may solve for @xmath49 . we later in fact impose the condition that the energy density in the pump system is a constant over @xmath52 , and calculate the resulting @xmath49 . since we are interested in a large amplitude pump wave , @xmath48 will typically be @xmath53 . however , since this field has no spatial dependence , it does not satisfy our wave equations . hence this is an approximation to a wave with a large wavelength . we must therefore modify the pump with an envelope of some wavelength . since the pump wave magnetic field points in the @xmath44-direction , the wavevector should point perpendicular to this to satisfy @xmath54 . for a planar geometry plasma , we choose the axes such that the wave varies in the @xmath55-direction , with wavenumber @xmath56 . by denoting the pump magnetic field ( and all subsequent pump fields ) with a 0 " superscript , we have @xmath57\hat{\bf z}. \label{movpump}\ ] ] in a cylindrical plasma , with @xmath58 denoting the radial distance from the cylindrical axis , we have @xmath59\hat{\bf z } , \label{movpumpj}\ ] ] where @xmath60 is the 0th order bessel function of the first kind . note that @xmath61 in eqs.([movpump ] ) and ( [ movpumpj ] ) . in the absence of a wave , we shall suppose that the plasma is stationary , with ion densities @xmath62 and charge ratios of @xmath63 . the effect of pumping will be to modify the velocity , density and charge imbalance to order @xmath48 , so that @xmath64 these quantities define the pump wave , and we now proceed to solve the equations ( [ movcont])([moveuld ] ) to order @xmath48 ( i.e. , the linear solution ) . by inspecting the form of the magnetic field in eq.([movpump ] ) [ or ( [ movpumpj ] ) ] , and eqs.([moveul ] ) and ( [ moveuld ] ) , we note that the velocity can not have a @xmath44-component . hence , the velocities of the ion species must be of the form @xmath65 for planar waves . from the resulting pair of equations we ascertain the form of each velocity component and density : @xmath66 where @xmath67 and @xmath68 are amplitudes independent of space and time , to be determined . then , from the continuity equations , we have @xmath69 for the cylindrical plasma case , we make the substitutions : @xmath70 , @xmath71 , @xmath72 and @xmath73 , where @xmath74 is the azimuthal angle and @xmath75 is the 1st order bessel function of the first kind . then it is found that all the relationships ( [ movtransvel])([movconamp ] ) still hold . it follows that the modes in the planar and cylindrical plasmas satisfy the same dispersion equation . upon substitution of eqs.([movv_x])([movconamp ] ) into eqs.([movcont])([moveuld ] ) , written to first order in @xmath48 , the following dispersion equation of the pump wave is obtained : @xmath76 here , we have defined @xmath77 where @xmath78 is a hybrid cutoff frequency as @xmath79 . also we have introduced @xmath80 , @xmath81 , @xmath82 and @xmath83 which allow us to specify the cold / warm plasma regimes , with @xmath84 and @xmath85 related through : @xmath86 . note that as @xmath87 ( i.e. , a single species ) we are left with the nondispersive relation @xmath88 where @xmath89 is the combined sound speed in this limit . this is the familiar fast magnetoacoustic wave characteristic for @xmath90 . however , in the presence of a secondary ion species the fast magnetoacoustic wave gains an additional mode and the relationship is dispersive , see fig.[mov1 ] . versus normalized pump wavenumber @xmath91 with @xmath92 and @xmath93 . ( a ) the warm plasma , with @xmath94 and @xmath95 . ( b ) the cold plasma , with @xmath96.,width=377 ] versus normalized pump wavenumber @xmath91 with @xmath92 and @xmath93 . ( a ) the warm plasma , with @xmath94 and @xmath95 . ( b ) the cold plasma , with @xmath96.,width=377 ] it is easy to show that the parameters @xmath97 and @xmath98 are always positive , while @xmath99 for the cold plasma and is positive if any of @xmath100 are nonzero . although fig.[mov1 ] is for a plasma in which the secondary ion species is positive , the same basic qualitative features are present when the second ion species is negative . for the warm plasma the effect of the @xmath101 term in the dispersion relation is to cause the lower branch to increase without bound as @xmath102 , as indicated in fig.[mov1](a ) . physically , it can be thought of as the fast magnetoacoustic mode being converted into an acoustic mode for large @xmath56 due to the inclusion of pressure . however , if the plasma is cold ( i.e. , @xmath99 ) then the lower branch experiences a resonance as @xmath102 , as indicated in fig.[mov1](b ) . in this limit @xmath97 becomes the square of a resonance frequency , given by @xmath103 which gives rise to a forbidden frequency region between @xmath104 and @xmath78 , since @xmath105 whenever @xmath106 . the ratio of the @xmath55-velocity amplitudes @xmath107 to the magnetic field parameter @xmath49 are of particular importance to our later discussion , so we briefly discuss them here . in terms of the frequency and wavenumber , we find , @xmath108 where @xmath109 finally , to complete the set of pump quantities , we find that the velocity in the @xmath110-direction is given by @xmath111 with @xmath112 given similarly . this is quite different to the single ion species case , where @xmath113 . the behavior of @xmath114 is dramatically altered by the choice of branch from the dispersion relation ; see fig.[mov2 ] , where for brevity we plot @xmath115 only . -velocity amplitude and @xmath110-velocity amplitude to magnetic field @xmath116 versus normalized pump wavenumber @xmath117 , with @xmath118 and @xmath95 . the solid curve is for the upper branch of the dispersion relation , the dashed curve is for the lower branch of the dispersion relation . note that on the vertical axis we plot the absolute value of this ratio , since we are interested here in the ratio of _ consequently one of the two dashed curves experiences a discontinuity in its first derivative when it touches the horizontal axis.,width=377 ] we see that as @xmath119 , the @xmath55-component approaches a finite positive value while the @xmath110-component approaches zero for the upper branch of the dispersion relation . on the other hand the @xmath55-component approaches large values and the @xmath110-component passes through zero for the lower branch of the dispersion relation . note that the two branches stemming from the upper branch of the dispersion relation become singular as @xmath120 . this motivates the need for a normalization condition relating the pump field amplitudes of the velocities @xmath67 , magnetic field @xmath49 , and density @xmath68 . a reasonable way to proceed is to assume that as we vary @xmath47 the energy density @xmath121 in the pump system is fixed . the appropriate relation in a fluid description for general magnetoacoustic waves is ( e.g. @xcite ) , @xmath122 where @xmath123 , and the summation in the final term is over each plasma species . in our approximations the electric energy term may be neglected compared to the magnetic energy . by switching to dimensionless units and setting @xmath124 , then all our variables are given uniquely in terms of the basic plasma parameters . by using the expressions for @xmath115 and @xmath125 we may solve for @xmath49 . the result is @xmath126 [ using @xmath127 from eq.([movf1 ] ) ] . a plot of @xmath49 versus wavenumber is given in fig.[mov3 ] for each branch @xmath128 . versus normalized pump wavenumber @xmath117 due to the normalization condition ( [ movbnorm ] ) , with @xmath118 and @xmath95 . the solid curve is for the upper branch of the dispersion relation , the dashed curve is for the lower branch of the dispersion relation.,width=377 ] it is important to note that @xmath129 as @xmath120 for the upper branch . it follows that all the velocity amplitudes are well behaved near @xmath130 . the remaining pump quantities to solve for are the charge ratios @xmath131 . these may be obtained by imposing the charge neutrality condition on both the background plasma and on the perturbations to the equilibrium . this is justified since we are interested only with frequencies that are much less than that of the electron plasma frequency . this means the electrons have sufficient time to respond to the perturbations from the equilibrium and neutralize the plasma at each point in space . thus , on the perturbed plasma we impose the condition , @xmath132 where we have ignored the effects of dust ( or ion ) charging , that is , @xmath4 is constant . it is then straightforward to obtain the charge ratios in terms of the densities , and subsequently in terms of the velocity amplitudes and @xmath49 , as follows : @xmath133 this implies that the amount of charge that resides on each species is temporally coupled to the other . using the derived set of quantities characterizing the pump wave we may , in principle , investigate the possibility of excited modes propagating at arbitrary angles with respect to the @xmath44-axis , as in ref.@xcite . however , by removing the @xmath55-dependence in the pump waves our calculations become much more tractable , with the requirement that our excited waves are restricted to _ parallel _ propagation along the @xmath44-axis , see ref.@xcite . for an infinitely extended medium , the perpendicular wavenumber of the excited waves is then zero . however , this would clearly be inconsistent with wavenumber matching of a propagating pump with finite perpendicular wavenumber . we therefore envisage this analysis to apply to a standing wave ( in the transverse direction ) pump in a finite geometry system , such as a cylindrical or toroidal laboratory plasma , or an astrophysical magnetic flux tube ; by seeking localized solutions for distances transverse to the magnetic field such that @xmath134 in a planar plasma , or @xmath135 in a cylindrical plasma , we can recover the initially proposed form for the magnetic field eq.([movinitb ] ) . in other words , we are taking @xmath55 small rather than @xmath136 small , to ensure that the fields will look approximately uniform in the perpendicular direction , i.e. that the wavelength is effectively very long on the length scale of interest . in the parallel direction , however , the pump fields are uniform , we place no such restriction on distances and excited wavelengths , and we can apply wavenumber matching of the excited wave fields in that direction , as discussed in section vi . then in the region near the @xmath44-axis we have @xmath137_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}b_{0 } ( 1+\bar{\varepsilon}b\cos(\omega_{0}t ) ) \hat{\bf z},\label{movmag2}\\ \left[{\bf v}_{1}^{(0)}\right]_{k_{0}x\sim{\bar{\varepsilon}}}{&\!\!}={&\!\!}{\bf 0 } , \,\,\,\,\,\,\,\,\ , \left[{\bf v}_{2}^{(0)}\right]_{k_{0}x\sim{\bar{\varepsilon}}}={\bf 0},\\ \left[\rho^{(0)}_{1,2}\right]_{k_{0}x\sim{\bar{\varepsilon}}}{&\!\!}={&\!\!}\rho_{01,2 } ( 1+\bar{\varepsilon}\frac{k_{0}}{\omega_{0 } } a_{x1,2}\cos(\omega_{0}t))\label{movrho2},\\ \left[\delta^{(0)}_{1,2}\right]_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}\delta_{01,2}(1+\bar{\varepsilon}(b-\frac{k_{0}}{\omega_{0}}a_{x1,2})\cos(\omega_0t)),\,\,\,\,\,\,\,\end{aligned}\ ] ] correct to @xmath138 . since the magnetic field has now simply a periodic variation in time it resembles a canonical parametric pump . the problem then resembles a parametric amplifier , such as a harmonic oscillator with a time - varying spring constant . for use in the excited wave equations we must also compute the derivatives near the @xmath44-axis , given here : @xmath139_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}{\bf 0},\\ \left[\frac{\partial{\bf v}^{(0)}_{1}}{\partial x } \right]_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}\bar{\varepsilon}k_{0}(a_{x1}\sin(\omega_{0}t ) , a_{y1}\cos(\omega_{0}t),0),\label{movvelderiv}\\ \left[\frac{\partial{\bf v}^{(0)}_{2}}{\partial x } \right]_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}\bar{\varepsilon}k_{0}(a_{x2}\sin(\omega_{0}t ) , a_{y2}\cos(\omega_{0}t),0),\,\,\,\,\,\,\,\,\label{movvelderiv2}\\ \left[\frac{\partial{\rho}^{(0)}_{1,2}}{\partial x } \right]_{k_{0}x\sim\bar{\varepsilon}}{&\!\!}={&\!\!}\left[\frac{\partial{\delta}^{(0)}_{1,2}}{\partial x } \right]_{k_{0}x\sim\bar{\varepsilon}}=0.\label{movrhoderiv}\end{aligned}\ ] ] note that the velocity has a finite derivative , even though the velocity itself is zero in this regime . next we wish to test the stability of the self consistent linear solution given in eqs.([movmag2])([movrhoderiv ] ) , now regarded as a finite - amplitude pump wave , to the excitation of waves propagating along the magnetic field direction . we must however be careful since this is only a linearized pump solution . to proceed we will adopt the basic methodology of refs.@xcite , but shall attempt to refine the argument . the basic technique is to perturb each quantity , i.e. , @xmath140 , @xmath141 , @xmath142 , @xmath143 , by an arbitrarily small amount as measured by an expansion parameter @xmath144 . for example , take an arbitrary field , say @xmath97 , with a known linear representation of the pump field : @xmath145 . then refs.@xcite proceed by perturbing this quantity in the following way : @xmath146 however , since @xmath147 has neglected terms of @xmath148 , the term of @xmath149 is even more so negligible in this expansion ( recall that @xmath144 is arbitrarily small ) . the problem lies in the fact that we are ultimately interested in the stability of the _ exact _ pump solution , say @xmath150 , with @xmath151 thus , we shall perform the following expansion instead : @xmath152 upon substitution into our set of nonlinear differential equations ( [ movcont])([moveuld ] ) , we obtain a set of equations of the following structure ( no approximations ) : @xmath153 where @xmath154 are differential operators acting on their respective arguments which we may treat as functions ( for brevity we have suppressed their dependence on fields other than @xmath97 ) . now @xmath155 by definition of an exact pump solution . also , the term of @xmath156 can certainly be neglected . next , we taylor expand @xmath157 and write @xmath158 we are then permitted to neglect this term of @xmath159 when inserted back into eq.([movabcd ] ) . this gives us a set of equations for the perturbation @xmath160 , in terms of the @xmath161 , to sufficient accuracy , i.e. , @xmath162 note that @xmath161 has served two purposes . first , it gives us an approximate description of the corresponding exact solution @xmath163 , and second , it allows us to investigate its stability without ever having to find @xmath164 . our task then is to ascertain the primed variables which are the excited fields . as stated previously , we are interested only in plane waves that travel parallel to the @xmath44-axis . in this case , any longitudinal components will decouple completely and merely describe a linear acoustic wave , so we may set the @xmath44-component of the primed velocity to zero . hence , both primed velocities have the following form : @xmath165 thus , @xmath166 , and from each of the two continuity equations we have @xmath167_{k_0 x\sim{\bar{\varepsilon}}},\ ] ] where we have imposed the @xmath134 [ or @xmath135 ] condition . hence we have @xmath169 , i.e. , there are no perturbations in the densities of the excited fields . from this it also follows that @xmath170 . by taking the derived expressions for @xmath20 , @xmath171 , @xmath172 and @xmath173 and substituting into eqs.([movmag1 ] ) & ( [ moveul ] ) , and using the stability analysis procedure conveyed in eq.([movpertmeth ] ) , we obtain @xmath174_{k_{0}x\sim\bar{\varepsilon } } \label{movmag3 } \\ { & \!\!}\rho_{01}{&\!\!}\delta_{01 } \frac{\partial{\bf v}_{1}'}{\partial t}- \rho_{02}\omega_{2}({\bf v}_{1}'-{\bf v}_{2 } ' ) \times\hat{\bf z}-\frac{1}{\mu_{0 } } ( \nabla\times{\bf b}')\times{\bf b}_{0}\nonumber\\ { & \!\!}={&\!\!}\!{\bar{\varepsilon}}\left[-\delta_{01}\rho_{01}v_{x1}'\frac{\partial\bar{v}_{1}}{\partial x } + \omega_{2}\bar{\rho}_{2}({\bf v}_{1}'-{\bf v}_{2}')\times\hat{\bf z}\right]_{k_{0}x\sim{\bar{\varepsilon}}}.\end{aligned}\ ] ] here the terms of @xmath159 , which occur on the right hand side , should be thought of as driving terms from the pump wave . note that there exist two further equations of motion under the index interchange @xmath175 . note also that there are no acoustic terms on the left hand sides of these equations because of the decoupling of the longitudinal motions : however there are still acoustic influences in the pump fields on the right hand sides of the equations . now in order to treat the @xmath55 and @xmath110 components of the @xmath20 and @xmath171 vectors on an equal footing we form a complex vector out of each component of these partial differential equations , utilizing the variables @xmath176 . the spatial variation is assumed to be periodic in the @xmath44-direction with wavenumber @xmath177 , as follows : @xmath178 . given this , the following pair of linear ordinary differential equations are obtained : @xmath179},\label{movbcomplex}\\ \!{&\!\!}{&\!\!}\delta_{01}\rho_{01}\frac{\partial v_{\pm 1}}{\partial t}\pm i\rho_{02}\omega_{2}(v_{\pm 1}-v_{\pm 2})-i \frac{k b_{0}}{\mu_{0}}b_{\pm}\nonumber\\ { & \!\!}={&\!\!}{\bar{\varepsilon}}\big{[}-\delta_{01}\rho_{01}k_{0}v_{x1}'\left ( a_{x1}\sin(\omega_{0}t)\pm i a_{y1}\cos(\omega_0 t)\right)\nonumber\\ { & \!\!}\mp{&\!\!}i \omega_{2}\rho_{02}\frac{k_0}{\omega_0}a_{x2}\cos(\omega_{0}t)(v_{\pm 1}-v_{\pm 2})\big{]}. \label{movcomplex}\end{aligned}\ ] ] let us define the sense of polarization in reference to the screw sense of the fields in the direction of propagation in the @xmath44-direction . then @xmath180 corresponds to a left hand circularly polarized wave for positive frequencies and a right hand circularly polarized wave for negative frequencies , while @xmath181 corresponds to a right hand circularly polarized wave for positive frequencies and a left hand circularly polarized wave for negative frequencies . also , @xmath182 is defined similarly . with these definitions @xmath183 may be eliminated entirely from these expressions . note that the presence of a second moving ion species introduces coupling through terms of the form : @xmath184 , and an additional term on the right hand side in eq.([movcomplex ] ) due to density variations . in an analogous way to the classic driven pendulum problem @xcite , we have found that the velocities of the two species satisfy a pair of generalized mathieu equations . furthermore , the two species and magnetic field are strongly coupled . to proceed , we move from the time domain to the frequency domain , under the fourier transform : @xmath185 where @xmath40 is the frequency of the excited waves . using the linearity of our pair of differential equations , we compute the fourier transform of both . upon eliminating @xmath182 and @xmath186 in favor of @xmath187 , we obtain : @xmath188 } , \label{movfourdisp}\end{aligned}\ ] ] where , @xmath189},\,\,\,\,\,\ , \label{movhdef}\\ n_{\pm}\!(a_2,d_1,\epsilon){&\!\!}:={&\!\!}\pm\frac{k_0}{4}\epsilon(1-\delta_{01})\omega_{1 } ( d_1+a_{y1 } ) -\frac{k_0}{4}\frac{\omega_1\delta_{01}\omega}{\delta_{02}\omega_2 } \frac{a_2\pm a_{y2}}{1\mp\frac{\omega}{\omega_2}},\end{aligned}\ ] ] and @xmath190 . eq.([movfourdisp ] ) is obviously not an algebraic expression from which we can uniquely obtain the dispersion equation between @xmath40 and @xmath177 . instead it provides us with a functional relationship between the fourier transforms with arguments @xmath40 , @xmath191 and @xmath192 . nevertheless , we may proceed by noting that the right hand is @xmath159 , which is small . for the left hand side to be small we must ensure that the frequency is near a root of the polynomial on this side with a correction of @xmath138 , i.e. , @xmath193 where @xmath194 is defined to satisfy , @xmath195 these define the naturally occurring modes in the absence of pumping , ( i.e. , if @xmath196 then @xmath197 ) . the solutions of these algebraic expressions give the dispersion relations characterizing plane transverse waves propagating parallel to the background magnetic field in a two species plasma . the last term of @xmath198 is only present when a secondary moving ion - like species resides in the plasma . it is an additional hall - type term , associated with the relative motion of the two charged species . in the one species limit this relation simplifies to @xmath199 and we are left with the familiar fast and slow ( ion - cyclotron ) transverse waves @xcite . the important feature of the two - species result in eq.([movcubic ] ) is that it is a cubic in @xmath194 and hence another mode of excitation has been added . moreover , since we may choose @xmath200 or @xmath201 it follows that there are a total of six solutions ; three are left hand circularly polarized and three are right hand circularly polarized . however , only three of these are physically different . this is because if @xmath194 satisfies @xmath202 , then @xmath203 satisfies @xmath204 , and vice versa . we may then concentrate on the positive frequency solutions . if the secondary species is positively charged then there are two left hand modes of excitation @xcite . these we denote by @xmath205 and @xmath206 , where the _ s _ " denotes slow , and the _ f _ " denotes fast . there is also a single right hand mode , denoted by @xmath207 . the corresponding dispersion relations are plotted in fig.[mov4](a ) . versus normalized wavenumber @xmath208 of the three natural modes . ( a ) the second species is positively charged , with @xmath92 and @xmath209 . ( b ) the second species is negatively charged , with @xmath210 and @xmath211.,width=377 ] versus normalized wavenumber @xmath208 of the three natural modes . ( a ) the second species is positively charged , with @xmath92 and @xmath209 . ( b ) the second species is negatively charged , with @xmath210 and @xmath211.,width=377 ] note that the right hand mode intersects with the fast left hand mode . also , the two left hand modes both experience a resonance as @xmath212 . for @xmath206 it is given by @xmath213 , and for @xmath205 it is given by @xmath214 ( where we have assumed @xmath215 ) . for a negatively charged secondary species there are also three modes of excitation , however the combination of polarizations has changed . in this case there are two right hand modes @xmath216 and @xmath217 , and a single left hand mode @xmath218 ; see fig.[mov4](b ) . here @xmath217 has a resonance at @xmath219 , and @xmath218 has a resonance at @xmath213 . in the dusty plasma case , further properties of these modes were investigated in refs.@xcite . note that the upper curves ( fast modes ) , regardless of the sign of the species , experience a nonzero cutoff frequency which coincides with that of the pump frequency , see eq.([movomega_c ] ) . from the above analysis we see that @xmath194 may be any one of @xmath205 , @xmath206 , @xmath207 for a positively charged secondary species , or @xmath217 , @xmath216 , @xmath218 for a negatively charged secondary species . since the frequency describing the excited wave is a perturbation of a natural mode frequency , we denote the change by @xmath74 which is allowed to be complex , where @xmath220 now by returning to the fourier transform relationship in eq.([movfourdisp ] ) we can see what effect @xmath74 ( i.e. , pumping ) will have on @xmath221 . without any pumping ( i.e. , @xmath196 ) the spatial solutions would be a linear superposition of three pure monochromatic exponentials . hence the fourier transform would be a sum of three dirac-@xmath173 functions ; see fig.[mov5](a ) for a representation . versus frequency @xmath40 . ( a ) no pump , giving three dirac - delta functions . ( b ) the effect of a pump , broadening the spectrum . note that we have chosen to illustrate the idea using only the right hand modes ( i.e. , @xmath222 ) , for a plasma with a negatively charged secondary species.,width=377 ] versus frequency @xmath40 . ( a ) no pump , giving three dirac - delta functions . ( b ) the effect of a pump , broadening the spectrum . note that we have chosen to illustrate the idea using only the right hand modes ( i.e. , @xmath222 ) , for a plasma with a negatively charged secondary species.,width=377 ] the effect of pumping is to modify this and provide frequency shifts . the modified solution will have some broadened frequency spectrum , since it is the case that dirac-@xmath173 functions do not solve the fourier transformed equations in the presence of a pump . these types of modifications will occur near each natural frequency ; see fig.5(b ) for a simplified representation of this effect . in fact an exact solution would be more complicated with the identification of harmonics etc , but this figure should illuminate the key feature of the interaction . this is described as a _ parametric interaction _ between the excited fields due to the pump fields . in order to proceed we consider the case in which the interaction is greatest , as in refs.@xcite . it is clear that this effect will be greatest when the right hand side of eq.([movfourdisp ] ) is large , which is the resonance condition . this occurs when one of the fourier transforms on the right hand side is resonant , in other words when their arguments are near one of the @xmath194 s , as conveyed in fig.[mov5 ] . since the right hand side of eq.([movfourdisp ] ) involves terms of the form : @xmath223 and @xmath224 , it follows that there will be a large parametric interaction when @xmath192 _ or _ @xmath191 satisfies this . now , as mentioned earlier we are restricting our attention to positive @xmath40 ( without loss of generality ) . under this condition , it is found that only @xmath191 can satisfy this . moreover , there are three choices for this interaction : a left hand mode interacting with a right hand mode , a left hand mode interacting with a left hand mode , or a right hand mode interacting with a right hand mode . to illustrate , suppose the secondary species is negatively charged and @xmath225 ( i.e. , @xmath226 ) . then we can have @xmath227 or @xmath228 ( left hand mode interacting with right hand mode ) or @xmath229 ( left hand mode interacting with left hand mode ) . in other words , the resonance condition is that the pump frequency satisfies @xmath230 respectively , which is a statement of conservation of energy . that is , the resonance condition describes the coupling of a pump wave with two daughter waves . moreover , if @xmath231 then we can have @xmath232 , @xmath233 or @xmath234 and the same conditions for resonance apply . similar rules apply for @xmath235 and for a plasma where the secondary species is positive . let us now address the issue of the treatment of the wavenumber @xmath177 in our pair of interacting waves . first , note that in the @xmath44-direction the _ pump _ wavenumber @xmath136 is zero . for the natural modes we have that a wave given by @xmath236 is physically equivalent to a wave given by @xmath237 . thus our two interacting waves will be given by @xmath238 and @xmath239 , where @xmath240 are any combination of @xmath241 . hence the wavenumbers associated with each interacting wave are equal and opposite , i.e. , @xmath242 which is a statement of conservation of momentum in the @xmath44-direction . the fact that the right hand side is zero reflects the spatial uniformity of the pump wave . this tells us that if the approximately spatially uniform standing pump wave decays , then it does so into two daughter waves of equal wavelength travelling in opposite directions . since the wavenumber magnitudes are equal we may just consider the wavenumber @xmath177 without referring to the sense of polarization . this is a direct generalization of the previous investigations in @xcite . with the above framework we now proceed to solve for @xmath74 from the fourier transform relationship . we discuss the method involved in obtaining @xmath74 in the case where a left handed wave interacts with a right handed wave , denoted @xmath243 . here we may assume , without loss of generality , that @xmath244 in eq.([movcorrection ] ) , where @xmath245 is any one of @xmath246 . this will interact with @xmath247 ( i.e. , @xmath248 may be any one of @xmath249 ) . following the methodology of ref.@xcite we form another equation corresponding to ( [ movfourdisp ] ) under the transformation @xmath250 . in the resulting two equations , we neglect the obviously nonresonant terms @xmath223 and @xmath251 . additionally , due to our particular choice of interaction , i.e. , left - right , we neglect @xmath252 and @xmath253 , and retain only @xmath254 and @xmath255 . we find the following : @xmath256\nonumber\\ { & \!\!}{&\!\!}\times\ , \left[y_{+}\!(+\chi_{l})+j_{+}\!(-a_{x1},-a_{x2},+a_{x1},+\chi_{l},-1)\right],\,\,\,\,\,\,\,\,\ , \label{movmaindisp}\end{aligned}\ ] ] where @xmath257 in these expressions we implicitly have that @xmath258 and @xmath40 , are related by : @xmath259 and @xmath260 . in the one ion species analysis , it is at this point that the conservation of energy and momentum rules are used to explicitly obtain @xmath261 . the procedure is to use @xmath198 to solve for @xmath262 in terms of @xmath177 . these solutions are then added together and equated to @xmath47 . the resultant expression is then solved for @xmath177 and subsequently substituted into the expressions for @xmath262 . this process explicitly shows that the resonance condition , for a given value of pump frequency @xmath47 , uniquely determines the frequencies and wavenumbers of the excited waves . now , this is all possible since the solution of @xmath198 in the one species limit is merely a quadratic , see eq.([movquad ] ) . however , in our case we must solve a cubic . as such , the resulting expressions for @xmath261 are too complicated to be reproduced here . let us just note that it is still true that @xmath47 determines them uniquely . in order to solve for @xmath243 in eq.([movmaindisp ] ) we taylor expand @xmath263 around @xmath264 to order @xmath48 , then use the fact that @xmath265 , to obtain @xmath266 where @xmath267 while @xmath268 can be treated similarly . hence @xmath243 is obtained from : @xmath269\nonumber\\ { & \!\!}\,\times{&\!\!}\left[y_{+}\!(+\chi_{l})+j_{+}\!(-a_{x1},-a_{x2},+a_{x1},+\chi_{l},-1)\right]\nonumber\\ { & \!\!}\times{&\!\!}\left[h\!(\chi_{l})\,h\!(-\chi_{r})\right]^{-1}.\,\,\,\,\,\,\,\,\,\,\,\end{aligned}\ ] ] although this was derived assuming we were near a natural frequency of a left hand mode , we obtain the same expression for @xmath270 for the right hand case ( viz . @xmath271 ) with the modification that @xmath272 is of the opposite sign . this technique may also be applied to both the left - left interactions and the right - right interactions . this includes the possibility of a slow wave interacting with a fast wave ( e.g. , @xmath205 and @xmath206 ) , and waves of equal frequency interacting . these sorts of interactions occur only because of the velocity derivative terms in ( [ movvelderiv ] ) & ( [ movvelderiv2 ] ) . in obtaining @xmath74 we must retain and neglect the appropriate selection of fourier transforms according to the choice made . here we introduce @xmath273 and @xmath274 as two ( possible equal ) left hand modes , and @xmath275 and @xmath276 as two ( possible equal ) right hand modes . the corresponding frequency changes are obtained from @xmath277\nonumber\\ { & \!\!}\,\,\,\,\times{&\!\!}\left[j_{-}\!(-a_{x1},-a_{x2},+a_{x1},+\chi_{l1},-1)\right]\nonumber\\ { & \!\!}\,\,\,\,\times{&\!\!}\left[h\!(\chi_{l1})\,h\!(-\chi_{l2})\right]^{-1},\\ \phi_{rr}^2{&\!\!}={&\!\!}\left[j_{-}\!(+a_{x1},+a_{x2},-a_{x1},-\chi_{r2},-1)\right]\nonumber\\ { & \!\!}\,\,\,\,\times{&\!\!}\left[j_{+}\!(-a_{x1},-a_{x2},-a_{x1},+\chi_{r1},+1)\right]\nonumber\\ { & \!\!}\,\,\,\,\times{&\!\!}\left[h\!(\chi_{r1})\,h\!(-\chi_{r2})\right]^{-1},\end{aligned}\ ] ] for the left - left and right - right interactions , respectively . it is worth noting that in other types of parametric problems , such as modulational and beam instabilities , a graphical approach is useful in classifying the instabilities ; they occur where two normal mode lines cross and reconnect @xcite . in those cases the number of interacting modes is finite , and the nonlinear dispersion relation can be obtained in closed form . in our case , however , we can not obtain the nonlinear dispersion relation in closed form , due to the infinite number of interacting modes ; the decay interaction of the two excited modes with the pump is postulated in the first approximation , and the classification of the instability is relatively straightforward . from the expressions for @xmath278 it is easy to see that these quantities are real valued when we are in a frequency - wavenumber regime such that a pump wave can propagate . hence @xmath74 is either purely real or purely imaginary . if @xmath74 is real then what we have found is a correction to the natural mode frequency which adds an extra fourier component to the field expansions . if @xmath74 is imaginary then the daughter waves will either undergo exponential growth or decay . moreover , since there will exist both a positive and a negative imaginary frequency solution and since the frequency is conserved , this implies that one of the daughter waves will grow and the other will decay . in such cases the presence of one exponentially growing daughter wave implies that anywhere in which @xmath279 is a region of instability . in this case @xmath74 is the growth rate . in fig.[mov6 ] we plot the full complement of ( normalized ) squared frequency changes . for the left - right interaction plotted against pump frequency @xmath280 . in both plots the second species is positive , with @xmath281 and @xmath282 . the number densities vary in ( a)(b ) , with @xmath283 in ( a ) , and @xmath284 in ( b ) . also , @xmath285 , @xmath286 , @xmath287 , @xmath288 " denotes the upper - slow , lower - slow , upper - fast and lower - fast combinations , respectively . here @xmath289 " denotes the single species result.,width=377 ] for the left - right interaction plotted against pump frequency @xmath280 . in both plots the second species is positive , with @xmath281 and @xmath282 . the number densities vary in ( a)(b ) , with @xmath283 in ( a ) , and @xmath284 in ( b ) . also , @xmath285 , @xmath286 , @xmath287 , @xmath288 " denotes the upper - slow , lower - slow , upper - fast and lower - fast combinations , respectively . here @xmath289 " denotes the single species result.,width=377 ] that is , we plot @xmath290 versus pump frequency @xmath280 for the upper @xmath291 " and lower @xmath292 " branches of the pump dispersion relation for each mode of interaction . this figure is for a warm plasma in which the second heavy species is positively charged . the same basic features are present in the negatively charged case . all curves labelled @xmath293 " , i.e. , @xmath285 " and @xmath287 " , correspond to the combination between @xmath207 and @xmath205 , that is the _ slow _ interaction . the curves labelled @xmath294 " , i.e. , @xmath286 " and @xmath288 " , are for the _ fast _ interaction between @xmath207 and @xmath206 . in fig.[mov6](a ) , where @xmath295 , the full spectrum of perturbations to the natural modes is seen . the @xmath286 interaction is unstable and monotonically decreases , approaching the single species result @xmath289 for large @xmath47 . hence this is indeed consistent with the single ion analysis , given by @xmath296 the @xmath288 interaction is also unstable , however it experiences a minimum ( corresponding to a _ maximum _ growth rate ) , about which it turns over and approaches zero . for our choice of parameters , in this figure , this occurs near @xmath297 . moreover the intersection with @xmath289 occurs at @xmath298 . next , we see that the @xmath285 interaction , is small and positive , and hence it is weakly stable . note , however , that if higher order terms are included in the field expansions then this may in fact prove to be weakly unstable . finally , we see that the @xmath287 interaction , which is only present in a _ warm _ plasma , is strongly unstable , with a growth rate that approaches infinity at a rate considerably faster than @xmath289 or @xmath286 . as we decrease the number density of the second species in fig.[mov6](b ) to @xmath299 some interesting features appear . the @xmath286 curve shows a sharply varying growth rate at just above the cutoff frequency . the @xmath288 minimum shifts to a lower value ( lying beyond the range of the plot ) . the @xmath285 interaction has actually become unstable in the frequency region just above the cutoff frequency . also , the manner in which our bi - ion result approaches the single species result as @xmath300 is interesting : the envelope of the three modes @xmath287 , @xmath285 , and @xmath286 form @xmath289 in the domains @xmath301 , @xmath302 and @xmath303 , respectively . in fig.[mov7 ] we plot the range of left - left interactions for the positively charged secondary species case . for the left - left interaction plotted against pump frequency @xmath280 . here the second species is positive , with @xmath304 and @xmath284 . also , @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath310 denotes the upper - slow - slow , lower - slow - slow , upper - fast - fast , lower - fast - fast , upper - slow - fast and lower - slow - fast combinations , respectively.,width=377 ] as can be seen in fig.[mov4](a ) this case has two left handed branches ( slow and fast ) . hence we may form the slow - slow , fast - fast and slow - fast combinations . this combined with the lower and upper branches of the pump dispersion relation gives a total of six interactions . of these six we find that all are unstable , except the @xmath310 " combination which experiences a point of maximum stability at just below @xmath311 . note that the @xmath306 " interaction decreases without bound as @xmath312 this is acceptable given that for the slow - slow case the excited wavenumber @xmath177 experiences a resonance as @xmath313 ( with pump wavenumber small ) . although it lies below the range of the plot the @xmath308 " interaction starts at a finite value , when @xmath314 . also note the curious feature wherein the @xmath309 " interaction has a corresponding growth rate which is large in the range @xmath315 only . finally , in fig.[mov8 ] we plot the range of right - right interactions for the negatively charged secondary species case . for the right - right interaction plotted against pump frequency @xmath280 . here the second species is negative , with @xmath316 and @xmath317 . also , @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath310 " denotes the upper - slow - slow , lower - slow - slow , upper - fast - fast , lower - fast - fast , upper - slow - fast and lower - slow - fast combinations , respectively.,width=377 ] again by inspection of fig.[mov4](b ) it makes sense that there are six values of @xmath318 . this plot shares many features in common with the previous plot . however , an important change is that the @xmath306 interactions has a minimum and approaches zero as @xmath212 . also , the @xmath309 maximal growth rate has shifted to much higher values of pump frequency . the @xmath310 interaction has stability as @xmath52 increases . hence , we have seen that the introduction of a second species adds a range of extra behaviors , even with a small number density . note that in the special case where the frequencies of different modes coincide , the selection of resonant fourier transforms to generate these plots is actually invalid . in particular , this occurs at the point seen in fig.[mov4](b ) where @xmath319 . at that point there are actually more terms resonant than have been accounted for . however , since this occurs on a set of measure zero , it has been ignored . this paper has outlined an investigation into a certain class of instabilities in a bi - ion or dusty plasma , allowing full mobility of all species and the inclusion of pressure . the results should be applicable to a wide range of laboratory and space plasmas where secondary ion species or charged dust grains occur . the immediate consequence of the presence of an extra heavy species is its increase or reduction of the number of free electrons in the plasma through the charge neutrality condition . we obtained a pump wave with spatial variation transverse to the background magnetic field , which contained an upper and lower branch , and made the approximation of spatial uniformity of this pump . the stability of this structure was investigated by perturbing the pump wave , as we looked for parametric instabilities wherein the pump wave generates excited waves . three natural transverse modes were found to be excited propagating parallel to the background magnetic field ; two left handed when the second species is positive and two right handed when the second species is negative . two out of each of these three are the modified fast and slow ( ion - cyclotron ) modes . there were three basic types of interactions allowed ; left - right , left - left and right - right , each of which has several combinations dependent on both the choice of branch from the pump wave ( upper and lower ) and the choice of natural mode ( slow and fast ) , giving rise to twelve interactions in total for any particular plasma . we derived the frequency changes to the natural modes , corresponding to growth rates when the perturbation to the frequency was imaginary . we can summarize the main features of the instabilities for the left right interactions ( opposite sense circular polarization ) as follows : \(i ) of the four combinations allowed , three are unstable when the number density of the second massive species is substantial , and all four are unstable when the number density of the second massive species is low . \(i ) these interactions involve six and two combinations respectively when the second species is positive , and the reverse when the second species is negative . this follows simply from the assortment of handedness of natural modes . \(ii ) of these only the @xmath310 interaction is stable , while the others display interesting behaviors . in particular , in the positive case , the @xmath306 curve approaches @xmath322 as @xmath323 where the excited wavenumber @xmath177 experiences a resonance , while its growth rate is maximal at @xmath52 just below @xmath324 in the negative case . \(iii ) in contrast to the left right interactions , the upper pump mode can give the strongest instabilities . the @xmath309 interaction has a large growth rate between @xmath325 and @xmath326 in the positive second species case , and at higher values of pump frequency in the negative second species case . in several of the above interactions the growth rate starts at a nonzero value this occurs in the cases where the starting value of @xmath52 have a corresponding nonzero value for @xmath136 , so the velocity derivative terms are nonzero . a possibility for further work would be to allow the excited waves to be fully oblique , rather than simply parallel to the background magnetic field . in doing so we would be able to remove the long pump wavelength assumption , which may lead to interesting features . in the single ion case this was treated in ref.@xcite , albeit without pressure . it is anticipated that implementation of the accompanying additional wavenumber matching rules , in both the @xmath44 and @xmath55 directions , would be algebraically complicated . 02 l. spitzer , jr . , _ physical processes in the interstellar medium _ , ( wiley , new york , 1978 ) . w. pilipp , t.w . hartquist , o. havnes , and g.e . morfill , astrophys . j. * 314 * , 341 ( 1987 ) . sakai , t.kawata , k. yoshida , k. furusawa and n.f . cramer , astrophys . j. * 537 * , 1063 ( 2000 ) . sakai , r. minamizuka , t.kawata , and n.f . cramer , astrophys . j. * 550 * , 1075 ( 2001 ) . bray , l.e . cram , c.j . durrant and r.e . loughhead , _ plasma loops in the solar corona _ , ( cambridge university press , cambridge , 1991 ) . d. montgomery and r. harding , phys . lett . * 23 * , 670 ( 1966 ) . g. vahala and d. montgomery , phys . fluids * 14 * , 1137 ( 1971 ) . n.f . cramer , plasma phys . * 17 * , 967 ( 1975 ) e. mjlhus , j. plasma phys . * 19 * , 437 ( 1978 ) . g. morales , phys . of fluids * 19 * , 849 ( 1976 ) . n.f . cramer , plasma phys . * 18 * , 749 - 760 ( 1976 ) . m. p. hertzberg , n. f. cramer and s. v. vladimirov , phys . plasmas * 10 * , 3160 ( 2003 ) . s.v . vladimirov and n.f . cramer , phys . e * 54 * , 6762 ( 1996 ) . cramer , _ the physics of alfvn waves _ , ( wiley - vch , berlin , 2001 ) . vladimirov , phys . plasmas * 1 * , 2762 ( 1994 ) . n.f . cramer , f. verheest , s.v . vladimirov , phys . plasmas * 9 * , 4845 ( 2002 ) . vladimirov , phys . e * 50 * , 1422 ( 1994 ) . cramer and s.v . vladimirov , phys . scr . * 53 * , 586 ( 1996 ) . k. mio , t. ogino , k. minami , and s. takeda , j. phys . * 41 * , 265 ( 1976 ) . e. mjlhus , j. plasma phys . * 16 * , 321 ( 1976 ) . spangler and j.p . sheerin , j. plasma phys . * 27 * , 193 ( 1982 ) . e. mjlhus and j. wyller , phys . * 33 * , 442 ( 1986 ) . ovenden , h.a . shah , and s.j . schwartz , j. geophys . res . * 88 * , 6095 ( 1983 ) . melrose , _ plasma astrophysics _ , vol . 1 , ( gordon and breach , new york , 1980 ) . pippard , _ the physics of vibration _ , vol . 1 , ( cambridge university press , cambridge , 1978 ) . d.a . mendis and m. rosenberg , ann . . astrophys . * 32 * , 419 - 463 ( 1994 ) . f. verheest and p. meuris , phys . lett . * a 198 * , 228 - 232 ( 1995 ) . k. nishikawa , j. phys . japan * 24 * , 916 - 922 ( 1968 ) . m. longtin and b.u.o . sonnerup , j. geophys . res . * 91 * , 6816 - 6824 ( 1986 ) . l. gomberoff , k. gomberoff and a.l . brinca , j. geophys . 107 * , a7 , ssh 5 - 1 - 7 ( 2002 ) .

we investigate the structure and nuclear region of the black hole ( bh ) hosting galaxy henize 2 - 10 . surface brightness ( sb ) profiles are analyzed using magellan / megacam- and-band images . excluding the central starburst , we find a best - fit two component srsic profile with , 260 pc , and , 1 kpc . integrating out to our outer most data point ( kpc ) , we calculate and . the corresponding enclosed stellar mass is m , larger than previous estimates . apart from the central pc , with blue colors and an irregular morphology , the galaxy appears to be an early - type system . the outer color is quite red , , suggesting a dominant old population . we study the nuclear region of the galaxy using archival gemini / nifs-band adaptive optics spectroscopy and _ hubble space telescope _ imaging . we place an upper limit on the bh mass of from the nifs data , consistent with that from the-radio - x - ray fundamental plane . no coronal lines are seen , but a br source is located at the position of the bh with a luminosity consistent with the x - ray emission . the starburst at the center of henize 2 - 10 has led to the formation of several super star clusters , which are within pc of the bh . we examine the fate of the nucleus by estimating the dynamical masses and dynamical friction timescales of the clusters . the most massive clusters ( ) have myr , and thus henize 2 - 10 may represent a rare snapshot of nuclear star cluster formation around a pre - existing massive bh .

1408.4446

the nuclei of galaxies are typically inhabitated by both massive black holes ( bhs ) and nuclear star clusters ( nscs ) . bhs have been found to be ubiquitous in galaxies more massive than the milky way , but at lower masses the picture is less clear . bhs have been found in galaxies up to 100 times lower mass than the milky way @xcite , but the occupation fraction of bhs in these galaxies remains uncertain @xcite . nscs are very common in fainter galaxies ; about 75% of galaxies with stellar masses between @xmath2110@xmath22 and 10@xmath23 m@xmath10 have nscs @xcite . the formation and evolution of bhs and nscs is not well understood . bhs are thought to originally form in the early universe from direct collapse or the remnants of population iii stars , and then grow through merger - induced or secular gas accretion , or through accretion of stellar material . an indirect way of probing the formation of these objects is by examining how their mass scales with the properties of their host galaxies . these scaling relationships have been studied for both bhs and nscs , and their masses have been found to correlate strongly @xcite . initially , @xcite and @xcite found that nscs and bhs scale similarly with the mass of their host galaxies . however , this claim has been disputed by more recent work @xcite . in the context of bh and nsc formation , henize 2 - 10 is a particularly interesting object . it is undergoing a major nuclear starburst with a star formation rate 1.9 @xmath24 . a vast majority of this star formation is taking place in a very small region around the center of the galaxy with a diameter of 120 pc ( 3@xmath25 ) . there are many super star clusters ( sscs ) @xmath26 m@xmath10 forming in this nuclear starburst , and these have been age - dated to be @xmath275 myr @xcite . a luminous nuclear x - ray point source was found by @xcite . @xcite suggested it may be an intermediate - mass bh or ultraluminous x - ray source . stronger evidence for the existence of an accreting massive bh was recently found by @xcite , who showed that the x - ray source was coincident with a radio source . the luminosities of this source are consistent with a @xmath28 bh lying on the fundamental plane @xcite . @xcite find that alternative explanations for this source are not plausible , and conclude that there is likely an accreting bh . the source is not coincident with any of the known sscs and parsec - scale non - thermal radio emission is detected from the source with very long baseline interferometry observations @xcite . in this paper , we focus on three aspects of what this interesting galaxy can tell us about bh and nsc formation . first , we examine the extended structure and stellar mass of the galaxy and place it in a broader context . next , we examine the kinematics of the nuclear region , and use this information to look at whether an nsc can form from the existing population of sscs found in the nucleus . finally , we examine the infrared spectra at the location of the putative bh for any signs of bh accretion visible in the nir . henize 2 - 10 is a blue compact dwarf or wolf - rayet galaxy , with an absolute @xmath29-band magnitude of -19.07 ( @xmath30 ) . due to its low galactic latitude ( @xmath31 ) and significant foreground contamination , he 2 - 10 s larger scale structure has global structural properties are not fully understood . deep imaging of the galaxy in the nir by and the optical by suggest the galaxy has an extended red halo around the central starburst . @xcite published a de vaucouleurs sb profile fit to v band data from 6.75@xmath25 to 40@xmath25 ; here we present fits to higher quality imaging data covering a larger radial extent . the total dynamical mass of the galaxy was estimated from poorly resolved hi kinematics to be 2.9@xmath32 within 2.1 kpc and the total gas mass is @xmath33 @xcite . here we investigate the extended structure of henize 2 - 10 by fitting srsic functions to its sb profiles based on ground - based imaging data in @xmath0- and @xmath1-bands . many kinematic observations have also been made of the starburst region of henize 2 - 10 . the gas and stellar kinematics show dramatically different components . the gas is clearly rotating within the central 8@xmath25 as traced by hi , co and nir molecular lines . the rotation speeds are as high as 30 - 80 km @xmath34 , while the velocity dispersion of this component is typically @xmath27 50 km @xmath34 . on the other hand , the stellar component within the central 6@xmath25 is non - rotating and dispersion dominated . the velocity dispersion of this component is @xmath18 45 km @xmath34 which implies a dynamical mass of @xmath35 m@xmath10 within the central 6@xmath25 . we investigate the global and internal kinematics of the sscs within the nucleus using gemini / nifs observations . because of the coexistence of a bh and sscs in the central 100 pc of henize 2 - 10 , we expect they may migrate to the center due to dynamical friction and form a nsc . this idea was first introduced to explain the nucleus of m31 @xcite , and more recently has been re - examined in the context of dwarf elliptical galaxies @xcite , spiral galaxies @xcite and in the presence of a massive bh @xcite . in order to examine this phenomenon in the context of henize 2 - 10 , we examine the ssc masses , combining the nifs and archival _ hubble space telescope ( hst ) _ imaging to model their dynamical friction timescales . this paper is organized into 6 sections . in section 2 , we describe the observations and data reduction . the extended structure of henize 2 - 10 will be discussed in section 3 . our results on the future of the sscs at the center of henize 2 - 10 are presented in section 4 . in section 5 , we examine the nir spectroscopy data and report a possible signature of the bh accretion based on br@xmath17 luminosity . we conclude in section 6 . we assume a distance to henize 2 - 10 of 9 mpc ; the physical scale assuming this distance is @xmath18 43 pc @xmath36 . unless otherwise indicated , all quantities quoted in this paper have been corrected for a foreground extinction @xmath37 ( @xmath38 , @xmath39 ; * ? ? ? images of henize 2 - 10 were taken on nov . 21 , 2011 with megacam on the magellan / clay telescope . the seeing was poor ( 1.5 - 2@xmath25 ) , but the night was photometric . five 100s exposures were taken in @xmath0- , @xmath1- , and @xmath40-band at an airmass of @xmath18 1.15 ; the i band data suffered from fringing and poor sky subtraction and thus was not used in our analysis . the images were dithered to remove the chip gaps . final , flat - fielded images were created by the pipeline at the smithsonian astrophysical observatory telescope data center ; we made further corrections to the sky subtraction on each chip to ensure a flat field . specifically , we fit a gradient to the residual background in each chip to a linear function along the horizontal direction and subtracted this off . we then matched the background in neighboring fields to ensure as flat a field as possible before combining the images using swarp @xcite . flux calibration was obtained through observations of equatorial sdss fields in each band , and are thus on the sdss system . gemini / nifs data of henize 2 - 10 was taken in april 2010 using altair adaptive optics for a program ( pi : usuda ) to examine the sscs and gas conditions within the galaxy . we downloaded the k - band data and accompanying calibration observations from the gemini science archive . the total data included 9 on - source and 9 off - source exposures , each with an exposure time of 300s . the data were taken at airmass between 1.5 and 1.9 . data reduction was performed as outlined by @xcite , with telluric calibration performed using an a0v star ( hip45037 ) taken at similar airmass . flux calibration of the spectroscopic cube was performed by using a zeropoint derived from telluric stars and a 2mass @xmath41 filter response curve ; the flux calibration is expected to be good to @xmath18 10% . final cubes were constructed from 8 of the 9 sky - subtracted on - source cubes with good image quality . the line spread function in each pixel was determined by using a sky cube created with the same dither pattern as the science data cube ; the median fwhm is 4.15(r @xmath18 5500 ) , with a variation from 3.4 and 4.9 across the map . the fwhm of the ao corrected psf in the final cube is quite good ; the brightest ssc , which we refer to as cltc1 following @xcite , has a measured fwhm of @xmath42 . from _ hst / hrc _ images ( 2.3 ) , we measured a true fwhm of @xmath43 , suggesting the true psf is @xmath44 fwhm ( 3 pixels in our final cubes ) . the signal - to - noise ratio ( snr ) per pixel ranges from 5 - 65 across the cube . the nifs data cubes were astrometrically aligned to previous measurements using _ hst _ data aligned to 2mass astrometry as in @xcite . the absolute error in astrometry is thus @xmath45 , and the astrometric error relative to the nuclear x - ray source is 0@xmath4613 in ra and 0@xmath4633 in dec , while the alignment of the nuclear radio source is @xmath471 ( supplementary materials ; * ? ? ? all velocities are corrected to the barycentric velocity ; the correction applied was -20.35 km @xmath34 . we use _ hst _ data in two filters in this paper : ( 1 ) _ acs / hrc _ data in the _ f814w _ filters taken in november 2005 ( program i d : 10609 , pi : vacca ) and ( 2 ) _ nicmos nic2 _ data in the _ f205w _ filter taken in october 2006 ( program i d : 10894 , pi : johnson ) . due to the high resolution of the _ f814w _ data , we used it to select the brightest star clusters in the starburst region as shown in fig . these images are used in combination with the gemini / nifs data to determine the masses of sscs and their possible dynamical friction timescales of forming an nsc at the center of henize 2 - 10 ( 4 ) . we used the procedure described in @xcite to derive stellar kinematics in the central starburst region of henize 2 - 10 . before deriving kinematics , we bin spectra using the voronoi binning method @xcite to ensure a signal - to - noise , snr , of @xmath1825 in each bin . next , we determined the radial velocity , @xmath48 , the velocity dispersion , @xmath49 , the skewness , @xmath50 , and the kurtosis , @xmath51 , in the wavelength region from 2.28 @xmath52 m to 2.40 @xmath52 m including the position of co absorption bandheads by using the ppxf code of @xcite . this code requires stellar templates ; we used high resolution templates from @xcite of eight stars with spectral types between g and m and including all luminosity classes . the program finds the best matching template from the set of templates , convolves them with the line spread function determined from fits to sky lines in each pixel , and finds the best light - of - sight velocity distribution ( losvd ) which is described in term of gauss - hermite series . errors on the losvd are calculated from monte carlo simulations adding in gaussian random error to each pixel and redetermining the velocities . these radial velocity and dispersion errors range from 0.3 km @xmath34 to 25 km @xmath34 ; we note that systematic errors due to sky and background subtraction , telluric correction , and template mismatch are likely of order a few km @xmath34 ; more discussion of the systematic errors is presented in 4.2 . based on visual inspection , errors in dispersion larger than 20% were unreliable , and therefore we eliminate these bins from our analysis . although we do not cover the full central region , the median velocity in the higher snr regions of our maps is 876 km @xmath34 or @xmath184 km @xmath34 offset from the systemic velocity of 872@xmath536 km @xmath34 given by . we therefore use the marquart systemic velocity value throughout this paper . [ cofig ] shows the velocity and dispersion maps of the stellar component derived from the co bandhead . we note that our data do not cover the entire nuclear region of henize 2 - 10 ; the expected location of the bh is near the western edge of our maps and is marked . by a red cross in fig . [ cofig ] . no peak in total intensity or change in dispersion or velocity is seen at the location of x - ray and radio source associated with the bh . the kinematic maps show two distinct regions . the stars clusters are blue shifted relative to the surrounding stars by 20 km @xmath34 to 30 km @xmath34 . the dispersions in these regions are low , @xmath54 km @xmath34 . the clusters are apparently rotating in the same sense as the gas component , with an amplitude somewhat less than the gas . outside of the clusters , the kinematics have a higher dispersion of @xmath18 45 km @xmath34 and do nt appear to have strong rotation . the kinematics of these non - cluster stars are consistent with the lack of rotation and @xmath18 45 km @xmath34 dispersion observed over the central 6@xmath25 region by . thus it appears the dispersion profile of the non - cluster component within the inner @xmath186@xmath25 is nearly flat . we note that in the central region where we are observing , the seeing limited measurements of find a dispersion of @xmath1828 km @xmath34 , likely due to the influence of the lower dispersion star clusters . we discuss dispersion measurements of the most massive cluster , cltc1 , in 4.2 . to examine the extended structure and nature of henize 2 - 10 , we constructed sb profiles from the @xmath0- and @xmath1-band magellan / megacam images ( 2.1 ) . we first masked out all foreground stars and background galaxies by using @xmath55 to determine the shape , size , and position of each source . each object s size was increased by a factor of 2 while masking . the sb profiles were then derived using the @xmath56 routine to extract fluxes , major semiaxis , etc . while extracting the fluxes , we use the best fit position angle and ellipticity from ; the ellipticity is fixed at @xmath57 , and the position angle is fixed at @xmath58 . the magellan @xmath1-band images with the contours are presented in fig . [ contourfig ] . the @xmath1-band sb profile is shown in fig . [ sersicfig ] . the total luminosity of the galaxy in @xmath0- and @xmath1-band between 0@xmath25 and 100@xmath25 is @xmath59 and @xmath60 . the effective ( half - light ) radius of the overall galaxy is 30@xmath460 ( 1.3 kpc ) and 32@xmath460 ( 1.4 kpc ) in @xmath0- and @xmath1- band , respectively . the central most region of henize 2 - 10 is dominated by a young starburst ; to focus on the extended structure and minimize the affect of this starburst we only fit the profile beyond 2@xmath25 . we note that because our psf fwhm is smaller than this radius , the affects of the psf on our sb profile fits are minimal . the outer boundary of our fits are set by the error in our sky subtraction ; this uncertainty in the sky subtraction is shown by black solid lines in fig . [ sersicfig ] ; beyond 70@xmath25 the sky subtraction makes the sb profile less reliable . the outer parts were fit by a double - srsic profile for both @xmath0- and @xmath1-images from @xmath61 ( 86 pc ) to @xmath62 ( 4.3 kpc ) from the center . here we used the non - linear least squares idl mpfit function to implement the fitting procedure . henize 2 - 10 s ( sb ) profile in both bands is best fit by two srsic components , an inner component with @xmath63 and @xmath64 ( 260 pc ) and an outer component with @xmath65 and @xmath66 ( 1 kpc ) . the full results and errors in each band are shown in table [ table : morphohe2 - 10 ] . [ sersicfig ] shows both the best double- and single - srsic fit profiles in r - band . the single - srsic fit to the r - band image provides a worse fit , but has @xmath67 ( a de vaucouleurs profile ) and @xmath68 ( 538 pc ) . in addition to having larger residuals , the single - sersic fit ( 1 ) does not well match the observed color profile , ( 2 ) the @xmath69 are @xmath1825% larger than that of the double - srsic profiles , and ( 3 ) over - predicts the amount of light even with the overlying starburst component at the center of the galaxy . therefore we prefer the double - sersic fit and use this fit throughout the paper . we note that while @xcite do not provide the effective radius of their fits , both our single and double - sersic fits are better fits to the data than the de vaucouleurs profile fit they show in their fig . 3 of @xcite . specifically , their fit is poor at radii less than 14@xmath25 . to estimate the effect of our choice of inner and outer boundaries on the fit , we varied the inner boundary from 2@xmath25 to 10@xmath25 and outer boundary from 70@xmath25 to 110@xmath25 for both @xmath0- and @xmath1-images ; this resulted in a variation of the inner srsic index and effective radius of 1% and the outer fitted srsic sb profile less than 5% . our results therefore appear robust to the range over which the profile is fitted . the profiles are fit well by the two srsic components model , with residuals less than @xmath18 20% except for an outer starburst component seen at a radius of @xmath70 . the colors throughout the region ( except for the starburst ) are consistent with a @xmath711 gyr old population and thus the sb profile we fit appears to describe the older underlying component of henize 2 - 10 ; this is especially true for the outer component , which dominates beyond @xmath1810@xmath25 . to test our results , we compared our profiles with those in bessel - johnson @xmath72 filters using conversions between sloan and johnson filter sets @xcite . the transformed profiles agree from the center to 90@xmath25 to within 10% , beyond this distance , a somewhat larger systemic discrepancy of @xmath1820% exists between the @xmath0- and @xmath29-band . this discrepancy does not affect our profile fits which are performed only out to 100@xmath25 . we also fit all 5 filters @xmath72 in the same range of radii from 2@xmath25 to 100@xmath25 with the similar double - srsic profile which we obtained from our @xmath0- and @xmath1-band data . the profiles fit well with residuals less than @xmath1820 % in each band . this suggests that the two component structure we derive from our data is robust . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] [ table:3partmass ] to determine the total mass of henize 2 - 10 , we split the galaxy into multiple components ; for the inner component we derived the mass using kinematics , while at intermediate and large radii , we determined the mass - to - light ratio @xmath73 from the integrated color of the galaxy . these results are summarized in table [ table:3partmass ] . in the outermost of the galaxy beyond 20@xmath25 , the color is nearly constant , @xmath74 . using the dependence of mass - to - light ratio , @xmath73 , on the color as describing in @xcite , we calculated @xmath75 , @xmath76 , respectively at large radii . the luminosity beyond 20@xmath25 is @xmath77 and @xmath78 corresponding to a total outer mass of @xmath79 and @xmath80 from the @xmath0- and @xmath1-band . due to the large number of foreground sources requiring masking , the uncertainty in the sky background is substantial at large radii ( see fig . [ sersicfig ] ) this is the dominant source of error in our luminosity estimate and these errors are included in the luminosity and mass uncertainties listed in tables [ table : morphohe2 - 10 ] and [ table:3partmass ] . at intermediate radii , between 6@xmath25 and 20@xmath25 , the color and thus the mass - to - light ratio varies significantly . the annulus colors , @xmath81 , are converted into annulus mass - to - light ratios using the @xcite relations . these vary between 1.0 and 2.5 in @xmath1-band and vary between 1.0 and 3.5 in @xmath0-band . the total mass between 2@xmath25 and 20@xmath25 are @xmath82 and @xmath83 for @xmath0- and @xmath1-band . the error bars on these data points include the uncertainty on integrated luminosity within this annulus and the scatter @xmath81 , which are converted into the mass - to - light ratio uncertainties @xcite . between radii of @xmath61 and @xmath84 , we can estimate masses using both the photometric method we used at larger radii , and kinematic estimates . we estimate the virial mass based on the dispersion in the inner @xmath84 of @xmath85 km @xmath34 measured by and their virial mass estimator . we derive a mass @xmath86 between @xmath61 and @xmath84 . this dynamical mass agrees well with the photometric mass estimates in this same annulus ( table [ table:3partmass ] ) ; for our final calculations we use the @xmath1-band based photometric mass within this annulus . because of the lack of resolution in our megacam images , the inner color and sb profile within @xmath61 from the center of henize 2 - 10 are difficult to determine . therefore , we obtained an estimation of the mass of the inner starburst component from kinematics . our gemini / nifs kinematics show the constant 45 km @xmath34 dispersion seen by continues to within the central @xmath61 . this corresponds to a dynamical mass of @xmath87 within 2@xmath25 , which we can add to our photometric estimates from larger radii . the @xmath1-band luminosity of this region is @xmath88 , and therefore the mass - to - light ratio in this region drops to @xmath89 as we might expect for a young population . summing the different radial components in table [ table:3partmass ] together , henize 2 - 10 has a total luminosity within 100@xmath25 ( 4.3 kpc ) of @xmath90 ( @xmath91 ) . the total stellar mass in this aperture is @xmath92 , corresponding to an average mass - to - light ratio of @xmath93 . [ m2l_massfig ] shows the resulting mass and mass - to - light ratio profile of the galaxy in the @xmath1-band , @xmath94 including the dominant error from the sky determination . the 100@xmath25 aperture may not enclose all the galaxy s light and mass ; integrating our sb profile fits to infinity yields @xmath95 or about 10% more light than within our aperture . the mass derived from @xmath0-band is consistent within the errors with the @xmath1-band mass . due to the reduced effects of extinction and the good match of our @xmath1-band profile to the profiles , we quote final numbers on the mass based on the @xmath1-band data . we note that while the inner starburst region ( @xmath96 ) contributes about 23% of henize 2 - 10 s @xmath1-band luminosity , it contributes about @xmath18 3% to the total stellar mass of the galaxy . a dynamical estimate of henize 2 - 10 s total mass within 2.1 kpc was derived from hi kinematics to be 2.7@xmath32 by @xcite . our mass estimate within this same aperture is 6.4@xmath97 m@xmath10 from @xmath1-band , thus implying inclinations of @xmath9838@xmath99 . this translates into an expected ellipticity of @xmath100 assuming a thin disk . this is consistent with the ellipticity of 0.15 of the outer isophotes in our ellipse fits and that of , suggesting that the gas disk is aligned with the outer stellar component . other previous photometric mass estimates of henize 2 - 10 have suggested a lower mass for henize 2 - 10 @xmath101 with factor of 3 uncertainties . these estimates are based on k - band absolute magnitudes derived from @xmath102 over an unspecified aperture and from 2mass @xmath103 @xcite . these translate to a @xmath41-band luminosity of @xmath104 . comparing our integrated magnitudes , the color is @xmath105 , bluer than expected for a predominantly old population . however , the color at radii beyond @xmath106 between our data and is @xmath107 and the profile appears to get redder in the inner part of the galaxy likely due to hot dust ; thus it appears the integrated @xmath41-band luminosities are underestimated . this may be due to use of a smaller aperture than our data . we also note that @xmath41 band @xmath108 are more uncertain than optical @xmath109s due to the poorly understood contributions of supergiant and agb stars ( e.g. , * ? ? ? * ; * ? ? ? although we can not separate out the masses of the two best - fit srsic components exactly , we use the @xmath94 profile to estimate the mass for both components . for the inner component , the integrated luminosity and the average mass - to - light ratio are @xmath110 ( @xmath111 ) , thus the inner component mass is @xmath112 . similarly , we obtained the integrated luminosity @xmath113 ( @xmath114 ) , and the mass @xmath115 for the outer component . apart from the central starburst component , henize 2 - 10 appears to be an early - type galaxy . this is suggested by ( 1 ) its central dispersion dominated kinematics , ( 2 ) the consistently red color of its outskirts . we argue below that its outer component srsic index ( @xmath65 ) is also consistent with this interpretation . it is useful to compare the morphology and luminosity of henize 2 - 10 to other early - type galaxies to better understand its nature and whether or not the presence of a massive bh is suprising . to facilitate this comparison , we placed our fitted models on diagrams of early - type galaxies from @xcite . we plotted each component of our double - srsic fits on the plot of absolute magnitude vs. srsic @xmath116 and effective radius ; the inner component is clearly an outlier component in both plots , but the outer component lies at the boundary between what kormendy considers spheroidal and elliptical galaxies . one possible interpretation of this is that henize 2 - 10 was a typical early - type galaxy before some event formed the nuclear component . if this is the case , the stellar populations even outside the nucleus should be somewhat younger than the surrounding envelope . in the future , this may result in an age gradient similar to that seen in m32 @xcite and other lower mass early - type galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . we can also consider the galaxy as a single component ( despite a worse fit to the profile ) , in which case our single - srsic fits place it firmly amongst the elliptical galaxies with a high srsic index . assuming our preferred two component model , the outer component is typical of an old early - type galaxy . the inner component , with its varying @xmath109 , likely represents more recent star formation ( perhaps the older portion of the current central starburst ) , and thus its position off the early - type galaxy locus is unsurprising . the single component fit overpredicts the flux in the center . given that most of this flux is in a young starburst it is clear that a single @xmath116=4 srsic profile does not accurately describe the old population or mass profile of the galaxy . the effective radius and srsic index therefore are unlikely to indicate much about its physical origin . in the context of henize 2 - 10 being a relatively massive @xmath117 m@xmath10 early - type galaxy , it is not surprising that henize 2 - 10 hosts a bh . while dynamical evidence for bhs in galaxies of this mass is almost non - existent @xcite , agn in this mass range and lower have been found @xcite . analyzing these observations , @xcite and @xcite suggest that the occupation fraction of bhs in early - type galaxies around @xmath118 m@xmath10 is high , @xmath1890% . this implies that the bh may be a pre - existing object , and not a result of the intense starburst taking place in the galaxy . finally , henize 2 - 10 appears to be very isolated , with no known companions within @xmath119 km @xmath34 and 2 degrees ( @xmath18 0.3 mpc ) of the galaxy . thus the most plausible explanation for henize 2 - 10 s relatively steep srsic index and current starburst is that it is a late stage merger of two other smaller galaxies as suggested by @xcite and . henize 2 - 10 contains several super star clusters ( sscs ) in the central starburst region and a massive bh with approximative mass of @xmath28 ( with an uncertainty of a factor of @xmath120 ) at its center @xcite . we expect the bh and sscs orbits to decay due to the gravitational interaction with the surrounding galaxy . there are two popular models of nsc formation : ( 1 ) dissipative model where the nsc forms _ in situ _ from gas accreted into the center and ( 2 ) the dissipationless model where migration of massive star clusters to the center forms the nsc @xcite . the lack of an existing nsc combined with the presence of a large number of massive young clusters suggests that henize 2 - 10 may be an example of ongoing dissipationless nuclei formation . we examine the expected dynamical friction timescale of this merging in this section based on models by @xcite . assuming that sscs have king density profiles , orbit within a galaxy with a power - law profile @xmath121 , and accounting for mass loss due to the galactic tidal field , the dynamical friction timescale for sscs to reach the center of the galaxy from dynamical friction mechanism is ( eq . 32 of * ? ? ? * ) : @xmath122 with the central density , @xmath123 , the scale radius , @xmath124 , the coulomb logarithm , @xmath125 , the cluster mass , @xmath126 , the central dispersion velocity of ssc , @xmath127 km @xmath34 , and the initial distance , @xmath128 , of the ssc from the galactic center . the @xmath17 dependent coefficient , @xmath129 for @xmath130 corresponding to the analytical formula can be found in @xcite . @xmath131 where @xmath132 . @xmath133 has a value from 2 to 7 @xcite . we note that the equation ( 1 ) does not take into account the presence of a bh ; however , given the low mass ( @xmath134 @xmath135 ) and small sphere of influence ( a few pc ) of the bh , clusters should reach a radius similar to the size of typical nscs before feeling the effects of the bh . we also caution that in some cases , dynamical friction appears to be less efficient than predicted from theory . most notably , @xcite suggest that the faintness of nuclei in low mass de galaxies ( @xmath136 ) indicates inefficient dynamical friction , perhaps due to supernova feedback or mismatches of the dark matter halo with the visible baryonic component . given the much higher luminosity of henize 2 - 10 and the likely dominance of baryonic matter near the center of the galaxy , it is in a regime where @xcite find evidence for dynamical friction working as expected . we apply these dynamical friction timescale results to explore the possibility of forming a nsc at the center of henize 2 - 10 due to the migration of sscs . before we can do this , we need to determine several parameters for the galaxy and clusters , which we do in the following subsections . this includes a determination of the cluster galactocentric positions , profiles , and masses , as well as the density profile of the galaxy . we derive these properties using a combination of high resolution _ acs / hrc _ imaging and gemini / nifs spectroscopy . we identified the brightest 11 sscs in the nuclear region using the _ hst _ images as shown in fig . [ sscs ] . these clusters are also seen in the gemini / nifs observations , but are not as well resolved ; the brightest four clusters are clearly visible in both data sets and allow us to astrometrically align our gemini / nifs image with the _ hst _ data . four of the sscs identified here were previously identified by @xcite ; we label these cltc1 - 4 . the brightest ssc is cltc1 , and this cluster is shown at the center of our maps . seven other bright sscs are also identified in the _ hst _ imaging , and we number these 5 through 11 . the coordinates of all clusters are given in table [ table : cltc1 - 4 ] . for purposes of our dynamical friction timescales calculation , we assumed the galactocentric radius of the clusters is their projected distance to the bh . given the compact region over which clusters are found , it is likely that this distance is within @xmath1850% of their true galactocentric distance . for all 11 clusters , we performed king model fits to the _ hst acs / hrc _ @xmath137 images . we followed the procedure described in @xcite assuming concentration parameter , @xmath138 . we fit the data with the ishape rountine which is part of the baolab software @xcite . to fit each ssc s size ( the core radius or effective radius ) in the image , ishape needs a corresponding psf , and a charge diffusion kernel image . the psf is generated by the tiny tim routine @xcite and oversampled by a factor of 10 for use with ishape . the charge diffusion kernel is generated from header file of distorted psf image when we ran the @xmath139 task of tiny tim routine . the core radius , @xmath140 , and effective radius , @xmath141 , from the best fit king model are shown in column 4 and 5 of table [ table : cltc1 - 4 ] . integrated fluxes of the clusters in both @xmath137 and @xmath142 are calculated using aperture photometry ( see 4.2 ) . the luminosity in @xmath137 is given in column 6 , while the color @xmath143 in vega system of integrated fluxes of the two bands is given in column 9 . this color is useful for evaluating the similarities in stellar population and reddening among the clusters ( see 4.3 ) . ccccccccc sscs&@xmath144 & @xmath145&@xmath146&@xmath140&@xmath147&@xmath148&@xmath149 & @xmath143 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) + & ( km @xmath34 ) & ( km @xmath34 ) & ( pc ) & ( pc)&@xmath150&@xmath151 & & + & & & & & & & & + cltc1 @xmath152 & & & & & & & + @xmath153 & 17.6 @xmath53 1.6 & 20.1 @xmath53 0.6 & 3.43 @xmath53 0.07 & 1.27 @xmath53 0.03 & 3.59 @xmath53 0.93 & 2.30 @xmath53 0.60 & 0.06 @xmath53 0.02 & 2.2 + & & & & & & & & + & & & & & & & & + cltc2 @xmath154 & & & & & & & + @xmath155 & 13.8 @xmath53 1.2 & & 2.07 @xmath53 0.03 & 0.77 @xmath53 0.01 & 1.45 @xmath53 0.56 & 0.92 @xmath53 0.61 & & 2.0 + & & & & & & & & + cltc3 @xmath156 & & & & & & & + @xmath157 & 15.4 @xmath53 1.3 & & 2.07 @xmath53 0.09 & 1.00 @xmath53 0.02 & 1.78 @xmath53 0.74 & 1.14 @xmath53 0.78 & & 2.1 + & & & & & & & & + cltc4 @xmath158 & & & & & & & + @xmath159 & 14.4 @xmath53 1.1 & & 1.89 @xmath53 0.11 & 0.71 @xmath53 0.05 & 1.42 @xmath53 0.53 & 0.91 @xmath53 0.60 & & 1.9 + & & & & & & & & + & & & & & & & & + 5 @xmath160 & & & & & & + @xmath161 & 11.0 @xmath53 1.0 & & 1.42 @xmath53 0.13 & 0.53 @xmath53 0.07 & 0.63@xmath53 0.12 & 0.40 @xmath53 0.21 & & 1.7 + & & & & & & & & + 6 @xmath162 & & & & & & + @xmath163 & 11.0 @xmath53 1.5 & & 1.42 @xmath53 0.17 & 0.53 @xmath53 0.04 & 0.63 @xmath53 0.12 & 0.40 @xmath53 0.21 & & 1.8 + & & & & & & & & + 7 @xmath164 & & & & & & + @xmath165 & 9.1 @xmath53 1.1 & & 2.38 @xmath53 0.34 & 0.89 @xmath53 0.21 & 0.72 @xmath53 0.12 & 0.46 @xmath53 0.25 & & 2.0 + & & & & & & & & + 8 @xmath166 & & & & & & + @xmath167 & 12.5 @xmath53 0.8 & & 1.24 @xmath53 0.04 & 0.46 @xmath53 0.02 & 0.70 @xmath53 0.12 & 0.45 @xmath53 0.24 & & 1.8 + & & & & & & & & + 9 @xmath168 & & & & & & + @xmath169 & 9.3 @xmath53 0.6 & & 1.00 @xmath53 0.03 & 0.37 @xmath53 0.01 & 0.32 @xmath53 0.07 & 0.20 @xmath53 0.11 & & 2.1 + & & & & & & & & + 10 @xmath170 & & & & & & + @xmath171 & 11.6 @xmath53 1.3 & & 1.45 @xmath53 0.05 & 0.46 @xmath53 0.02 & 0.70 @xmath53 0.12 & 0.45 @xmath53 0.24 & & 2.0 + & & & & & & & & + 11 @xmath172 & & & & & & + @xmath173 & 6.2 @xmath53 0.9 & & 2.21 @xmath53 0.07 & 0.37 @xmath53 0.01 & 0.32 @xmath53 0.07 & 0.20 @xmath53 0.11 & & 1.6 + & & & & & & & & + [ table : cltc1 - 4 ] @xcite estimated masses of the 4 brightest sscs , cltc1 - 4 , using stellar population model fits of ultraviolet and blue _ hst / stis _ spectra , finding a mass for clct1 of @xmath18 4.5 @xmath17410@xmath175 m@xmath10 . however , these fits have significant uncertainties due to dust extinction . here , we measured the dispersion of the brightest cluster , cltc1 , from our gemini / nifs data and used it to determine a more robust virial mass estimation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? we could only determine the internal dispersion of cltc1 because of galaxy contamination in the other fainter clusters . . [ ssc1st ] fig . [ ssc1st ] shows the full spectrum and co absorption bandhead of cltc1 from gemini / nifs data after subtracting an estimation of the background light of the galaxy . we estimated the galaxy background value by taking the median value of pixels surrounding cltc1 but avoiding contamination from neighboring clusters . for our gemini / nifs data of the sscs , only the spectrum of cltc1 was cleanly separated from the surrounding galaxy . specifically , using annuli surrounding the clusters , we find that the background contamination in cltc1 is @xmath1820% , while in the rest of the clusters it is @xmath7145% . as for the galaxy as a whole , we fit the kinematics in the co absorption bandhead to estimate velocity dispersion of cltc1 using ppxf ( see 2.4 ) . as the cluster is nearly unresolved in our data , we sum the light within an 0@xmath462 radius aperture over which the source is well above the galaxy background . our resulting best - fit dispersion for this spectrum is @xmath176 km @xmath34 . given that this aperture extends out @xmath182@xmath174 the effective radius of the cluster , we assume that this dispersion is reflective of the cluster s global velocity dispersion @xmath177 , which is used in virial mass estimation . to assess the robustness of this dispersion , we conducted several tests : ( 1 ) we determined the dispersion value in just the central pixel of the cluster to be @xmath178 km @xmath34 . ( 2 ) we varied the pixels used for background subtraction and find dispersion results that vary from 17 to 21 km @xmath34 . ( 3 ) we fit the data to no dispersion model and find that the resulting reduced @xmath69 is higher than the best - fit model by 20% . these tests suggest that our dispersion value is clearly resolved by the data , but that the systematic error is more like @xmath183 km @xmath34 . the cluster s radial velocity is blue - shifted by 6 km @xmath34 relative to the systemic velocity of the galaxy . the virial mass is calculated using @xcite : @xmath179 where @xmath177 is the global velocity dispersion , @xmath180 is the half mass radius , and @xmath181 , @xmath141 is the effective radius or half - light radius . the structural parameter , @xmath182 , of globular clusters ( gcs ) is 2.5 @xcite . we determined @xmath180 or @xmath183 from king model fits as discussed above . based on the measured dispersion of cltc1 , we estimated a mass of 2.3 @xmath184 . assuming a 3 km @xmath34 error on the dispersion , this translates to a @xmath185% error ( 0.6@xmath184 ) on the mass . for cltc1 , @xcite found a mass of 4.5 @xmath186 from their spectral fitting , assuming a starburst 99 @xcite model with a salpeter initial mass function ( imf ) and with lower and upper masses of 1 @xmath187 and 100 @xmath187 , respectively . this mass is a factor of 5@xmath174 lower than our dynamical mass estimate of 2.3 @xmath184 . this discrepancy could be due to the assumption by @xcite of a 1 m@xmath10 lower limit on the imf mass in their models which will lead to an order of magnitude underestimate in the total stellar mass relative to a kroupa imf @xcite . in this section , we made rough estimates of the masses of the other optically visible clusters in the nuclear region using our derived m / l for cluster cltc1 . while this assumption is an undoubtedly oversimplification , it is a reasonable one for two reasons . first , the star clusters studied in this region have all been found to have ages @xmath27 5 myr @xcite , as might be expected from the prodigious star formation rate of @xmath188 . second , our measurements of the color ( @xmath143 ) between @xmath137 and @xmath142 bands ( table [ table : cltc1 - 4 ] ) in most of the clusters suggests they have similar ages and extinctions because the colors change slightly over the nifs field of view ( fov ) and _ hst _ data . we noted that colors were calculated from a fixed @xmath189 aperture in each cluster ; no correction was made for the differing psf between the bands ; however , the difference in encircled energies for a point source is relatively small ( 75% in _ f814w _ , 65% for _ f205w _ ) , and likely quite similar for all clusters . the clusters luminosities used here are the integrated king model luminosities which are obtained from the ishape rountine . each cluster s luminosity is ouput as a parameter , we then correct for galactic extinction to get the total luminosity shown in column 6 of table [ table : cltc1 - 4 ] . the mass - to - light ratio of cltc1 in @xmath137 estimated using the virial mass estimate from 4.2 gives a @xmath190 . this value is quite consistent with the mass - to - light ratio in h band of 0.05 found for the larger starburst region by . assuming this constant mass - to - light ratio for all the sscs , we derived masses as given in table [ table : cltc1 - 4 ] . we found a total mass in young clusters in the nuclear region to be @xmath187.4 @xmath184 . this nuclear region overlaps with region a from @xcite , who derived a total mass between @xmath191 from uv spectra . this large discrepancy might arise from their assumed imf , which like @xcite used salpeter imf between 1 - 100 @xmath135 ( see above ) . ccccc sscs & @xmath128 & @xmath192&@xmath193 & @xmath194 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) + & ( pc ) & ( pc ) & ( pc ) & @xmath195 + & & & & + cltc1 & 64.8 @xmath53 1.4 & 7 & 6 & @xmath196 0.5 + & & & & + & & & & + cltc2 & 36.9 @xmath53 1.4 & 5 & 4 & @xmath197 0.2 + & & & & + cltc3 & 20.4 @xmath53 1.4 & 2 & 7 & @xmath198 0.04 + & & & & + cltc4 & 11.2 @xmath53 1.4 & 2 & 3 & @xmath199 0.01 + & & & & + 5 & 86.4 @xmath53 1.4 & 7 & 2 & @xmath200 10 + & & & & + 6 & 77.6 @xmath53 1.4 & 6 & 2 & @xmath201 7 + & & & & + 7 & 114.5@xmath53 1.4 & 6 & 6 & @xmath202 1 + & & & & + 8 & 130.7@xmath53 1.4 & 7 & 3 & @xmath203 3 + & & & & + 9 & 74.1 @xmath53 1.4 & 4 & 2 & @xmath204 5 + & & & & + 10 & 70.2 @xmath53 1.4 & 4 & 2 & @xmath205 2 + & & & & + 11 & 55.8 @xmath53 1.4 & 2 & 5 & @xmath201 7 + & & & & + [ table : ssc_dynafriction ] in addition to the cluster parameters , we require the central density profile of the galaxy to determine the dynamical friction timescales . following @xcite , we characterize the central starburst region of henize 2 - 10 with a simple power - law density profile , @xmath121 . there are three parameters we need to specify : the central stellar mass density , @xmath206 , the scale radius of the bulge , @xmath207 , and power - law index , @xmath17 . due to the presence of the central starburst , measuring the central profile of the galaxy is not straightforward . while our sb profile fits did not include the central region , we can extrapolate the expected density from those profiles to determine the central density profile . we are thus assuming the central density is dominated by the extrapolation of the older stellar populations as captured by our two - srsic component fit , and not the young starburst which likely contributes only a small fraction of the mass despite dominating the optical luminosity at the center . to facilitate deprojection of the sb profile to a 3d mass density profile , we create a multiple gaussian expansion ( mge ) model based on the two component sb profile presented in 3 . in order to translate this luminosity profile into a mass profile , we use the observed kinematics in the central region to constrain the mass - to - light ratio . specifically , we created a jeans axisymmetric model @xcite to predict the dispersion in the central region based on the mge model . we fix the mass - to - light ratio of the outer component s srsic profile to @xmath208 , while varying that of the inner srsic component to match the observed dispersion . the best - fit requires a @xmath209 for this inner component . we then fit the deprojected mge density model to the @xmath121 profile within the central 3@xmath25 . the best fit parameters are : @xmath210 , @xmath211 pc , and @xmath212 . now that we have a model for the central potential in the galaxy , we can determine whether each cluster is currently tidally limited and whether the clusters will reach the center without being disrupted by the gravitational field of the galaxy and bh . specifically , we use ( eq . 23 of * ? ? ? * ) to calculate each cluster limiting tidal radius @xmath213 : @xmath214^{-1/2}\ ] ] where @xmath1 is the projected distance between each cluster and the position of the bh , and @xmath215 is the bh s mass . the projected distance of each ssc is listed in the column 2 of table [ table : ssc_dynafriction ] . @xmath216 is ssc s velocity dispersion which is estimated from the mass and profile estimate listed in the column 2 of table [ table : morphohe2 - 10 ] , except for cltc1 where we use our measurement . we find that in each cluster the limiting radius is larger than the best - fit core radius , but smaller than the tidal radius ( we fit all data to a @xmath138 king model , thus @xmath217 listed in column 3 of table [ table : ssc_dynafriction ] ) . this means each cluster is already being limited by the gravitational potential of the galaxy , and thus should have a dynamical friction timescale given by equation 1 which incorporates the affects of mass loss as the cluster moves inwards . the clusters may be disrupted by the external gravitational field before reaching the center of the galaxy ; following @xcite we assume this occurs when the limiting radius is equal to the core radius of each cluster . this galactocentric radius is given in column 4 of table [ table : ssc_dynafriction ] ; each cluster should survive until it is within the central 10 pc ( and in most cases the central 5 pc ) . given that typical nscs have effective radii of @xmath184 pc ( e.g. , * ? ? ? * ) , each cluster is likely to add at least some of its mass to a forming nsc if their dynamical friction timescales are short enough . internal processes could also shorten the lifetimes of these clusters ; substantial natal gas or stellar wind mass loss could cause the clusters to dissolve . however , the clusters are already optically visible and thus have likely lost their natal gas . furthermore , stellar wind mass loss represents @xmath2730% of the clusters mass within the first 500 myr of a cluster @xcite , and thus this effect is unlikely to dissolve these massive clusters . finally , we expect dissolution from two body relaxation to be very slow ; the clusters have relaxation times of @xmath218 yr ; dissolution due to dynamical evolution occurs over many relaxation times @xcite . finally , we calculate the clusters dynamical fricition timescales using eq . ( 1 ) ( eq . 32 of * ? ? ? these timescales are given in column 5 of table [ table : ssc_dynafriction ] ; for the most massive clusters these timescales are @xmath2710@xmath219 years . the most massive cluster , cltc1 has a timescale of just 170 myr . thus a nuclear cluster well in excess of @xmath220 m@xmath10 is likely to form within a few hundred myr . the end product will have the appearance of a nsc ; the faintest nscs in @xmath117 m@xmath10 galaxies typically have nscs of @xmath220 m@xmath10 @xcite . because the timescale of formation is so short relative to the age of the universe , the state we are seeing in henize 2 - 10 should be rare . henize 2 - 10 thus provides an interesting snapshot of nsc formation in action and constraints on the coevolution of nscs and bhs . while there have been previous suggestions that massive clusters may provide the seeds for bhs ( e.g. , * ? ? ? * ) , if there is any coevolution between the two components in henize 2 - 10 it would be due to merger fed mutual growth as modeled by @xcite . thus while the nsc is forming , the bh may also grow through gas accretion or the accretion of stars @xcite . however , as we argued in 3.3 , it appears likely based on the galaxy s overall mass and early - type nature that a bh was already present in the galaxy , and that the growth of the nsc is occurring after bh formation . this would suggest that the formation of the nsc and bh , at least in this galaxy , is largely uncorrelated . there are several ways an accreting bh could be detected from our @xmath14-band nifs spectrum : ( 1 ) dynamical modeling of the stellar velocity dispersion in the vicinity of bh ( e.g. , * ? ? ? * ) , ( 2 ) detection of coronal line emission such as , which requires extreme emission from the region around the accreting bh to excite ( e.g. , * ? ? ? * ) , or ( 3 ) detection of other emission lines , such as br@xmath17 , which might be associated with x - ray radiation from the accreting region . to determine if the bh may be dynamically detectable from our kinematics observations we first consider the sphere of influence of the bh . the mass of the bh was estimated to be @xmath221 m@xmath10 by @xcite based on the fundamental plane which provides order of magnitude estimates of bh masses ( e.g. , * ? ? ? the dispersion in the inner region of the galaxy outside of the star clusters is @xmath222 km @xmath34 . using this dispersion , the sphere of influence ( soi ) is given as @xmath223 pc or @xmath224 . this is comparable to the psf fwhm of @xmath225 and thus it may be possible to see the effect of a bh on the kinematics . we note however that there is considerable uncertainty in the bh mass estimate and therefore in the sphere of influence . there is no clear detection of a dispersion peak or enhanced rotation in the vicinity of the bh ; in fact , the dipsersion is lower at the location of the bh than in the surrounding areas ( fig . [ cofig ] ) . to better understand the expected signal from the bh , we use our jeans model for the center of the galaxy derived in 4.4 to simulate our nifs observations with the addition of a bh with mass between 0 and 10@xmath226 m@xmath10 . we find that for a bh of mass @xmath227 we would expect an increase in dispersion of @xmath228 km @xmath34 ; this is smaller the 1@xmath49 random errors on our measurement and thus a bh of this mass can not be ruled out . we can place a firm upper limit of @xmath15 on the bh which would increase the dispersion by @xmath187 km @xmath34 and thus be clearly detectable at the 3@xmath49 level in our data . another possibility is that the bh is significantly obscured and the stars we are seeing are in the foreground of the bh . if this were the case , the obscuring material would need to be on scales larger than the sphere of influence . we also searched for coronal line emission in the vicinity of the bh and did not find a convincing detection . this line has an ionization energy of 285 ev , thus requiring x - ray emission likely only generated by an agn ( e.g. , * ? ? ? the only other coronal line in our bandpass is the line which falls on top of the co bandhead region and thus is challenging to detect at weak flux levels . we place a 3@xmath49 upper limit on the luminosity of @xmath229 @xmath230 . fig . [ brgamma ] shows the br@xmath17 emission map within our nifs fov . this map shows a clear enhancement of br@xmath17 emission at the agn s position ( red cross is the bh s suggested location with an astrometric uncertainty of @xmath180@xmath461 ) . to examine this br@xmath17 peak in more detail , we create a nuclear spectrum around this source with a radius of @xmath231 and sky annulus is about 10% of the total br@xmath17 flux . this spectrum is shown in fig . [ agn ] ; the main panel shows the spectrum , while the small upper right panel shows a zoom - in of the dominant emission line , br@xmath17 . no broad br@xmath17 emission is seen . the total br@xmath17 luminosity is @xmath232 @xmath34 . we can compare this luminosity to the x - ray luminosity to evaluate whether the agn could be the source of the observed br@xmath17 line or not . the best fit power law model for the source gives an x - ray luminosity is @xmath233 @xmath34 @xcite . this best fit also gives @xmath234 which corresponds to @xmath235 . we can convert the x - ray luminosity to an expected br@xmath17 luminosity ; derives an empirical linear relationship of x - ray to @xmath236 luminosity for a sample of 47 agns and 30 qsos : @xmath237 ; it appears that this error bar is misprinted ; we recalculated the error on the relation by examining the scatter in the data and found a zeropoint error of 0.85 , therefore we assume a relation @xmath238 . we also note that this relation may not hold at low accretion rates or at low masses ( e.g. , for henize 2 - 10 ) . proceeding with this caveat , the predicted @xmath239 luminosity ranges from @xmath240 @xmath34 to @xmath241 @xmath34 . next , assuming case b recombination and a temperature of @xmath242 k , the ratio of br@xmath17/h@xmath243 is 0.01 @xcite . thus the predicted br@xmath17 luminosity is between @xmath244 @xmath34 and @xmath245 @xmath34 . the error in this determination is dominated by the error on the conversion between @xmath246 and @xmath239 ; the conversion from h@xmath247 to br@xmath17 is similar over the full range of reasonable temperatures ( i.e. , 5,00020,000 k ) . we now estimate whether this predicted br@xmath17 luminosity is consistent with the observed luminosity given the extinction values from the x - ray data and observed from line - emission ratios by . with the observed and the predicted br@xmath17 luminosity above , the ratio @xmath248 ; therefore , assuming the @xcite extinction law , the extinction suggested is @xmath249 or corresponding to a hydrogen column of @xmath250 @xcite . the extinction maps of show values between @xmath251 ; nearest the location of the agn , the average extinction is approximately @xmath252 . the hydrogen column also overlaps with the poorly constrained hydrogen column fit from the x - ray spectrum by @xcite of @xmath234 . thus it is plausible that the br@xmath17 emission we observe is coming from accretion onto the bh . if the br@xmath17 emission is indeed from the bh , it allows us to measure the velocity of the bh relative to the galaxy as a whole . fitting the spectrum shown in fig . [ agn ] ( with r = @xmath231 ) , the br@xmath17 radial velocity is ( @xmath253 km @xmath34 redshifted relative to systemic velocity of the galaxy and its velocity dispersion is ( @xmath254 ) km @xmath34 taking into account the local instrumental dispersion . we assume the systemic velocity of the galaxy is @xmath18872@xmath536 km @xmath34 . we have examined the overall morphology and nuclear regions of henize 2 - 10 using ground - based photometric data , adaptive optics gemini / nifs data and _ hst _ data . our primary findings are : 1 . henize 2 - 10 is well fit by a two component srsic profile . the inner srsic profile has @xmath63 and @xmath255 ( @xmath256 pc ) , and the outer srsic profile has @xmath65 and @xmath257 ( @xmath18 1kpc ) . the absolute magnitude of henize 2 - 10 within 4.3 kpc is @xmath258 , @xmath259 in @xmath0-band and @xmath1-band . 2 . the total stellar mass of henize 2 - 10 within 4.3 kpc is @xmath260 as derived from the @xmath1-band , a factor of @xmath261 times higher than previous estimates based on the @xmath14-band luminosity . 3 . apart from the inner starburst , henize 2 - 10 appears to be typical early - type galaxy . its outer color @xmath262 is consistent with an old population . furthermore , it is non - rotating and dispersion - dominated near the center and has an outer srsic index consistent with other early - type galaxies of similar luminosity . we estimate the dynamical mass of the brightest ssc in henize 2 - 10 to be @xmath263 m@xmath10 , higher than previous estimates . we use this mass estimate to derive new masses for 11 clusters at the center of henize 2 - 10 , assuming a constant @xmath109 for all the clusters . the timescale for dynamical friction is @xmath264 years for the sscs at the center of henize 2 - 10 . an nsc of mass @xmath265 m@xmath10 should be formed within a few hundred myr ; this cluster mass would be typical for galaxies of henize 2 - 10 s mass . thus , we are seeing nsc formation in progress in this galaxy , and this formation appears to be independent of the formation of the bh in this system . 6 . while there are few examples of @xmath266 star - forming galaxies with observational evidence for a massive bh ( e.g. , * ? * ; * ? ? ? * ) , the occupation fraction of massive bhs in early - type galaxies of this mass is likely @xmath1890% @xcite . therefore , it is reasonable that henize 2 - 10 hosts a massive bh . our observations allow us to place a firm upper limit on the bh mass of @xmath267 . we do not detect coronal emission from the bh , but br gamma emission consistent with the x - ray emission is detected at the location of the bh . we would like to thank maureen conroy at the sao / tdc for providing us the magellan / megacam data of @xmath0- , @xmath40- , and @xmath1-bands , micheva genoveva for generously sharing us with her sb profiles in 5 bessel - johnson filters @xmath72 , and the university of utah , physics and astronomy department for supporting this work . the authors also thank fabio antonini for helpful discussions . support for amy e. reines was provided by nasa through the einstein fellowship program , grant pf1 - 120086 . , e. , mellier , y. , radovich , m. , et al . 2002 , in astronomical society of the pacific conference series , vol . 281 , astronomical data analysis software and systems xi , ed . d. a. bohlender , d. durand , & t. h. handley , 228 , j. e. , hook , r. n. , & stoehr , f. 2011 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 8127 , society of photo - optical instrumentation engineers ( spie ) conference series

in this paper , we develop a method of performing the one - point statistics of a perturbed density field with a multiresolutional decomposition based on the discrete wavelet transform ( dwt ) . we establish the algorithm of the one - point variable and its moments in considering the effects of poisson sampling and selection function . we also establish the mapping between the dwt one - point statistics in redshift space and real space , i.e. the algorithm for recovering the dwt one - point statistics from the redshift distortion of bulk velocity , velocity dispersion , and selection function . numerical tests on n - body simulation samples show that this algorithm works well on scales from a few hundreds to a few h mpc for four popular cold dark matter models . taking the advantage that the dwt one - point variable is dependent on both the scale and the shape ( configuration ) of decomposition modes , one can design estimators of the redshift distortion parameter ( ) from combinations of dwt modes . comparing with conventional estimators , such as quadrupole - to - monopole ratio , the dwt estimators are scale - decomposed . it is useful to consider scale - dependent effects . when the non - linear redshift distortion is not negligible , the quadrupole - to - monopole ratio is a function of scale . this estimator would not work without adding information about the scale - dependence , such as the power - spectrum index or the real - space correlation function of the random field . the dwt estimators , however , do not need such extra information . moreover , the scale - decomposed estimators would also be able to reveal the scale - dependence of the bias parameter of galaxies . numerical tests show that the proposed dwt estimators are able to determine robustly with less than 15% uncertainty in the redshift range .

astro-ph0211159

the one - point probability distribution function ( pdf ) of a random mass density field @xmath3 , or the counts - in - cells ( cic ) statistics of a discrete random distribution such as galaxies , is probably the first statistics used to reveal the clustering feature of galaxy distribution . in the famous work by edwin hubble ( 1934 ) , he showed that the frequency distribution of galaxy count @xmath4 in angular cells is not gaussian , but lognormal . this result indicates that the pdf of galaxy distribution is fundamental in characterizing the cosmic mass and velocity fields . although current samples applied for large - scale structure study have much deeper redshift and much wider angular size than available in hubble s time , the one - point statistics is still frequently applied . this is because the one - point distribution and its moments contain complete information of the field , which might not be easily detected by other conventional methods . unlike the fourier amplitude , cic is not subject to the central limit theorem . it can detect the non - gaussianity of a field consisting of randomly distributed non - gaussian clumps , while the pdf of the fourier amplitudes is still gaussian due to the central limit theorem ( fan and bardeen 1995 ) . even the 2nd moment of one - point distribution is different from the 2nd moment of the fourier decomposition power spectrum . the former contains perturbations on scales larger than the size of the observed sample , while the latter does not . therefore , one - point statistics is applied on various samples of large - scale structures , including galaxy surveys ( e.g. hamilton 1985 ; alimi et al 1990 ; gaztaaga 1992 ; szapudi et al 1996 ; kim & strauss 1998 ) , transmitted flux of quasars ly@xmath5 absorption spectrum ( e.g. meiksin & bouchet 1995 ; gaztanaga & croft 1999 ; zhan & fang 2002 ) , and n - body simulation samples ( e.g. coles & jones 1991 ; taylor & watts 2000 ) . one - point statistics of a density fluctuation field @xmath6 is given by the distribution of the one - point variable @xmath7 , which is a sampling of the field by a window function @xmath8 around position @xmath9 and on scale @xmath10 , i.e. @xmath11 the variable @xmath7 is actually a mean value of the field at position @xmath9 and on scale @xmath10 . the distribution of @xmath7 gives a pdf description of the field . eq.(1 ) is a space(@xmath9)-scale(@xmath10 ) decomposition of the field @xmath6 with bases @xmath12 . most popular windows @xmath8 are gaussian and top - hat filters . the gaussian windows generally are not orthogonal , i.e. the function @xmath8 does not satisfy @xmath13 , if @xmath14 . thus , the variables @xmath7 are either incomplete or redundant . in turn , these may lead to 1 ) loss of information of the field @xmath6 ; and 2 ) false correlation . it is possible to construct an orthogonal and complete set of bases from top - hat windows . however , they are not localized in the fourier space . the index @xmath10 does not refer to a well defined scale @xmath15 . as a consequence , the one - point statistics with conventional windows are not suitable for problems with a scale - dependence . for instance , the redshift distortion on the one - point statistics can be properly estimated only if we know how the redshift distortion depends on the scale and shape of the window function . therefore , a better algorithm for the one - point statistics is needed , for example , to recover the real - space rms density fluctuation @xmath16 from redshift distortion . in this paper , we show that these problems can be solved with a multiresolutional decomposition via the discrete wavelet transform ( dwt ) . the bases of the dwt decomposition are complete , orthogonal , and localized in both physical and fourier spaces . it was shown in the last few years that the dwt can be employed as an alternative representation in most conventional statistics of cosmic mass and velocity fields , including power spectrum ( pando & fang 1998 ; fang & feng 2000 ; yang et al 2001a , 2002 ) , high order correlations ( pando & fang 1998 ; pando , feng & fang 2001 ; feng , pando & fang 2001 ) , bulk and pair - wise velocity ( zhan & fang 2002 ; yang et al . 2001b ) , and identification of halos and clusters ( xu , fang & wu 2000 ) . the dwt representation is also able to reveal statistical features , which might not be easily detected without a set of decomposition bases localized in both physical and scale spaces . with the dwt analysis , the intermittency of the fluctuations of quasars ly@xmath5 transmitted flux has been detected ( jamkhedkar , zhan & fang 2000 ; pando , feng & fang 2000 ; zhan , jamkhedkar & fang 2001 ; pando et al . this property can not be simply detected by the popular @xmath8 or fourier decompositions . we also show that the dwt decomposition is very useful for one - point statistics . we establish the algorithm of the one - point variable and its moments in considering various corrections , such as poisson sampling and selection function . to demonstrate the advantage of the dwt one - point statistics , we show that one can map between the one - point statistics in real and redshift spaces scale - by - scale , taking account the distortion due to bulk velocity , velocity dispersion and selection function . with these results one can construct @xmath1 ( redshift distortion parameter ) estimators with moments of the dwt one - point variables . unlike conventional estimators , the dwt estimators do not need extra - information or _ ad hoc _ assumption when the non - linear redshift distortion is not negligible . this paper is organized as follows . 2 presents the algorithm for the pdf of one - point variables with the dwt space - scale decomposition . 3 discusses the second moment of the one - point variables in considering the effects of poisson sampling and selection function . in 4 we develop the theory of redshift distortion of the dwt one - point variables . 5 tests the theoretical results in 4 on n - body simulation samples . the emphases are the redshift - to - real - space mapping of the diagonal and off - diagonal second moments . 6 shows the application in estimating the redshift distortion parameter . finally , the conclusions and discussions are given in 7 . the appendix provides relevant formulae for quantities defined in 4 . we release the codes for calculating these quantities via anonymous ftp at samuri.la.asu.edu/pub/zhan/dwtcic.tgz . as emphasized in 1 , a key problem for one - point statistics is to have proper window functions @xmath17 , and define the variable with eq.(1 ) . in the dwt analysis , this is done by the so - called scaling function . let us first briefly introduce the dwt - decomposition of a random field . for the details of the mathematical properties of the dwt see mallat ( 1989a , b ) ; meyer ( 1992 ) ; daubechies ( 1992 ) , and for physical applications see fang & thews ( 1998 ) . to simplify the notation , we first consider a one - dimensional ( 1-d ) density fluctuation @xmath18 on a spatial range from @xmath19 to @xmath20 . we first divide the space @xmath20 into @xmath21 segments labeled by @xmath22 . each segment is of size @xmath23 . the index @xmath24 can be a positive integer . it stands for a length scale @xmath23 . the higher the @xmath24 , the smaller the length scale . even though we often refer some properties to a ( range of ) @xmath24 in the analyses below , it must be read as an association of the properties with the length scale @xmath23 . the index @xmath25 is for position , and it corresponds to the spatial range @xmath26 . that is , the space @xmath20 is decomposed into cells @xmath27 . in the dwt analysis , each cell @xmath27 supports two compact functions : the scaling function @xmath28 and the wavelet @xmath29 . one example of such functions is the daubechies 4 ( d4 ) wavelets , which have to be constructed recursively ( daubechies 1992 ; see also fang & thews 1998 ) . the properties of d4 wavelets are listed below and in the appendix a.1 . we show , in fig . 1 , examples of the d4 scaling function and wavelet , and their fourier transform @xmath30 and @xmath31 . one can see from fig . 1 that both the scaling function and the wavelet are localized in fourier space as well as in physical space . generally , @xmath28 and @xmath29 are localized in cell @xmath27 , @xmath30 is localized in the scale range @xmath32 , and @xmath31 in the range @xmath33 , where @xmath34 . obviously , the scaling function @xmath28 is a window function on scale @xmath24 and around the segment @xmath25 . it can be used to measure the mean field in cell @xmath27 @xmath35 where @xmath36 is called scaling function coefficient ( sfc ) , given by @xmath37 in analogous to eq.(1 ) , eqs.(2 ) and ( 3 ) show that the sfc @xmath36 or @xmath38 can be employed as the variable for one - point statistics . 1 also demonstrates that wavelets are admissible , i.e. @xmath39 , and therefore , they are used to measure the fluctuations of a field with respect to its mean in cells @xmath27 @xmath40 where @xmath41 is called the wavelet function coefficient ( wfc ) . the scaling function @xmath28 and the wavelet @xmath29 satisfy a set of orthonormal relations as @xmath42 where @xmath43 is the kronecker delta function . to be consistent with eq.(5 ) , the scaling function @xmath28 is normalized to @xmath44 with these properties , a 1-d density field @xmath18 can be decomposed into @xmath45 the second term on the r.h.s of eq.(9 ) contains only the information of the field @xmath18 on scales equal to and less than @xmath23 . therefore , the @xmath21 variables ( sfcs ) @xmath36 , with @xmath46 , completely describe the behavior of the field on scales larger than @xmath23 . for instance , if the resolution of the sample is @xmath47 , @xmath18 can be described by the @xmath48 sfcs @xmath49 . because of the orthogonal relation eqs.(5 ) and ( 7 ) , the @xmath21 variables ( sfcs ) @xmath36 are independent and irredundant . in a functional space consisting of functions , which are smoothed on scales equal to and less than @xmath23 , the completeness of the scaling function @xmath28 can be expressed as @xmath50 where @xmath51 is the dirac delta function . all the above - mentioned results can be easily generalized to three - dimensional ( 3-d ) fields . let us consider a 3-d distribution @xmath6 in a volume @xmath52 to @xmath53 . similar to the 1-d case , we divide the volume @xmath54 into cells @xmath55 , where @xmath56 refers to the length scale @xmath57 , and @xmath58 refers to the spatial range of the cell , @xmath59 and @xmath60 , where @xmath61 . the one - point statistical variable of the field @xmath6 is @xmath62 where the 3-d scaling function ( 3-d window ) @xmath63 is given by a direct product of 1-d scaling functions as @xmath64 one can also construct 3-d wavelets @xmath65 as a direct product of 1-d scaling functions and wavelets ( see yang et al 2002 ) , and generalize eq.(9 ) into 3-d as @xmath66 since the second term on the r.h.s . is not needed in calculating the one - point statistics , therefore , we do not show the details of @xmath67 , but only give the orthogonal relations @xmath68 if we consider only functions smoothed on scales equal to and less than the scale of @xmath69 in each dimension , the set of bases @xmath63 is orthogonal and complete . in one - point statistics , the statistical feature of the field @xmath6 is characterized by the moments of the distribution of @xmath70 . the @xmath71th moment is @xmath72 , where @xmath73 stands for an ensemble average . using eq.(11 ) , we have @xmath74 for a homogeneous field , these moments are @xmath75-independent . moreover , when the fair sample hypothesis " ( peebles 1980 ) holds , or equivalently , when the random field is ergodic , the @xmath76 sfcs @xmath70 , @xmath60 , @xmath61 for a given scale @xmath69 can be considered as @xmath76 independent measurements , because they are measured by projecting onto the mutually orthogonal basis @xmath63 . accordingly , the @xmath76 sfcs form a statistical ensemble on the scale @xmath69 . this ensemble represents actually the one - point distribution of @xmath70 over the dwt modes at a given scale @xmath69 . thus , the average over @xmath75 is a fair estimation of the ensemble average , i.e. @xmath77 we now consider the 2nd moment @xmath78 . from eq.(15 ) , the 2nd moment is determined by the usual two point correlation function @xmath79 . we define a dimensionless 2nd moment on scale @xmath69 by @xmath80 using eqs.(12 ) and ( 15 ) , one can rewrite eq.(17 ) as @xmath81 where @xmath82 is the fourier transform of the basic scaling function ( see eq.(a1 ) in appendix a ) . the term @xmath83 in eq . ( 18 ) is the fourier power spectrum defined by @xmath84 where @xmath85 with @xmath86 , and @xmath87 one can compare the 2nd moment @xmath88 with the dwt power spectrum given by fang & feng ( 2000 ) as @xmath89 where @xmath90 is the fourier transform of the basic wavelet . therefore , both the dwt power spectrum and the dwt one - point statistics can be calculated with a single dwt decomposition . the former relies on the wavelet , while the latter on the scaling function . the fourier transform of the scaling function @xmath82 is non - zero in @xmath91-space where @xmath92 , and @xmath90 is mainly in @xmath93 ( fig . 1 and eq.(a4 ) ) . therefore , the dwt power spectrum @xmath94 is actually a banded fourier power spectrum in the wavenumber range @xmath95 , while the 2nd moment @xmath96 contains all powers with @xmath97 . it is well known that the 2nd moment of one - point statistics , such as @xmath88 or @xmath16 , is sensitive to long wavelength behavior of the perturbations . this can also be seen from the relation between the scaling functions and wavelets as follows @xmath98 that is , the factor @xmath99 in eq.(22 ) extracts all powers on scales larger than @xmath23 . eq.(22 ) holds only for compactly supported discrete wavelets such as daubechies wavelets . we now consider distributions of discrete objects , such as simulation particles and observed or mock galaxies . if the position measurement of particles or galaxies is perfectly precise , the number density distribution of these samples can be written as @xmath100,\ ] ] where @xmath101 is the total number of the particles or galaxies , @xmath102 is the position of the @xmath71th galaxy , @xmath103 is its weight , and @xmath104 is the selection function , which is given by the mean number density of galaxies when galaxy clustering is absent , and @xmath6 is the density fluctuation in the distribution . one can subject @xmath105 to a dwt decomposition . similar to eq.(13 ) , we have @xmath106 where @xmath107 the dwt one - point variables of @xmath6 is now given by @xmath108 \phi_{\bf j , l}({\bf x } ) d{\bf x}.\ ] ] since the selection function varies slowly , we have @xmath109 where @xmath110 is the mean of the selection faction in the cell @xmath55 . the algorithm for @xmath110 is given in next subsection . substituting eq.(27 ) into eq.(26 ) we have @xmath111 the second term on the r.h.s . is due the normalization of the scaling function eq.(8 ) . by definition of equation ( 23 ) , selection function @xmath104 is the galaxy distribution if galaxy clustering @xmath6 is absent . in the plane - parallel approximation , selection function depends only on @xmath112 , i.e. the coordinate in the redshift direction or the line - of - sight ( los ) . thus , from equation ( 23 ) the selection function @xmath113 can be approximated by an average of @xmath105 over the plane @xmath114 , which depends upon the geometry of a real survey . in a simple case , for example , a mock survey in a simulation box , it reads @xmath115 with eq.(24 ) , eq.(29 ) yields @xmath116 by definition , @xmath110 is the mean of @xmath113 in the cell @xmath55 . using eq . ( 30 ) , we have @xmath117 that is , the selection function can be approximately expressed by the sfc of the ( observable ) galaxy distribution @xmath105 . the observed galaxy distributions @xmath105 are considered to be a poisson sampling with an intensity @xmath118 $ ] . in this case , the characteristic function of the galaxy distribution @xmath105 is @xmath119 \rangle_p= \exp\left \ { \int d{\bf x } n({\bf x})[e^{iu({\bf x})}-1 ] \right \},\ ] ] where @xmath120 is the average for the poisson sampling . the @xmath71-point correlation functions of @xmath105 are given by @xmath121_{u=0}.\ ] ] we have then @xmath122 and @xmath123 substituting @xmath118 $ ] into eq.(35 ) , we have @xmath124 subjecting eq.(36 ) to a scaling function projection , and using eq.(27 ) , we have approximately @xmath125 the last term in eq.(37 ) is due to the normalization eq.(8 ) . from eq.(37 ) , the 2nd moment @xmath88 is given by @xmath126 -1.\ ] ] the second term on the r.h.s . under the summation is the correction due to poisson noise , and the rest is the normalized 2nd moment of @xmath6 . one can also calculate the poisson correction for higher order moments with eq.(33 ) . actually , astronomical data often gives only the distribution in redshift space , i.e. @xmath127= \bar{n}^s({\bf s})[1+\delta^s({\bf s})],\ ] ] where @xmath128 is the radial ( @xmath129 ) component of the velocity of the @xmath130th galaxy , @xmath131 is the selection function in redshift space , and @xmath132 is the hubble constant at the corresponding redshift . thus , we can only calculate the cic moments in redshift space . for a given mass field @xmath6 , the galaxy velocity @xmath133 is a random field with mean @xmath134 where @xmath135 is the average over the ensemble of velocities . the mean velocity @xmath136 is also called bulk velocity at @xmath137 , which is assumed irrotational ( bertschinger & dekel 1989 ; dekel , bertschinger & faber 1990 ; dekel et al . 1999 ) . in linear regime , the bulk velocity is related to the density contrast by @xmath138 where the parameter @xmath139 at present , i.e. redshift @xmath140 , and @xmath141 is the linear bias parameter . the _ rms _ deviation of velocity @xmath133 from the bulk velocity @xmath136 is @xmath142 ^ 2\rangle_v = [ \sigma^{v_i}({\bf x})]^2 , \mbox { $ i=1,2,3$}.\ ] ] in order to express the scale dependence of the bulk velocity @xmath143 and the variance @xmath144 , we can also decompose the velocity field @xmath133 with a dwt , i.e. @xmath145 obviously , the variance @xmath144 on scale @xmath69 is given by @xmath146^{1/2}.\ ] ] although @xmath147 is @xmath75-dependent , @xmath148 should be @xmath75-independent if the random field @xmath133 is ergodic . on large scales , say @xmath149h@xmath0mpc , the velocity field @xmath133 is roughly _ gaussian _ , and it can be described by its mean and variance , which are generally scale - dependent . because of eq.(39 ) , the directly measurable dwt variable is not given by eqs . ( 26 ) or ( 28 ) , but @xmath150 \phi_{\bf j , l}({\bf s})d{\bf s } \\ \nonumber & = & \frac{1}{\bar{n}^s_{\bf j , l } } \sum_{m=1}^{n_g}w_m \phi_{\bf j , l}({\bf x}_m + \hat{\bf r } v_r({\bf x}_m)/h ) - \sqrt{\frac{l_1l_2l_3}{2^{j_1+j_2+j_3}}}.\end{aligned}\ ] ] where @xmath151 is calculated in the same way as eq.(31 ) , but with @xmath152 to replace @xmath105 . in this section , we do not consider the effect of selection function , i.e. @xmath153 . in the plane - parallel approximation , i.e. @xmath154 is along the @xmath155-direction , eq.(45 ) becomes @xmath156 if the velocity field is gaussian , subjecting eq.(46 ) to an average over the ensemble of velocities , we have @xmath157 e^{v_3({\bf x})\frac{1}{h } \frac{\partial}{\partial x_3 } + \frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2 } \phi_{\bf j , l}({\bf x } ) d{\bf x } - \sqrt{\frac{l_1l_2l_3}{2^{j_1+j_2+j_3}}}.\end{aligned}\ ] ] for clarity , the angle brackets @xmath158 are dropped hereafter without causing any confusion . if we consider only the linear effect of the bulk velocity , equation ( 47 ) is approximately @xmath159 \left [ 1+v_3({\bf x})\frac{1}{h } \frac{\partial}{\partial x_3 } \right ] e^{\frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2 } \phi_{\bf j , l}({\bf x } ) d{\bf x } - \sqrt{\frac{l_1l_2l_3}{2^{j_1+j_2+j_3}}}.\ ] ] neglecting the terms of the order of @xmath160 , and using the linear relation between @xmath6 and @xmath136 , we have @xmath161 \frac{\partial } { \partial x_3 } \right\ } e^{\frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2 } \phi_{\bf j , l}({\bf x } ) d{\bf x } - \sqrt{\frac{l_1l_2l_3}{2^{j_1+j_2+j_3 } } } \\ \nonumber & = & \int e^{\frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2 } \phi_{\bf j , l}({\bf x } ) d{\bf x } + \sum_{{\bf l } ' } \epsilon_{{\bf j , l } ' } \int \phi_{{\bf j , l}'}({\bf x } ) e^{\frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2 } \phi_{\bf j , l}({\bf x } ) d{\bf x}\\ \nonumber & & + \beta \sum_{\bf l ' } \epsilon_{\bf j , l ' } \int \phi_{\bf j , l'}({\bf x } ) e^{\frac{1}{2}(\sigma_{\bf j}^{v_3}/h)^2 \left ( \frac{\partial } { \partial x_3 } \right)^2}\nabla^{-2 } \frac{\partial^2}{\partial x_3 ^ 2 } \phi_{\bf j , l}({\bf x } ) d{\bf x } - \sqrt{\frac{l_1l_2l_3}{2^{j_1+j_2+j_3}}}.\end{aligned}\ ] ] in the last step , integration by parts for @xmath162 and the completeness relation equation ( 10 ) are used . the summation of @xmath75 in eq.(49 ) is over @xmath163 . to simplify eq.(49 ) , we define the following matrices @xmath164 where the differential operators @xmath165 and @xmath166 are @xmath167 obviously @xmath168 depends on @xmath169 , not @xmath75 or @xmath170 individually . appendix a provides the algorithms to calculate @xmath168 . thus , eq.(49 ) becomes @xmath171 using the so - called `` partition of unity '' ( fang & feng 2000 ) , one can show that @xmath172 therefore , eq.(52 ) gives @xmath173 this is the mapping of one - point variables between real and redshift spaces . the second moment of one - point statistics in redshift space is @xmath174 [ \sum_{\bf l''}(\gamma^{1,0}_{\bf j , l , l '' } + \beta \gamma^{1,1}_{\bf j , l , l '' } ) \epsilon_{\bf j , l '' } ] \big\rangle.\ ] ] one can show that , even for a weakly non - linear field , both @xmath175 and @xmath176 are symmetric and quasi - diagonalized with respect to @xmath75 and @xmath170 , and @xmath177 should be @xmath75-independent . the wavelet counterpart , i.e. @xmath178 , has been shown by pando , feng & fang ( 2001 ) . thus , eq.(55 ) becomes @xmath179 using the completeness relation equation ( 10 ) , we have @xmath180 thus , eq.(56 ) gives @xmath181 where we have used @xmath182 . therefore , we have finally the mapping of @xmath88 between real and redshift spaces @xmath183 in linear approximation , it is reasonable to estimate the effect of redshift distortion of @xmath152 and @xmath131 separately . in this case , one can still use eq.(54 ) as the mapping from @xmath184 to @xmath185 . we only need to study the effect of the mapping between @xmath131 and @xmath104 , which is given by @xmath186 \rangle_v \simeq \bar{n}({\bf x } ) + \frac{1}{h}v_r({\bf x } ) \hat{\bf r}\cdot \nabla n^g({\bf x}),\ ] ] where we have used @xmath187 . with the plane - parallel approximation of selection function ( 3.2 ) , eq.(60 ) becomes @xmath188 where we have dropped @xmath189 for @xmath131 . > from equation ( 41 ) , @xmath190 can be represented by @xmath6 , so we have @xmath191 \\ \nonumber & \simeq & \bar{n}(x_3 ) \left \ { 1 - \beta\frac{d\ln \bar{n}(x_3)}{dx_3 } \frac{\partial}{\partial x_3 } \nabla^{-2}\left [ \frac{n^g ( { \bf x})}{\bar{n}(x_3 ) } - 1 \right ] \right \}.\end{aligned}\ ] ] combining eqs.(54 ) and ( 62 ) , we have finally @xmath192 \frac { \epsilon_{\bf j , l'}}{\bar{n}_{\bf j , l ' } } + \left . \beta \frac{d\ln \bar{n}(x_3)}{dx_3}\right|_{\bf j , l } \sum_{l_3 } q_{\bf j , l , l ' } \frac{\epsilon_{\bf j , l'}}{\bar{n } _ { \bf j , l ' } } , \ ] ] where @xmath193 stands for the mean value of @xmath194 in the cell @xmath55 , and @xmath110 is given by eq.(31 ) . in the last term of eq.(63 ) , @xmath195 and @xmath196 , and the summation runs only over @xmath197 . the coefficient @xmath198 is defined by @xmath199 the calculation of @xmath198 is given in appendix a. because all @xmath75-diagonal elements of @xmath200 are zero ( appendix a ) , and @xmath201 are quasi-@xmath75-diagonal , the first and the second terms on the r.h.s . of equation ( 63 ) are not correlated . we have then @xmath202 ^ 2 \sum_{\bf l ' } q^2_{\bf l , l ' } \left \langle \left ( \frac{\epsilon_{\bf j , l'}}{\bar{n}_{\bf j , l ' } } \right ) ^2 \right \rangle.\end{aligned}\ ] ] for a uniform field , @xmath203 and @xmath204 are @xmath75-independent . thus , equation ( 65 ) gives @xmath205 ^ 2 \sum_{\bf l ' } q^2_{\bf 0 , l ' } \right ) d_{\bf j}.\ ] ] using the inequality equation ( a16 ) , we can show that if @xmath206 we have @xmath207 ^ 2 \sum_{l_3-l'_3 } q^2_{{\bf j } , l_3-l'_3 } < \beta^2 s^2_{\bf j}.\ ] ] that is , the selection function term in equation ( 66 ) is even less than the second order terms @xmath208 if the selection function is slowly varying with @xmath112 . we use n - body simulation samples to test the dwt algorithms for recovering the one - point statistics from redshift space to real space . the model parameters are listed in table 1 . notice that the simulation boxes are relatively large to minimize the effect on the bulk velocities due to the missing power at long wavelengths ( tormen & bertschinger 1996 ) . we use a modified p@xmath209 m code ( jing & fang 1994 ) to evolve @xmath210 ( @xmath211 for lcdm2 ) cold dark matter ( cdm ) particles in a periodic cube of length @xmath20 on each side . the linear power spectrum is given by the fitting formula in bardeen et al . zeldovich approximation is applied to set up the initial perturbation . the particles evolve 600 ( 800 for lcdm2 ) integration steps from @xmath212 down to @xmath213 for all the models . table 1 [ cols="^,^,^,^,^,^,^,^",options="header " , ] in addition to @xmath1 at @xmath214 , we can also test the redshift distortion algorithms at high redshifts . in this case , @xmath1 is a function of @xmath215 , @xmath216 and @xmath217 as ( lahav et al . 1991 ) @xmath218^{0.6}.\ ] ] the simulation code is modified to generate light - cone outputs from @xmath219 to @xmath220 by a similar method in the _ hubble volume simulations _ ( evrard et al . 2001 ) . instead of producing spherical light - cones , we apply the plane - parallel approximation in which a plane ( light - front ) sweeps through the simulation box at the speed of light . let the los be along the @xmath112-axis , and the position of the light plane at time @xmath221 be @xmath222 . the position and velocity of a particle is recorded when it crosses the plane , i.e. when the position @xmath223 of the particles satisfies @xmath224 since the time step @xmath225 in the simulations is finite , it is computationally impractical to use eq.(70 ) directly . the position of the plane in time interval from step @xmath130 to step @xmath226 is approximately @xmath227,\ ] ] where @xmath5 is from 0 to 1 . on the other hand , for a particle we have @xmath228 , where @xmath229 is the velocity of the particle along the los . thus , if we can find a solution of @xmath230 from the following equation @xmath231,\ ] ] we have @xmath232 , i.e. the particle crosses the plane at time @xmath233 . we then record its position @xmath234 and velocity @xmath235 . the accuracy is satisfactory . for example , the difference between @xmath236 and @xmath237 is typically less than 50 @xmath238kpc out to redshift @xmath239 in the lcdm1 model . in the actual simulations , the integration variable is @xmath240 , the cosmic expansion factor , instead of @xmath221 . since the size of the simulation box is less than the distance swept by the plane from redshift @xmath219 to 0 , the motion of the plane is realized by periodic extensions of the simulation box when the plane meets the boundary . this treatment should not have significant effects on our analyses , because the simulations already impose a periodic boundary condition , and the largest scale analyzed for each simulation is only a quarter of its box size . once the real - space light - cone output is obtained , the observed redshift of a particle is given by @xmath241 , where @xmath215 is the actual redshift , and @xmath242 is the peculiar velocity . to avoid negative @xmath243 caused by peculiar velocities at low redshifts , a lower cut is set to @xmath244 . this results in less than a hundred particles bearing a negative @xmath243 in each run , which are then removed from the samples . three light - cone outputs along the three orthogonal axes are produced for each simulation , which effectively increases the number of realizations by a factor of 3 . before testing the theory of recovering @xmath88 ( eq.(59 ) ) , we need to study the factors @xmath245 , @xmath246 , and @xmath247 , because the first two are the key of the redshift distortion of bulk velocity , while the last one the distortion of velocity dispersion . below we drop @xmath248 in subscripts for simplicity . by the definition of eq.(50 ) , the factors @xmath249 and @xmath250 are given by @xmath251 we plot @xmath249 and @xmath250 in fig . 2 for the d4 wavelet . it should be pointed out that @xmath252 depends only on the _ shape _ , not the scale @xmath69 of the scaling function . for instance , @xmath253 , because the 3-d cells of @xmath254 have the same shape , i.e. all the ratios of length : width : height of these cells are 2:2:1 , even though they have different volumes ( scales ) . for cubic cells , i.e. @xmath255 , we have @xmath256 , and @xmath257 . the factor @xmath258 is given by @xmath259 it depends on both the scaling function and the velocity dispersion @xmath260 , and therefore it is model - dependent . the velocity variance @xmath260 is scale - dependent . it is shown in fig . 3 for all the models , which is similar to fig . 6 in yang actually , @xmath261 is given by all the power of pairwise velocity on scales less than the scale of @xmath69 , and therefore , it is larger on larger scales . to avoid crowding the figure , we only show @xmath262 for lcdm1 in fig . 4 . other modes , such as @xmath263 , follow @xmath262 closely . an interesting feature seen in fig . 4 is that @xmath264 and @xmath265 are almost equal to 1 from @xmath266 to 6 , which corresponds to spatial scales from 200 to 12.5 h@xmath0 mpc on the celestial sphere , and 200 to 100 h@xmath0 mpc along the los . even for @xmath267 , the dependence on @xmath24 is only mild . moreover , this property is model - independent , i.e. for other models , we also have the @xmath24-independence of @xmath264 and @xmath265 . the factor @xmath263 is sensitive to @xmath261 ( as a variable ) when the scale along the los is small , but insensitive when the scale is large . for example , when @xmath261 varies from 0 to 500 km s@xmath0 , @xmath265 drops from 1 to 0.97 , i.e. the change is only 3% , while @xmath268 from 1 to 0.46 , i.e. the change is by a factor of 2 . if we try to use a scale - independent @xmath269 to recover the real space power spectrum , we should chose @xmath269 to give a good fitting on small scales , because the value of @xmath269 does not significantly affect the large scales . in addition to @xmath270 , @xmath258 depends on @xmath271 and @xmath272 ( collectively @xmath273 ) through @xmath260 , which is less than 500 km s@xmath0 ( fig . 3 ) . therefore , when the los scale is above 100 h@xmath0 mpc , @xmath258 is almost a constant with respect to @xmath273 . on the other hand , when the scale is below 50 h@xmath0 mpc , @xmath258 shows a mild dependence on @xmath273 due to the scale dependence of @xmath260 . in a word , we benefit from the dwt representation to see the different behavior of the two types of the redshift distortion . the former is sensitive to the shape of the mode , not the scale , while the latter just the contrary . we neglect the effect of selection function here for the simulation samples , which is similar to that on the dwt power spectrum tested in yang et al . ( 2002 ) . since the matrix @xmath168 is quasi - diagonalized with respect to @xmath274 , we have approximately @xmath275 . thus , eq.(59 ) yields @xmath276 in eq.(75 ) , the contributions of the operators @xmath165 and @xmath166 to the redshift distortion are separated . usually terms of @xmath166 ( @xmath249 and @xmath250 ) are called linear redshift distortion , and that of @xmath165 ( @xmath258 ) non - linear distortion . these terminologies may cause a confusion . here `` linear '' means only that @xmath166 is from the linear term of @xmath190 in the approximation eq.(48 ) , and `` non - linear '' means that @xmath165 is non - linear of the velocity dispersion @xmath277 . it does not imply that the terms of @xmath166 is enough for estimating the redshift distortion of a field that is in the linear regime . since the velocity dispersion @xmath269 is non - zero in the linear regime , the non - linear distortion term of @xmath165 may also play a non - negligible role even when the field is linear . in this paper , we follow the tradition to call the terms of @xmath166 and @xmath165 the linear and non - linear effects of redshift distortion , respectively . however , it should be kept in mind that these names do not reflect whether the field is linear or non - linear . for @xmath69-diagonal modes , equation ( 75 ) reduces to @xmath278 . the factors within the parentheses are known as the linear redshift - to - real space mapping for two - point correlation function ( kaiser 1987 ) . figure 5 shows the recovery of @xmath69-diagonal @xmath88 . it is evident that the recovery equation ( 75 ) works very well on all scales and redshifts considered , except for the scdm model at @xmath279 . due to greater @xmath1 parameters , the distortions of the tcdm and the scdm models are generally stronger than that of the rest . for these two models , @xmath280 is almost parallel to @xmath88 , which indicates a very weak non - linear redshift distortion on all scales . this is consistent with the fact that the velocity dispersions of the two models are small on all scales compared to the others . since the volume of a cubic cell with @xmath281h@xmath0mpc on each side is approximately the same as that of a sphere with a radius of @xmath282h@xmath0mpc , the values of @xmath283 are consistent with @xmath284 at corresponding redshifts for each model . the differences , nevertheless , are due to the difference in the window functions . similar to @xmath88 , one can generalize @xmath16 to @xmath285 , which is the _ rms _ density fluctuation in a sphere of radius @xmath10 . since the behavior of @xmath286 is the same as @xmath287 , one can recover the real space @xmath285 from the redshift space @xmath288 in a similar way as eq.(75 ) , i.e. @xmath289^{1/2 } \sigma_r$ ] , where @xmath24 is chosen to match the volumes of the dwt cell and the spherical top - hat window of radius @xmath10 . figure 6 demonstrates the recovery of @xmath285 for all the models . it is a coincidence that @xmath290 for the low density models . figures 7 and 8 give the recovery of off - diagonal @xmath88 . this recovery works well on scales above 50 h@xmath0 mpc . on scales less than 50 h@xmath0 mpc , the error is noticeable but still small . this is partially due to the approximation of the quasi - diagonality of the covariance @xmath291 in equation ( 56 ) . moreover , the separation between the factors involving operators @xmath165 and @xmath166 in eq.(75 ) is not perfect as well . the errors of recovering @xmath88 can be more clearly seen via a comparison with the recovery of the dwt power spectrum from redshift space @xmath292 to real space @xmath293 . it is ( yang et al . 2002 ) @xmath294 where @xmath295 is defined by eq.(50 ) but with replacement of @xmath296 to @xmath297 . the factors @xmath298 and @xmath299 also satisfy @xmath300 and @xmath301 . the non - linear term @xmath302 behaves the same as @xmath258 , but with a little stronger dependence on @xmath69 . we find that the recovery of @xmath88 by eq.(75 ) has an error within 10% in most cases , while the dwt power spectrum recovery eq.(76 ) is accurate to 5% or better . this is because that the quasi - diagonality of the sfc s covariance @xmath303 is poorer than the quasi - diagonality of the wfc s covariance @xmath304 . as we know from the dwt analysis , the covariance @xmath305 is always quasi - diagonal for a gaussian field , regardless the power spectrum . on the other hand , the quasi - diagonality of the covariance @xmath306 is not a generic mathematical property , but only an approximation . all the above algorithms can be applied to real samples as well . however , there are several problems to be considered in the analysis on real data . data assignment on grid _ to calculate the moments of counts - in - cells , one has to cast a grid on the data set either explicitly or implicitly , and assign the data in cells defined by the grid . the arbitrariness of the assignment may lead to significant error ( colombi , bouchet & schaeffer 1995 ) . one way to reduce this error is to shift the grid . it is well known that this assignment may also result in spurious features of the power spectrum on scales around the nyquist frequency of the grid . in the dwt analysis , one can use different grids , but no shift is needed . for a given grid , the assignment is realized by the same scaling function used for the data decomposition . since scaling functions are orthogonal and complete in each given scale @xmath69 , the spurious features and false correlations can be completely avoided ( fang & feng 2000 ) . to test the effect of the assignment on the one point statistics , we calculate @xmath287 for a snapshot sample of lcdm1 with randomly shifted grids . the snapshot sample is take at @xmath307 . in fig . 9 , snap-0 represents @xmath287 without grid shifting , while snap-32 and snap-1024 , respectively , the average over 32 , and 1024 random shifts for grids on all scales . no significant deviation is detected among the results . _ 2 . boundary condition and edge effect _ in a real survey , the edge effect is unavoidable due to the geometry ( e.g. szapudi & colombi 1996 ) . since the dwt bases are localized in physical space , the effect of edges can be effectively suppressed by dropping modes that are close to the boundary of the samples . in other words , when calculating @xmath88 , the averaging in eq.(17 ) runs only over modes that are not significantly affected by the edge effect , and the normalization factor @xmath308 is replaced by the volume over which the average takes place . this method has been tested numerically in the power spectrum detection ( pando & fang 1998b ) . it shows that the power spectrum can be fairly reconstructed regardless whether applying a periodic boundary or zero padding outside of the samples . the treatment of dropping edge modes has also been successfully employed to obtain the power spectrum for the las campanas redshift survey of galaxies , which has a slice - like geometry ( yang et al . when the interested scale is close to the dimension of the sample , one can not afford to drop all the edge modes , and then a more thorough treatment is needed . the edge effect exists in our simulation sample because the light - cone output is not periodic along the redshift axis , while we have used periodic d4 wavelets in the analysis . for comparison , the real - space light - cone @xmath287 from fig . 5 is included in fig . 9 . there is no significant difference between the results from the snapshot , which is truly periodic , and that from the light - cone around the same redshift . the slightly larger standard deviation of the light - cone result at 200 @xmath238mpc ( a quarter of the simulation box ) is due to a larger fraction of edge cells . non - poisson sampling _ in 3.3 , we have considered the correction for poisson sampling . it is sufficient for simulation samples . however , it may not be typical for real samples . for instance , some galaxy catalogs may be given by sub - poisson sampling on small scales , or small halos ( e.g. bullock , wechsler & somerville , 2002 ) . the sub - poisson distribution is simply due to very low mass of the considered halo , so that it can not host any additional objects . in other words , the sub - poisson distribution is significant on small scales on which galaxies are anticorrelated . therefore , this sub - poisson sampling is similar to the sub - poisson distribution of the fermi - dirac statistics ( one state can host no more than one particle ) . the effect of this sub - poisson has been extensively studied in quantum optics ( e.g. martin & landauer , 1992 ) . it is possible to extend the poisson sampling eq.(35 ) to include fermi - like sampling . the algorithm for a modified poisson sampling has been developed by jamkhedkar , bi and fang ( 2001 ) ( see their append b ) . we first consider a @xmath1 estimator without non - linear effect . for instance , @xmath309 ( fig . 4 ) , i.e. the non - linear effects are small above 100 h@xmath0 mpc along the los , eq.(75 ) gives @xmath310 it is easy to construct a @xmath1 estimator with eq.(77 ) , because the quantities @xmath88 , @xmath280 , and @xmath311 are not rotationally invariant , but they satisfy the following symmetries with respect to the triple indices @xmath312 . 1 . if the cosmic density and velocity fields are statistically isotropic , @xmath88 in real space is invariant with respect to cyclic permutations of index @xmath56 , i.e. @xmath313 2 . in the plane - parallel approximation , i.e. the coordinate @xmath112 is in the redshift direction , we have @xmath314 @xmath315 using eqs.(77)-(80 ) , we have a @xmath1 estimator as follows @xmath316 eq.(81 ) looks very similar to the @xmath1 estimator with quadrupole - to - monopole ratio , or the multipole moments of two - point correlation function . however , eq.(81 ) contains not only the information of shape ( like multipole moments ) , but also scales . all the dwt quantities in eq.(77 ) are on scale @xmath69 . that is , the @xmath1 in eq.(77 ) refers to mode @xmath56 . therefore , if @xmath1 is scale - dependent , the @xmath1 estimated by eq.(81 ) is its value referring to @xmath69 . on the other hand , if @xmath1 is scale - free , the values of @xmath1 given by estimator eq.(81 ) with different mode @xmath69 should be the same . figure 10 plots the results in the case of a scale - free @xmath1 for the lcdm1 model using estimators with @xmath317 . the results indeed show that the estimated @xmath1 is independent of the @xmath69 used in the estimator , and it follows the theoretical curves in the entire redshift range considered . therefore , we expect that the estimator ( 81 ) would be useful to study the scale - dependence of bias parameters of galaxies . for comparison , fig 11 shows the @xmath1 estimated from quadrupole - to - monopole ratio as @xmath318 where @xmath319 and @xmath320 are the quadrupole moment and the monopole moment of the power spectrum @xmath321 respectively ( cole , fisher & weinberg 1994 , hereafter cfw ; hamilton 1997 ) . we average the results from wavelength 50 h@xmath0mpc to 400 h@xmath0mpc for each realization . thus the error bars are among different realizations of each model . as noticed in cfw , this estimator gives a lower @xmath1 than the true value at @xmath214 even for wavelengths @xmath322 h@xmath0mpc . in addition , we find that it progressively overestimates @xmath1 at higher and higher redshifts . it is suggested in cfw that the underestimate at short wavelengths is due to the impact of non - linear gravitational clustering . however , this explanation is still difficult to apply for the underestimate at long wavelengths , and the overestimate at high redshift , as cosmic field on long wavelengths or high redshifts should be linear . this discrepancy occurs because the quadrupole - to - monopole ratio eq.(82 ) is scale - dependent even when the field is in linear regime . this scale - dependence is given by the non - linear redshift distortion , or the distortion of velocity dispersion . as mentioned in 5.3 , @xmath277 is non - zero in linear clustering regime , the non - linear redshift distortion ( or the distortion of @xmath277 ) should be considered regardless whether the field is linear or non - linear . therefore , the discrepancy in cfw can not be eliminated for long wavelengths or high redshift . to solve this problem , extra information , such as the power - spectrum index of the density fluctuations , or the real - space correlation function , is added in the quadrupole - to - monopole ratio estimator eq.(82 ) ( peacock et al . 2001 ) . different from the quadrupole - to - monopole ratio , the dwt @xmath1 estimators are scale - decomposed . it is useful to consider scale - dependent effect . the dwt estimators are able to consider the non - linear redshift distortion ( @xmath277 redshift distortion ) without assuming extra information . actually , the result with estimator eq.(81 ) has already shown the effect of the non - linear redshift distortion . we can see from fig . 10 that the @xmath1 estimated by eq.(81 ) are slightly dependent on @xmath24 , and it is progressively higher in the order @xmath323 . this can be explained with fig . 4 , which shows that the factor @xmath258 is less than one , and is smaller for a smaller scale . thus the estimator eq.(81 ) , which ignores the factor @xmath258 in eq.(77 ) , leads to a progressively higher @xmath1 in the order @xmath323 . thus , a simplest way to estimate the non - linear effect is to take an average over @xmath1 given by eq.(81 ) with different @xmath24 . the left panel of figure 12 presents the averaged @xmath1 from fig . 10 for the four models . the error bars contain the contribution of the non - linear effect . that is , the dwt algorithm can do a self - test on whether the result is largely affected by the non - linear effect . more delicate @xmath1 estimators can be constructed if we consider the following properties : the non - linear factor @xmath258 depends only on @xmath270 , but independent of @xmath273 if the los scale is above 100 h@xmath0mpc , while the linear factors @xmath249 and @xmath250 depends only on the shape of the dwt mode ( 5.2 ) . these properties apply to @xmath302 , @xmath298 , and @xmath299 as well . thus , we can combine modes with similar @xmath270 but different shapes to cancel the non - linear factors , such as @xmath324 thus , from eq.(76 ) , we have a @xmath1 estimator as @xmath325 this estimator is similar to that in yang et al . it can be used for any pairs ( @xmath24 @xmath326 @xmath327 ) even when the scale of @xmath24 is small . the right panel of figure 12 shows that for the four models , the @xmath1 estimated by eq.(84 ) has no more than about 15% error at all redshifts @xmath328 . one can also use the modes @xmath329 , @xmath330 to replace modes @xmath331 and @xmath332 in eqs.(83 ) and ( 84 ) , which gives similar results to fig . that is , the weakly non - linear redshift distortion can be considered by the dwt @xmath1 estimators on scales until about @xmath333 h@xmath0 mpc when the quadrupole - to - monopole estimator shows significant errors ( fig . the counterpart of eq.(84 ) with @xmath88 , @xmath334 is not as good as eq.(84 ) . it causes an error of about 40% in @xmath1 because again the quasi - diagonality of the sfc s covariance is poorer than that of wfc s . since a dwt decomposition of a random field yields variables for calculating both the power spectrum and one - point statistics , the @xmath1 estimations with eqs.(81 ) and ( 84 ) can be done in the same time . an accurate @xmath1 estimator requires the knowledge of a precise recovery of @xmath293 or @xmath88 , which accounts for the non - linear redshift distortion due to the velocity dispersion , and even the second order effect of the bulk velocity ignored in eq.(48 ) . in other words , the non - linear redshift distortion exists even on scales that are linear in the sense of structure formation , and it must be corrected to achieve a reliable and accurate @xmath1 . we have developed the one - point statistics of a perturbed density field with the multiresolutional decomposition based on discrete wavelet transform . since the scale and shape of the dwt bases are well defined , this frame work is very effective to deal with problems of how the one - point distribution and its moments depend on the scale and shape of the window function . with this property , we have established the algorithm of one - point variable and its moments in considering the effects of poisson sampling and selection function . we have also established the algorithm for recovering the dwt one - point statistics from the redshift distortion due to bulk velocity , velocity dispersion and selection function . because the recovery of the real - space dwt one - point variable and its moments can be realized scale - by - scale , one can design @xmath1 estimators which are sensitive to the scale - dependence of @xmath1 , for instance , caused by the scale - dependence of bias parameter of galaxies . these @xmath1 estimators are effective in avoiding the difficulty caused by the scale - dependence of the non - linear redshift distortion . compared with conventional @xmath1 estimators ( peacock et al 2001 ) , the dwt @xmath1 estimators do not need to assume that the velocity dispersion is scale - independent , or to add extra information , such as the power - spectrum index or the real - space correlation function of the field . numerical tests by n - body simulation samples show that the proposed estimators can yield the correct value of @xmath1 with about 15% uncertainty for all popular cdm models in the redshift range @xmath335 . since dwt decomposition contains two sets of bases , the scaling function @xmath28 and wavelet @xmath29 , a dwt decomposition of a density field @xmath18 actually yields variables for one - point statistics @xmath336 as well as the variables for calculating the power spectrum @xmath337 ( fang & feng 2000 ) . in this sense , one can say that the dwt decomposition unifies the algorithm of cic and power spectrum , which are only two aspects of the statistics with the dwt variables sfcs and wfcs respectively . for a finite size sample , these two aspects of statistics can play different roles : the former contains information of the perturbations on scales larger than the size , while the latter does not . however , the latter generally has diagonalized covariance , while the former does not . therefore , the @xmath1 estimators with wfcs are better than sfcs , while sfcs are useful to estimate the effect of perturbations on large scales . actually , the dwt decomposition can provide more types of statistics variables . for 2-d or 3-d samples , we have variables defined as @xmath338 @xmath339 obviously , these variables are sampled partially by the scaling function @xmath28 , and partially by the wavelet @xmath29 . statistics with these variables are not typical cic , or power spectrum . they are , however , useful to study the one - point statistics , or power spectrum of 2-d and 3-d samples . with the method developed in this paper , it is not difficult to calculate various corrections ( poisson sampling , selection function , redshift distortion ) on the one - point statistics with variables ( 86 ) or ( 87 ) . hz is grateful to daniel eisenstein for extensive discussions on this paper and facilitating the lcdm2 simulation . hz would also like to thank david burstein for hosting the utility codes in this paper . let us consider the plane - parallel approximation , i.e. coordinate @xmath112 is in the redshift direction . by definition eq.(50 ) , we have @xmath340 because 1-d scaling function @xmath28 is given by dilating and translating the basic wavelet @xmath341 as @xmath342 the fourier transform of @xmath29 is @xmath343 and @xmath344 where @xmath82 is the fourier transform of the basic scaling function @xmath345 the function @xmath346 is shown in fig . 1 . if @xmath354 is independent of @xmath137 , @xmath258 is @xmath355.\ ] ] in the plane - parallel approximation , we have @xmath356.\ ] ] the summation of equations ( a8 ) and ( a9 ) also runs only over numbers of @xmath357 . using the results for @xmath252 and @xmath258 , it is easy to find @xmath358 \exp[-(a/2)(\sigma^{v_3 } k_3/h)^2 ] |\hat{\phi}(u_1)\hat{\phi}(u_2)\hat{\phi}(u_3)|^2.\end{aligned}\ ] ] it is obvious from equation ( a10 ) that @xmath176 is symmetric with respect to @xmath75 and @xmath170 , and it depends only on the difference @xmath359 . consequently , @xmath360 is independent of @xmath75 . in addition , the elements @xmath360 are dominant over @xmath176 with @xmath361 , because the latter sums over oscillating terms . since @xmath367\sin[2\pi u'_3(l_3-l_3 ' ) ] \simeq \left \ { \begin{array}{ll } 1/2\pi u_3 & \mbox{if $ u_3=u'_3 $ } \\ 0 & \mbox{otherwise } \end{array } \right .,\ ] ] we have @xmath368 ^ 2 [ \gamma^{0,1}_{\bf j}]^2.\ ] ]

microlensing detections of cool planets are important for the construction of an unbiased sample to estimate the frequency of planets beyond the snow line , which is where giant planets are thought to form according to the core accretion theory of planet formation . in this paper , we report the discovery of a giant planet detected from the analysis of the light curve of a high - magnification microlensing event moa 2010-blg-477 . the measured planet - star mass ratio is and the projected separation is in units of the einstein radius . the angular einstein radius is unusually large mas . combining this measurement with constraints on the `` microlens parallax '' and the lens flux , we can only limit the host mass to the range . in this particular case , the strong degeneracy between microlensing parallax and planet orbital motion prevents us from measuring more accurate host and planet masses . however , we find that adding bayesian priors from two effects ( galactic model and keplerian orbit ) each independently favors the upper end of this mass range , yielding star and planet masses of and at a distance of kpc , and with a semi - major axis of au . finally , we show that the lens mass can be determined from future high - resolution near - ir adaptive optics observations independently from two effects , photometric and astrometric .

1205.6323

gravitational microlensing is an important method to detect extrasolar planets @xcite . the method is sensitive to planets not easily accessible to other methods , in particular cool and small planets at or beyond the snow line @xcite , and free - floating planets @xcite . the snow line represents the location in the protoplanetary disk beyond which ices can exist @xcite and thus the surface density of solids is highest @xcite . according to the core accretion theory of planet formation @xcite , the snow line plays a crucial role because giant planets are thought to form in the region immediately beyond the snow line . therefore , microlensing planets can provide important constraints on planet formation theories , in particular by measuring the mass function beyond the snow line @xcite . a major component of current planetary microlensing experiments is being carried out in survey and follow - up mode , where survey experiments are conducted in order to maximize the event rate by monitoring a large area of the sky one or several times per night , while follow - up experiments are focused on events alerted by survey observations to densely cover planet - induced perturbations . in this mode , high - magnification events are important targets for follow - up observations . this is because the source trajectories of these events always pass close to the central perturbation region and thus the sensitivity to planets is extremely high @xcite . in addition , the time of the perturbation can be predicted in advance so that intensive follow - up observation can be prepared . this leads to an observational strategy of monitoring high - magnification events as intensively as possible , regardless of whether or not they show evidence of planets . as a result , the strategy allows one to construct an unbiased sample to derive the frequency of planets beyond snow line @xcite . for the alternative low - magnification channel of detection , see for instance @xcite . in this paper , we report the discovery of a giant planet detected from the analysis of the light curve of a high - magnification microlensing event moa 2010-blg-477 . due to the high magnification of the event , the perturbation was very densely covered , enabling us to place constraints on the physical parameters from the higher - order effects in the lensing light curve induced by finite source effects as well as the orbital motion of both the lens and the earth . we provide the most probable physical parameters of the planetary system , corresponding to a jupiter - mass planet orbiting a k dwarf at about 2 au , the system lying at about 2 kpc from earth . the event moa 2010-blg-477 at coordinates ( ra , dec)=(@xmath8 ( j2000.0 ) , @xmath9 was detected and announced as a microlensing alert event by the microlensing observations in astrophysics @xcite collaboration on 2010 august 2 ( @xmath10 ) using the 1.8 m telescope of mt . john observatory in new zealand . the event was also observed by the optical gravitational lensing experiment @xcite using the 1.3 m warsaw telescope of las campanas observatory in chile . real - time modeling based on the rising part of the light curve indicated that the event would reach very high magnification and enabled us to predict the time of the peak . followed by this second alert , the peak of the light curve was densely covered by follow - up groups of the probing lensing anomalies network ( planet ) @xcite , microlensing follow - up network ( @xmath11fun ) @xcite , robonet @xcite , and microlensing network for the detection of small terrestrial exoplanets ( mindstep ) @xcite . more than twenty telescopes were used for the follow - up observations , including planet 1.0 m of south african astronomical observatory ( saao ) in south africa , planet 1.0 m of mt . canopus observatory in tasmania , australia , planet 0.6 m perth observatory in australia , planet 0.4 m astep telescope at dome c , antarctica , @xmath11fun 1.3 m smarts telescope of the cerro tololo inter - american observatory ( ctio ) , chile , @xmath11fun 0.4 m of auckland observatory , @xmath11fun 0.36 m of farm cove observatory ( fco ) , @xmath11fun 0.36 m of kumeu observatory , @xmath11fun 0.4 m of possum observatory , @xmath11fun 0.4 m of vintage lane observatory ( vlo ) , @xmath11fun 0.3 m of molehill astronomical observatory ( mao ) , all in new zealand , @xmath11fun 0.46 m of wise observatory , israel , @xmath11fun 0.8 m of teide observatory at canary islands ( iac ) , spain , @xmath11fun 0.6 m of pico dos dias observatory , brazil , robonet 2.0 m of faulkes north ( ftn ) in hawaii , u.s.a . , robonet 2.0 m of faulkes south ( fts ) , australia , robonet 2.0 m of liverpool ( lt ) at canary islands , spain , mindstep 1.54 m danish telescope at la silla , chile , and mindstep 1.2 m monet north telescope at mcdonald observatory , u.s.a .. please note that this paper reports the first microlensing observations from antarctica . unfortunately , the data quality of these pioneering observations was not high enough to be included into the models , but we give a short overview in appendix [ app : astep ] . in order to better constrain the second - order effects , new observations were taken at the @xmath11fun 1.3 m smarts telescope at ctio during the 2011 campaign . to better characterize the lensed source star , spectroscopic observations were conducted near the peak of the event ( @xmath12 ) by using the b&c spectrograph on the 2.5 m du pont telescope at las campanas observatory in chile . the resolution was @xmath13 , corresponding to @xmath14 . a comparison between the observed spectrum and synthetic ones was conducted to derive the effective temperature , gravity and metallicity of the source star . the usable part of the spectrum is only @xmath15 due to some scattered light issues with the instrument , particularly a problem for this faint star . we started by fitting standard stars using h@xmath16 , mgb , and nad , but found that fits with h@xmath16 alone resulted in the most accurate temperatures , so this is the diagnostic we used . the derived parameters of the source star are @xmath17 k for @xmath18 and a solar metallicity . the corresponding fit to the h@xmath16 line is shown in figure [ fig : spectrum ] . the data collected by the individual groups were initially reduced using various photometry codes developed by the individual groups . for the data sets from saao , fts , possum , canopus , perth , danish and monet north , we use photometry from rereductions obtained with the pysis package , described in more details in appendix [ app : pysis ] . figure [ fig : lcurve ] shows the light curve of the event . the event is highly magnified with a peak magnification @xmath19 . outside of the region @xmath20 $ ] , the light curve is consistent with a standard single - lens curve @xcite . the perturbation is composed of two spikes at @xmath21 and 5420.9 and two bumps at @xmath22 and 5422.4 . as in the great majority of planetary microlensing events , we are able to measure the `` angular einstein radius '' ( projected on the sky ) @xmath23 , but not the projected einstein radius ( projected on the observer plane ) @xmath24 . consequently , it is challenging to estimate the physical parameters of the lens . the two radius quantities are related to those ( * ? ? ? * see ) by @xmath25 where @xmath26 , @xmath27 is the lens mass in @xmath28 , and @xmath29 is the lens - source relative parallax ( @xmath30 and @xmath31 are the lens and source distances ) . since @xmath23 is well measured , the product @xmath32 is also well determined , but in the present case , the ratio @xmath33 is poorly constrained , and so it is difficult to estimate @xmath27 alone . in this section , we provide an overview of the various techniques that we use to place constraints on the individual quantities @xmath34 and @xmath27 . we will show that here @xmath23 is large enough to enable a substantial constraint from upper limits on the lens flux . that is , from equation ( [ eqn : theta_e+pi_e ] ) , large @xmath23 implies large @xmath27 or @xmath34 , both of which lead to brighter lenses . this will lead to the unambiguous conclusion that the lens is nearby , @xmath35kpc , with mass @xmath36 . the light curve of this event enables stronger constraints than is usually the case because we are able to obtain a measurement of one of the microlens parallax component . the microlens parallax is actually a vector , @xmath37 , with the magnitude given by equation ( [ eqn : theta_e+pi_e ] ) and the direction by the lens - source relative motion @xcite . one component of the parallax only weakly constrains the scalar @xmath38 , but it constrains the direction of proper motion , which will be very important for future observations ( see below ) . moreover , one parallax component actually does provide a robust lower limit on the mass @xmath39 and distance @xmath40kpc . to proceed further , we must apply a bayesian analysis , but here as elsewhere we are more fortunate than is typical . as always , there are bayesian priors from a galactic model , and in this case these strongly prefer the upper end of the mass / distance range permitted by the light curve . but in addition , the reason that only one dimension of @xmath37 can be robustly measured is that the other dimension is degenerate with orbital motion of the planet @xcite . thus , it is necessary to fit simultaneously for both parallax and orbital motion . bayesian priors on the orbital motion then _ independently _ also prefer the upper end of the allowed mass / distance parameter space . finally , we predict that high - resolution imaging could measure the mass and distance of the lens through two independent effects , photometry and astrometry . the first is obvious : different mass / distance combinations yield different color / magnitude measurement . the main point here is that the large value of @xmath23 virtually guarantees that the lens will be detectable . the second is less obvious : different mass / distance combinations also predict different _ directions _ of source - lens relative proper motion , which can be measured as the lens and source are separated over the next several years . the photometric error bars of the various data sets from individual observatories are generally not accurate enough to be taken at face value . they need a rescaling to reproduce the dispersion of contiguous data points in a given night . at the same time , outliers must be identified and removed . these are important preliminary steps to modeling , because the weight of a given data set to constrain the model depends on how large its error bars are , compared to other data sets . we describe our adopted noise model and rescaling factors for each observatory data set in appendix [ app : rescaling ] . we first test a static binary lens model . the corresponding parameter set includes the three single lens parameters : the einstein time scale , @xmath41 , the time of the closest lens - source approach , @xmath42 , and the lens - source separation at that moment , @xmath43 , and it also includes the three binary parameters : the mass ratio of the companion to its host star , @xmath44 , the projected separation between the lens components in units of the einstein radius , @xmath45 , and the angle between the source trajectory and the binary axis , @xmath16 . since the light curve exhibits caustic - crossing features , we need to consider the modification of magnifications caused by the finite - source effect @xcite . this requires us to include an additional parameter : the normalized source radius @xmath46 where @xmath47 is the angular source radius . evaluation of @xmath48 from the model together with the measurement of @xmath47 ( see section [ sec : source ] ) will then yield @xmath23 and thus the product @xmath32 ( see equation [ [ eqn : theta_e+pi_e ] ] ) . since the event moa 2010-blg-477 involves caustic crossings and approaches , one must compute magnification affected by the finite - source effect . computation of finite magnifications is based on the numerical ray - shooting method . in this method , a large number of rays are uniformly shot from the image plane , bent according to the lens equation , and land on the source plane . the lens equation is represented by @xcite @xmath49 where @xmath50 , @xmath51 , and @xmath52 are the complex notations of the source , lens , and image positions , respectively , @xmath53 denotes the complex conjugate of @xmath54 , and @xmath55 are the masses of the individual lens components , and @xmath27 is the total mass of the lens system . then , the magnification is computed as the ratio of the number density of rays on the source plane to the density of the image plane . for the initial search for solutions in the space of the grid parameters , we accelerate the computation by using the `` map making '' method @xcite . in this method , a magnification map is made for a given set of @xmath56 and then it is used to produce numerous light curves resulting from different source trajectories instead of re - shooting rays all over again . we further accelerate the computation by using a semi - analytic hexadecapole approximation for finite - magnification computation @xcite in the region where the source location is not very close to the caustic . in computing finite magnifications , we consider the effect of limb darkening of the source star surface by modeling the specific intensity as @xcite @xmath57 } } \ ] ] where @xmath58 is a limb - darkening coefficient ( hereafter ldc ) , @xmath59 is the total flux from the source star , @xmath60 is the angle between the direction toward the observer and the normal to the stellar surface . from the @xmath61 improvement we find that the limb - darkening effect is clearly detected . we compute the ldcs for equation [ eqn : eq4 ] as accurately as possible , including a proper treatment of the effect of extinction . we refer the interested reader to appendix [ app : lldc ] for details . we then test if second - order effects are present in the residuals of the light curve . these effects may have several origins : orbital motion of the earth around the sun @xcite , which induces a deviation of the lens - source motion from rectilinear , orbital motion of the planet about the lens star , orbital motion of the source star if it is a binary , and possible additional objects ( planets or stars ) in the lens system . in fact we will limit our study to the first two effects , namely microlens parallax and planet orbital motion . a binary source in which the companion is either not lensed or too faint to contribute to the light curve can mimic the parallax effect due to the acceleration of the source , in a so - called `` xallarap '' effect . this effect is described for instance in @xcite and tested in @xcite . however , it implies in its simplest form 5 additional parameters versus 2 each for parallax and planet orbital motion . given the fact that our detection of second - order effects , although clear , remains relatively marginal , it would be difficult to trust 5 additional parameters constrained by the light curve residuals from a standard model . moreover , binary stars generally have long periods , the effect of which will remain undetected in the short lapse of time when our events are observed ( months versus years ) . testing for a third body , either from a circumbinary planet or a second planet orbiting the lens star , is beyond the scope of this paper , and again would imply a large number of additional parameters to fit , without any guarantee to obtain meaningful results . in the short history of planetary microlensing , it has been evidenced only once @xcite , and in that case only because a feature in the light curve could not be fitted by a single planet model . there is no such feature in the present light curve . as the domain of possible configurations to test if we add a third body is vast , the effort expended would be incommensurate with the potential scientific return , at least until we get independent evidence that such a modeling effort is necessary : see section [ sec : highres ] . as discussed in section [ sec : overview ] , the microlens parallax is characterized by a 2-dimensional vector @xmath37 whose magnitude is given by equation ( [ eqn : theta_e+pi_e ] ) and whose direction is that of the lens - source relative motion on the plane of the sky . hence , there are two parameters @xmath62 , the components of this vector in equatorial coordinates . the planet orbital motion affects the light curve in two different ways . first , it causes the binary axis to rotate or , equivalently , makes the source trajectory angle change in time @xcite . second , it causes the projected binary separation , and thus the magnification pattern , to change in time . considering that the time scale of the lensing event is of order a month while the orbital period of typical microlensing planets is of order years , the rates of change in @xmath16 and @xmath45 can be approximated as uniform during the event and thus the orbital effect is parameterized by @xmath63 and @xmath64 where @xmath65 and @xmath66 are the rates of change in the source trajectory angle and projected binary separation in units of @xmath67 , respectively , and @xmath68 is a reference time . as explained in @xcite , the effect of planet orbital motion is similar to the microlensing parallax effect in the sense that the deviations caused by both effects are smooth and long lasting . then , if the deviation caused by the planet orbital motion is modeled by the parallax effect alone , the measured parallax would differ from the correct value . therefore , the orbital motion effect is important not only to constrain the orbital properties of the lens , but also to precisely constrain the lens parallax and thus the physical parameters of the lens system . lcccc @xmath61 & 6420.051 & 6411.450 & 6369.345 & 6365.664 + @xmath42 ( hjd ) & 5420.93685 & 5420.93702 & 5420.93773 & 5420.93915 + @xmath43 & -0.003562 & -0.003540 & -0.003372 & + 0.003404 + @xmath41 ( days ) & 42.55 & 42.94 & 46.57 & 46.92 + @xmath45 & 1.12282 & 1.12338 & 1.12385 & 1.12279 + @xmath44 & 0.0023808 & 0.0023694 & 0.0022075 & 0.0021809 + @xmath16 ( rad ) & 3.67605 & 3.67528 & 3.68023 & 2.60087 + @xmath48 & 0.0006429 & 0.0006403 & 0.0005861 & 0.0005764 + @xmath69 & 0.4111 & 0.4082 & 0.3757 & 0.3731 + @xmath70 & 997.3 & 997.9 & 1045.8 & 1059.8 + @xmath71 & & + 0.37 & + 0.27 & + 0.77 + @xmath72 & & + 0.012 & + 0.02 & -0.11 + @xmath66 ( @xmath67 ) & & & + 0.64 & + 0.86 + @xmath65 ( @xmath67 ) & & & + 0.06 & -1.28 + @xmath73 ( yr ) & & & 14.25 & 4.59 + @xmath68 ( hjd ) & & 5421 & 5421 & 5421 + for each tested model , we search for the solution of the best - fit parameters by minimizing @xmath61 in the parameter space . in the initial search for solutions , we divide the parameters into two categories . parameters in the first category are held fixed during the fitting , and parameter space is searched with a grid . for the parameters in the second category , solutions are searched by using a downhill approach . we choose @xmath45 , @xmath44 , and @xmath16 as the grid parameters because they are related to the light curve features in a complex way , where a small change in the parameter can result in dramatic changes in the light curve . on the other hand , the remaining parameters are more directly related to the features of the light curve and can thus be searched by using a downhill approach . for the downhill @xmath61 minimization , we use a markov chain monte carlo method . brute - force search throughout the grid - parameter space is also needed to test the possibility of the existence of local minima that result in degenerate solutions . for the light curve of moa 2010-blg-477 , we find that the other local @xmath61 minima have @xmath61 values much larger than the best fit by @xmath74 and are therefore not viable solutions . once the space of the grid parameters around the solution is sufficiently narrowed down , we allow the grid parameters to vary in order to pin down the exact location of the solution and to estimate the uncertainties of the parameters . modeling was also done independently using the method of @xcite , and this analysis reached the same conclusions . this independent analysis also uses a slightly different implementation of the planetary orbital motion parameters @xcite . equations [ eqn : eq1 ] and [ eqn : eq2 ] describe the orbital velocities , which are the first order contribution of orbital motion . to second order , we have only one component of acceleration because this must be directed towards the host star . but , one additional parameter is also all that is needed to describe a circular orbit , so we can add the planetary orbital period , @xmath73 as a parameter and replace the constant velocities ( in polar coordinates ) of equations [ eqn : eq1 ] and [ eqn : eq2 ] with the projection into the plane of the sky of the circular orbit described by @xmath45 , @xmath16 , @xmath75 , @xmath65 , and @xmath73 . unlike the case of ogle 2006-blg-109lb , c @xcite , the value of @xmath73 does not have an influence on the light curve model @xmath61 values . however , @xmath73 is still useful because it can be used to help constrain the other orbital parameters to values consistent with a physical orbit . ( this is an issue because it is quite possible to have @xmath75 and @xmath65 values that are not consistent with a bound orbit , and this is a simple way to ensure that this is not the case ) . if we assume that @xmath76 is known , then we can calculate the lens system mass , @xmath77 , which follows from equation [ eqn : theta_e+pi_e ] . this also allows us to determine @xmath34 , but we can not determine the lens and source distances , @xmath30 and @xmath31 , separately . however , we can use the @xmath73 value to determine the orbital semi - major axis , under the assumption that the orbit is circular . this allows us to determine the einstein radius , @xmath78 in physical units , and since @xmath79 , we have a second relation between @xmath31 and @xmath30 . but , since the source star is very likely to be in the galactic bulge , we also have approximate knowledge of @xmath31 . therefore , we can apply a constraint on the value of @xmath31 implied by the light curve parameters , @xmath80kpc . for this event , @xmath73 is not really constrained by the light curve measurements , so this constraint serves to force @xmath73 toward a value consistent with a circular orbit for a bulge source . this constraint on the source distance also serves to enforce a constraint on the orbital velocity parameters , @xmath75 and @xmath65 . they must also be consistent with a circular orbit for a bulge source . parameters that satisfy this constraint are also consistent with most orbits with moderate eccentricity , @xmath81 . but , these parameters are not consistent with orbits with the highest possible transverse velocities . in fact , the light curve measurements marginally favor implausibly large @xmath75 and @xmath65 values corresponding to orbits that are either unbound or just barely bound . these barely bound orbits with large @xmath75 and @xmath65 values have high eccentricities that just happen to have been observed with motion in the plane of the sky during the brief time near periapsis . the best fit model that is consistent with a bound orbit is such a model , which has an orbit with an extremely low _ a priori _ probability . this low _ a priori _ probability makes such a model much more unlikely than the best fit model with the circular orbit constraint , so we report the best fit model with the circular orbit constraint as the best fit model " in table [ tab : one ] . however , the unlikely models with values of @xmath75 and @xmath65 do tend to have better light curve @xmath61 values , and they also cover a large volume of parameter space . so , they should not be ignored in our consideration in the range of possible physical parameters for the moa 2010-blg-477l planetary system . therefore , we do not enforce this source distance constraint in the mcmc runs that we use to estimate the distribution of likely physical parameters for this system . instead , we apply a more general constraint on the orbital and galactic parameters of the lens system as discussed in appendix [ app : jacobi ] . in table [ tab : one ] , we present the results of modeling along with the best - fit parameters for the 3 tested models . the best - fit light curve is presented in figure [ fig : lcurve ] . in figure [ fig : geom ] , we also present the geometry of the lens system . it is found that the perturbation near the peak of the light curve was caused by the source crossings and approaches of the caustic produced by the binary system with a low - mass companion . the measured mass ratio between the lens components is @xmath0 and thus the companion is very likely to be a planet . the measured projected separation between the lens components is @xmath1 , which is close to the einstein radius . as a result , the caustic is resonant , implying that the caustic forms a single closed curve composed of 6 cusps . the perturbations at @xmath82 and @xmath83 are produced by the source crossings of one of the star - side tips of the caustic . the bumps at @xmath84 and @xmath85 are caused by the source s approach close to the weak and strong cusps on the side of the host star , respectively . the most important constraints on the lens mass and distance come from the measurement of @xmath23 , which can be rewritten from equation ( [ eqn : rhodef ] ) as @xmath86 where @xmath87 is the intrinsic ( dereddened ) flux of the source and @xmath88 is its intrinsic @xmath89-band surface brightness . a key point is that the surface brightness does not depend at all on the microlens model ( just on the source color and/or spectrum ) . hence , @xmath23 depends on the model only through the parameter combination @xmath70 , where @xmath69 is the instrumental source flux ( @xmath90 corresponds to magnitude 18 ) . it is clear from table [ tab : one ] that this parameter combination varies very little . the biggest uncertainties are therefore in measuring the surface brightness and measuring the offset between the instrumental flux @xmath69 and the dereddened source flux @xmath87 . traditionally , these are measured simultaneously by determining the offsets of the color and magnitude , respectively , of the source from the clump on an instrumental cmd @xcite . an instrumental cmd from ctio is shown in figure [ fig : cmd ] . from it , we can read instrumental magnitude in @xmath89 of 19.07 . the instrumental @xmath91 color is measured more accurately from a regression of @xmath92 flux vs. @xmath89 flux , and gives @xmath93 . comparing to the instrumental clump position , we find @xmath94 and @xmath95 , assuming @xmath96_{0,\rm clump } = ( 1.06,14.42)$ ] . the color error is determined empirically from a sample of microlensed dwarfs with spectra @xcite , while the magnitude error comes primarily from the error in fitting the clump centroid and the assumed galactocentric distance of 8 kpc ( both about 0.1 mag ) . these lead to an estimate @xmath97 . we can compare the dereddened color estimate to the color deduced from the high - resolution spectrum reported in section [ sec : obs ] , from which we measured an effective temperature of @xmath98k , which corresponds to @xmath99 . these determinations are marginally consistent . we adopt the first to maintain the general practice of microlensing papers , but note that if we adopted the mean of the two estimates , the inferred value of @xmath47 ( and so @xmath23 ) would rise by 2% . finally , we evaluate @xmath100 this value is unusually large and implies that the lens must be very massive or very close . specifically @xmath101 the model gives a measurement of the light coming not only from the source , but also from any other stars in the same point spread function , generally called blended light . this light may come from the lens itself , a companion to the lens , a companion to the source , or any unrelated star on the same line of sight , but not participating to the amplification process . the ogle - iii image of this field , displayed in figure [ fig : ogle ] , shows two stars at the target position . their @xmath89-band magnitudes in the ogle - iii photometric catalogue is @xmath102 for the brighter star , numbered 119416 , and @xmath103 for the fainter one with number 119534 . they are separated by @xmath104 , enough to be separated by psf photometry at good sites such as ctio or las campanas ( ogle ) in chile . a difference image analysis ( dia ) made on ctio images shows that the microlensed star does not correspond to the position of either star ( red circle in figure [ fig : ogle ] ) , but is displaced by 1 ctio pixel ( @xmath105 ) from the brighter star . now , the blended light in @xmath89-band as measured from dophot ctio photometry by the model is @xmath106 , and this corresponds precisely to the flux from the brighter star among the two ogle stars , which is not separated from the microlensed target at the scale of the ctio seeing ( typically @xmath107 ) . knowing that the blended flux comes from an unrelated blended star , the light from the lens must be smaller . we can rigorously conclude that the lens has less than half the light in the observed blend . otherwise , the lens and blend would be separated by at least @xmath108 in the ogle image , and so would have been at least marginally resolved . an additional argument showing that the lens is faint enough to remain undetected comes from its large relative proper motion ( @xmath109 mas / yr ) . dia analysis of ogle - iii good seeing images separated by 3.3 years shows no residual at the target position , which proves that no detected star has moved during this period . combining this limit with equation ( [ eqn : thetae2 ] ) yields strong constraints on the lens . the lens must be closer than the source , and so be at or closer than the galactocentric distance , and suffer the same or less extinction . these imply @xmath110 , and so @xmath111 , which corresponds to @xmath112 @xcite . then , even the @xmath113 limit from equation ( [ eqn : thetae2 ] ) implies @xmath114 mas and so @xmath115kpc . but @xmath116 then implies @xmath117 , which corresponds to @xmath118 . cycling through this argument one more time yields @xmath119kpc , @xmath120 , @xmath121 . if the lens were in front of some of the dust , this argument would become still stronger . however , at the relatively high latitude of this field ( @xmath122 ) , most of the dust probably lies in front of 3 kpc , and in any case , there is no basis for adopting a more optimistic assumption about the dust . table [ tab : one ] shows that parallax alone improves the fit by @xmath123 , while including both parallax and lens orbital motion improves it by @xmath124 . however , while _ detection _ of these effects is therefore unambiguous , we can not fully disentangle one from the other . for each effect , one of its two parameters is well determined while the other is highly degenerate with one parameter from the other effect . as first discovered by @xcite and further analyzed by @xcite , weak detections of parallax and orbital motion lead to a strong degeneracy between @xmath125 ( the component of @xmath37 perpendicular to the projected position of the sun ) and @xmath65 ( the component of orbital motion perpendicular to planet - star axis ) , while @xmath126 and @xmath75 are well constrained . the impact of this is illustrated in panel ( a ) of figure [ fig : bayes1 ] , which shows @xmath61 values by color as a function of @xmath127 . first , it is clear that the parallax vector is almost completely degenerate along a line that is @xmath128 west of north , within @xmath129 of the predicted orientation of @xmath130 ( @xmath131 ) . second , the light curve excludes @xmath132 at @xmath133 . as we have @xmath134 this corresponds to @xmath135 . thus , combining constraints from this section and from section [ sec : flux ] , and using equation ( [ eqn : thetae2 ] ) , we have @xmath136 note that , because the parallax contours pass through the origin , parallax provides no additional constraint at low @xmath38 , i.e. , at high mass . a similar diagram is given for the slightly disfavored @xmath137 solution in figure [ fig : bayes2 ] . equation ( [ eqn : massdistlims ] ) defines the limits of what can be said rigorously about the lens mass and distance based on current data . as we discuss in section [ sec : highres ] , the lens could almost certainly be detected by high - resolution imaging , which would probably completely resolve the uncertainty in equation ( [ eqn : massdistlims ] ) . in the meantime , we can perform a bayesian analysis based on a galactic model and constraints from a keplerian orbit . figure [ fig : bayes1 ] shows that each of these constraints separately favors small values of @xmath38 and hence , relatively large masses and distances ( within the limits set by equation [ [ eqn : massdistlims ] ] ) , and so tend to reinforce each other . panel ( a ) shows the raw results of the mcmc . panel ( b ) shows the same chain , post - weighted by the bayesian prior due to the galactic model and the flux constraint . the latter , which implies @xmath118 ( section [ sec : flux ] ) is responsible for the circular `` holes '' at the centers of panels ( b ) and ( d ) . as discussed in detail by @xcite this includes terms not only reflecting the density of lenses along the sight and the expected distribution of proper motions , but also a jacobian transforming from the `` natural microlensing variables '' to the physical system of the galaxy . it is this last term that actually dominates , particularly in the `` near field '' ( @xmath138kpc ) permitted by the constraints , and is roughly @xmath139 ( @xcite , equation [ 18 ] ) . this weighting is so severe that one must ask whether the result of the weighting is plausibly compatible with the raw @xmath61 from the light curve . in this case , the solutions favored by the bayesian post - weighting are compatible with the raw @xmath61 minimum at better than @xmath113 , so there is no major conflict . panel ( c ) shows the results of post - weighting the chain only by the orbital jacobian . this is described in detail by @xcite for the case of chains with complete orbital solutions . these contain two orbital parameters ( called @xmath140 and @xmath141 ) in addition to the two first - order orbital parameters considered here ( @xmath75 and @xmath65 ) . these higher order parameters would be completely unconstrained in the present problem , so we simply resample the chain with a uniform integration over these two parameters . panel ( c ) clearly also favors more massive , more distant lenses ( although not the most massive ) , but it is not immediately obvious why . appendix [ app : jacobi ] details the reasons behind this result . finally , we note that , as expected , the combined effect of these two priors shown in panel ( d ) is stronger than either separately . why do the galactic model and kepler constraints each favor more distant ( and more massive lenses ) than the light curve alone ( see fig . [ fig : bayes1 ] ) ? the galactic model constraint is virtually guaranteed to favor more distant lenses because , as mentioned above , most of the weighting is simply due to a coordinate transformation from microlensing to physical coordinates , and there is more phase space at larger distances . since the distance errors are fairly large , this effect will be relatively strong . by contrast , the kepler constraint could have just as easily favored more - distant as less - distant lenses . the most likely explanation is then , simply , that the light curve prediction of the distance is too close by 1.7 sigma , and the constrained value ( panel d ) is a better estimate . this conjecture is testable by future observations , as described in the next section . follow - up observations are important to check the predictions of our models . in the microlensing field , the idea of doing follow - up observations preceded the detection of planets , going back at least to 1998 in the case of macho - lmc-5 , with the corresponding hst observations published in @xcite . then , it was successfully applied to derive more accurate parameters of the planetary systems , using hst or adaptive optics systems at keck or vlt , for instance in @xcite . in the future , it could be used to study planets in the habitable zones of nearby dwarf stars , as suggested by @xcite . while the bayesian analysis strongly favors a lens close to @xmath142 , i.e , the upper limit permitted by the measurements , there is no airtight evidence against lower - mass lenses . fortunately , this question can almost certainly be resolved by high - resolution imaging . the @xmath143 source magnitudes are @xmath144 , @xmath145 , and @xmath146 . these come from ctio @xmath147-band measurements , taken simultaneously with @xmath89 and @xmath92-band observations . a linear regression between both fluxes gives a very accurate instrumental color of @xmath148 . after converting @xmath147 to the 2mass photometric system and @xmath89 to the ogle - iii system by use of common stars in the ctio field , the color becomes @xmath149 . subtracting this color from the @xmath89 magnitude of the source in the ogle - iii system ( @xmath150 ) returns the above value for @xmath147 . as we now show , the lens brightness must be similar to the source brightness . as can be judged from figure [ fig : ogle ] , the field is relatively sparse , at a galactic latitude of @xmath151 , so these two stars ( for the moment superposed ) will very likely be the only two stars in their immediate high - resolution neighborhood . let us consider three examples consistent with equation ( [ eqn : thetae2 ] ) , @xmath152 , @xmath153 , @xmath154 . these would have lens absolute magnitudes @xmath155 according to @xcite and so @xmath156 . of course , the extinction would be different at these different distances , but the entire column to the source is only @xmath157 . therefore , a 0.5 @xmath28 lens star will be as bright as the source , and even a lens star at the bottom of the main sequence will produce an easily detectable amount of light ( @xmath158 mag ) over the expected source magnitude . since all @xmath159 combinations produce similar @xmath147 magnitudes , such a measurement would , by itself , have little predictive power . but these various scenarios would yield substantially different @xmath160 colors , which would add discriminatory power . time has been allocated on various large telescopes to observe the field of this event , detect and measure the light coming from the source and lens stars . in addition , because of the relatively large proper motion , @xmath161 mas / yr the lens and source could be separately resolved within about 5 years . this would then yield the angle of proper motion ( @xmath162 ) and so ( from figure [ fig : bayes1 ] ) the amplitude of @xmath38 @xcite . we can also consider follow - up observations with the hubble space telescope , which will be able to detect the lens - source relative proper motion as early as 2012 @xcite . the implied absolute magnitudes for the three examples given above ( @xmath163 , 0.5 , and @xmath164 ) are @xmath165 , and the implied extinction free magnitudes are @xmath166 . since the extinction in the foreground of the lens is @xmath167 , this implies that the host star should be detectable with at least 9% of the @xmath89-band flux of the source star over the full range of main sequence host star masses . finally , the esa satellite gaia to be launched in 2013 will image the galactic bulge down to @xmath168 mag , so it may detect the lensing object and will certainly measure its proper motion if it does . in conclusion , these additional observations should reveal whether our choice of second - order effects ( microlensing parallax and planet orbital motion ) correspond to the reality . if we find contradictory results to our predictions , then it will be a strong argument to conduct the modeling of additional effects mentioned in section [ sec : second ] , such as xallarap effect from a binary source , or involvement of a third body . in order to select the best among the three competing models ( standard , parallax only , orbital motion and parallax ) , a simple comparison of @xmath61 values is not enough , because more refined models use more parameters . taking into account these additional parameters by normalizing the @xmath61 estimate by the number of degree of freedom is not the proper way to select the best model . a vast literature exists about model selection , and an application of different criteria to astrophysics is described in @xcite . a simple way to take care of the larger number of parameters is to use the akaike information criterion ( aic ) @xcite , which introduces a penalty to the @xmath61 by adding twice the number of additional parameters . different criteria , such as the bayesian information criterion ( bic ) @xcite or the deviation information criterion , introduced by @xcite , can also be used . formulas are given below , where @xmath169 is the number of additional parameters , @xmath170 the number of data points , and @xmath171 the @xmath61 of the average parameter set @xmath172 . @xmath173 as there are 7 parameters in the standard model , 9 in the parallax only model and 11 in the orbital motion and parallax model , we see that our observed difference in @xmath61 of 8.6 in the parallax only model is only marginally significant according to the aic ( the expected difference is 4 ) , while the observed difference of 54.4 ( for the @xmath174 ) or 50.7 ( for the @xmath137 ) is clearly an improvement of the orbital motion and parallax model over the standard one ( the expected difference is 8) . similar results are obtained using dic , while the difference between models returned by bic is less significant . however , it is important to note that @xmath169 strictly corresponds to the number of additional parameters only in the case of a linear regression problem . here , we clearly have non - linear fits , so we should compute an `` effective '' number of parameters , which is difficult to estimate . the above conclusion should therefore not be taken as a quantitative one . a confirmation of the detection of second - order effects comes from the fact that in the orbital motion and parallax model , the degeneracy between @xmath137 and @xmath174 models is clearly broken . it is instructive to plot both second - order effects vs each other , separately for the @xmath137 and @xmath174 solutions . this is done in figure [ fig : secorder ] , where the orbital motion @xmath175 is plotted vs. the parallax effect @xmath38 . if we remember that the ratio of the projected kinetic to potential energy must be smaller than 1 to get bound orbits , and that this ratio is proportional to @xmath176 , where the proportionality constant depends on @xmath38 as given by equation [ eqn : gamma0 ] , it is easy to interpret these diagrams . when @xmath38 increases , the proportionality factor decreases , so that if @xmath175 remains small enough , the bound orbit condition is respected . in the @xmath174 diagram ( left side ) , small values of @xmath38 ( say below 0.17 ) have a proportionality factor larger than 1 . as @xmath175 is nearly 1 for these solutions , they are ruled out , and the light curve confirms it , as few small @xmath61 solutions ( black and red points ) lie there . larger @xmath38 are also excluded , as they correspond to large @xmath175 values , although the light curve would favor such solutions . the only surviving region in this diagram is around @xmath177 , corresponding to lens masses of half a solar mass , where the proportionality factor is about 0.4 and @xmath175 slightly exceeds 1 . in the @xmath137 diagram ( right side ) , although this is slightly disfavored by the light curve ( @xmath178 for the best chain without the circular orbit constraint ) , there is a region where @xmath175 is about constant at 0.7 for @xmath38 varying from 0 to 0.4 . the bound solutions correspond to the larger values of the @xmath38 domain ( smaller proportionality factor ) , and they therefore agree with the range found in the @xmath174 diagram . we therefore conclude that both solutions agree , and give bound orbits when the lens mass is about half solar , corresponding to a lens distance of about 1.6 kpc . if we now move to the post - bayesian analysis , we see that this solution favored by the light curve has some tension with the galactic model constraint , because nearby lenses are rarer than more distant ones . but if we move to more distant lenses , we get many chains with unbound orbits or high eccentricities . by the way , the @xmath174 solution , which has a lower @xmath61 than the alternate @xmath137 solution , is also the one where more chains correspond to unbound orbits . there is therefore a tension between galactic and keplerian priors , and the issue will only be solved photometrically , by measuring the light coming from the source and the lens . this will be the subject of a forthcoming article about this event . we conclude by giving the 1-d distributions of lens mass , lens distance , and planet orbit semi - major axis . the mass function for the lenses involved in these plots include main - sequence stars , brown dwarfs , but also white dwarfs , neutron stars and black holes , which may have large masses without violating the lens flux limit constraint . for the ms and bd stars , we adopt the following slopes of the present - day mass function @xmath179 : @xmath180 between 0.03 and 0.7 @xmath28 , @xmath181 between 0.7 and 1.0 @xmath28 , and @xmath182 above . for the remnants , we adopt gaussian distributions , whose mean value , standard deviation , and fraction of total mass with respect to ms and bd stars below @xmath183 are given in table [ tab : massdist ] . lllr wd & 0.6 & 0.07 & 22/69 + ns & 1.35 & 0.04 & 6/69 + bh & 5.0 & 1.0 & 3/69 for details about the choice of these numbers , please refer to @xcite . in each diagram ( see figure [ fig:1dplots1 ] for mass and flux , and figure [ fig:1dplots2 ] for distance and semi - major axis ) , the black curves show the full mass function , while the red curves show the mass function truncated at @xmath184 . for ms stars , this limit is imposed by the lens flux constraint , and will be refined once we obtain the adaptive optics photometry of the individual stars in the field . wd at this mass are extremely rare ; jovian planets around pulsars ( ns ) have not been found , despite very extensive searches ; and super - jupiter planets orbiting bh are a priori unlikely . let us first consider the lens mass distribution : the no - flux - limit ( black ) curve shows a huge spike at expected ns position and a smaller bump corresponding to bh . note that for these bumps , the @xmath137 solution dominates , despite its @xmath61 handicap . this is because the galactic model very strongly favors distant lenses , primarily because of the volume factor , and this overwhelms the modest preference of the light curve for nearby lenses . because @xmath23 is roughly fixed , these distant lenses are massive . this preference is much stronger in the @xmath137 solution , which can be seen in its rapid rise beginning at @xmath185 . note that the wd peak ( at @xmath186 ) is clearly visible , especially in the @xmath174 solution . the lens distance distribution basically looks at this same situation from the standpoint of distance . the new notable feature is that both ms and ns peaks are in the disk , while the bh bump is in the bulge . and the semi - major axis distribution peaks at about 2 - 3 au . > from these diagrams , we can estimate a most probable value of lens mass , distance and semi - major axis , and an asymmetric standard deviation read at 50% of the distribution corresponding to the red bold curves . we get a star and planet mass of @xmath4 and @xmath5 , respectively , at a distance of @xmath6 kpc , and with a semi - major axis of @xmath7 au . as a final note , it could be said that more complex models are worth exploring : the geometry of the caustic crossing , where the source passes close to the three - cusps tail of the caustic , is extremely sensitive to a third body ( second planet or binary companion to the lens star ) . a similar geometry where two planets were detected is described in @xcite . these models could be investigated in a forthcoming paper , once we get the lens flux measurement from adaptive optics . we acknowledge the following sources of support : creative research initiative program ( 2009 - 0081561 ) of national research foundation of korea ( ch ) ; grants jsps20340052 and jsps22403003 for moa ; czech science foundation grant gacr p209/10/1318 ; the french polar institute ( ipev ) and the italian antarctic programme ( pnra ) for the logistics and data transmission at concordia . ogle project has received funding from the european research council under the european community s seventh framework programme ( fp7/2007 - 2013 ) / erc grant agreement no . part of the computer work was performed using hpc resources from calmip ( grant 2011-p1131 ) . e. bachelet gratefully acknowledges the chungbuk national university for a one - month stay where most of this work was prepared . gaudi and a. gould acknowledge support from nsf ast-1103471 . gaudi , a. gould , and r.w . pogge acknowledge support from nasa grant nng04gl51 g . work by j.c . yee is supported by the national science foundation graduate research fellowship under grant no . 2009068160 . t.c . hinse acknowledges support from the krcf young scientist research fellowship program in south korea . the planet collaboration acknowledges the financial support of anr holmes and pnps grants . astep was financed through the help of anr , ipev , cnrs , observatoire de la cte dazur . astep 400 is a 40 cm newtonian telescope installed at the concordia base , located on the dome c plateau in antarctica @xcite . although the aim of the project concerns transiting planets , the ability to observe near - continuously during the antarctic winter and the excellent weather on site @xcite imply that the telescope can usefully complement microlensing observations from other sites , even though the declination of the fields and their crowding make the analysis difficult . the observations with astep 400 started on august 12 , 2010 , 15:27 ut after the first alert was sent by email and phone to concordia . on august 14 , 23:59 ut , the observations were stopped because the magnification had become too small for useful observations . the weather conditions were excellent . however , the seeing conditions were poor ( 3 - 4 " ) , mostly due to the low - declination of the field and location of the telescope on the ground . the 229 mb of data corresponding to 500x500 cropped images were transmitted to nice through satellite connexion around august 17 - 18 for an in - depth analysis . although the data of this run were not good enough for being used in this study , this pioneering test will serve for improving the thermics of the acquisition system and get higher quality future observations of microlensing targets . the pysis3.0 difference image analysis package is fully described in @xcite . it is based on the original isis package @xcite but the kernel used to transform the reference image to the current image before subtraction is no longer analytic , but numerical . this allows dealing with images whose psf can not be assimilated to a sum of gaussian profiles . the numerical kernel has been introduced in image subtraction by @xcite . one of the regular problems encountered in difference image analysis is the choice of the best possible reference image . for astrometry , the best seeing image is generally a good choice , if the sky background is not too high . for photometry , it is our experience that stacking good images improves the result , but only if these images have been acquired during a short time slot , to avoid light variations of the target and slow variations due for instance to small changes in the flat - field . in order to choose good reference images , we use a suite of astromatic software @xcite , namely sextractor and psfex . sextractor builds catalogues of sources from all images of a given telescope , with their characteristics , and psfex derives a model of the psf of these images , from which we extract a few numbers to estimate the image quality ( seeing , ellipticity , number of stars ) . this allows an almost automatic selection of the templates , with a final verification by eye to check the selected images . once the images have been subtracted using these templates , the photometry of the target is done by iteratively centering the psf on the light maximum close to the center of the image . this method enables the photometering of faint stars in the glare of nearby much brighter stars . the standard procedure of rescaling error bars so that the @xmath187 of each telescope data set is of order unity is acceptable if the resulting error bars roughly correspond to the dispersion of the data points at a given time for this data set . there is therefore an interplay between finding the correct model and rescaling the error bars , because a too large rescaling factor reduces the constraint from a given data set and allows the model to shift from the correct one . our procedure has been to use rescaling factors which look plausible given the telescope size and site quality , trying to get a @xmath187 of order unity only if the dispersion in successive data points is well reproduced by this rescaling scheme . this generally involves two parameters to modify the original photometric error bars @xmath188 . one is a minimal error @xmath189 to reproduce the dispersion of very bright ( or in this case , highly magnified ) sources due to fundamental limitations of the photometry , such as flat fielding errors . the other is the classical multiplicative rescaling factor @xmath190 . the adopted formula is : @xmath191 the balance between both rescaling factors is given by comparing the cumulative distribution of @xmath187 , ordered from magnification given by the model , to a standard cumulative distribution for gaussian errors . an example for ogle and moa distributions is given in figure [ fig : cumul ] . table [ tab : two ] gives the adopted values of both parameters for each telescope . it must also be noted that for some amateur telescopes , data were binned before this rescaling process . finally , after the initial modeling was conducted , it was realized that three data sets could not accommodate the condition @xmath192 ( auckland and vlo ) or a positive source flux ( perth ) . they were therefore removed and the final models are based on 16 data sets . it is generally difficult to treat the limb - darkening effect accurately , because it varies from one telescope data set to the other , primarily due to the different photometric bands involved . second - order effects include hardware - specific variations of the spectral response ( filters , ccds ) , and for very broad bands , the atmospheric and interstellar extinctions . we adopt the linear limb - darkening approximation using the formalism given by equation [ eqn : eq4 ] , introduced by @xcite and @xcite . this is a common formalism in the microlensing community , but the more widely used formalism is based on the following equation @xmath193\ ] ] where @xmath194 is the specific intensity at disk center and @xmath195 is the linear limb - darkening coefficient . we compute the values of @xmath195 from a stellar atmosphere model from the kurucz atlas9 grid @xcite using the method described by @xcite . as we converted the @xmath195 values to our ldc parameter @xmath58 , we give the @xmath195 values in table [ tab : two ] , and recall the conversion relation @xmath196 lccccrllc ogle & @xmath89 & 0.4399 & 0.4363 & & 157 & 2.4 & 0.010 & ogle dia + ctio & @xmath89 & 0.4395 & 0.4375 & & 133 & 1.5 & 0.010 & dophot + ctio & @xmath92 & 0.5931 & 0.5908 & & 21 & 2.0 & 0.015 & dophot + auckland & w12 & 0.5016 & 0.4843 & y & 5 & 1 & 0.003 & dophot + fco & unfiltered & 0.5550 & 0.5150 & y & 4 & 1.7 & 0.003 & dophot + fts & @xmath197 & 0.4547 & 0.4524 & & 68 & 3.0 & 0.002 & pysis + kumeu & w12 & 0.5016 & 0.4852 & y & 17 & 3.5 & 0.003 & dophot + perth & @xmath89 & 0.4325 & 0.4297 & & 26 & 3.0 & 0.003 & pysis + ftn & @xmath197 & 0.4547 & 0.4524 & & 72 & 3.8 & 0.006 & dandia + possum & w12 & 0.5204 & 0.5031 & & 77 & 5.4 & 0.003 & pysis + saao & @xmath89 & 0.4261 & 0.4217 & & 2364 & 2.2 & 0.003 & pysis + saao & @xmath92 & 0.6114 & 0.6086 & & 3 & 1.2 & 0.003 & pysis + vlo & unfiltered & 0.5442 & 0.5022 & y & 3 & 1 & 0.003 & dophot + wise & unfiltered & 0.5522 & 0.5131 & y & 9 & 0.8 & 0.003 & dophot + canopus & @xmath89 & 0.4355 & 0.4335 & & 50 & 1.5 & 0.010 & pysis + danish & @xmath89 & 0.4394 & 0.4370 & & 167 & 2.2 & 0.003 & pysis + moa & red & 0.4754 & 0.4694 & & 3985 & 1.05 & 0 & moa dia + lt & @xmath197 & 0.4543 & 0.4513 & & 41 & 4.7 & 0 & dandia + monet n & @xmath89 & 0.4414 & 0.4392 & & 130 & 4.3 & 0.004 & pysis for the source characteristics , we adopt the spectroscopic result ( @xmath198 k , @xmath18 , [ fe / h]=0.0 ) . we thus get an estimate of the limb - darkening effect for each data set . table [ tab : two ] shows the adopted values of @xmath195 for each individual telescope and band combination , both with and without interstellar extinction correction , together with the final number of data points after rejection of possible outliers , and details about the image reduction process . as can be seen , these coefficients vary even if the photometric band is supposedly the same . see for instance the cousins @xmath89-band and sdss @xmath197-band , or the wratten # 12 filter ( w12 ) used at some amateur telescopes to mimic an @xmath199 band . more details about the derivation of these values for each telescope can be found in @xcite . we have also computed the effect of interstellar extinction on the ldcs , as it is not a priori negligible : we typically find differences of a few @xmath200 for a data set with a filter and a few @xmath201 for unfiltered ones . the magnitude of the effect could then be neglected for filtered data sets , but not for unfiltered ones or very broad band filters . to give an idea , we compare its effect to the uncertainty of our spectroscopic determination of the effective temperature of the source , about 150 k. for the ogle @xmath89-band , the difference between ldcs with and without interstellar extinction correction for an adopted extinction of 1.2 mag corresponds to a shift in effective temperature of 40 k , while for the possum broad w12 filter , it corresponds to about 200 k , and for an unfiltered data set , it gives a shift of about 500 k. figure [ fig : anal ] shows several quantities plotted against @xmath202 . in each case the mean and standard deviation of all chain links within a given @xmath202 bin are calculated . the bottom panel shows the behavior of @xmath65 ( @xmath203 ) . since @xmath202 is very similar to @xmath125 , this reflects the degeneracy between the perpendicular components of earth orbital motion ( parallax ) and lens orbital motion , which is analyzed in some detail by @xcite and @xcite . in the present case , the correlation is quite tight . the top panel shows the lens mass @xmath204 . since @xmath23 is nearly constant for different links in the chain , the mass scales @xmath205 and so is peaked near @xmath206 where @xmath38 is near its minimum ( see figure [ fig : bayes1 ] ) . note that @xmath210 is the ratio of the so - called projected kinetic to projected potential energy , which has a strict upper limit of unity for bound orbits . we do enforce this limit , but the main impact of the orbital motion is more subtle . first look at @xmath208 . in the region away from @xmath211 , we have both @xmath212 ( equation [ eqn : gamma0 ] ) and ( very roughly ) @xmath213 ( bottom panel ) . hence , @xmath208 is approximately constant in these two regimes . much of the @xmath214 region is at or above the physical limit @xmath215 , which is why these values are disfavored in figure [ fig : bayes1](c ) . now examine @xmath207 . by itself , @xmath216 does not vary much with @xmath202 , so the form of the structure is basically just @xmath217 , which scales @xmath218 away from zero . the point is , however , that the overall scale ( which is set by the measurement of @xmath216 ) is small , so that except near @xmath211 , @xmath207 is extremely close to zero . naively , this would seem to be disfavored by the virial theorem , but how does the jacobian `` know '' about this ? for simplicity of exposition let us consider circular orbits . for these , @xmath219 implies that the planet is exactly in the plane of the sky : either the orbit is exactly face - on ( so this is always true ) , or the orbit just happens to be passing through the plane of the sky at the time of the event peak . the jacobian is `` unhappy '' about either alternative because there is very little kepler - parameter space relative to chain - variable space at such orbital configurations . finally , note that in the immediate neighborhood of @xmath211 ( where @xmath27 reaches its highest values ) , @xmath220 is frequently at or above its physical limit ( @xmath221 ) , which is exacerbated by the relatively high values of @xmath216 . this is reponsible for the modest suppression of extremely low parallaxes ( high masses ) in figure [ fig : bayes1 ] . thus , both kepler and galactic+flux priors separately predict a relatively high lens mass , near the limit of what is permitted by the flux constraint .

in two - tier networks comprising a conventional cellular network overlaid with shorter range hotspots ( e.g. femtocells , distributed antennas , or wired relays ) with universal frequency reuse , the near - far effect from cross - tier interference creates dead spots where reliable coverage can not be guaranteed to users in either tier . equipping the macrocell and femtocells with multiple antennas enhances robustness against the near - far problem . this work derives the maximum number of simultaneously transmitting multiple antenna femtocells meeting a per - tier outage probability constraint . coverage dead zones are presented wherein cross - tier interference bottlenecks cellular and femtocell coverage . two operating regimes are shown namely 1 ) a cellular - limited regime in which femtocell users experience unacceptable cross - tier interference and 2 ) a hotspot - limited regime wherein both femtocell users and cellular users are limited by hotspot interference . our analysis accounts for the per - tier transmit powers , the number of transmit antennas ( single antenna transmission being a special case ) and terrestrial propagation such as the rayleigh fading and the path loss exponents . single - user ( su ) multiple antenna transmission at each tier is shown to provide significantly superior coverage and spatial reuse relative to multiuser ( mu ) transmission . we propose a decentralized carrier - sensing approach to regulate femtocell transmission powers based on their location . considering a worst - case cell - edge location , simulations using typical path loss scenarios show that our interference management strategy provides reliable cellular coverage with about femtocells per cellsite .

0902.3210

wireless operators are in the process of augmenting the macrocell network with supplemental infrastructure such as microcells , distributed antennas and relays . an alternative with lower upfront costs is to improve indoor coverage and capacity using the concept of _ end - consumer _ installed femtocells or home base stations@xcite . a femtocell is a low power , short range ( @xmath1 meters ) wireless data access point ( ap ) , functioning in service provider owned licensed spectrum , which provides in - building coverage to home users and transports the user traffic over internet based backhaul such as cable modem . because of the proximity of users to their aps , femtocells provide higher spatial reuse of spectrum and cause less interference to other users . the spatial reuse ( in @xmath2 ) is readily expressible by the area spectral efficiency ( ase ) @xcite which is a measure of the total obtainable network throughput per unit hz per unit area . previous studies@xcite have shown a nearly @xmath3x improvement in overall spatial reuse when moving from a macrocell - only network to a two - tier underlay with @xmath4 femtocells per cellsite . in addition to improved spatial reuse , cellular operators desire to operate both cellular and indoor femtocell users in the same bandwidth ( termed _ universal frequency reuse _ ) , as is assumed in this paper , for cost effectiveness and flexible deployment @xcite . with shared spectrum , practical challenges stem from the absence of coordination across tiers @xcite due to scalability issues . because femtocells are consumer deployed in their self - interest and because of reasons of security and limited backhaul capacity , these femtocells will potentially offer privileged coverage only to licensed , subscribed indoor users . this paper assumes _ closed access _ ( ca ) , which means only licensed home users within radio range can communicate with their own femtocell . the drawback of such a co - channel closed access deployment of femtocells is that cross - tier interference becomes the capacity - limiting factor . for example , a cellular user located at the edge of their macrocell may experience unacceptable interference from an actively transmitting femtocell in vicinity . commercial femtocell offerings ( such as verizon s `` home network expander '' ) provide both public access and closed access operation which are user configurable . our results provide a networkwide performance benchmark in a closed access setting . the motivation behind this paper is to understand how exploiting the available degrees of freedom through multiple antenna transmission influences coverage and spatial reuse in a two - tier network with universal frequency reuse . we consider both single - user ( su ) multiple antenna transmission and multiuser ( mu ) multiple antenna transmission ( fig . [ fig : twotiermutransmission ] ) employed by the macrocell basestation ( bs ) and femtocell aps . array gain resulting from su transmission provides robustness against cross - tier interference . multiuser transmission , on the other hand , increases the number of simultaneous transmissions at the expense of reduced signal strength per user terminal and potential inter - user interference . given a multiple antenna transmission strategy ( su or mu ) , let @xmath5 denote the maximum density ( in femtocells per square meter ) of simultaneously transmitting femtocells denoted as the _ maximum femtocell contention density _ that guarantees a certain minimum per - tier quality of service ( qos ) requirement . given a certain minimum per - tier target signal - to - interference ratio ( sir ) equaling @xmath6 , the qos requirement stipulates that the instantaneous sir at each user should exceed @xmath6 with a probability of at least @xmath7 , where @xmath8 is a design parameter . since the signal power for a cellular user decays as @xmath9 ( @xmath10 being the distance from the macrocell bs and @xmath11 is the outdoor path loss exponent ) , satisfying its qos requirement requires @xmath12 to be a monotonically decreasing function of @xmath10 . conversely , satisfying the qos requirement for a femtocell user at @xmath10 necessitates @xmath12 to be monotonically decreasing as @xmath13 . this paper characterizes near - far effects and the resulting per - tier coverage by defining two quantities of interest namely the _ no - coverage femtocell radius _ and the _ cellular coverage radius_. the no - coverage femtocell radius @xmath14 determines the minimum sir feasible femtocell distance from the macrocell ( see fig . [ fig : femtomacro_coverageradii ] ) . any femtocell user within @xmath15 meters from the macrocell experiences an outage probability greater than @xmath8 due to excessive cellular interference . this suggests that any user at @xmath16 should communicate with the macrocell because of its potentially higher cellular sir . the cellular coverage radius @xmath17 denotes the maximum sir feasible distance from the macrocell up to which a cellular user can satisfy its outage probability constraint in the presence of hotspot interference . since there is no coordination between tiers for managing interference , providing greater spatial reuse using femtocells trades off the coverage radii and vice - versa . because the cellular network serves as the primary network to mobile outdoor users , it is desirable to maximize @xmath17 in the presence of hotspot interference . assuming that each tier employs either transmit beamforming ( bf ) [ for su transmission ] or linear zero - forcing precoding [ for mu transmission ] with transmission powers @xmath18 and @xmath19 in each resource ( eg . frequency sub - band ) , this work poses the following questions : * what is the maximum femtocell contention density @xmath12 as a function of the location @xmath10 with respect to ( w.r.t ) the macrocell bs , the ratio @xmath20 , the transmission strategy ( su vs. mu transmission ) , the number of transmit antennas per macrocell bs and femtocell ap , the target per - tier sir @xmath6 , the maximum outage probability @xmath8 and the path loss exponents ? * given an average of @xmath21 transmitting femtocells per cell - site , how much cellular coverage can the macrocell bs provide to its users ? * how does the no - coverage femtocell radius vary with su and mu transmission strategies ? * how should femtocells adapt their transmission power for ensuring reliable cellular coverage ? prior research in tiered networks have mainly considered an operator planned underlay of a macrocell with single / multiple microcells@xcite . a microcell has a much larger radio range ( 100 - 500 m ) than a femtocell , and generally implies centralized deployment , i.e. by the service - provider . this allows the operator to either load balance users or preferentially assign high data rate cellular users to the microcell @xcite because of its inherently larger capacity . in contrast , femtocells are consumer installed and the traffic requirements at femtocells are user determined without any operator influence . consequently , decentralized strategies for interference management may be preferred@xcite . the subject of this work is related to huang _ _ et al.__@xcite which derives per - tier transmission capacities with spectrum underlay and spectrum overlay . in contrast to their work which assumes relay - assisted cell - edge users , our work proposes to improve coverage by regulating femtocell transmit powers . et al.__@xcite have derived transmission capacities in an _ ad hoc _ network with spatial diversity . our work has extended this analysis to a cellular - underlaid _ ad hoc _ network . finally , related works on cognitive radios ( cr ) include ( but not restricted to ) analyzing sensing - throughput tradeoffs@xcite for computing optimal sensing time by cr users and limit transmit powers of cr users@xcite . the differentiating aspect of our work is a decentralized femtocell transmit power selection scheme which ensures a per - tier outage probability below a desired threshold . given @xmath22 antennas at the macrocell bs and @xmath23 antennas per femtocell ap , a maximum tolerable per - tier outage probability @xmath8 and path loss exponents @xmath11 ( outdoor cellular transmission ) and @xmath24 ( during indoor - to - outdoor femtocell transmission ) respectively , this work provides the following contributions . we derive coverage zones wherein cross - tier interference prevents users in each tier from satisfying their qos requirements . single - user macrocell transmission is shown to increase the cellular coverage radius by a factor @xmath25 relative to mu transmission . single - user femtocell transmission is shown to decrease the no - coverage femtocell radius @xmath14 by a factor of @xmath26 relative to mu femtocell transmission . this suggests that su transmission results in superior coverage in either tier . we also show that femtocell performance is regulated by cellular interference and hotspot interference is negligible in comparison . we derive the maximum femtocell contention density @xmath27 at distance @xmath10 from the macrocell bs . two distinct operating regimes are shown namely a cellular - limited regime , wherein femtocell users are primarily affected by cellular interference and hotspot - limited regime wherein both cellular and hotspot users are affected by hotspot interference . regime 1 determines the coverage provided to femtocell users , while regime 2 determines @xmath27 and the cellular coverage radius . in regime 2 , su macrocell transmission is shown to increase @xmath27 by a factor of @xmath28 ( where @xmath29 is the gamma function ) relative to mu transmission . femtocells maximize their area spectral efficiency by choosing their transmission strategy depending on @xmath24 , with su transmission being desirable with considerable hotspot interference ( @xmath30 ) . this suggests that per - tier su transmission is preferable from a spatial reuse perspective . we propose a carrier - sensing approach in which a femtocell chooses its transmit power depending on its distance from the macrocell bs for minimizing cross - tier interference . this strategy provides reliable cellular coverage with up to 60 femtocells per cell site ( with typical cellular parameters ) . assume a central macrocell @xmath31 using @xmath22 antennas to service a geographical region @xmath32 , assumed as a circular disc with radius @xmath33 and area @xmath34 . each femtocell is equipped with @xmath23 antennas . femtocell users are located on the circumference of a disc of radius @xmath35 centered at their femtocell ap . both cellular users and femtocell users are assumed to be equipped with single - antenna receivers . in a given time / frequency slot , each macrocell [ resp . femtocell ] employs its @xmath22 [ resp . @xmath23 ] antennas for serving @xmath37 cellular [ resp . @xmath38 indoor ] users . although user selection has a potentially beneficial impact , it is not considered in this work for analytical tractability . this paper employs a stochastic geometry framework for modeling the random spatial distribution of the underlying femtocells . hotspot locations are likely to vary from one cell site to another , and be opportunistic rather than planned , so an analysis that embraces instead of neglecting randomness should provide more accurate results and more plausible insights . the randomly located femtocells are assumed to be distributed according to a spatial poisson point process ( sppp ) @xmath39 ( see @xcite for background , prior works include @xcite ) . provided @xmath39 is a homogeneous sppp ( or the intensity @xmath12 in femtocells per square meter stays constant over @xmath32 ) , the average number of actively transmitting femtocells is readily obtained as @xmath40 femtocells per cellsite . because of near - far effects inherent to a two - tier network , the maximum hotspot intensity varies with the location @xmath10 in the cell - site . the signal decay encountered using terrestrial propagation to users in either tier is represented using a distance based path loss model . temporal amplitude variations of the complex vector downlink channel are modeled as frequency - flat rayleigh fading e.g. each frequency sub - band in frequency division multiple access ( fdma ) transmission with individual complex entries distributed as @xmath41 . for analytical simplicity , this work does not consider random lognormal shadow fading . instead , we shall assume a fixed partition loss encountered during outdoor - to - indoor and indoor - to - indoor wireless propagation . shown below , we use the imt-2000 channel model ( * ? ? ? * page 26 ) for modeling indoor losses ( based on the cost231 indoor model ( * ? ? ? * page 44 ) ) and outdoor path losses . the decibel path loss between @xmath31 and cellular user @xmath42 is modeled as @xmath43 where @xmath11 is the outdoor path loss exponent , @xmath44 represents the fixed decibel loss during outdoor propagation , @xmath45 is the carrier frequency in mhz and @xmath10 is the distance between @xmath31 and its user . ( with path loss @xmath46 ) . to keep the analysis general , this work parameterizes the outdoor path loss exponent . ] we model each femtocell as a point object , hence all indoor users served by a given femtocell experience identical path loss from cellular interference . this decibel path loss is given as @xmath47 where @xmath48 designates the fixed decibel path loss , @xmath10 is the distance between @xmath31 and the femtocell and @xmath49 equals the decibel wall partition loss during outdoor - to - indoor wireless propagation . the decibel path loss between a femtocell to its licensed , subscribed indoor users is modeled as @xmath50 where @xmath51 db models the fixed propagation loss in decibels between the femtocell to its desired user , @xmath52 represents the indoor path loss exponent . given a transmitting femtocell , any cellular user located at distance @xmath10 will experience cross - tier interference with decibel path loss modeled as @xmath53 . here , the fixed decibel path loss is designated by @xmath54 , while @xmath24 denotes the path loss exponent during indoor - to - outdoor wireless propagation . the decibel path loss of the hotspot interference caused by a transmitting femtocell at another femtocell is given as @xmath55 where @xmath56 denotes the fixed decibel path loss ( the factor of @xmath57 models the double wall partition loss during indoor to indoor propagation ) and @xmath10 is the distance between the two femtocells . assume that the macrocell @xmath31 serves @xmath37 users . define @xmath58 as the channel from @xmath31 to cellular user @xmath59 with its entries distributed as @xmath60 . the direction of each vector channel is represented as @xmath61 . designate @xmath62^{\dagger } \in \mathbb{c}^{u_c \times t_c}$ ] as the concatenated matrix of channel directions , where the symbol @xmath63 denotes conjugate transpose . [ as : as1 ] perfect channel state information ( csi ) is assumed at the central macrocell [ resp . femtocells ] regarding the channels to their own users . although we acknowledge that imperfect channel estimation plays a potentially significant role , we defer its analysis for subsequent research and instead employ as[as : as1 ] for analytical tractability . [ as : as2 ] for analytical tractability , interference from neighboring macrocell bss is ignored . this work assumes linear zero - forcing ( zf ) precoding transmission because it has low complexity , yet achieves the same multiplexing gain as higher complexity schemes such as dirty - paper coding . with zf precoding transmission , macrocell bs @xmath31 chooses its precoding matrix @xmath64_{1 \leq i \leq u_c}$ ] as the normalized columns of the pseudoinverse @xmath65 . similarly , each femtocell @xmath66 serves @xmath38 users with the channel directions between @xmath67 to its individual users represented as @xmath68^{\dagger } \in \mathbb{c}^{u_f \times t_f}$ ] , where @xmath69 with the entries of @xmath70 distributed as @xmath41 . with zf precoding , the columns of the precoding matrix @xmath71_{1 \leq i \leq u_f } \in \mathbb{c}^{t_f \times u_f}$ ] equal the normalized columns of @xmath72 . consider a reference femtocell @xmath73 at distance @xmath10 from the macrocell @xmath31 . during a given signaling interval , the received signal at femtocell user @xmath42 at distance @xmath35 w.r.t @xmath73 is given as @xmath74 where the vectors @xmath75 and @xmath76 designate the transmit data symbols for users in @xmath31 and @xmath67 , which satisfy @xmath77 \leq p_c$ ] and @xmath78 \leq p_f$ ] respectively ( assuming equal power allocation ) and @xmath79 represents background noise . the term @xmath80 [ resp . @xmath81 designates the downlink vector channel from the interfering macrocell bs @xmath31 [ resp . interfering femtocell ap @xmath67 ] to user @xmath42 . neglecting receiver noise for analytical simplicity , the received sir for user @xmath42 is given as @xmath83 for successfully decoding the message intended for user @xmath42 , @xmath84 should be greater than equal to the minimum sir target @xmath6 . for clarity of exposition , we define @xmath85 user @xmath42 can successfully decode its signal provided @xmath84 is at least equal to its minimum sir target @xmath6 . combining and , the probability of successful reception is given as @xmath86 } = \mathbb{p}{\left [ |\mathbf{g}_0^{\dagger } \mathbf{w}_{0,0}|^2 \geq \gamma \mathcal{q}_f \left(\frac{\mathcal{p}_f}{u_c}||\mathbf{f}_{0}^{\dagger}\mathbf{v}||^2 + \frac{1}{u_f}\sum_{j \in \pi_f } ||\mathbf{g}_{0,j}^{\dagger}\mathbf{w}_j||^2 |x_{0,j}|^{-\alpha_{fo}}\right ) \right]}.\end{aligned}\ ] ] @xmath87 note that @xmath88 and the expression @xmath89 characterizes the relative strength of cellular interference . as @xmath90 increases ( or @xmath91 ) , user @xmath42 experiences progressively poor coverage due higher cellular interference . conversely , as @xmath92 , @xmath84 is limited by interference from neighboring femtocells . for satisfying the femtocell qos requirement , @xmath93 } \geq 1-\epsilon$ ] in equation . because of cellular interference , any femtocell user within @xmath94 meters of @xmath31 ( @xmath14 being the no - coverage femtocell radius ) can not satisfy their qos requirement . as long as @xmath95 , a femtocell user can tolerate interference from _ both _ cellular transmissions and hotspot transmissions . before computing @xmath14 , we provide the following definitions . [ defn : betacdf ] given a beta distributed random variable @xmath96 with two positive shape parameters @xmath97 and @xmath98 , denote its cumulative distribution function ( cdf ) @xmath99 $ ] namely the regularized incomplete beta function as @xmath100 . [ defn : invbetacdf ] given a beta distributed random variable @xmath96 with cdf @xmath101 , denote its inverse cdf @xmath102 as that value of @xmath103 for which @xmath104 . [ th : nocoveragefemtodist ] _ any femtocell @xmath73 within @xmath105 meters of the macrocell @xmath31 can not satisfy its qos requirement @xmath8 , where @xmath14 is given as _ @xmath106^{-1/\alpha_c}\textrm { , where } k \triangleq \frac{a_{fi}}{a_{f , c}}r_f^{-\alpha_{fi}}.\end{aligned}\ ] ] refer to appendix [ pf : theorem1 ] . [ pr : incompletebetamonotonicity ] _ the inverse function @xmath107 is monotonically increasing with @xmath97 and monotonically decreasing with @xmath98 for any @xmath108 . _ we use the following two expansions ( * ? ? ? * page 29 ) for @xmath100 . @xmath109 where the gamma function @xmath110 for any positive integer @xmath111 . relation ( a ) shows that @xmath112 monotonically decreases with @xmath97 . the equivalent relation ( b ) shows that @xmath100 is monotone increasing w.r.t @xmath98 . consequently , the inverse function @xmath107 monotonically increases with @xmath97 and monotonically decreases with @xmath98 . with su femtocell transmission ( @xmath113 ) , the no - coverage radius @xmath114 is strictly smaller than the no - coverage radius @xmath115 with mu transmission ( @xmath116 ) . this follows by applying proposition [ pr : incompletebetamonotonicity ] to in theorem [ th : nocoveragefemtodist ] . [ co : nocoveragefemtodist ] _ with @xmath117 as defined in theorem [ th : nocoveragefemtodist ] and @xmath118 , the reduction in the no - coverage radius using a su transmission strategy at femtocells relative to mu transmission to @xmath119 users [ resp . single antenna transmission ] is given as _ @xmath120^{1/\alpha_c } \approx \left [ \frac{\epsilon^{1 - 1/t_f}}{t_f } \right]^{1/\alpha_c}. \notag \\ \frac{d_{f,\textrm{su}}}{d_{f,1 \textrm { antenna } } } & = \left [ \left ( \frac{1-\epsilon^{1/t_f}}{\epsilon^{1/t_f } } \right ) \frac{\epsilon}{1-\epsilon}\right]^{1/\alpha_c } \approx \epsilon^{\frac{1}{\alpha_c}\left ( 1 - 1/t_f \right)}. \notag\end{aligned}\ ] ] with su femtocell transmission [ resp . mu transmission to @xmath119 users ] and @xmath118 user , the incomplete beta function @xmath121 simplifies as @xmath122 where @xmath123 and @xmath124 respectively . therefore , the no - coverage distances in are respectively given as @xmath125^{-1/\alpha_c } , \ d_{f,\textrm{mu } } = \left[\frac{k}{\gamma } \frac{p_f / t_f}{p_c } \left(\frac{\epsilon}{1-\epsilon}\right)\right]^{-1/\alpha_c}.\end{aligned}\ ] ] assuming small @xmath8 , su femtocell transmission consequently reduces @xmath14 by a factor of approximately @xmath126 relative to mu transmission . a similar argument shows that su transmission reduces @xmath14 by a factor of approximately @xmath127 relative to single antenna transmission . [ re : re1 ] for fixed @xmath23 , @xmath128 and @xmath129 , the no - coverage femtocell radius @xmath14 in scales as @xmath130 . decreasing @xmath14 by a factor of @xmath111 requires increasing @xmath19 by @xmath131 decibels . this suggests that a graph of @xmath14 versus @xmath132 ( see fig . [ fig : femtosdma_vs_subf ] ) is a straight line on a log - log scale with slope @xmath133 . next , we derive the maximum obtainable spatial reuse from multiple antenna femtocells when they share spectrum with cellular transmissions . mathematically , the _ maximum femtocell contention density _ satisfying is expressed as @xmath134 [ th : optimalfemtodensity ] _ in a two - tier network , the maximum femtocell contention density @xmath27 at distance @xmath10 from the macrocell @xmath31 , which satisfies ( in the small-@xmath8 regime ) is given as @xmath135\end{aligned}\ ] ] where @xmath136 , @xmath137 is given by , @xmath90 is given by , and @xmath138^{-1}\end{gathered}\ ] ] where @xmath139 denotes the beta function and @xmath140 whenever @xmath119 . _ refer to appendix [ pf : theorem2 ] theorem [ th : optimalfemtodensity ] provides the maximum femtocell contention density at @xmath10 considering both cross - tier cellular and hotspot interference from neighboring femtocells . alternatively , given an average of @xmath12 transmitting femtocells per square meter , can be inverted ( numerically ) to obtain the minimum @xmath10 which guarantees that is feasible . theorem [ th : optimalfemtodensity ] provides two fundamental operational regimes depending on the hotspot location relative to the macrocell . assuming @xmath141 , a necessary condition for @xmath142 in is @xmath143 , or @xmath144 in is upper bounded as @xmath145 indeed , from theorem [ th : nocoveragefemtodist ] , a femtocell can not guarantee reliable coverage to its users because of excessive cross - tier interference , whenever the above condition is violated . as @xmath92 or @xmath146 , the sir at any femtocell located at @xmath10 is primarily influenced by hotspot interference . consequently , @xmath147 in approaches the limit @xmath148 given as @xmath149 the limit @xmath150 determines the maximum contention density in the special case of an _ ad hoc _ network no cellular interference of homogeneously distributed transmitters equipped with multiple antennas @xcite . their work shows that @xmath150 and @xmath151 scales with @xmath23 and @xmath128 as @xmath152 , \ \mathcal{c}_f\mathcal{q}_f^{\delta_f } \sim \theta(u_f^{\delta_f } ) . \end{aligned}\ ] ] further , @xmath153 and @xmath154 is bounded as @xmath155 @xcite . we shall now consider two cases in the hotspot - limited regime ( @xmath146 ) . first , with multiuser transmission to @xmath119 femtocell users and using , the femtocell area spectral efficiency ( in @xmath2 ) which is given as @xmath156 scales according to @xmath157 . with su transmission , the ase scales as @xmath158 . this suggests that in path loss regimes with @xmath30 , higher spatial reuse is obtainable ( order - wise ) provided femtocells employ their antennas to transmit to just one user . in contrast , mu femtocell transmission provides higher network - wide spatial reuse ( order - wise ) only when hotspot interference is significantly diminished ( @xmath159 ) . we now consider a reference cellular user @xmath42 at distance @xmath10 from their macrocell @xmath31 . during a given signaling interval , neglecting background noise , the received signal at user @xmath42 is then given as @xmath160 where @xmath161 \leq p_c$ ] and @xmath162 \leq p_f$ ] represent the transmit data symbols for users in each tier . further , @xmath163 and @xmath164 respectively denote the distance and the downlink vector channel from the interfering femtocell @xmath67 to user @xmath42 . the received sir for user @xmath42 is given as @xmath165 for successfully decoding user @xmath42 s signal , @xmath166 should be greater than equal to the minimum sir target @xmath6 . define @xmath167 . then , the probability of successful reception at @xmath42 is given as @xmath168 } = \mathbb{p}{\left [ \frac{|\mathbf{h}_0^{\dagger}\mathbf{v}_0|^2}{\frac{1}{u_f}\sum_{f_j \in \pi_f } ||\mathbf{e}_{j}^{\dagger}\mathbf{w}_j||^2 |x_{j}|^{-\alpha_{fo } } } \geq \mathcal{q}_c \gamma \right]}.\end{aligned}\ ] ] both the desired channel powers denoted as @xmath169 and the interfering marks @xcite given by @xmath170 follow a chi - squared distribution with @xmath171 and @xmath172 degrees of freedom respectively . using @xcite , the maximum femtocell contention density @xmath173 for which satisfies the maximum outage probability constraint @xmath174 of a cellular user is given as @xmath175 where @xmath136 as before and @xmath151 is given by . from @xcite , @xmath176 is bounded as @xmath177 where the upper bound is a good approximation for @xmath176 ; for example , with @xmath178 , @xmath118 and @xmath179 , the term @xmath180 equals @xmath181 while the upper bound equals @xmath182 . [ re : lambdafscalingcellular ] since varies as @xmath183 , approximating @xmath176 by the upper bound in shows that the maximum contention density for single user beamforming given as @xmath184 is proportional to @xmath185 . with @xmath186 transmitted users , the maximum femtocell contention density denoted as @xmath187 is proportional to @xmath188 therefore , su transmission increases the maximum hotspot density by a factor of @xmath189^{\delta_f}$ ] . with @xmath190 users ( implying @xmath191 ) , one obtains @xmath187 to be proportional to @xmath192 , so that @xmath193 equals @xmath194 . given an average of @xmath12 femtocells per sq . meter , inverting yields the maximum distance up to which the cellular outage probability lies below @xmath8 . this cellular coverage radius @xmath17 is given as @xmath195 since @xmath17 varies as @xmath196 , increasing the cellular coverage radius by a factor of @xmath111 necessitates increasing @xmath18 by @xmath197 decibels relative to @xmath19 . [ re : cellularcoveragescaling ] in , @xmath17 is proportional to @xmath198 . with su transmission [ resp . mu transmission to @xmath190 users ] at the macrocell and applying , the cellular coverage distance @xmath17 scales with @xmath22 as @xmath199 . this suggests that su macrocell transmission provides coverage improvement by a factor of @xmath200 ( order - wise ) relative to mu transmission . in this section , we provide design interpretations of the preceding results derived in sections [ se : sirfemtoanlys ] and [ se : sircellranlys ] in realistic path loss scenarios . we shall use the system parameters given in table [ tbl : sysprms ] and the path loss model described in section [ sec : channelmodel ] . [ fig : femtosdma_vs_subf ] plots the normalized no - coverage femtocell radius @xmath14 versus @xmath132 on a log - log scale . assuming a reference femtocell placed at distance @xmath10 w.r.t the macrocell bs , figs . [ fig : femtodensityvsdistance]-[fig : femtodensityvspathlossexponent ] plot the maximum number of simultaneous femtocell transmissions given as @xmath201 where @xmath202 , considering su and mu femtocell transmissions and different values of @xmath24 . shown below are the three key observations . in the cellular - limited regime with @xmath203 antennas , fig . [ fig : femtodensityvsdistance ] shows that su transmission obtains a nearly @xmath204x reduction in the no - coverage femtocell radius @xmath14 w.r.t single antenna transmission . next , both figs . [ fig : femtosdma_vs_subf]-[fig : femtodensityvsdistance ] show that su transmission reduces @xmath14 by a factor of nearly @xmath205x relative to mu transmission . both these observations agree with the predicted improvements in corollary [ co : nocoveragefemtodist ] . this indicates that su transmission significantly improves hotspot coverage . in figs . [ fig : femtodensityvsdistance]-[fig : femtodensityvspathlossexponent ] , @xmath21 increases from zero ( at the no - coverage femtocell radius ) to greater than @xmath206 femtocells per cell - site within a few meters outside the no - coverage radius . this step - like transition from the cellular - limited to the hot - spot limited regime suggests that cross - tier cellular interference is the capacity - limiting factor even in densely populated femtocell networks and interference between femtocells is negligible because of the proximity of home users to their aps and double wall partition losses . in the hotspot - limited regime with @xmath179 , su transmission consistently outperforms mu transmission . for example , with @xmath203 antennas , there is a nearly @xmath207x spatial reuse gain ( @xmath208 with su transmission versus @xmath209 with mu transmission ) . in a scenario in which hotspot interference is _ significantly diminished _ [ fig : femtodensityvspathlossexponent ] with @xmath210 and @xmath211 antennas ) , mu transmission to @xmath212 hotspot users provides a marginally higher spatial reuse relative to su transmission . the conclusion is that achieving the multiplexing benefits of mu transmission requires relative isolation ( or large @xmath24 ) between actively transmitting femtocell aps . [ fig : femtocelltransmissions_cellularref ] plots the maximum number of transmitting femtocells @xmath213 as a function of the cellular user distance @xmath10 . with @xmath214 and a desired @xmath215 femtocells / cellsite , su macrocell transmission provides a normalized cellular coverage radius @xmath216 . in contrast , the coverage provided by mu transmission is only @xmath217 , resulting in a coverage loss of @xmath218x relative to su transmission ( remark [ re : cellularcoveragescaling ] gives an order - wise loss of @xmath219 ) . with su transmission and @xmath220 , a cellular user at @xmath221 can tolerate interference from nearly @xmath222 femtocells / cellsite . in contrast , with mu transmission , @xmath21 reduces to nearly @xmath223 femtocells / cellsite . the observed improvement in the maximum femtocell contention density ( equaling @xmath224x ) is well approximated by the predicted improvement ( @xmath225x ) in remark [ re : lambdafscalingcellular ] . the preceding observations reveal since mu performance is significantly limited by residual hotspot interference , the macrocell should maximize cellular coverage by transmitting to just a single cellular user and that femtocells should adapt their transmit powers depending on their location in order to ensure reliable cellular coverage . to motivate carrier - sensing at femtocells , fig . [ fig : femtodensityvsdistance ] shows that even with @xmath226 , a femtocell at normalized distance @xmath227 can tolerate hotspot interference from greater than @xmath228 neighboring femtocells . this suggests that in dense femtocell deployments , @xmath229 can be increased to minimize hotspot interference without violating the qos requirement at femtocells . this section presents a carrier - sensing interference management strategy for choosing the femtocell transmission power provided there is a cellular user in its vicinity . [ as : as3 ] each cellular user periodically transmits over a set of uplink pilot slots ( time or frequency resource ) with power @xmath230 for communicating their channel information to the macrocell . [ as : as4 ] each femtocell is capable of inferring its distance from its closest macrocell bs ( either through gps or measuring the average received power of the macrocell bs transmission , with prior calibration ) . during carrier - sensing , each femtocell performs energy detection while monitoring uplink pilot cellular transmissions . in the absence of a cellular user , the femtocell maintains a constant transmit power @xmath19 . when the detected energy of a cellular user exceeds a threshold , the femtocell chooses its @xmath19 based on its location @xmath10 within the underlying macrocell . we shall first derive the minimum required sensing distance @xmath231 such that any transmitting femtocell located within @xmath232 meters of the cellular user violates its maximum outage probability requirement . define the notation @xmath233 to denote a circular region of radius @xmath231 containing @xmath234 femtocells . given an intensity of @xmath12 femtocells per square meter and assuming the cellular user @xmath42 is located at normalized distance @xmath10 w.r.t the macrocell , its outage probability is lower bounded as @xmath235 & \overset{(a)}\geq \mathbb{p}[\mathrm{sir}_c(b_0,d ) \leq \gamma , |\mathcal{b}(d_{\textrm{sense}})| > 0 ] \\ & \overset{(b)}= \mathbb{p}\left[\mathrm{sir}_c(b_0,d ) \leq \vert |\mathcal{b}(d_{\textrm{sense}})| > 0 \right ] \cdot ( 1-e^{-\lambda_f \pi d_{\textrm{sense}}^2 } ) \notag \\ & \overset{(c ) } > \mathbb{p}\left[\mathrm{sir}_c(b_0,d ) \leq \gamma \big \vert |\mathcal{b}(d_{\textrm{sense}})| = 1 , r = d_{\textrm{sense}}\right ] \cdot(1-e^{-\lambda_f \pi d_{\textrm{sense}}^2 } ) \notag\end{aligned}\ ] ] where step ( a ) in is a lower bound as it ignores the event of zero hotspots present within @xmath233 . step ( b ) rewrites ( a ) in terms of the conditional probability . finally , step ( c ) is a lower bound because it considers the event that @xmath236 and the hotspot is located exactly at @xmath237 meters ( thereby experiencing the highest path loss ) . a necessary condition for ensuring @xmath238 \leq \epsilon$ ] is that the right hand side in step ( c ) in consisting of the product of two probabilities should be less than @xmath8 . the first term represents the outage probability from interfering hotspots due to the time - varying channel powers and the second term represents the probability that @xmath233 is non - empty . assuming large @xmath12 ( or @xmath239 ) , a reasonable choice for selecting @xmath231 is to set the conditional outage probability @xmath240 $ ] , given exactly one interfering femtocell ap @xmath73 at distance @xmath231 from the cellular user , to equal @xmath8 . evaluating this probability , @xmath241 & = \mathbb{p}\left[\frac{|\mathbf{h}_0^{\dagger}\mathbf{v}_0|^2}{||\mathbf{e}_{0}^{\dagger}\mathbf{w}_0||^2 } \leq \frac{\gamma \mathcal{q}_c d_{\textrm{sense}}^{-\alpha_{fo}}}{u_f } \right ] \\ & \overset{(a)}= \mathbb{p}\left[z \leq \theta \frac{u_f}{t_c - u_c + 1 } \right ] \notag \\ & \overset{(b)}=\mathcal{i}_{\frac{\theta}{\theta + 1}}(t_c - u_c + 1 , u_f ) \notag\end{aligned}\ ] ] where @xmath167 as before , while the terms @xmath242 and @xmath243 denote the chi - squared distributed desired and interfering channel powers , as given earlier in . step ( a ) in follows by defining @xmath244 and defining the normalized ratio @xmath245 which is f - distributed @xcite . step ( b ) follows by substituting the cdf of the f - distributed r.v @xmath246 . the minimum required sensing radius at @xmath10 is consequently given as @xmath247^{1/ \alpha_{fo}}.\end{aligned}\ ] ] using the numerical values in table [ tbl : sysprms ] , fig . [ fig : femtosensingrangevsplexponents ] plots @xmath231 for different values of the path loss exponents @xmath11 and @xmath24 as well as different cellular user locations @xmath10 . assuming su transmission in both tiers , @xmath226 db and @xmath248 , a minimum sensing range @xmath249 meters is required at the cell - edge ( @xmath250 ) . next , the following lemma derives bounds on @xmath20 that satisfy the per - tier outage probability requirements at distance @xmath10 from the macrocell . _ given a mean intensity of @xmath12 femtocells per square meter and minimum per - tier sir target @xmath6 , satisfying the per - tier outage probability requirement at distance @xmath10 from the macrocell necessitates @xmath251 to be bounded as @xmath252 \leq p_c / p_f \leq ( p_c / p_f)_{\textrm{ub}}[d]$ ] , which are given as @xmath253 & = \gamma \left(\frac{a_{c , f}}{a_{c}}\right ) \left(\frac{u_c}{d^{-\alpha_c}}\right)\left(\frac{\mathcal{c}_f \lambda_f}{\epsilon \mathcal{k}_c}\right)^{1/\delta_f } \\ \label{eq : pcoverpfub } \left(\frac{p_c}{p_f}\right)_{\textrm{ub}}[d ] & = \left(\frac{1}{\gamma}\right ) \left(\frac{a_{fi}}{a_{f , c}}\right)\left(\frac{u_c r_f^{-\alpha_{fi}}}{u_f d^{-\alpha_c}}\right ) \frac{\mathcal{i}^{-1}(\tilde{\epsilon};t_f - u_f+1,u_c)}{1-\mathcal{i}^{-1}(\tilde{\epsilon};t_f - u_f+1,u_c)}\end{aligned}\ ] ] where @xmath136 as before , @xmath176 is given by , @xmath137 is given by , @xmath151 is given by and @xmath254 _ a lower limit on @xmath20 is obtained by computing the minimum @xmath18 required to satisfy the outage probability requirement for a cellular user at distance @xmath10 w.r.t @xmath31 . combining and yields @xmath255 in . conversely , given a femtocell user at distance @xmath10 w.r.t @xmath31 , an upper limit for @xmath20 is obtained by computing the minimum required @xmath19 for satisfying @xmath256 \leq \epsilon$ ] . substituting the upper bound for @xmath154 and inverting to compute @xmath257 yields . inspecting and reveals that the difference between the decibel upper and lower bounds is constant for all @xmath10 . [ fig : twotier_pcoverpfbounds ] plots @xmath258 and @xmath257 for different normalized @xmath10 . at a cell - edge ( @xmath259 ) location , the bounds on the required @xmath260 are given as @xmath261 db . assume that each femtocell monitors a set of pilot slots ( we assume time - slotted transmission in the subsequent discussion ) and employs energy detection@xcite . we briefly describe the sensing procedure below and refer to @xcite for further details . let @xmath262 denote the sensing time ( number of sensing time slots times the slot duration ) and @xmath263 designate the sensed bandwidth . given a received signal @xmath264 in the pilot slots and @xmath265 being complex gaussian noise process with power @xmath266 per complex dimension , define the following hypotheses namely @xmath267 : absence of cellular user [ @xmath268 and @xmath269 : presence of an active cellular user [ @xmath270 . the femtocell compares the energy detector output @xmath271 against a threshold @xmath272 for inferring the presence ( or absence ) of a cellular user . define @xmath273 to equal the time - bandwidth product @xmath274 ( assumed to be an integer ) . the average sensed pilot signal - to - noise ratio ( snr ) at the femtocell is given as @xmath275 where @xmath10 denotes the distance of the cellular user from the femtocell . the noise power @xmath276 is chosen with reference to a cell - edge user obtaining an average downlink snr @xmath277 . assuming selection combining ( sc ) is used at the @xmath23 available diversity branches for choosing the maximum snr branch , the detection probability @xmath278 and the false - alarm probability @xmath279 are respectively given as @xcite @xmath280 & = t_f \cdot \sum_{i=0}^{t_f-1 } \frac{(-1)^{i}}{i+1 } \binom{t_f-1}{i}p_{d,\textrm{ray}}\left(\frac{\overline{\gamma}}{i + 1}\right ) , p_{\textrm{false } } = \mathbb{p}[y > \lambda | \mathcal{h}_0 ] = \frac{\mathbf{\gamma}(2m,\lambda)}{\mathbf{\gamma}(2 m ) } \notag \\ \textrm{where } p_{\textrm{detect , ray}}(\overline{\gamma})&=\frac{\mathbf{\gamma}(2m-1,\lambda)}{\mathbf{\gamma}(2m-1)}+e^{-\frac{\lambda}{(1+m\overline{\gamma})}}\left(1+\frac{1}{m\overline{\gamma}}\right)^{2m-1 } \left[1-\frac{\mathbf{\gamma}(2m-1,\frac{\lambda m \overline{\gamma}}{(1+\lambda m \overline{\gamma})})}{\mathbf{\gamma}(2m-1)}\right].\end{aligned}\ ] ] here , @xmath281 is the upper incomplete gamma function . because of the complex baseband signal model , there is a factor of @xmath57 discrepancy in with respect to @xcite . [ fig : femtosensingrangevstw ] plots the maximum femtocell sensing range @xmath231 versus different values of the time - bandwidth product @xmath273 . for example , with @xmath282 dbm ( @xmath283 db below the maximum ut transmit power ) , and probabilities @xmath284 and @xmath285 respectively , obtaining a sensing range of @xmath286 meters requires a minimum time - bandwidth product @xmath287 . this section reports the results of computer simulations using the system parameters in table [ tbl : sysprms ] . the simulation consisted of @xmath228 different random drops of femtocell hotspots with @xmath228 trials per drop to simulate rayleigh fading . additive white gaussian noise power was chosen to obtain an average cell - edge snr of @xmath288 db . single - user transmission is assumed in either tier considering its superior coverage and spatial reuse performance . with an average of @xmath215 femtocells per cell - site , we evaluate whether the @xmath289 percentile outage capacity ( @xmath290 ) satisfies the minimum required per - tier spectral efficiency @xmath291 b / s / hz ( or nearly @xmath292 b / s / hz for @xmath293 ) . during carrier - sensing , each femtocell can detect active cellular users within a sensing radius equaling @xmath294 meters ( determined using computer simulations ) , which exceeds the minimum required sensing range of @xmath295 meters obtained in section [ subse : minsensingrange ] . we consider both a fixed @xmath20 ( without carrier - sensing or power control at femtocells ) and a location based selection of @xmath20 ( wherein femocells adjust their @xmath19 upon sensing a cellular user ) . under ambient conditions ( no detected cellular user ) , a fixed @xmath226 db is chosen . upon sensing a cellular user , a femtocell chooses its @xmath19 such that @xmath296 = 0.7(p_c / p_f)_{\textrm{ub}}[d,\textrm{db}]+ 0.3(p_c / p_f)_{\textrm{lb}}[d,\textrm{db}]$ ] , which are given in - . two scenarios are considered namely a cellular user is placed at normalized distances ( @xmath297 and @xmath298 ) w.r.t the macrocell . the cdfs of the achievable cellular data rates have been reported . a reference hotspot is placed at normalized distances ( @xmath299 and @xmath300 respectively ) from the macrocell . a reference cellular user is co - linearly placed at a distance @xmath301 w.r.t the hotspot . the conditional cdfs of the hotspot data rates ( assuming idealized sensing ) have been reported . [ fig : cellularratecdfplots ] shows the cdfs of the obtained cellular data rates for @xmath215 femtocells / cell - site . without carrier - sensing , the @xmath289 percentile outage capacities are below @xmath302 b / s / hz . by employing carrier - sensing , the @xmath289 percentile outage capacities ( corresponding to @xmath290 ) equal @xmath303 b / s / hz and @xmath304 b / s / hz for cellular user locations of @xmath297 and @xmath298 respectively . thus , our scheme ensures uniform cell - edge coverage with large numbers of femtocells . [ fig : femtocellratecdfplots ] shows the cdfs of the obtained femtocell data rates with carrier - sensing and transmit power control at femtocell aps . the lowest outage capacity is obtained when the femtocell is located close to the no - coverage zone ( @xmath305 ) . because the location - based power control scheme monotonically decreases transmit power with femtocell distance from @xmath31 , the outage capacities monotonically decrease with @xmath10 . the @xmath289 percentile outage capacities are respectively equal to @xmath306 and @xmath307 b / s / hz for the different femtocell locations given above which exceeds the minimum desired target spectral efficiency of @xmath292 b / s / hz . in two - tier cellular systems with universal frequency reuse , cross - tier interference will likely be the main obstacle preventing uniform coverage . this paper has derived analytical expressions for the coverage zones in such a tiered architecture with spatial diversity considering the number of antennas , the maximum tolerable outage probability accounting for path loss and rayleigh fading . single - user transmission in either tier is analytically shown to provide significantly superior coverage and spatial reuse while performance of multiple - user transmission suffers from residual cross - tier interference . for providing uniform cellular coverage , we have proposed a location - assisted power control scheme for regulating femtocell transmit powers . this scheme is fully decentralized and provides uniform cellular and hotspot coverage on the cell - edge , as opposed to randomized hotspot transmissions without carrier - sensing . these results motivate deploying closed - access tiered cellular architectures while requiring minimal network overhead . the probability of successful reception in can be upper bounded as @xmath308 } \leq \mathbb{p}{\left [ |\mathbf{g}_0^{\dagger } \mathbf{w}_{0,0}|^2 \geq \gamma \mathcal{q}_f \left(\frac{\mathcal{p}_f}{u_c}||\mathbf{f}_0^{\dagger}\mathbf{v}||^2 \right ) \right ] } = \mathbb{p}{\left [ |\mathbf{g}_0^{\dagger } \mathbf{w}_{0,0}|^2 \geq \kappa ||\mathbf{f}_0^{\dagger}\mathbf{v}||^2 \right]}.\end{aligned}\ ] ] the term @xmath309 is distributed as a chi - squared random variable ( r.v ) @xmath310 with @xmath311 degrees of freedom denoted as @xmath312 . to prove this claim , whenever @xmath113 ( beamforming to a single user ) , @xmath313 , therefore , @xmath309 is distributed as @xmath314 . when @xmath315 , @xmath316 which equals the product of two independent r.vs which are distributed as @xmath317 ( see ( * ? ? ? * theorem 1.1 ) for proof ) and @xmath314 respectively . in ( * pages 169 - 170 ) , it is shown that @xmath309 is distributed as a @xmath312 r.v . the probability density function of @xmath309 is given as @xmath318 where @xmath110 for any positive integer @xmath111 . similarly , the r.v @xmath319 is the sum of @xmath129 r.vs , wherein each term @xmath320 equals the squared modulus of a linear combination of @xmath22 complex normal r.vs , which is exponentially distributed . consequently , @xmath321 is distributed as a @xmath322 r.v . define @xmath323 . then @xmath246 is the ratio of two independent @xmath324 r.vs with @xmath311 and @xmath325 degrees of freedom respectively . therefore , @xmath246 follows a canonical @xmath326-distribution@xcite and @xmath327 is an f - distributed r.v with parameters @xmath311 and @xmath325 respectively . substituting @xmath90 in and taking the complement of , one obtains @xmath328 } \geq \mathbb{p}{[z \leq \kappa ] } = \mathbb{p}{\left[\frac { z \ u_c}{t_f - u_f+1 } \leq \kappa \frac{u_c}{t_f - u_f+1}\right ] } = \mathcal{i}_{\frac{\kappa}{\kappa+1}}(t_f - u_f+1,u_c ) . \notag\end{aligned}\ ] ] a necessary condition for meeting the qos requirement @xmath8 for indoor users served by @xmath73 is given as @xmath329 substituting the definition of @xmath90 in , one obtains @xmath14 . this completes the proof . using , the probability of successful reception @xmath93}$ ] is given as @xmath330 } \textrm { , where } i_{f , c}=\frac{\mathcal{p}_f}{u_c}||\mathbf{f}_{0}^{\dagger}\mathbf{v}||^2 , \ i_{f , f}=1/u_f \sum_{f_j \in \pi_f \setminus f_0 } ||\mathbf{g}_{0,j}^{\dagger}\mathbf{w}_j||^2 |x_{0,j}|^{-\alpha_{fo}}. \notag\end{aligned}\ ] ] the interference from neighboring femtocells @xmath331 is a poisson shot - noise process ( @xcite ) with independent and identically distributed marks@xcite . the distributions of the signal powers and marks of the interferers are chi - squared with degrees of freedom given as @xmath332 and @xmath333 respectively . consequently , @xmath334 } & \overset{(a)}= \int_{0}^{\infty } \sum_{k=0}^{t_f - u_f } \frac{(s \mathcal{q}_f\gamma)^{k}}{k ! } e^{-s\mathcal{q}_f\gamma } \textrm{d } \mathbb{p}{(i_{f , c}+i_{f , f } \leq s ) } \\ \label{eq : outagep3b } & \overset{(b)}=\sum_{k=0}^{t_f - u_f}\frac{(-\mathcal{q}_f\gamma)^{k}}{k ! } \frac{\textrm{d}^k}{\textrm{d}\theta^k}\mathcal{l}_{i_{f , c}}(\theta)\mathcal{l}_{i_{f , f}}(\theta ) \big\rvert_{\theta = \mathcal{q}_f\gamma}\end{aligned}\ ] ] where step @xmath335 follows by conditioning on @xmath336 and computing the complementary cumulative distribution ( ccdf ) of @xmath337 . for deriving step @xmath338 , with @xmath339 , the integral in @xmath335 corresponds to the laplace transform ( lt ) of the r.v @xmath340 given as @xmath341 $ ] evaluated at @xmath342 ( originally derived in @xcite ) . next , since @xmath343 and @xmath331 are independent r.vs , the lt of their sum decouples as the product of their lts @xmath344 \mathbb{e}[e^{-i_{f , f}\theta}]$ ] . finally , for any @xmath345 , we have the identity @xmath346 = ( -1)^{k}f^{(k)}(s)$ ] , where @xmath347 represents the @xmath111th derivative of @xmath348 ( this technique is borrowed from @xcite ) . the lts of @xmath343 and @xmath331 may be written as @xmath349 = \frac{1}{(1+\mathcal{p}_f\theta / u_c)^{u_c}}\end{aligned}\ ] ] @xmath350 & \overset{(a)}= \exp \left \lbrace -\lambda_f \int_{\mathbb{r}^2 } 1-\mathbb{e}_{s}[e^{-\theta \frac{s}{u_f } |x|^{-\alpha_{fo } } } ] \ \textrm{d}x \right \rbrace \notag , \textrm{where } s \sim \chi^2_{2u_f}\notag \\ & \overset{(b)}=\exp \left \lbrace -\pi \lambda_f \delta_f \left(\frac{\theta}{u_f}\right)^{\delta_f } \sum_{k=0}^{u_f-1 } \binom{u_f}{k}b(k+\delta_f , u_f - k-\delta_f ) \right \rbrace \notag \\ & \overset{(c)}= \exp(-\lambda_f \mathcal{c}_f \theta^{\delta_f})\end{aligned}\ ] ] where follows from the lt of a chi - squared r.v with @xmath325 degrees of freedom . in , step @xmath335 represents the lt of a poisson shot - noise process with independent and identically distributed marks @xmath351 equaling @xmath352 in our case . defining @xmath353 , steps @xmath338 and @xmath354 follow from @xcite . substituting and in leads to the following requirement for the success probability @xmath355 using the leibniz rule , the @xmath111th derivative of @xmath356 is given as @xmath357 where @xmath358 is the coefficient of @xmath359 in the expansion of @xmath360 . considering the low outage regime , we shall evaluate the @xmath111th derivative of @xmath361 using a first - order taylor series approximation around @xmath362 . then , for all @xmath363 , the @xmath111th derivatives of @xmath364 and @xmath361 are individually given as @xmath365 combining with and and substituting @xmath342 leads to @xmath366 next , substituting @xmath367 from and performing a first - order taylor series expansion of @xmath368 , the above expression simplifies as @xmath369 note that @xmath370 . using proposition [ pr : incompletebeta ] , we have the identity @xmath371.\end{aligned}\ ] ] with straightforward algebraic manipulation , it is easily shown that @xmath372 using , we now define @xmath154 in , where @xmath373 whenever @xmath119 ( since does not contain derivative terms ) . by substituting and @xmath154 in and discarding ( for small @xmath12 ) the @xmath374 terms ( which are @xmath375 ) , the upper bound on @xmath12 is given as @xmath376 since the _ maximum contention density _ @xmath377 maximizes the number of simultaneous femtocell transmissions , @xmath377 satisfies with equality . this completes the proof . multiplying the left hand side of by @xmath383 , @xmath384 the coefficient of @xmath385 , where @xmath386 , is given as @xmath387 ( using the combinatorial identity ( * ? ? ? * page 22 , ( 3.2 ) ) ) . consequently , we have @xmath388 next , by definition of the incomplete beta function and using @xmath389 , @xmath390 where step @xmath335 follows by definition , while step @xmath338 follows by replacing the index @xmath391 in step @xmath335 by @xmath392 . combining and gives the desired result . l. t. w. ho and h. claussen , `` effects of user - deployed , co - channel femtocells on the call drop probability in a residential scenario , '' in _ proc . , ieee international symp . on personal , indoor and mobile radio comm . 2007 , pp . 15 . t. e. klein and s .- j . han , `` assignment strategies for mobile data users in hierarchical overlay networks : performance of optimal and adaptive strategies , '' _ ieee journal on sel . areas in comm . _ , vol . 22 , no . 5 , pp . 849861 , june 2004 . v. chandrasekhar , j. g. andrews , t. muharemovic , z. shen , and a. gatherer , `` power control in two - tier femtocell networks , '' _ submitted , ieee trans . on wireless comm . _ , 2008 , [ online ] available at ` http://arxiv.org/abs/0810.3869 ` . v. chandrasekhar and j. g. andrews , `` uplink capacity and interference avoidance in two - tier femtocell networks , '' _ to appear , ieee trans . on wireless comm . _ , 2009 , [ online ] available at ` http://arxiv.org/abs/cs.ni/0702132 ` . k. hamdi , w. zhang , and k. ben letaief , `` power control in cognitive radio systems based on spectrum sensing side information , '' in _ proc . , ieee international conf . on comm . _ , june 2007 , pp . 51615165 . m. haenggi , j. g. andrews , f. baccelli , o. dousse , and m. franceschetti , `` stochastic geometry and random graphs for the analysis and design of wireless networks , '' _ submitted , ieee journal on sel . areas on comm . _ , 2009 . y. akyildiz and b. d. rao , `` statistical performance analysis of optimum combining with co - channel interferers and flat rayleigh fading , '' in _ proc . , ieee global telecomm . conference _ , vol . 6 , san antonio , tx , usa , 2001 , pp . 36633667 . e. sousa and j. silvester , `` optimum transmission ranges in a direct - sequence spread - spectrum multihop packet radio network , '' _ ieee journal on sel . areas in comm . _ , vol . 8 , no . 5 , pp . 762771 , june 1990 . with su transmission as a function of the distance @xmath10 . shaded region shows feasible @xmath20 s at location @xmath10 , which satisfies the per - tier outage probability requirement for different average numbers of femtocells per cell - site.,width=480 ]

around 1923 , lie cartan introduced affine connections on manifolds and defined the main related concepts : torsion , curvature , holonomy groups . he discussed applications of these concepts in classical and relativistic mechanics ; in particular he explained how parallel transport with respect to a connection can be related to the principle of inertia in galilean mechanics and , more generally , can be used to model the motion of a particle in a gravitational field . in subsequent papers , lie cartan extended these concepts for other types of connections on a manifold : euclidean , galilean and minkowskian connections which can be considered as special types of affine connections , the group of affine transformations of the affine tangent space being replaced by a suitable subgroup ; and more generally , conformal and projective connections , associated to a group which is no more a subgroup of the affine group . around 1950 , charles ehresmann introduced connections on a fibre bundle and , when the bundle has a lie group as structure group , connection forms on the associated principal bundle , with values in the lie algebra of the structure group . he called _ cartan connections _ the various types of connections on a manifold previously introduced by . cartan , and explained how they can be considered as special cases of connections on a fibre bundle with a lie group as structure group : the standard fibre of the bundle is then an homogeneous space ; its dimension is equal to that of the base manifold ; a cartan connection determines an isomorphism of the vector bundle tangent to the the base manifold onto the vector bundle of vertical vectors tangent to the fibres of the bundle along a global section . these works are reviewed and some applications of the theory of connections are sketched . [ multiblock footnote omitted ]

1401.8272

it is a great honor for me to be invited at the seventh international conference on geometry and topology of manifolds , dedicated to the mathematical legacy of charles ehresmann . i enjoyed with great pleasure the hospitality of the mathematical research and conference center of the polish academy of sciences , and i address my warmest thanks to the organizers and to the supporting institutions . around 1923 , lie cartan @xcite introduced the notion of an _ affine connection _ on a manifold . that notion was previously used , in a less general setting , by h. weyl @xcite and rests on the idea of parallel transport due to t. levi - civita @xcite . at the very beginning of @xcite , even before defining explicitly affine connections , lie cartan explains how that concept can be used in newtonian and einsteinian mechanics . he shows that the _ principle of inertia _ ( which is at the foundations of mechanics ) , according to which a material point particle , when no forces act on it , moves along a straight line with a constant velocity , can be expressed locally by the use of an affine connection . under that form , that principle remains valid in ( curved ) einsteinian space - times . cartan even shows that by a suitable adjustment of the connection , a gravity force ( that means , an acceleration field ) can be accounted for , and becomes a part of the geometry of space - time . that result expresses the famous _ equivalence principle _ used by einstein for the foundations of general relativity . as shown by cartan , it is valid for newtonian mechanics as well . then lie cartan presents a thorough geometric study of affine connections ; he defines their curvature and torsion , and discusses the parallel displacement of a frame along a closed loop . he introduces euclidean , galilean and minkowskian connections , for which the group of affine transformations is replaced by a suitable subgroup . in @xcite he introduces more general types of connections associated to transformation groups which are no more subgroups of the group of affine transformations . cartan s ideas were fully formalized by charles ehresmann in the framework of connections on fibre bundles , which he introduced in @xcite . in section 2 we briefly present cartan s intuitive ideas about connections . then in section 3 we describe ehresmann connections on fibre bundles , and in section 4 cartan connections as seen by ehresmann . in section 5 we present with more details examples of cartan connections , including affine , projective and conformal connections . in section 6 , following cartan , we show how a gravitational force can be included in the geometry of space - time by the use of a suitable connection , and we briefly present other applications of connections : geometric quantization , phases in mechanics , nonholonomic or active constraints , maxwell s equations , yang - mills fields . cartan writes in @xcite : une varit connexion affine est une varit qui , au voisinage immdiat de chaque point , a tous les caractres dun espace affine , et pour laquelle on a une loi de reprage des domaines entourant deux points infiniment voisins : cela veut dire que si , en chaque point , on se donne un systme de coordonnes cartsiennes ayant ce point pour origine , on connat les formules de transformation ( de mme nature que dans lespace affine ) qui permettent de passer dun systme de rfrence tout autre systme de rfrence dorigine infiniment voisine . approximate translation : a manifold with an affine connection is a manifold whose properties , in the neighborhood of each point , are those of an affine space , and on which there is a law for fitting together the neighborhoods of two infinitesimally nearby points : it means that if , on a neighborhood of each point , we have chosen cartesian coordinates with that point as origin , we know the transformation formulae ( of the same nature as those valid in an affine space ) which allow to go from a reference frame to another reference frame with an infinitesimally nearby origin . in fact , given a smooth manifold @xmath2 , it is not on a neighborhood of each point @xmath3 that cartan considers a local affine structure . rather , at each point @xmath3 , he considers the tangent space @xmath4 endowed with its natural affine space structure . and he writes : la varit sera dite _ connexion affine _ lorsquon aura dfini , dune manire dailleurs arbitraire , une loi permettant de reprer lun par rapport lautre les espaces affines attachs deux points infiniment voisins quelconques @xmath5 et @xmath6 de la varit ; cette loi permettra de dire que tel point de lespace affine attach au point @xmath6 correspond tel point de lespace affine attach au point @xmath5 , que tel vecteur du premier espace est parallle ou quipollent tel vecteur du second espace . en particulier le point @xmath6 lui - mme sera repr par rapport lespace affine du point @xmath5 @xmath7 . approximate translation : the manifold will be said to be endowed with an _ affine connection _ once we have defined , in an arbitrary way , a law allowing to localize one with respect to the other the affine spaces attached to two infinitesimally nearby points @xmath5 and @xmath6 of that manifold ; that law will tell us which point of the affine space attached to @xmath6 corresponds to a given point of the space attached to @xmath5 , and will tell us whether a vector living in the first space is parallel , or equipollent , to a vector living in the second space ; in particular , the point @xmath6 itself will be localized in the affine space attached to @xmath5 @xmath7 . cartan then explains with more details how to define an affine connection on a @xmath8-dimensional manifold @xmath2 ; of course his definition extends easily to manifolds of any dimension . he considers , at each point @xmath3 , an affine frame of the affine tangent space @xmath4 , with as origin the point @xmath5 itself ( identified with the null vector at @xmath5 ) , and with the linear basis @xmath9 as basis . in order to define the law which links the affine spaces tangent to the manifold @xmath2 at two infinitesimally nearby points @xmath5 and @xmath6 , he write the relations @xmath10 these equations mean that the point @xmath6 , origin of @xmath11 , infinitesimally near @xmath5 , must be identified with the point @xmath12 of the affine space @xmath4 . similarly , the vectors @xmath13 , @xmath14 , @xmath15 of @xmath11 must be identified with the vectors @xmath16 of @xmath4 . equations @xmath17 must be understood as equalities between differential @xmath18-forms on a @xmath19-dimensional space with , as coordinates , the @xmath8 coordinates which specify a point on @xmath2 , and @xmath20 more coordinates on which depend the affine frames of a @xmath8-dimensional affine space . in fact , these differential @xmath18-forms live on the principal bundle of affine frames of the affine tangent spaces to the maniflold @xmath2 . the meaning of equations @xmath17 is the following : let @xmath21 be an affine frame of the affine tangent space @xmath4 and let @xmath22 be a vector tangent at @xmath23 to the space of affine frames . the infinitesimal displacement of that affine frame defined by the vector @xmath22 is parallel , with respect to the connection , if and only if @xmath24 , where @xmath25 is the @xmath18-form of the connection ( in the sense of ehresmann , as introduced later ) . equation @xmath24 , when explicitly written , leads to the @xmath20 scalar equations @xmath17 because the @xmath18-form @xmath25 takes its values in the @xmath20-dimensional lie algebra of the group of affine transformations of @xmath26 . this will become clear with the works of charles ehresmann @xcite . euclidean connections , galilean connections , minkowskian connections , @xmath7 , considered by cartan , appear as special cases of affine connections , obtained by replacing the group of affine transformations of an affine space by an appropriate subgroup . for example , on a @xmath8-dimensional manifold @xmath2 endowed with a riemannian metric @xmath27 , instead of general affine frames @xmath21 with the contact point @xmath5 as origin of the affine tangent space @xmath4 , cartan considers only orthonormal frames , which satisfy @xmath28 now the dimension of the space of orthonormal affine frames is 9 ( instead of 15 for the space of general affine frames ) . by differentiation of the above relations we see that the @xmath18 forms @xmath29 and @xmath30 of equations @xmath17 must satisfy @xmath31 only 6 of the 12 scalar equations @xmath17 are now independent ( in agreement with the dimension , 6 , of the lie algebra of the group of affine orthogonal transformations of @xmath26 ) . other types of connections were defined and discussed by . cartan @xcite , with transformation groups which are no more subgroups of the group of affine transformations : conformal connections , projective connections , @xmath7 the general idea underlying the notions of such connections on a manifold @xmath2 is to use , as a local model of that manifold , an homogeneous space of the same dimension as @xmath2 . a copy of that homogeneous space is attached at each point of the manifold , and considered as tangent at that point to the manifold . the connection is essentially a law which indicates how these homogeneous spaces are glued together . of course , affine connections appear as the special case in which the homogeneous space which is attached to each point of the manifold is an affine space , more precisely the affine space tangent to the manifold at that point . for conformal connections on an @xmath32-dimensional manifold , . cartan @xcite writes : attachons chaque point @xmath33 de cette varit un espace conforme @xmath32 dimensions , @xmath7 la varit sera dite _ connexion conforme _ si nous nous donnons une loi ( dailleurs arbitraire ) permettant de rapporter , dune manire conforme , lespace conforme attach au point @xmath33 de la varit lespace conforme attach au point infiniment voisin @xmath34 . approximate translation : let us link an @xmath32-dimensional conformal space to each point @xmath33 of our manifold . that manifold will be said to be endowed with a _ conformal connection _ when we have specified , in an arbitrary way , how to tie ( or maybe identify ) the conformal space linked at point @xmath33 with the conformal space linked to the infinitesimally nearby point @xmath34 . in @xcite , cartan writes : lide fondamentale se rattache la notion de _ paralllisme _ que m. t. levi - civita a introduite de manire si fconde . les nombreux auteurs qui ont gnralis la thorie des espaces mtriques sont tous partis de lide fondamentale de m. levi - civita , mais , semble - t - il , sans pouvoir la dtacher de lide de _ vecteur_. cela na aucun inconvnient quand il sagit de varits connexion affine @xmath7 mais cela semblait interdire tout espoir de fonder une thorie _ autonome _ de varits connexion conforme ou projective . en fait , ce quil y a dessentiel dans lide de m. levi - civita , cest quelle donne un moyen pour raccorder entre eux deux petits morceaux infiniment voisins dune varit , et cest cette ide de _ raccord _ qui est fconde . approximate translation : the fundamental idea stems from the notion of _ parallelism _ introduced by m. t. levi - civita in such a fruitful way . the many authors who genralized the theory of metric spaces all started from the fundamental idea of m. levi - civita , but , seemingly , without freeing it from the idea of _ vector_. that does not matter as long as one deals with manifolds with affine connections @xmath7 but that seemed to forbid any hope to build an _ autonomous _ theory of manifolds with conformal or projective connections . in fact , the main thing in m. levi - civita s idea is that it allows to glue together two small , infinitesimally nearby pieces of a manifold , and it is that idea of gluing which is most fruitful . let @xmath35 be a locally trivial smooth fibre bundle with base @xmath36 , standard fibre @xmath37 , and canonical projection @xmath38 . for each @xmath39 , the fibre at @xmath40 , @xmath41 , is diffeomorphic to @xmath37 . ehresmann @xcite defines an _ infinitiesimal connection _ on that bundle as follows . [ definfinitesimalconnection ] an _ infinitesimal connection _ on the locally trivial bundle @xmath35 is a vector sub - bundle @xmath42 of @xmath43 , complementary to @xmath44 , _ i.e. _ such that for each @xmath45 , @xmath46 which satisfies the additional condition : @xmath47given any smooth path @xmath48 in @xmath36 going from a point @xmath49 to another point @xmath50 , and any @xmath51 , there exists a smooth path @xmath52 in @xmath53 , called the _ horizontal lift _ of @xmath48 through @xmath54 , such that @xmath55\,.\ ] ] more generally any smooth path @xmath52 in @xmath53 is said to be _ horizontal _ if for each @xmath56 , @xmath57 . [ remparalleltransport ] under the assumptions of definition 1 , let @xmath48 be a smooth path in @xmath36 , defined on an interval @xmath58 . for each @xmath59 and each @xmath60 , its horizontal lift @xmath52 through @xmath54 is unique and defined on @xmath58 . for each pair @xmath61 , the map @xmath62 , @xmath63 is a diffeomorphism of the fibre @xmath64 onto the fibre @xmath65 , called the _ parallel transport _ of the fibres of @xmath35 along the smooth path @xmath48 . by looking only at paths defined on finite closed intervals , and taking their end points , we see that the connection @xmath42 determines a homomorphism of the groupoid of smooth paths in @xmath36 which joint two points in @xmath36 into the groupoid of diffeomorphisms of a fibre or @xmath35 onto another fibre . [ defdevelopment ] let @xmath52 be a smooth path , defined on an interval @xmath58 , with values in the total space @xmath53 of a locally trivial , smooth bundle @xmath35 equipped with a connection @xmath42 . let @xmath66 be the corresponding smooth path in the base manifold @xmath36 . let @xmath59 and , for each @xmath56 , let @xmath67 be the parallel transport of the fibre @xmath64 onto the fibre @xmath68 along the path @xmath69 . the smooth path in the fibre @xmath64 , @xmath70 is called the _ development _ of the path @xmath52 in the fibre @xmath64 . the development in @xmath64 of a path @xmath71 in @xmath53 is a constant path if and only if @xmath52 is horizontal . we assume now that @xmath35 is a smooth fibre bundle @xmath72 with a lie group @xmath0 as structure group . let us first recall that notion , due to ehresmann , and explain the notations . for simplicity we consider only smooth bundles , although in @xcite ehresmannn , more generally , defines topological bundles . the _ total space _ @xmath53 and the _ base _ @xmath36 of the bundle are two smooth manifolds , and there is a smooth submersion @xmath38 of the total space onto the base . for each @xmath39 , @xmath41 is a smooth submanifold of @xmath53 called the _ fibre over _ @xmath40 . the _ standard fibre _ @xmath37 is a smooth manifold and the _ structure group _ @xmath0 is a lie group of diffeomorphisms of @xmath37 , _ i.e. _ a lie group @xmath0 which acts on the manifold @xmath37 on the left in such a way that the map @xmath73 , @xmath74 , is smooth . that action is assumed to be effective , _ i.e. _ such that for each @xmath75 other than the unit element @xmath76 , the corresponding diffeomorphism @xmath77 of the standard fibre @xmath37 is not the identity ; therefore an element @xmath75 is fully determined by the corresponding diffeomorphism of @xmath37 . an _ atlas _ @xmath78 of the bundle @xmath72 is defined by the following data : ( iii ) a family @xmath79 of open subsets of @xmath36 such that @xmath80 , for each @xmath81 , a diffeomorphism @xmath82 such that for each @xmath83 , @xmath84 ( where @xmath85 is the first projection ) and that for each pair @xmath86 with @xmath87 , @xmath88 with @xmath89 , @xmath90 , and where @xmath91 is a smooth map which takes its values in the lie group @xmath0 . two atlases @xmath78 and @xmath92 of the bundle @xmath72 are said to be _ equivalent _ if their union is still an atlas of that bundle . the @xmath0-bundle structure of @xmath72 is determined by the choice of an equivalence class of atlases . diffeomorphisms @xmath93 which belong to an atlas of that equivalence class are called _ admissible local trivializations _ of the bundle @xmath72 . for each @xmath39 , the fibre @xmath41 is diffeomorphic to @xmath37 . more precisely there are _ admissible diffeomorphisms _ of @xmath37 onto @xmath94 , obtained by means of an admissible local trivialization @xmath95 , with @xmath96 : @xmath97 for each @xmath98 , if @xmath99 is an admissible diffeomorphism , another diffeomorphism @xmath100 is admissible if and only if there exists @xmath75 such that @xmath101 . in ehresmann s notation @xmath72 , we have already explained what are the total space @xmath53 , the base @xmath36 , the standard fibre @xmath37 and the structure group @xmath0 . we still have to define @xmath102 : it is the set of all admissible diffeomorphisms of the standard fibre @xmath37 onto the fibre @xmath41 , for all @xmath39 . in @xcite , ehresmann proves that @xmath102 is a smooth manifold . more precisely it is the total space of a smooth fibre bundle @xmath103 with base @xmath36 , standard fibre @xmath0 , structure group @xmath104 ( that means the lie group @xmath0 considered as acting on itself by left translations ) . the fibre @xmath105 over a point @xmath39 is the set of admissible diffeomorphisms of @xmath37 onto @xmath41 . in ehresmann s notations @xmath106 is the set of admissible diffeomorphisms @xmath107 of the standard fibre @xmath0 onto the fibre @xmath108 of @xmath102 over some point @xmath39 . in fact , @xmath106 can be identified with @xmath102 , since an element @xmath109 , previously considered as a diffeomorphism of @xmath37 onto @xmath94 , can also be considered as a diffeomorphism @xmath107 of @xmath0 onto @xmath108 by writing @xmath110 the smooth fibre bundle @xmath111 is said to be a _ principal bundle _ because its structure group @xmath0 is also its standard fibre . more precisely , it is called _ the principal bundle associated to _ the bundle @xmath112 , and it is fully determined by that bundle . conversely , let @xmath111 be a principal bundle with structure lie group @xmath104 , and @xmath37 be a smooth manifold on which the lie group @xmath0 acts on the left . there exist a bundle @xmath72 , unique up to an isomorphism , with structure lie group @xmath0 , standard fibre @xmath37 and @xmath111 as associated principal bundle . the bundles obtained in that way are said to be _ associated _ to the given principal bundle @xmath111 . [ remrightaction ] there is a natural , free action on the right of the structure lie group @xmath0 on the total space @xmath102 of the principal bundle @xmath111 , @xmath113 whose orbits are the fibres @xmath108 , @xmath39 . [ defgconnection ] let @xmath72 be a fibre bundle with a lie group @xmath0 as structure group . an infinitesimal connection @xmath42 on that bundle , in the sense of definition [ definfinitesimalconnection ] , is said to be _ compatible with the structure group _ @xmath0 , and is called a _ @xmath0-connection _ , if the parallel transport along any smooth path in @xmath36 starting from any point @xmath114 and ending at any other point @xmath115 , is of the form @xmath116 , with @xmath117 and @xmath118 . a @xmath0-connection on the bundle @xmath72 determines a unique @xmath104-connection @xmath119 on the associated principal bundle @xmath111 such that , for any smooth path @xmath48 in the base @xmath36 , any @xmath120 in the interval of definition of that path and any @xmath121 , the horizontal lift through @xmath122 of the path @xmath48 with respect to the connection @xmath119 is the path @xmath123 where @xmath124 is the parallel transport along the smooth path @xmath48 , for the @xmath0-connection @xmath42 on @xmath72 . conversely , a @xmath104-connection @xmath119 on the principal bundle @xmath111 determines a unique @xmath0-connection @xmath42 on @xmath72 such that , for any smooth path @xmath48 in the base @xmath36 , any @xmath120 in the interval of definition of that path and any @xmath60 , the horizontal lift through @xmath54 of the path @xmath48 with respect to the connection @xmath42 is the path @xmath125 where @xmath126 is any horizontal lift in @xmath102 of the path @xmath48 for the @xmath104-connection @xmath119 , and where @xmath127 is the unique element in @xmath37 such that @xmath128 . a @xmath104-connection on a principal bundle @xmath111 is a vector sub - bundle @xmath119 of the tangent bundle @xmath129 , complementary to @xmath130 , _ i.e. _ such that , for each @xmath131 , @xmath132 which remains invariant by the right action of @xmath104 defined in remark [ remrightaction ] , _ i.e. _ which is such that for any @xmath131 and @xmath75 , @xmath133 where @xmath134 is the lift to vectors of the map @xmath135 , @xmath136 . conversely , a vector sub - bundle @xmath137 of @xmath129 complementary to @xmath130 which remains invariant by the right action of @xmath104 is a @xmath0-connection on @xmath111 : by using the @xmath0-invariance , one can prove that condition @xmath47 of definition [ definfinitesimalconnection ] is automatically satisfied . the next proposition introduces the very important notion of _ connection form_. [ propconnectionform ] let @xmath119 be a @xmath104-connection on the principal bundle @xmath111 . there exists a unique @xmath18-form @xmath25 on @xmath102 , with values in the lie algebra @xmath138 of @xmath0 , such that for all @xmath131 , @xmath139 and that for each @xmath140 , @xmath141 we have denoted by @xmath142 the fundamental vector field on @xmath102 associated to @xmath143 , _ i.e. _ the vector field whose value , at an element @xmath131 , is @xmath144 the @xmath18-form @xmath25 is called the connection form of the @xmath104-connection @xmath119 . a connection on a principal bundle compatible with its structure group can be defined by its connection form , as shown by the next proposition . [ proppropertiesconnectionform ] the connection form @xmath25 of a @xmath104-connection @xmath137 on the principal bundle @xmath111 has the following properties : ( iii ) for each @xmath145 ( the lie algebra of @xmath0 ) and each @xmath131 , @xmath146 where @xmath142 is the fundamental vector field on @xmath102 associated to @xmath143 ; for each @xmath75 , the pull - back @xmath147 of the form @xmath25 by the right translation @xmath148 , is @xmath149 conversely , any @xmath18-form on @xmath102 which takes its values in the lie algebra @xmath138 of @xmath0 , and which satisfies these two properties , is the connection form of a @xmath104-connection on the principal bundle @xmath111 . we will see in this section that connections on fibre bundles with a lie group as structure group offer a very nice setting for a rigorous presentation of cartan connections . [ propprincipalsubbundle ] let @xmath72 be a smooth bundle with a lie group @xmath0 as structure group . we assume that @xmath37 is an homogeneous space @xmath1 , with @xmath150 a closed subgroup of @xmath0 . we denote by @xmath151 the image of the unit element @xmath152 by the projection @xmath153 . moreover , we assume that there exists a smooth section @xmath154 of the bundle @xmath72 . let @xmath155 be the subset of elements @xmath131 such that @xmath156 then @xmath155 is the total space of a principal bundle @xmath157 with @xmath150 as structure group . [ defcartanconnection ] under the assumptions and with the notations of the previous proposition , let @xmath25 be the connection form of a @xmath0-connection on @xmath102 , and let @xmath158 be the @xmath18-form induced on @xmath155 by @xmath25 . we assume that for any @xmath159 and any vector @xmath160 , we have @xmath161 moreover , we assume that @xmath162 such a connection on the bundle @xmath163 ( or on the associated principal bundle @xmath111 ) is called a _ cartan connection_. [ reminducedformofcartanconnection ] the assumptions and notations are those of proposition [ propprincipalsubbundle ] and definition [ defcartanconnection ] . although defined on the total space @xmath155 of the principal bundle @xmath157 , the form @xmath164 is not a connection form on that bundle in the sense of charles ehresmann , since it takes its values in the larger lie algebra @xmath138 , not in the lie algebra @xmath165 of @xmath150 . however , the connection @xmath18-form @xmath25 , defined on @xmath102 , is fully determined by the form @xmath164 induced on @xmath155 : its values on a vertical vector tangent to @xmath102 ( maybe not tangent to @xmath155 ) at a point of @xmath155 is known , since such a vector is the value of a fundamental vector field associated to an element of the lie algebra @xmath138 ; and its values at points of @xmath102 outside @xmath155 can be deduced from its values on @xmath155 by using the formula @xmath166 moreover , let @xmath157 be a principal bundle whose structure group @xmath150 is a closed subgroup of a lie group @xmath0 , and let @xmath164 be a @xmath18-form defined on @xmath155 , taking its values in the lie algebra @xmath138 , satisfying the two properties : ( iii ) for each @xmath167 ( the lie algebra of @xmath150 ) and each @xmath168 , @xmath169 where @xmath170 is the fundamental vector field on @xmath155 associated to @xmath171 ; for each @xmath172 , the pull - back @xmath173 of the form @xmath164 by the right translation @xmath174 , is @xmath175 given such a @xmath18-form , called a _ cartan connection form _ , we can build a principal bundle @xmath176 with @xmath0 as structure group , and a connection form @xmath25 on @xmath102 , in such a way that @xmath155 is a subset of @xmath102 and that @xmath164 is the form infuced on @xmath155 by the connection form @xmath25 . we set @xmath177 , quotient of the product @xmath178 by the equivalence relation for which two elements @xmath179 and @xmath180 in @xmath178 are equivalent if there exists @xmath172 such that @xmath181 and @xmath182 . we identify @xmath155 with the set of equivalence classes of elements in @xmath183 , and we define @xmath25 on @xmath102 as indicated above . [ remgeneralizedcartan ] the first condition in the above definition , @xmath184 implies @xmath185 , since @xmath186 and @xmath187 . for a cartan connection , the second condition in the above definition imposes the equality , @xmath188 . when the first condition of the above definition is kept and the second condition is dropped , ehresmann calls @xmath25 a connexion de cartan au sens large , _ i.e. _ a generalized cartan connection . [ remparallelizable ] the existence of a cartan connection implies that @xmath155 is parallelizable , since @xmath164 determines an isomorphism of @xmath189 onto @xmath190 . the next proposition indicates another important consequence of the existence of a cartan connection : the tangent bundle @xmath191 to the base manifold is isomorphic to the bundle of vertical vectors , tangent to the fibres of the bundle @xmath72 along the image of the global section @xmath192 . [ propvectorbundleisomorphism ] the assumptions and notations are those of proposition [ propprincipalsubbundle ] and definition [ defcartanconnection ] . let @xmath39 , @xmath193 . there exists a unique vector @xmath194 such that , for any @xmath195 and @xmath196 which satisfies @xmath197 , we have @xmath198 the map @xmath199 so obtained is a vector bundle isomorphism , called the soldering isomorphism , of the tangent bundle @xmath191 onto the vector bundle @xmath200 of vertical vectors , tangent to the fibres of @xmath53 along the image @xmath201 of the global section @xmath192 . in the formula which defines @xmath202 we have identified the lie algebra @xmath138 with the space @xmath203 , tangent to the lie group @xmath0 at its unit element @xmath76 . we have only to show that @xmath204 depends only on @xmath205 , not on the choices of @xmath206 and @xmath207 . if we keep @xmath207 unchanged and replace @xmath206 by @xmath208 , @xmath209 is the value at @xmath207 of a fundamental vector field associated to an element of the lie algebra @xmath210 , therefore @xmath211 and @xmath212 . now if we replace @xmath207 by @xmath213 , with @xmath172 , we may replace @xmath206 by @xmath214 , so @xmath215 since @xmath216 . [ remtangency ] for each point @xmath39 , the soldering isomorphism determined by a cartan connection on the bundle @xmath163 allows us to identify the space @xmath217 , tangent to the base @xmath36 at @xmath40 , with the space @xmath218 , tangent to the fibre @xmath94 at @xmath219 . it is in that precise sense that the soldering isomorphism allows us to consider that for each @xmath39 , the base manifold @xmath36 is tangent to the fibre @xmath94 , the point @xmath39 being in contact with the point @xmath220 . [ defdevelopmentcartan ] the assumptions are those of definition [ defcartanconnection ] . let @xmath48 be a smooth path in the base manifold @xmath36 . the _ development _ of that path in the fibre @xmath64 is the development ( in the sense of definition [ defdevelopment ] ) of the smooth path in @xmath53 , @xmath221 [ remdevelopmentcartan ] the development @xmath222 of the path @xmath48 in the fibre @xmath64 can be obtained as follows . let @xmath223 be a smooth path in @xmath155 such that , for each @xmath56 , @xmath224 such a path always exists ( and is not unique ) . it is not horizontal for the connection @xmath119 ( except when @xmath48 is constant ) since @xmath225 does not vanish when applied to nonzero vectors tangent to @xmath155 . let @xmath126 be the horizontal lift in @xmath102 of the path @xmath48 , through the point @xmath226 . there exists a unique path @xmath227 in @xmath0 such that @xmath228 and that , for each @xmath56 , @xmath229 an easy calculation shows that he development of @xmath48 in @xmath64 is the path @xmath230 where @xmath231 is the image of the unit element @xmath76 by the projection @xmath232 . moreover , the map @xmath233 is the vector bundle isomorphism @xmath202 of proposition [ propvectorbundleisomorphism ] . that property is in agreement with the fact than when the fibre @xmath64 is considered as tangent to the base @xmath36 at @xmath234 , the tangent vectors to the path @xmath48 at @xmath234 and to its development at @xmath235 should be related by the isomorphism @xmath202 let @xmath150 be a closed subgroup of a lie group @xmath0 such that the left action of @xmath0 on the homogeneous space @xmath236 is effective . we consider the trivial @xmath0-bundle @xmath72 with @xmath237 , @xmath238 , the fibres being the subsets @xmath239 ( in other words , the projection @xmath240 is the first projection ) . the associated principal @xmath0-bundle is the trivial bundle @xmath111 with @xmath241 , an element @xmath242 being considered as the diffeomorphism of the standard fibre @xmath37 onto the fibre @xmath243 : @xmath244 the structure group @xmath0 acts on @xmath102 on the right by @xmath245 for the global section @xmath246 , we choose the diagonal map @xmath247 . the principal @xmath150-bundle @xmath157 has as its total space @xmath155 the subset of elements @xmath248 which map the point @xmath249 of @xmath37 into the image of the global section @xmath192 , _ i.e. _ which are such that @xmath250 . therefore @xmath155 can be identified with @xmath0 by means of @xmath251 . the fibre of @xmath155 over a point @xmath252 of @xmath36 is the coset @xmath253 . on the trivial bundle @xmath254 , we consider the integrable connection @xmath42 for which the parallel transport along any smooth path in @xmath36 with end points @xmath255 and @xmath256 is the map @xmath257 given a smooth path @xmath48 in @xmath36 , its development in the fibre @xmath258 is @xmath259 . the connection considered here being integrable , the soldering isomorphism integrates in the following sense : for each point @xmath40 in the base manifold @xmath36 , there is a natural diffeomorphism @xmath260 of the base @xmath36 onto the fibre @xmath239 , whose differential at @xmath40 is the soldering isomorphism . when we identify @xmath155 with @xmath0 as indicated above , the @xmath18-form @xmath164 induced on @xmath155 by the connection form is the left maurer - cartan form on @xmath0 which , for any @xmath75 and @xmath261 , yields @xmath262 ; of course @xmath263 if and only if @xmath264 , as required by definition [ defcartanconnection ] . let @xmath2 be a smooth @xmath32-dimensional manifold . its tangent bundle @xmath265 is generally considered as a vector bundle , with the linear group @xmath266 as structure group ; but here we consider it as an _ affine _ bundle , with the group @xmath267 of affine transformations of @xmath268 as structure group . we set @xmath269 , @xmath270 ; the standard fibre @xmath1 is identified with @xmath268 , considered as an affine space . the corresponding principal bundle @xmath271 has as total space the set @xmath102 of affine isomorphisms of @xmath268 onto the tangent space @xmath272 at some point @xmath273 . we take as global section @xmath192 of @xmath265 the zero section ( for each @xmath273 , @xmath219 is the zero vector at @xmath40 ) . let @xmath155 be the subset of elements in @xmath102 which map the origin of @xmath268 into the image of @xmath192 . in other words , an element @xmath274 is a _ linear _ isomorphism of @xmath268 onto the tangent space @xmath272 ( considered now as a vector space ) at some point @xmath273 , and @xmath155 is the total space of the @xmath150-principal bundle @xmath275 . let @xmath119 be a connection on the principal bundle @xmath271 . its connection form @xmath25 takes its value in the lie algebra of the affine group , which is a semi - direct product @xmath276 of the lie algebra @xmath277 of the linear group and of the lie algebra @xmath268 of the group of translations in @xmath268 . we may therefore write @xmath278 where @xmath279 takes its values in @xmath277 and @xmath280 in @xmath268 . in the same way , the form @xmath164 induced by @xmath25 on @xmath155 can be written @xmath281 where @xmath282 takes its values in @xmath165 and @xmath283 in @xmath268 . we see that @xmath282 is a connection form on the @xmath150-principal bundle @xmath275 ; in other words it is the connection form of a _ linear _ connection . the second term @xmath284 can be identified with a linear endomorphism @xmath285 as follows . let @xmath273 , @xmath286 , @xmath287 and @xmath288 such that @xmath289 . we set @xmath290 as in the proof of proposition [ propprincipalsubbundle ] , we see that @xmath291 only depends on @xmath292 , not on the choices of @xmath207 and @xmath293 , and that @xmath202 is a linear endomorphism of the tangent bundle @xmath265 . moreover , @xmath284 is fully determined by @xmath202 . conversely , let @xmath294 be the connection form of a linear connection , _ i.e. _ a connection on the principal bundle @xmath295 of linear frames , and @xmath202 be a linear endomorphism of @xmath265 . there exists a unique connection on the principal bundle @xmath271 of affine frames whose connection form @xmath296 is such that @xmath297 and that the the linear endomrphism of @xmath265 determined by @xmath284 is @xmath202 . so a connection on the principal bundle of affine frames can be identified with a pair made by a linear connection and a linear endomorphism of @xmath265 . the lie algebra of the affine group is a semi - direct product @xmath298 of the lie algebra @xmath277 of the linear group and the lie algebra @xmath268 of the group of translations @xmath268 . let @xmath232 be the canonical projection . when @xmath138 is identified with @xmath203 , @xmath1 and @xmath299 with @xmath268 , the second projection @xmath300 is simply the tangent map @xmath301 at the unit element . the connection form @xmath25 on the frame of affine bundles is a cartan connection in the sense of definition [ defcartanconnection ] if and only if @xmath164 has zero kernel , _ i.e. _ for each @xmath168 , @xmath302 . let @xmath303 and @xmath304 be the subspaces of @xmath305 spanned by the values at @xmath207 of fundamental vector fields associated , respectively , to elements in the lie subalgebras @xmath277 and @xmath268 of @xmath138 . we have the direct sum decompositions @xmath306 since @xmath307 we see that @xmath308 if and only if @xmath309 is the graph of a linear isomorphism of @xmath310 onto @xmath304 . using the above remark , we conclude that the connection form @xmath25 is a cartan connection if and only if the endomorphism @xmath285 determined by @xmath25 is an isomorphism . when that condition is fulfilled , @xmath202 is the soldering isomorphism . it is very natural to impose to @xmath202 to be the identity isomorphism of @xmath265 , since that ensures that any smooth path @xmath48 in @xmath2 has , at a point @xmath234 , the same tangent vector as its development in @xmath311 . affine connections considered by . cartan @xcite satisfy that property . contrary to ehresmann @xcite and lichnerowicz @xcite who call _ affine connection _ any connection on the principal bundle of affine frames , kobayashi and nomizu @xcite reserve that name for a cartan connection whose soldering isomorphism is the identity of @xmath265 , and speak of _ generalized affine connections _ for all other connections on the principal bundle of affine frames . in this subsection the standard fibre @xmath312 is the standard @xmath32-dimensional real projective space , quotient of @xmath313 by the equivalence relation of colinearity . the structure group will be the corresponding projective linear group @xmath314 , quotient of @xmath315 by the group of homotheties . we denote by @xmath316 the standard coordinates in @xmath317 . let @xmath318 be the straight line in @xmath317 defined by the equations @xmath319 , @xmath150 be the closed subgroup of @xmath0 made by equivalence classes of elements @xmath320 which map @xmath318 onto itself , and @xmath151 be the image of @xmath321 by the projection @xmath322 . the standard fibre @xmath312 will be identified with the homogeneous space @xmath1 . for each @xmath323 , let @xmath324 be the map @xmath325\mapsto[x_0,g(x_1,\ldots , x_n)]\,.\ ] ] the injective map @xmath326 will allow us to identify @xmath266 with a closed subgroup @xmath327 of @xmath328 , contained in the subgroup @xmath150 defined above.we observe that @xmath150 is the subgroup made by elements in @xmath0 which leave fixed the point @xmath151 , and that @xmath327 is the subgroup made by elements which , in addition , leave globally invariant the hyperplane at infinity ( image of the hyperplane defined by the equation @xmath329 by the projection @xmath330 ) . now let @xmath2 be an @xmath32-dimensional smooth manifold . we consider the direct sum of the trivial bundle @xmath331 and of the tangent bundle @xmath265 . we take the complementary part of the image of the zero section in the total space of that bundle and quotient it by the equivalence relation of colinearity in the fibres . we obtain a fibre bundle @xmath332 whose standard fibre is the projective space @xmath312 . the tangent bundle can be considered as a dense subset of the total space @xmath53 of that bundle , if we identify , for each @xmath273 and @xmath333 , the vector @xmath334 with the element in @xmath94 , equivalence class of @xmath335 . since the tangent bundle @xmath265 and the trivial bundle @xmath336 admit as structure group , respectively , the linear group @xmath337 and the trivial group @xmath338 , and since we have identified @xmath266 with a closed subgroup of @xmath328 , the bundle @xmath332 admits @xmath314 as structure group . following ehresmann s notations , we will denote it by @xmath339 , where @xmath102 is the principal bundle of projective frames in the fibres of @xmath332 . for each @xmath273 , we denote by @xmath219 the equivalence class of @xmath340 , where @xmath341 is the zero vector at @xmath40 . we may now define the subset @xmath155 of elements @xmath342 which map the element @xmath151 into the image of @xmath192 , and observe that it is the total space of a principal bundle @xmath275 . a cartan connection , in the sense of definition [ defcartanconnection ] , is a connection form @xmath25 on the principal bundle @xmath271 such that the form @xmath164 induced on @xmath155 has zero kernel . the soldering isomorphism can still be considered as an isomorphism of the tangent bundle @xmath265 , since we have identified @xmath265 with an open dense subset of @xmath53 . a _ projective connection _ is a cartan connection whose soldering isomorphism is the identity of @xmath265 . cartan and ehresmann have shown that projective connections exist on any manifold @xmath2 . we follow the presentation of kobayashi @xcite . on the space @xmath343 , with coordinates @xmath344 , let @xmath345 be the quadratic form @xmath346 its signature is @xmath347 . let @xmath348 be the subgroup of @xmath349 of elements which leave the quadratic form @xmath345 unchanged , and let @xmath350 be the cone , subset of @xmath343 , defined by the equation @xmath351 . let us call _ rays _ the straight lines through the origin in @xmath343 . any element @xmath352 maps a ray onto another ray , and a ray contained on @xmath350 on another ray contained in @xmath350 . the set of rays is the projective space @xmath353 . therefore the group @xmath354 acts on @xmath353 and that action leaves invariant the image @xmath355 of @xmath350 by the projection @xmath356 . that action is transitive on @xmath355 . we denote by @xmath357 the ray defined by the equations @xmath358 , and by @xmath359 its stabilizer . we will identify @xmath355 , which is called the _ mbius space _ , with @xmath360 . let @xmath361 be the affine hyperplane in @xmath343 defined by the equation @xmath362 . the affine injective map , defined on @xmath317 , with values in @xmath343 , @xmath363 with @xmath364 has @xmath361 as image and maps the sphere @xmath365 onto @xmath366 . since @xmath361 meets each ray contained in @xmath350 at a unique point , @xmath366 is diffeomorphic to @xmath355 . therefore , by composition with the projection @xmath356 , we obtain a diffeomorphism of @xmath367 onto the mbius space @xmath355 . let @xmath368 be the affine hyperplane in @xmath343 defined by the equation @xmath369 . it meets each ray contained in @xmath350 at a unique point , except the ray defined by the equations @xmath370 . the smooth injective map , defined on @xmath268 and with values in @xmath343 , @xmath371 with @xmath372 has @xmath373 as image . composed with the projection @xmath356 , that map yields a smooth injective map of @xmath268 into the mbius space @xmath355 , whose image is @xmath374 minus one point , the missing point corresponding to the ray defined by the equations @xmath370 . the image , by that map , of the origin of @xmath268 is the point @xmath375 . moreover , that map is equivariant with respect to the actions of the orthogonal group @xmath376 , on @xmath268 and on @xmath343 , with the convention that an element @xmath377 is identified with the element @xmath378 which maps @xmath379 on @xmath380 . when the mbius space @xmath355 is identified with the sphere @xmath367 , the injective map @xmath381 defined above is the inverse of the stereographic projection from @xmath367 minus its south pole onto @xmath268 . now let @xmath382 be an @xmath32-dimensional smooth riemannian manifold . we consider the direct sum of two copies of the trivial bundle @xmath383 and of the tangent bundle @xmath265 , in the following order : @xmath384 . to shorten the notation , we will denote that bundle by @xmath385 . each element of the total space of that bundle is a triple @xmath386 , with @xmath387 and @xmath388 and @xmath389 , its projection on the base @xmath2 being the projection of @xmath334 . let @xmath345 be the quadratic form , defined on the fibres of that bundle , @xmath390 and @xmath350 be the subspace of @xmath391 defined by the equation @xmath351 . we consider the projective bundle @xmath392 , quotient of @xmath393 by the equivalence relation of colinearity in the fibres . we have denoted by @xmath394 the image of the zero section in the bundle @xmath391 . the image of @xmath395 by the projection of @xmath393 onto @xmath396 is the total space of a fibre bundle @xmath397 , with base the manifold @xmath2 , and with standard fibre the mbius space @xmath355 . the tangent bundle @xmath265 can be considered as an open , dense subset of @xmath53 by identifying each vector @xmath389 with the equivalence class of @xmath398 . since the manifold @xmath2 is equipped with a riemannian metric , the tangent bundle @xmath265 admits @xmath376 as structure group . this is true for the bundle @xmath397 too . but as seen above , @xmath376 can be considered as a subgroup of @xmath348 , which acts on the standard fibre @xmath355 . the bundle @xmath397 therefore has @xmath399 as structure group and , using ehresmann s notation , we will denote it by @xmath400 . the total space @xmath102 of the corresponding principal bundle is the space of _ conformal frames _ on the riemannian manifold @xmath382 . for each @xmath273 , we denote by @xmath219 the equivalence class of @xmath401 , where @xmath341 is the zero vector at @xmath40 . the subset @xmath155 of elements @xmath342 which map the element @xmath357 into the image of @xmath192 is the total space of a principal bundle @xmath275 . a cartan connection , in the sense of definition [ defcartanconnection ] , is a connection form @xmath25 on the principal bundle @xmath271 such that the form @xmath164 induced on @xmath155 has zero kernel . the soldering isomorphism can still be considered as an isomorphism of the tangent bundle @xmath265 , since we have identified @xmath265 with an open dense subset of @xmath53 . conformal connection _ is a cartan connection whose soldering isomorphism is the identity of @xmath265 . cartan and ehresmann have shown that conformal connections exist on any manifold @xmath2 . the first very important application of the notion of connection is probably to be found in the theory of general relativity , in which the levi - civita connection associated to the pseudo - riemannian structure on space - time plays a key role . according to the title of his paper @xcite , sur les espaces connexion affine et la thorie de la relativit gnralise , . cartan was probably , for a large part , motivated by possible physical applications when he investigated the properties of connections . he explains how , in the framework of classical , non relativistic mechanics , a gravitation field ( _ i.e. _ an acceleration field ) can be included in the geometry of space - time by the use of an appropriate affine connection . we present here this idea in the more modern language of ehresmann . our manifold @xmath2 is the non relativistic space - time . for simplicity we assume that @xmath402 ( we take into account only one dimension for space ) . the choice of a galilean frame and of units for time and length allows us to identify @xmath2 with @xmath403 , with coordinates @xmath404 . the lie group @xmath0 , which will be called the _ affine galileo group _ , is the group of affine transformations of @xmath403 of the form @xmath405 an element @xmath27 of @xmath0 is therefore a triple @xmath406 , and the composition law of @xmath0 is @xmath407 let @xmath150 be the linear galileo group , _ i.e. _ the subgroup of elements of @xmath0 of the form @xmath408 . the homogeneous space @xmath236 can be identified with @xmath403 , with coordinates @xmath409 . the bundle @xmath339 is simply the tangent bundle @xmath265 , considered as an affine bundle , its structure group being restricted to the affine galileo group ( instead of the full affine group ) . the corresponding principal bundle @xmath271 is the bundle of affine galilean frames in @xmath265 . its total space is simply the product @xmath410 . the subset of elements in @xmath102 which map the origin of @xmath403 in the image of the zero section of @xmath265 is @xmath411 . it is the total space of a principal bundle @xmath275 . an affine galilean connection is determined by a connection form @xmath25 on @xmath102 , with values in the lie algebra @xmath138 of @xmath0 , which induces on @xmath155 a form @xmath164 with zero kernel . in addition we impose to that form to be such that the corresponding soldering isomorphism is the identity of @xmath265 . the lie algebra @xmath412 can be identified with @xmath413 , by means of the basis @xmath414 which corresponds to the coordinates @xmath415 on @xmath0 . taking into account the equivariance properties of @xmath25 and the fact that the corresponding soldering isomorphism is the identity of @xmath265 , we obtain @xmath416 where @xmath417 are the coordinates on @xmath410 , @xmath22 and @xmath293 being two smooth functions on @xmath2 ( therefore depending only on the coordinates @xmath404 ) . the development of the trajectory @xmath418 of a particle is a straight line if and only if the functions @xmath22 and @xmath293 , on which the connection form depend , satisfy @xmath419 we assume now that there exists on @xmath2 a gravity field @xmath27 ( which may eventually depend on time @xmath56 and space location @xmath40 ) . the equation of motion of a material particle submitted to that gravitational field is @xmath420 therefore , if we choose @xmath421 and @xmath422 , the development of the trajectory of any material particle submitted to the gravity field @xmath27 , but to no other forces , is a straight line . when @xmath27 is a constant , the corresponding connection is integrable : a nonlinear change of coordinates in space - time eliminates the gravity force ( in agreement with einstein s thought experiment in which an observer in a lift in free fall no more feels the gravity force ) . this is no more true when @xmath27 is not constant . it would be interesting to develop the example in which @xmath27 is the acceleration field in kepler s problem ( in a non relativistic space - time of dimension 3 ) ; for a suitably chosen connection , the developments of the keplerian trajectories of the planets should be straight lines . let @xmath423 be a symplectic manifold . a _ prequantization _ of that symplectic manifold is a principal bundle @xmath424 with base @xmath2 and with the circle @xmath425 as structure group , endowed with a connection @xmath18-form @xmath25 whose curvature is @xmath426 . since the lie algebra of @xmath425 can be identified with @xmath427 , the connection form @xmath25 can be considered as a _ contact form _ on @xmath33 . according to a theorem by b. kostant @xcite and j .- m . souriau @xcite , there exists a prequantization of @xmath423 if and only if the cohomology class of @xmath426 is integer . various uses of connections are made in the mathematical description of mechanical systems . let us consider a hamiltonian system , depending of some parameters , which for any fixed value of these parameters , is completely integrable . the motion of the system , for a fixed value of the parameter , is quasi - periodic on a lagrangian torus of phase space . at a certain time , the parameters vary slowly , describe a closed loop in the space of values of the parameters , and after taking again their initial values , remain constant . the motion of the system becomes again quasi - periodic on the same lagrangian torus , but with a change of phase ( the hannay and berry phase ) . this change of phase is interpreted as the holonomy of an ehresmann connection in the works of marsden , montgomery and ratiu @xcite . several different approaches have been used for the mathematical description of mechanical systems with constraints . in one of these approaches , the configuration space of the system is a smooth manifold and the constraints are described by a vector ( or sometimes an affine ) sub - bundle @xmath42 of the tangent bundle @xmath428 . the admissible motions of the system are smooth curves @xmath48 in @xmath345 such that , for all @xmath56 , @xmath429 j. koiller @xcite considered systems where the configuration space @xmath345 is the total space of a principal bundle over a base @xmath36 , with a lie group @xmath0 as structure group , and where the constraint @xmath42 is a connection on that principal bundle . let us consider a mechanical system in which some geometric constraints can be acted on , as a function of time , in order to control the motion of the system . for example , a cat in free fall can change the shape of her body to try ( generally with success ) to reach the ground on her feet . for the mathematical description of such systems @xcite , we use a manifold @xmath345 as configuration space , and a smooth submersion @xmath430 onto another manifold @xmath431 ( the space of shapes of the cat s body , or more generally the space of possible states of the active constraint ) . the dynamical properties of the system ( other than those used to change the value of the active constraint ) are described by a lagrangian @xmath432 . let @xmath433 be the dual bundle of the vertical sub - bundle @xmath434 . we can identify @xmath433 with the quotient bundle @xmath435 . let @xmath436 , @xmath437 and @xmath438 be the projections . when the lagrangian @xmath439 is @xmath440 there is on the bundle @xmath441 an ehresmann connection ( called the dynamical connection ) which can be used to determine the way in which an infinitesimal change of the state of the active constraint , represented by a vector tangent to @xmath431 , affects the motion of the mechanical system . let us call _ kinetic connection _ the ehresmann connection , on the bundle @xmath430 , for which the horizontal lift at @xmath442 of a vector @xmath443 is the unique @xmath444 , orthogonal ( with respect to @xmath27 ) to the vertical subspace @xmath445 , such that @xmath446 . the dynamic connection is characterized by the two properties : ( iii ) the horizontal lift at @xmath447 of a vector @xmath448 with respect to the dynamical connection projects on @xmath345 onto the horizontal lift at @xmath449 of @xmath334 with respect to the kinetic connection ; the horizontal lift to @xmath433 of any smooth vector field on @xmath431 , with respect to the dynamical connection , is an infinitesimal automorphism of the poisson structure of @xmath433 . the famous maxwell s equations are usually written ( see for example @xcite ) @xmath450 where @xmath451 is the electric field , @xmath452 the magnetic induction , @xmath453 the diplacement current , @xmath454 the magnetic field , @xmath455 the electric charge density and @xmath456 the current density . moreover there are constitutive equations which link @xmath451 and @xmath453 , @xmath452 and @xmath454 , @xmath457 let us introduce the @xmath458-form on space - time ( in which @xmath459 are the coordinates , with respect to some galilean reference frame , some chosen units of time and length and some orthonormal frame in space ) @xmath460 and @xmath461 let us set @xmath462 then we have @xmath463 where @xmath464 is the hodge operator on the @xmath465-dimensional pseudo - riemannian manifold space - time . the velocity of light in vacuum is @xmath466 maxwell s equations become @xmath467 maxwell s equations can be made even more beautiful : the electromagnetic @xmath458-form @xmath37 on space - time can be considered as the curvature form of a connection on a principal bundle , with space - time as base and the circle @xmath468 as structure group ( this idea was first introduced by h. weyl @xcite ) . the @xmath458-form @xmath37 should then be considered as taking its values in the lie algebra @xmath469 of @xmath470 . the connection form @xmath471 such that @xmath472 , is not unique : we may add a closed @xmath18-form ( gauge transformation ) . the first maxwell s equation , @xmath473 , is automatically satisfied . the second maxwell s equation becomes @xmath474 where @xmath318 is the covariant exterior differential operator . gauge theories generalize maxwell s theory of electromagnetism written in terms of connections . they use a principal bundle with space - time as base and a non - abelian group as structure group ( @xmath475 in the so - called _ standard model _ ) . they introduce a connection @xmath18-form @xmath471 on that bundle and lead , for the curvature form @xmath37 of the connection , to field equations similar to maxwell s equations : @xmath476 and @xmath477 where @xmath478 is the covariant differential with respect to the connection @xmath471 and @xmath479 a current which generalizes the @xmath465-dimensional current density of maxwell s theory . many thanks to paulette libermann , for her very careful reading of this text and her many suggestions for improvements . and again my warmest thanks to the organizers and supporting institutions of this great international conference .

we examine the influence of environmental interactions on simple quantum systems by obtaining the exact reduced dynamics of a qubit coupled to a one - dimensional spin bath . in contrast to previous studies , both the qubit - bath coupling and the nearest neighbor intrabath couplings are taken as the spin - flip xx - type . we first study the rabi oscillations of a single qubit with the spin bath prepared in a spin coherent state , finding that nonresonance and finite intrabath interactions have significant effects on the qubit dynamics . next , we discuss the bath - induced decoherence of the qubit when the bath is initially in the ground state , and show that the decoherence properties depend on the internal phases of the spin bath . by considering two independent copies of the qubit - bath system , we finally probe the disentanglement dynamics of two noninteracting entangled qubits . we find that entanglement sudden death appears when the spin bath is in its critical phase . we show that the single - qubit decoherence factor is an upper bound for the two - qubit concurrence .

1403.4900

the quantum dynamics of a single qubit or central spin coupled to a spin environment @xcite has been widely studied theoretically in several different areas , including quantum information sciences @xcite , quantum decoherence @xcite , and excitation energy transfer @xcite . one of the most promising candidates for quantum computation , solid - state spin systems , are inevitably coupled to their surrounding environment , usually through interactions with neighboring nuclear spins @xcite . the coupling of a qubit to a spin bath can in general lead to non - markovian behavior @xcite , causing the usual markovian quantum master equations to fail for such models . most recently , it was demonstrated for the first time that a spin bath can assist coherent transport in a two - level system @xcite . fully understanding the role played by a spin environment is an interesting and important issue . one commonly studied qubit - spin bath system is the so - called spin - star network @xcite , in which a preferred central spin is coupled homogeneously to a spin bath without intrabath interactions . a more realistic type of environment takes the form of quantum interacting spin chains @xcite , where the decay of the qubit s coherence is found to be related to the critical properties of the spin environments . most prior work making use of such an environment considered qubit - spin bath coupling of the ising form , which is spin conserving . as a result , it is much easier to analytically obtain the full dynamics of the system , in contrast to the situation where a spin - flip coupling is present . exceptions include refs . @xcite , where the authors considered the spin - flip xx - type qubit - bath coupling but with a spin bath having homogeneous self - interactions , and ref . @xcite , where the authors use @xmath0-dmrg to study the reduced dynamics of a qubit coupled locally to an xxz spin chain via the heisenberg - type qubit - bath interaction . it should be noted that , in general , both the spin - star network and the homogeneously coupled spin bath can be treated by introducing collective angular momentum operators which facilitates the analytical treatment . in this work , we will focus on a more realistic system with a ( not necessarily uniform ) spin - flip qubit - bath interaction as well as short range xx - type intrabath interactions . to our knowledge , the exact dynamics of such a model , which is one step closer to faithfully representing environmental spins interacting via fully general heisenberg - type interactions , has not been obtained before . the collapse and revival ( cr ) behavior of rabi oscillations of a qubit coupled to a single bosonic field mode , described by the jaynes - cummings ( jc ) model , is a fundamental consequence of field quantization and provides a much - studied illustration of the quantum nature of qubit - field systems @xcite . using a correspondence between the jc model and a spin - star network with a large number of spins , it is found in ref . @xcite that within a certain parameter regime , cr phenomena also appear in a qubit - big spin model . @xcite goes beyond the resonant jc model to the nonresonant dicke model , and notes that the dynamics depends on the sign of the detuning between the qubit and field frequency . in this work , we extend the model studied in ref . @xcite to the nonresonant case with a self - interacting spin bath modeled by the periodic xx spin chain . it is found that both nonzero detuning and the nearest neighbor coupling within the xx bath can have an effect on the qubit s dynamics . in particular , the interplay between nonresonance and intrabath interaction is able to reproduce cr behavior even for a spin bath with a relatively small number of sites . in addition , the dynamics of entanglement in many - body systems has recently been studied from different perspectives @xcite . as interacting quantum spin systems are believed to be paradigmatic for quantum information processing @xcite , their entanglement dynamics has attracted much attention @xcite . in prior works , the dynamical behavior of pairwise entanglement is found to be related to quantum phase transitions of the spin chains . another emerging focus is on the evolution of the entanglement of a pair of qubits exposed to noisy environments . in a seminal work , yu and eberly @xcite found that the markovian dynamics of the entanglement between two qubits coupled to individual bosonic baths can behave in sharp contrast to single qubit decoherence : the pairwise entanglement of two initially entangled two - level atoms suddenly disappears in a finite time proportional to the spontaneous lifetime of single atoms , while the single atom coherence only vanishes asymptotically . this phenomenon is called entanglement sudden death ( esd ) . more recently , it has been shown in the same setup that there exists a revival of the vanished entanglement if non - markovian effects are taken into account @xcite . as mentioned earlier , the non - markovian behavior caused by the spin environment may result in novel dynamics of the pairwise entanglement of two qubits each coupled to their own spin bath , as observed in a locally interacting qubit - spin - bath system @xcite . esd and subsequent revivals have also been observed to occur in two qubits when they are coupled to _ classical _ interacting spin baths @xcite and external fields @xcite , the latter of which has been demonstrated experimentally @xcite , and to stochastic noise sources @xcite . furthermore , understanding the relation between decoherence and disentanglement is believed to be of importance both for the foundations of quantum mechanics and practical applications in quantum information science @xcite . in this work , by taking the bath s initial state as the ground state of the xx chain , we first study the decoherence of a single qubit immersed in the xx bath . the decoherence dynamics is found to depend on the internal phases of the xx bath . the short time dynamics of the decoherence factor behaves like a gaussian , with the decay rate only depending on the filling number of the ground state of the bath . by considering two copies of our qubit - bath systems , we further study the disentanglement dynamics of the two initially entangled qubits coupled to their individual baths . we analytically show that the concurrence is bounded from above by the decoherence factor of a single qubit at all times . the initial bell state considered suffers from esd when the xx bath is in its critical phase . we also obtain the disentanglement time in the sudden death region and find that esd always occurs earlier than the onset of decoherence in a single qubit . the rest of the paper is structured as follows : in sec . ii , we introduce our model hamiltonian and describe how to obtain the exact qubit - bath time - dependent wavefunctions in the momentum space of the xx spin chain . the components of the bloch vector of the qubit are obtained by tracing out the bath degrees of freedom over these total wavefunctions . the results for the nonresonant and interacting cases are presented . in sec . iii , we study single qubit decoherence in a single qubit - bath system and disentanglement of two initially entangled qubits in two independent qubit - bath systems . conclusions are drawn in sec . our model consists of a single qubit coupled to a spin bath of @xmath1 spins-@xmath2 via the hamiltonian @xmath3 where @xmath4 and @xmath5 are the pauli matrices for spin @xmath6 in the spin bath and the central spin , respectively . @xmath7 is the energy difference of the two levels of the single qubit , and @xmath8 is the interaction between bath spins @xmath9 and @xmath6 . an external magnetic field @xmath10 in the spin bath is also included . the single qubit is coupled with spin @xmath6 in the spin bath via xx - type interactions with coupling strength @xmath11 . we introduce the collective angular momentum operator @xmath12 , where @xmath13 . note that our @xmath14 takes the same form as that in ref . @xcite and @xcite . however , there is no intrabath interaction in ref . @xcite and uniform intrabath interactions in ref . @xcite . in this work , we will choose as the spin bath a periodic one - dimensional chain with nearest neighbor interaction @xmath15 , namely , an xx spin chain with periodic boundary conditions . for this system , it can be easily checked that the total magnetization @xmath16 is a good quantum number . however , the total angular momentum @xmath17 of the spin bath is not conserved due to either the finite interaction @xmath18 or the inhomogeneous coupling @xmath11 . a spin coherent state of the spin bath , which lives in the @xmath19 subspace and is parameterized by the unit vector @xmath20 , can be written as @xcite @xmath21 with @xmath22 , and @xmath23 ( @xmath24 ) are the fully symmetric dicke states @xcite , which are simultaneous eigenstates of @xmath25 and @xmath26 with eigenvalues @xmath27 and @xmath28 . to study the rabi oscillations of the qubit , the initial state is chosen as the product state @xmath29 . ( the down state will be denoted by @xmath30 ) . it will be convenient to work in the interaction picture with respect to @xmath31 . the energy levels and eigenstates of @xmath32 can be obtained by using the jordan - wigner transformation @xmath33 , where @xmath34 are fermionic operators . @xmath35 then describes a qubit immersed in a spinless fermion bath , @xmath36 with the string operators @xmath37 . one can define two projection operators , @xmath38 and @xmath39 , which project onto subspaces where the total fermion number operator @xmath40 has even or odd eigenvalues @xmath41 . for even or odd @xmath41 , anti - periodic @xmath42 or periodic boundary conditions @xmath43 , respectively , are imposed on the fermions . as a result , we can introduce the following two sets of fourier transformations , @xmath44 where @xmath45 and @xmath46 are fourier modes with wave - numbers surviving in @xmath47 and @xmath48 , respectively . now @xmath49 is diagonalized as @xmath50 with @xmath51 the single particle spectrum . by direct calculation , we arrive at the interaction picture hamiltonian @xmath52,\end{aligned}\ ] ] where h.c . stands for the hermitian conjugate . since @xmath53 is a conserved quantity , it is sufficient to study time evolution from the states @xmath54 . in the interaction picture , the state evolved from the initial state in eq . ( [ initial ] ) then reads @xmath55 , with @xmath56 . in general , the structure of this state is highly complicated due to the non - conservation of the total angular momentum of the spin bath ( see appendix a ) . to this end , we represent the dicke states in terms of the momentum space fermion operators : @xmath57 where @xmath58 for even ( odd ) @xmath59 . here @xmath60 is the vacuum state of the fermions , which corresponds to the state with all bath spins in their down states @xmath61 . the function @xmath62 is the slater determinant made up of plane waves . therefore , @xmath63 can be written as a linear combination of free fermion states . in the following we treat even @xmath59 or odd @xmath59 separately . + + ( 1 ) @xmath64 . + it is easily seen that the most general form of @xmath63 is @xmath65 where @xmath66 and @xmath67 are coefficients to be determined by the time - dependent schrdinger equation @xmath68 . after a straightforward calculation ( see appendix b ) , we arrive at the following two sets of equations of motion for the coefficients @xmath69 and @xmath70 @xmath71 @xmath72 where the auxiliary function @xmath73 is defined to be @xmath74 here @xmath75 is the string of length @xmath76 , @xmath77 , where @xmath78 has been removed . note that the qubit - bath coupling configuration @xmath79 is completely incorporated into the the @xmath73-functions . for simplicity , we take uniform coupling @xmath80 in the following numerical calculations . in this case , eq . ( [ ff ] ) can be factorized as @xmath81 , where @xmath82 is the interaction - independent part of the @xmath73-function . ( [ d1 ] ) and ( [ b1 ] ) imply that @xmath10 and @xmath7 only enter the equations of motion through their sum @xmath83 , the detuning . the initial values of the @xmath69s and @xmath70s can be read off from eq . ( [ dicke ] ) : @xmath84 note that the above equations also include the case of @xmath85 , where there are no @xmath86-arguments for @xmath87 . the corresponding @xmath82 function is defined by @xmath88 . + + ( 2 ) @xmath89 . + similarly , the time - evolved states for odd @xmath59 are of the form @xmath90 where the coefficients @xmath91 and @xmath92 obey the same sets of equations of motion eqs . ( [ d1]-[b1 ] ) , except the number of arguments for the @xmath91s and @xmath92s change . to get an intuitive understanding of the dynamics , we first consider the non - interacting case @xmath93 . in this case the total angular momentum @xmath25 is conserved and the analytical expression for the bloch vector @xmath94 and the qubit purity @xmath95 can be easily calculated ( see appendix a ) . it turns out that @xmath96 and @xmath96 depend on both the detuning @xmath83 and the qubit energy difference @xmath7 , but @xmath97 and @xmath98 depend only on @xmath83 . 1 shows the dynamics of these four quantities in the resonant case @xmath99 , where clear cr behavior appears for the polarization dynamics @xmath97 . this observation has been recently made in ref . @xcite via a correspondence between the non - interacting qubit - spin - bath system and the jc model at large @xmath1 . as in the jc model @xcite , in the collapse regime @xmath100 the polarization only undergoes very small oscillations with nearly vanishing amplitudes , but is accompanied by a maximum of the purity . we will refer to such cr behavior as ` conventional ' cr dynamics observed in the qubit - field system . this behavior can be understood from examining the dynamics of @xmath96 and @xmath101 . for example , @xmath96 always vanishes for @xmath102 ( fig . [ res](a ) ) while @xmath101 reaches its maximum in the collapse regime , indicating the approximate creation of a pure state @xmath103 with the qubit pointing along the @xmath104 direction . we observe that @xmath7 controls the frequency of rotation of the bloch vector in the @xmath105 plane in the conventional collapse region . it was argued in ref . @xcite that the correspondence between the qubit - spin bath model and the jc model only holds for the parameter regime @xmath106 , and may break down for @xmath107 . in fig . [ res](d ) , we display the dynamics for @xmath108 with all the other parameters the same as in fig . [ res](a ) . we see that the cr dynamics still survives and that the behavior of @xmath97 and purity is almost the same as in fig . [ res](a ) , but with the qubit evolving into state @xmath109 in the collapse regime . dynamics of @xmath96 ( red ) , @xmath101 ( green ) , @xmath97 ( blue ) and purity @xmath98 ( black ) in the resonant case @xmath110 for four different sets of parameters : ( a ) : @xmath111 , ( b ) : @xmath112 , ( c ) : @xmath113 , ( d ) @xmath114 . other parameters : @xmath115 . ] ( a ) dynamics of @xmath96 ( red ) , @xmath101 ( green ) , @xmath97 ( blue ) and purity @xmath98 ( black ) in the nonresonant case . parameters : @xmath116 ; ( b ) magnification of the interval @xmath117 $ ] , where approximate pure states are sustained during the evolution . ] fig . [ nonres ] shows the results for the nonresonant and noninteracting case @xmath118 . from examining different values of the detuning @xmath83 , we observe that it controls both the amplitude and period of the oscillations of the envelope of @xmath97 . larger values of @xmath119 lead to longer periods and smaller amplitudes of these oscillations , as can also be seen from eq . ( [ szj0 ] ) . interestingly , the cr dynamics emerges even for a relatively small number of bath spins @xmath120 , which would not occur in the resonant case . however , this is not the conventional cr dynamics as seen in the resonant case . for @xmath121 , there is a collapse region for @xmath122 and @xmath123 at @xmath124 , where both the purity and @xmath97 suffer from rapid oscillations between @xmath125 and @xmath126 . more interesting dynamics appears at @xmath127 ( fig . [ nonres](b ) ) , where the purity undergoes small oscillations but remains close in absolute value to unity . we will refer to the behavior in both these regions as ` unconventional ' cr dynamics . unlike in the resonant case , where the pure state of the qubit rotates in the @xmath105 plane , here the qubit moves along the surface of the northern hemisphere of the bloch sphere . for finite @xmath18 , although a closed form solution to the equations of motion can not be obtained , we have been able to solve eqs . ( [ d1]-[b1 ] ) numerically for finite @xmath1 . to carry out the integration in a reasonable amount of time , it is necessary to solve for the auxiliary @xmath82-functions beforehand , and we were able to write a recursive function to do so . this is the most time consuming step in the numerics , and prevented us from examining systems with larger @xmath1 . based upon the zeros of the @xmath82-functions , one can also decouple the system of equations in both eqs . ( [ d1]-[b1 ] ) . doing so allows each component to be solved in parallel , resulting in a great speedup of the numerical integration . the three components of the bloch vector can be calculated from eq . ( [ psieven ] ) and eq . ( [ psiodd ] ) , and their expressions in terms of the coefficients @xmath69 , @xmath70 , @xmath91 and @xmath92 are listed in appendix a. numerical results for finite intrabath interaction and finite detuning with @xmath128 and @xmath129 are plotted in fig . [ jhw](a ) . comparing with fig . 2 , we see that conventional cr dynamics in @xmath97 reappears after introducing finite intrabath coupling . this is shown more clearly in fig . [ jhw](b ) . except for the facts that the oscillation center of @xmath97 moves to around @xmath125 , and that the peaks of the purity are below 1.0 in this case , the dynamics in the collapse region closely mimics that of in fig . this is an intriguing observation , considering that our spin bath contains only a relatively small number of spins ( @xmath120 ) . however , not every peak of the purity is accompanied by the collapse of @xmath97 , as can be seen from the first and third peaks in fig . [ jhw](a ) . in fig . [ jhw](c ) , we display the same plot for @xmath130 . we see that increasing the coupling strength @xmath131 causes the period between successive peaks of the purity to decrease . these revivals in the purity also appear to wash out more quickly than they do in fig . [ jhw](a ) . finally , we note that for @xmath93 the polarization dynamics @xmath97 is symmetric under changing the sign of the detuning @xmath132 for fixed @xmath7 ( cf . ( [ szj0 ] ) ) . however , this is not the case for finite @xmath18 and only the relative sign between @xmath83 and @xmath18 is relevant ( see the end of appendix a for an example of @xmath133 ) . the dynamics of the qubit in the nonresonant regime , with @xmath129 , when the interbath interactions are present . in ( a ) and ( b ) , @xmath134 and in ( c ) , @xmath135 . ( b ) displays the conventional collapse region which appears when the interactions are turned on even for @xmath120 . other parameters : @xmath136 , @xmath137 . ] in the previous section , we studied the reduced dynamics of a single qubit coupled to an interacting spin bath , with the bath initially prepared in the spin coherent state . now we consider two such copies of the qubit - bath system , between which there is no direct interaction : @xmath138 with @xmath139 and @xmath140 given by eq . ( [ single ] ) , and the upper index indicating the operators for copies @xmath141 or @xmath142 . as shown in ref . @xcite , the reduced dynamics of the two qubits can be determined completely from that of only one of the two copies . explicitly , let @xmath143 denote the reduced density matrix of the two qubits . assuming a separable initial state @xmath144 , @xmath143 can be written in the basis of the two qubits @xmath145 as @xmath146 where @xmath147 is determined by the dynamics of each part through @xmath148 for an initial state @xmath149 of copy @xmath150 . thus , in the following we focus on the dynamics of a single qubit coupled to a single bath described by eq . ( [ single ] ) , and drop the upper index @xmath150 for simplicity . obviously , the dynamics depends on the initial state of the bath @xmath151 . in this section , we will choose the ground state of the isolated xx chain as the bath s initial state . we set @xmath152 and @xmath153 henceforth . for a chain with a finite number of sites , lowering @xmath10 from the critical field @xmath154 to @xmath155 causes @xmath156 level crossings , which correspond to transitions between different parity sectors . this leads to the kosterlitz - thouless phase transition in the thermodynamic limit . correspondingly , the ground states are filled by @xmath157 @xmath158-fermions ( @xmath159-fermions ) for @xmath76 even ( odd ) . the detailed ground state structure of the periodic xx chain can be found in appendix c. in order to get a better understanding of the relationship between decoherence and disentanglement , which is believed to be of importance for both the foundation of quantum mechanics and practical applications of quantum information @xcite , we first study the decoherence dynamics of a single qubit . we suppose that initially the qubit is not entangled with the xx bath . that is , @xmath160 is the ground state of the xx chain . the coefficients @xmath161 and @xmath162 satisfy @xmath163 . note that @xmath164 is not an eigenstate of eq . ( [ single ] ) , so the evolution starting from @xmath165 is non - trivial . the spin - flip qubit - bath coupling will induce entanglement between the qubit and spins in the xx chain . the ground state @xmath164 characterized by the number of excitations @xmath41 will evolve into superpositions of states within subspaces with @xmath166 excitations due to the interaction term @xmath14 . let us first focus on the case of @xmath167 , where the excitation number @xmath168 . depending on the parity of the filling number @xmath76 , we will use indices ` @xmath169 ' or ` @xmath170 ' to indicate quantities corresponding to odd or even @xmath76 . for @xmath171 ( [ gc ] ) ) with @xmath76 odd , the most general form of @xmath172 will be @xmath173\prod^{m+1}_{l=1}c^\dag_{k_l}|0\rangle\nonumber\\ & & + \sum_{k_1< ... <k_{m+2}}a_1d(k_1, ... ,k_{m+2};t)|\bar{1}\rangle\prod^{m+2}_{l=1}d^\dag_{k_l}|0\rangle\nonumber\\ & & + \sum_{k_1< ... <k_{m}}a_{\bar{1}}c(k_1, ... ,k_{m};t)|1\rangle\prod^{m}_{l=1}d^\dag_{k_l}|0\rangle.\end{aligned}\ ] ] by similar calculations as in the spin coherent state case , we find that @xmath69 and @xmath70 obey the same set equations of motion as eqs . ( [ d1 ] ) and ( [ b1 ] ) . in addition , the equations of motion for @xmath174 and @xmath175 read @xmath176 @xmath177 the nonzero initial values of these variables can be read from eq . ( [ gd ] ) and eq . ( [ gc ] ) @xmath178 all other initial values of @xmath174 , @xmath69 , @xmath175 and @xmath70 vanish . for @xmath179 , the ground state is the fully polarized state with @xmath180 , which can be included in eq . ( [ psiio ] ) . the reduced density matrix of the qubit , and hence the @xmath181 factors in eq . ( [ rhow ] ) , can be obtained by tracing out the bath degrees of freedom . note that @xmath182=0 $ ] , so the trace can be taken over @xmath159-fermions and @xmath158-fermions independently : @xmath183 with the schrdinger picture state given by @xmath184 . by using eq . ( [ rhow ] ) , we obtain the @xmath181 factors @xmath185 with all other elements vanishing . a similar analysis can be carried out for @xmath76 even where @xmath186 ( see appendix c ) . we recognize the decoherence factor of a single qubit @xcite @xmath187 from eq . ( [ rhow ] ) as @xmath188 whose absolute value is bounded by @xmath189 , corresponding to complete decoherence and no loss of coherence , respectively . in fig . [ weak ] , we plot the temporal evolution of the decoherence factor @xmath190 for different values of nearest neighbor couplings @xmath131 in the weak qubit - bath coupling regime @xmath191 . the revival of coherence occurs since the bath is finite . the loss of coherence is modest for the smallest value of @xmath192 , and @xmath190 approaches zero only for values @xmath193 , namely , in the critical regime of the xx chain . as the intrabath interaction strength @xmath194 is increased , the coherence of the qubit is first suppressed , as can be seen by comparing the curves for @xmath195 and @xmath196 , and then enhanced , as can be seen by examining the curves for @xmath197 and @xmath198 . these observations indicate that the relationship between decoherence and interaction strength is not straightforward . [ gtalpha](a ) displays the short time behavior of the decoherence factor @xmath190 . it can be seen that when @xmath199 is small , @xmath190 decays as a gaussian @xmath200 for different values of @xmath194 as blue dots which were numerically fit to @xmath190 for small times . interestingly , we note that @xmath201 exhibits plateaus as a function of @xmath194 . this behavior can be understood from second - order time - dependent perturbation theory in the qubit - bath coupling @xmath202 . it turns out that the initial gaussian rate is given by ( see appendix d for the derivation ) the decoherence factor is shown for different values of nearest neighbor coupling @xmath203 in the weak coupling regime with @xmath204 . other parameters are : @xmath205 . ] ( a ) the short time behavior of the decoherence factor @xmath190 . for @xmath206 , the curves collapse onto five gaussians with decay rates that depend only the value of @xmath76 , the filling number . two curves within each group of two lines have been plotted with their own line style , and for @xmath207 , the five groups can be made out . ( b ) the dependence of the decay rate @xmath201 on the intrabath interaction strength @xmath194 . the red line is the theoretical calculation of eq . ( [ alpha ] ) , and the blue dots are numerical fits at small times to the decoherence factor for different values of @xmath203 . other parameters are : @xmath205 . ] the first maximum the decoherence factor reaches , @xmath208 , is plotted against the value of @xmath194 for different values of @xmath209 corresponding to different qubit - bath couplings . other parameters are : @xmath205 . ] @xmath210 where @xmath211 or @xmath212 for initial states with @xmath171 or @xmath213 . this perturbative result is displayed as the red set of plateaus in fig . [ gtalpha](b ) . note that the @xmath82-functions , and hence the rate @xmath201 , have nothing to do with the system s parameters and only depend on the filling number @xmath76 , which explains the presence of plateaus . in fig . [ gtalpha](a ) , two curves are plotted for each value of @xmath76 , and for @xmath207 the ten curves are seen to collapse into five groups corresponding to the five different values of @xmath76 . the first divergence within a group can be seen for the @xmath214 sector , where the curves for @xmath215 and @xmath198 separate past @xmath216 . the short time behavior of the decoherence factor for intermediate qubit - bath coupling , with @xmath217 , and strong qubit - bath coupling , with @xmath218 ( not shown here ) , is similar to that of weak qubit - bath coupling . in particular , they are also characterized by gaussian behavior . however , the behavior at longer times , as @xmath194 is increased , changes . in order to quantitatively compare the behavior of @xmath190 in these three regimes , in fig . [ rmax ] we plot the value of the first maximum of the decoherence factor @xmath208 ( aside from the initial value @xmath219 ) as a function of the intrabath interaction strength @xmath194 for the three coupling regimes examined above . this quantity is representative of the extent to which coherence is maintained in the qubit @xcite . for all three regimes , when the bath is in a polarized phase @xmath220 , the decoherence factor returns to unity after one oscillation , and indeed it appears that the periodic revival of the coherence continues for all times . all of the curves exhibit a sudden drop at @xmath221 , reflecting the transition to the critical phase of the bath . interestingly , in the critical region @xmath222 , @xmath208 displays markedly different behavior in each regime . for weak qubit - bath coupling , @xmath208 behaves non - monotonically and oscillates about high values . however , @xmath208 appears to monotonically increase for @xmath217 and @xmath218 , albeit very slowly for the latter . these results seem to show that strong intrabath interaction strength suppresses the decoherence of the qubit , a result has been observed for a qubit coupled to a spin bath with homogeneous self - interaction @xcite . as another means of assessing the effect of intrabath interactions on the coherence of the qubit , in fig . [ decmin ] we plot the time @xmath199 at which the decoherence factor @xmath190 reaches its first minimum as a function of @xmath194 for the three qubit - bath coupling regimes . we focus on the critical region @xmath222 , as the time of the first minimum for @xmath223 is much larger and does not display much variation . interestingly , the green curve , for which @xmath204 , initially decreases sharply with each successive sector and then displays a global minimum at @xmath224 , where the decoherence disappears the quickest . for intermediate and strong qubit - bath coupling @xmath217 and @xmath218 , as @xmath194 is increased and successive magnetization sectors of the bath are encountered , the time @xmath199 of the first minimum drops . but unlike the weak coupling case , @xmath199 increases monotonically within each sector , agreeing with previous results that strong interactions within the bath suppress decoherence of the qubit @xcite . the time @xmath199 at which the decoherence factor @xmath190 reaches its first minimum is plotted against the intrabath interaction strength @xmath194 for the three qubit - bath coupling regimes . other parameters are : @xmath205 . ] now , we turn to the study of the disentanglement of two qubits interacting with independent xx - baths . we focus on one type of initial state for the two - qubit system @xmath225 with @xmath201 real and @xmath226 . from eq . ( [ aabb ] ) , it follows that the time evolved reduced density matrix for the two qubits reads @xmath227 with @xmath228 where we have assumed that the two environments are identical , so that @xmath229 . we use the concurrence @xcite to measure the bipartite entanglement between the two qubits . the concurrence is defined as @xmath230 where @xmath231 is the largest eigenvalue of the matrix @xmath232 . the concurrence for state @xmath143 reads @xmath233 where @xmath190 is the single qubit decoherence factor in eq . ( [ rt ] ) . concurrence dynamics for the initial state @xmath234 for different values of nearest neighbor coupling @xmath203 . the corresponding decoherence factors , plotted in fig [ weak ] , are an upper bound for these curves . when entanglement sudden death is present , it always occurs before the qubits individual decoherence factors reach a minimum . other parameters are : @xmath235 . ] the time @xmath199 at which the concurrence first vanishes is plotted against intrabath interaction strength @xmath194 . other parameters are : @xmath205 . ] in fig . [ conc ] , we plot the evolution of concurrence as a function of @xmath199 from the maximally entangled bell state @xmath236 . in order to make a comparison between the disentanglement dynamics and the decoherence of a single qubit , we set all parameters to be the same as those in fig . the concurrence appears to be bounded from above by the corresponding decoherence factor @xmath190 for all time . for @xmath237 , the xx chain is in the fully polarized state along the @xmath238 direction . the concurrence shows regular oscillations about a high value and never vanishes in this regime . in fact , we have seen that the two - qubit concurrence @xmath239 exactly coincides with the single qubit decoherence factor @xmath190 here . this relation can be understood by examining eq . ( [ ct ] ) : for @xmath240 , we have @xmath241 ; while for @xmath242 , we have @xmath243 hence , the concurrence is always bounded from above by @xmath190 . when the spin bath is in a polarized state , we always have @xmath244 , as can be seen from eq . ( [ psiio ] ) . therefore , @xmath245 for the bell state with @xmath246 . a similar conclusion also holds for the other type of entangled state @xmath247 . this relationship between decoherence and disentanglement has also been observed before in spin - boson type models when the two qubits are coupled to separate bosonic baths @xcite . on the other hand , esd always exists in the critical regime @xmath193 . this is consistent with the result for two distant qubits coupled locally to an xxz spin chain via isotropic heisenberg qubit - bath coupling @xcite , where it was found that esd is absent in the ferromagnetic or polarized phase of the spin bath . we also observe that esd always occurs earlier than the minimum of the corresponding single - qubit decoherence factor , a result in agreement with the case of qubits coupled to independent bosonic baths @xcite . in the sudden death region , revival of the entanglement appears a period of time after disentanglement , which is also observed in ref . this revival phenomenon is induced by the non - markovian nature of the spin bath @xcite . in order to compare the entanglement dynamics with the decoherence of a single qubit , we plot the disentanglement time , which is defined as the time when the concurrence first vanishes , as a function of intrabath coupling @xmath194 in the sudden death region @xmath222 in fig . the time @xmath199 until esd occurs decreases as sectors with lower filling factors @xmath76 are encountered , but within each sector , @xmath199 increases as @xmath203 is increased , although this effect is only pronounced for weak qubit - bath coupling @xmath204 . this plot displays many similarities to fig . [ decmin ] . however , for the green curve with @xmath204 , increasing @xmath194 only causes an increase in @xmath199 with each sector , in contrast to the nonmonotonic behavior of the corresponding curve in fig . [ decmin ] . also , the intrabath coupling strengths at which the entanglement dies the fastest and at which the decoherence is minimized most quickly are not the same . in spite of these differences , there is a qualitative agreement between figs . [ decmin ] and [ esd ] , indicating that the decoherence dynamics of a single qubit and the entanglement dynamics of two noninteracting qubits are linked , especially when the qubits are strongly coupled to their respective baths . in this work , we studied the reduced dynamics of a specific qubit - spin - bath model . unlike the spin conserving ising - type qubit - bath coupling utilized in most previous works , we considered an xx - type spin - flip qubit - bath coupling , which complicates the analytical analysis , since the system - bath interaction term does not commute with the rest of the hamiltonian . in addition , we model interactions in the bath by introducing nearest - neighbor xx - type couplings among the bath spins . such a model may be more physical than the non - interacting ` spin star ' and homogeneously interacting spin baths which have been examined before , but is more difficult to treat analytically . however , by mapping this xx chain into momentum space via the jordan - wigner transformation , we have shown how to obtain the equations of motion for the time - dependent total wavefunctions in momentum space . the reduced dynamics of a single qubit is then obtained by tracing out the bath degrees of freedom . using the above results , we first studied the rabi oscillations of the qubit with the bath initially prepared in a spin coherent state . interestingly , the interplay between off - resonance and intrabath interactions was found to produce conventional collapse and revival behavior even for a relatively small spin bath size . we further discussed the bath - induced decoherence of a single qubit with the bath s initial state taken to be its ground state . we found that the decoherence properties of the qubit depend on the internal phases of the xx bath . specifically , the short time decay rate of the decoherence factor only depends on the filling number of the bath ground state . this result was confirmed through second order time - dependent perturbation theory . finally , we considered two independent copies of such qubit - bath subsystems and studied the disentanglement dynamics of two initially entangled qubits . the two qubits are always entangled if the xx bath is in its polarized state , whereas entanglement sudden death appears in the critical phase . qualitative similarities were observed between the time dependence of the two - qubit entanglement and the single - qubit decoherence factor , and we showed that the concurrence is bounded from above by the decoherence factor of the single qubit . however , for spin - boson type systems , it has been shown that this is not the case when the two - qubits share a common bosonic bath @xcite . we believe that such a relation between single - qubit decoherence and two - qubit disentanglement might also break down for two qubits coupled to a common xx bath . this deserves further study based on our model system . the xx spin chain in our model is equivalent to the one - dimensional bose - hubbard model in the hard - core limit @xcite , which can be realized using a cold atomic gas contained in an optical lattice @xcite . correspondingly , the qubit - spin - bath coupling @xmath14 can be mapped to a conventional spin - boson coupling within the rotating - wave approximation , @xmath248 , where @xmath249 are bosonic operators . regarding the methodology used in this work , we note that the only requirement for our formalism is the conservation of the total magnetization @xmath53 . thus , our method can also be applied to the nonuniform heisenberg type qubit - spin - bath coupling @xcite , where the total magnetization is conserved . our results suggest that turning on interactions among bath spins can have markedly different effects on the decoherence and entanglement properties of the central spins , depending on how strongly they are coupled to their environment . therefore , this work may be of relevance to all efforts aimed at using such systems to construct quantum information processing devices . + + * acknowledgements : * we acknowledge support from nsf grant no . che-1058644 and aro - muri grant no . w911nf-08 - 1 - 0124 . a.n . was supported by the program in plasma science and technology at princeton university . the calculations in this work were performed at the tigress high performance computer center at princeton . 99 n. v. prokofev and p. c. e. stamp , rep . . phys . * 63 * , 669 ( 2000 ) . a. hutton and s. bose , phys . rev . a * 69 * , 042312 ( 2004 ) . d. rossini , t. calarco , v. giovannetti , s. montangero , and r. fazio , phys . a * 75 * , 032333 ( 2007 ) . n. y. yao , l. jiang , a. v. gorshkov , z .- x . gong , a. zhai , l .- m . duan , and m. d. lukin , phys . lett . * 106 * , 040505 ( 2011 ) . j. j. pla , k. y. tan , j. p. dehollain , w. h. lim , j. j. l. morton , d. n. jamieson , a. s. dzurak , and a. morello , nature * 489 * , 541 ( 2012 ) . e. a. chekhovich , m. n. makhonin , a. i. tartakovskii , a. yacoby , h. bluhm , k. c. nowack , and l. m. k. vandersypen , nat . mater . * 12 * , 494 ( 2013 ) . r. j. warburton , nat . * 12 * , 483 ( 2013 ) . xiang , s. ashhab , j. q. you , and f. nori , rev . 85 * , 623 ( 2013 ) . b. urbaszek , x. marie , t. amand , o. krebs , p. voisin , p. maletinsky , a. hgele , and a. imamoglu , rev . phys . * 85 * , 79 ( 2013 ) . h. p. breuer , d. burgarth , and f. petruccione , phys . b * 70 * , 045323 ( 2004 ) . y. hamdouni , m. fannes , and f. petruccione , phys . b * 73 * , 245323 ( 2006 ) . m. bortz , s. eggert , c. schneider , r. stbner , and j. stolze , phys . b * 82 * , 161308 ( 2010 ) . d. stanek , c. raas , and gtz s. uhrig , phys . b * 88 * , 155305 ( 2013 ) . b. erbe and j. schliemann , phys . b * 81 * , 235324 ( 2010 ) . e. barnes , l. cywinski , and s. das sarma , phys . 109 * , 140403 ( 2012 ) . l. ratschbacher , c. sias , l. carcagni , j. m. silver , c. zipkes , and m. khl , phys . lett . * 110 * , 160402 ( 2013 ) . wang and s. takahashi , phys . rev . b * 87 * , 115122 ( 2013 ) . a. faribault and d. schuricht , phys . lett . * 110 * , 040405 ( 2013 ) . s. j. balian , g. wolfowicz , j. j. l. morton , and t. s. monteiro , phys . b * 89 * , 045403 ( 2014 ) . a. olaya - castro , c. f. lee , f. f. olsen , and n. f. johnson , phys rev . b * 78 * , 085115 ( 2008 ) . i. sinayskiy , a. marais , f. petruccione , and a. ekert , phys . * 108 * , 020602 ( 2012 ) . n. wu and y. zhao , j. chem . phys . * 139 * , 054118 ( 2013 ) . v. v. dobrovitski , g. d. fuchs , a. l. falk , c. santori , and d. d. awschalom , annu . . condens . matter phys . * 4 * , 23 ( 2013 ) . d. d. awschalom , l. c. bassett , a. s. dzurak , e. l. hu , and j. r. petta , science * 339 * , 1174 ( 2013 ) . h. krovi , o. oreshkov , m. ryazanov , and d. a. lidar , phys . a * 76 * , 052117 ( 2007 ) . e. ferraro , h .- breuer , a. napoli , m. a. jivulescu , and a. messina , phys . b * 78 * , 064309 ( 2008 ) . s. lorenzo , f. plastina , and m. paternostro , phys . a * 87 * , 022317 ( 2013 ) . s. dooley , f. mccrossan , d. harland , m. j. everitt , and t. p. spiller , phys . a * 87 * , 052323 ( 2013 ) . a. faribault and d. schuricht , phys . b * 88 * , 085323 ( 2013 ) . h. t. quan , z. song , x. f. liu , p. zanardi , and c. p. sun , phys . rev . lett . * 96 * , 140604 ( 2006 ) . x. z. yuan , h. s. goan , and k. d. zhu , phys . b * 75 * , 045331 ( 2007 ) . r. walters , s. r. clark , and d. jaksch , int . j. quant * 8 * , 271 ( 2010 ) . c. y. lai , j. t. hung , c. y. mou , and p. chen , phys . b * 77 * , 205419 ( 2008 ) . d. w. luo , h. q. lin , j. b. xu , and d. x. yao , phys . rev . a * 84 * , 062112 ( 2011 ) . g. l. giorgi and t. busch , phys . a * 86 * , 052112 ( 2012 ) . j. gea - banacloche , phys . . lett . * 65 * , 3385 ( 1990 ) . a. alvermann , l. bakemeier , and h. fehske , phys . a * 85 * , 043803 ( 2012 ) . l. amico , r. fazio , a. osterloh , and v. vedral , rev . phys . * 80 * , 517 ( 2008 ) . m. a. nielson and i. l. chuang , _ quantum computation and quantum information _ ( cambridge university , cambridge , england , 2000 ) . l. amico , a. osterloh , f. plastina , r. fazio , and g. m. palma , phys . a * 69 * , 022304 ( 2004 ) . v. subrahmanyam , phys . a * 69 * , 034304 ( 2004 ) . a. sende , u. sen , and m. lewenstein , phys . a * 72 * , 052319 ( 2005 ) . j. jing and z .- g . l , phys . b * 75 * , 174425 ( 2007 ) . g. b. furman , v. m. meerovich , and v. l. sokolovsky , phys . a * 77 * , 062330 ( 2008 ) . h. wichterich and s. bose , phys . a * 79 * , 060302 ( 2009 ) . z. chang and n. wu , phys . a * 81 * , 022312 ( 2010 ) . d. i. tsomokos , m. j. hartmann , s. f. huelga , and m. b. plenio , new . * 9 * , 79 ( 2007 ) . b. alkurtass , g. sadiek , and s. kais , phys . a * 84 * , 022314 ( 2011 ) . w. qin , c. wang , and g. l. long , phys . a * 87 * , 012339 ( 2013 ) . ting yu and j. h. eberly , phys . lett . * 93 * , 140404 ( 2004 ) . b. bellomo , r. lo franco , and g. compagno , phys . 99 * , 160502 ( 2007 ) . c. wang and q .- h . chen , new j. phys . * 15 * , 103020 ( 2013 ) . a. v. zampetaki and f. k. diakonos , phys . * 109 * , 090402 ( 2012 ) . r. lo franco , b. bellomo , e. andersson , and g. compagno , phys . a * 85 * , 032318 ( 2012 ) . xu , k. sun , c .- f . li , x .- y . xu , g .- c . guo , e. andersson , r. lo franco and g. compagno , nat . commun . * 4 * , 2851 ( 2013 ) . j. h. wilson , b. m. fregoso , and v. m. galitski , phys . b * 85 * , 174304 ( 2012 ) . d. tolkunov , v. privman , and p. k. aravind , phys . rev . a * 71 * , 060308 ( 2005 ) . k. ann and g. jaeger , phys . b * 75 * , 115307 ( 2007 ) . j. t. barreiro , p. schindler , o. ghne , t. monz , m. chwalla , c. f. roos , m. hennrich , and r. blatt , nat . * 6 * , 943 ( 2010 ) . kim , j .- c . lee , o. kwon , and y .- ho . kim , nat . * 8 * , 117 ( 2012 ) . f. t. arecchi , e. courtens , r. gilmore , and h. thomas , phys . rev . a * 6 * , 2211 ( 1972 ) . f. m. cucchietti , j. p. paz , and w. h. zurek , phys . a * 72 * , 052113 ( 2005 ) . a. relano , j. m. arias , j. dukelsky , j. e. garcia - ramos , and p. perez - fernandez , phys . a * 78 * , 060102(r ) ( 2008 ) . l. tessieri and j. wilkie , j. phys . a * 36 * , 12305 ( 2003 ) . c. m. dawson , a. p. hines , r. h. mckenzie , and g. j. milburn , phys . a * 71 * , 052321 ( 2005 ) . w. k. wootters , phys . lett . * 80 * , 2245 ( 1998 ) . d. p. s. mccutcheon , a. nazir , s. bose , and a. j. fisher , phys . a * 80 * , 022337 ( 2009 ) . f. benatti , r. floreanini , and m. piani , phys . lett . * 91 * , 070402 ( 2003 ) . j. jordan , r. ors , and g. vidal , phys . b * 79 * , 174515 ( 2009 ) . d. jaksch , c. bruder , j. i. cirac , c. w. gardiner , and p. zoller , phys . 81 * , 3108 ( 1998 ) . om mandel , m. greiner , a. widera , t. rom , t. w. hansch , and i. bloch , nature ( london ) * 425 * , 937 ( 2003 ) . in the non - interacting case @xmath93 with uniform coupling @xmath251 , the total angular momentum of the spin bath is conserved , so @xmath252 takes the simple form @xmath253 by applying the schrdinger operator eq . ( [ hi ] ) to the above equation , we obtain the following equations of motion for the coefficients @xmath254 and @xmath255 @xmath256 with initial conditions @xmath257 , @xmath258 and @xmath259 . the solutions are @xmath260,\nonumber\\ b_n(t)&=&-2i\tilde{g}_ne^{-\frac{i}{2}(h+\omega)t}\frac{\sin \frac{1}{2}\sqrt{4\tilde{g}_n^2+(h+\omega)^2}t}{\sqrt{4\tilde{g}_n^2+(h+\omega)^2}}.\end{aligned}\ ] ] the polarization dynamics is given by @xmath261\nonumber\\ & = & 1 - 8\sum^n_{n=0 } \frac{\tilde{g}^2_n|c_n|^2\sin^2\frac{t}{2}\sqrt{4\tilde{g}^2_n+(h+\omega)^2}}{4\tilde{g}^2_n+(h+\omega)^2 } .\nonumber\\\end{aligned}\ ] ] the other two components can be calculated directly from eq . ( [ j0 ] ) @xmath262,\nonumber\\ \langle\sigma_y(t)\rangle&=&-2\im[e^{-i\omega t}\sum^n_{n=1}c^*_{n-1}c_nb^*_{n-1}a_n].\end{aligned}\ ] ] we also monitor the purity dynamics of the qubit @xmath263 although @xmath96 and @xmath101 depend on both @xmath83 and @xmath7 , @xmath97 and @xmath98 depend only on @xmath83 . also , note that @xmath97 is symmetric under changing the sign of the detuning : @xmath132 . for finite @xmath18 and/or non - uniform coupling @xmath11 , the total angular momentum @xmath264 of the spin bath is not conserved . so @xmath63 will be driven into other @xmath265-subspaces under the action of @xmath266 : @xmath267 which complicates the analysis . in this case , the dynamics of the bloch vector can be obtained from eq . ( [ psieven ] ) and eq . ( [ psiodd ] ) as @xmath268,\nonumber\\ & & \langle\sigma_y(t)\rangle=-2\im[e^{-i\omega t}z(t)],\end{aligned}\ ] ] with @xmath269 and @xmath270\nonumber\\ & + & \sum^{\frac{n}{2}-1}_{n=0}|c_{2n+1}|^2[\sum_{k_1< ... <k_{2n+1}}|b'(k_1, ... ,k_{2n+1};t)|^2\nonumber\\ & & -\sum_{k_1< ... <k_{2n+2}}|d'(k_1, ... ,k_{2n+2};t)|^2].\end{aligned}\ ] ] unlike the non - interacting and uniform coupling case , the dynamics of @xmath97 is no longer symmetric under @xmath132 , as can be seen from the a simple example of @xmath133 ; i.e. , for a spin bath made up of only two spins , where an analytical expression can be obtained ( with equal qubit - bath coupling @xmath271 ) : @xmath272 with @xmath273 . note that only the relative sign between @xmath83 and @xmath18 is relevant . eqs . ( [ d1]-[b1 ] ) can be derived by inserting eq . ( [ psieven ] ) into the time - dependent schrdinger equation @xmath274 after acting with @xmath266 on @xmath63 , only two terms survive : @xmath275 @xmath276 can be calculated as @xmath277 with @xmath278\sum_{p_1< ... <p_{n+1}}\nonumber\\ & & s^*(p_1, ... ,p_{n+1};j , j_1, ... ,j_n)d^\dag_{p_1} ... d^\dag_{p_{n+1}}|0\rangle\nonumber\\ & = & \sum_{j_1<j_2< ... <j_{n+1}}[\sum^n_{l=1 } g_{j_l } s(k_1, ... ,k_n;j_1,j_2, ... ,\underline{j_l}, ... ,j_n)]\nonumber\\ & & \sum_{p_1< ... <p_{n+1}}s^*(p_1, ... ,p_{n+1};j_1,j_2, ... ,j_{n+1})d^\dag_{p_1} ... d^\dag_{p_{n+1}}|0\rangle\nonumber\\ & = & \sum_{p_1< ... ,p_{n+1};k_1, ... ,k_{n};\{g_j\})d^\dag_{p_1} ... d^\dag_{p_{n+1}}|0\rangle\nonumber\\\end{aligned}\ ] ] here @xmath75 is the string of length @xmath76 with the element @xmath78 removed from the sequence @xmath279 , and the coupling configuration dependent auxiliary @xmath73-function is given by eq . ( [ ff ] ) in the main text . by invoking eq . ( [ sch ] ) , we obtain eq . ( [ d1 ] ) . the equations of motion for the @xmath69s ( eq . ( [ b1 ] ) ) can be derived similarly . for a periodic xx chain described by @xmath49 with @xmath152 and @xmath280 , lowering @xmath10 from @xmath154 to @xmath155 will cause @xmath156 level crossings or parity changing at the following @xmath156 critical fields : @xmath281 note that @xmath282 , so that the region @xmath283 is divided into the following intervals : @xmath284 for fields within interval @xmath285 and with @xmath76 even , the ground state is filled by @xmath286 @xmath158-fermions @xmath287 similarly , for odd @xmath76 , the ground state is filled by @xmath286 @xmath159-fermions @xmath288.\end{aligned}\ ] ] for fields within interval @xmath289 , the ground state is always the fully polarized state with all spins pointing in the @xmath238 direction . in the fermionic picture , this state corresponds to the completely occupied state @xmath290 which is filled by the @xmath159-fermions and has an energy @xmath291 . depending on whether @xmath292 or @xmath293 , the ground state for fields within interval @xmath294 will be either @xmath295 or @xmath296 . for the initial state eq.([phi0 ] ) with @xmath186 ( even @xmath76 ) , @xmath297 has a similar form as eq . ( [ psiio ] ) , @xmath298\prod^{m+1}_{l=1}d^\dag_{k_l}|0\rangle\nonumber\\ & & + \sum_{k_1< ... <k_{m+2}}a_1d'(k_1, ... ,k_{m+2};t)|\bar{1}\rangle\prod^{m+2}_{l=1}c^\dag_{k_l}|0\rangle\nonumber\\ & & + \sum_{k_1< ... <k_{m}}a_{\bar{1}}c'(k_1, ... ,k_{m};t)|1\rangle\prod^{m}_{l=1}c^\dag_{k_l}|0\rangle.\end{aligned}\ ] ] the coefficients @xmath299 , @xmath91 , @xmath300 and @xmath92 satisfy a set of equations of motion of the same form as eq.([d1]-[b1 ] ) and eq.([c]-[a ] ) , except that @xmath76 is now even . similar expressions for the @xmath181-factor also hold , with @xmath174 , @xmath69 , @xmath175 and @xmath70 replaced by @xmath299 , @xmath91 , @xmath300 and @xmath92 . when the qubit - bath coupling is small compared with the external field @xmath10 and interaction @xmath18 between neighboring spins in the bath ( @xmath301 ) , then standard second order time - dependent perturbation theory can be applied . the time - dependent wavefunction in the interaction picture can be written to second order in the perturbation as @xmath302 with @xmath266 and @xmath165 given by eq . ( [ hi ] ) and eq . ( [ phi0 ] ) , respectively . for @xmath303even , the initial bath state @xmath304 , with @xmath305 direct calculation gives @xmath306 comparing with eq . ( [ psiio ] ) in the main text , we have @xmath307 for @xmath308 . from eq . ( [ w ] ) , we finally obtain @xmath309 for short times @xmath310 , we have latexmath:[\[\begin{aligned }

this paper establishes that so - called instrumental variables enable the identification and the estimation of a fully nonparametric regression model with berkson - type measurement error in the regressors . an estimator is proposed and proven to be consistent . its practical performance and feasibility are investigated via monte carlo simulations as well as through an epidemiological application investigating the effect of particulate air pollution on respiratory health . these examples illustrate that berkson errors can clearly not be neglected in nonlinear regression models and that the proposed method represents an effective remedy .

1308.2836

many statistical data sets involve covariates @xmath0 that are error - contaminated versions of their true unobserved counterpart @xmath1 . however , the measurement error often does not fit the classical error structure @xmath2 with @xmath3 independent from @xmath1 . a common occurrence is , in fact , the opposite situation , in which @xmath4 with @xmath5 independent from @xmath0 , a situation often referred to as berkson measurement error [ @xcite , @xcite , @xcite ] . a typical example is an epidemiological study in which an individual s true exposure @xmath1 to some contaminant is not observed , but instead , what is available is the average concentration @xmath0 of this contaminant in the region where the individual lives . the individual - specific @xmath1 randomly fluctuate around the region average @xmath0 , resulting in berkson errors . existing approaches to handle data with berkson measurement error [ e.g. , @xcite , @xcite ] unfortunately require the distribution of the measurement error to be known , or to be estimated via validation data , which can be costly , difficult or impossible to collect . ( in classical measurement error problems , the distribution of the error can be identified from repeated measurements via a kotlarski - type equality [ @xcite , @xcite ] . however , such results do not yet exist for berkson - type measurement error . ) a popular approach to relax the assumption of a fully known distribution of the measurement error is to allow for some adjustable parameters in the distributions of the variables and their relationships , and solve for the parameter values that best reproduce various conditional moments of the observed variables , under the assumption that this solution is unique . this approach has been used , in particular , for polynomial specifications [ @xcite ] and , more recently , for a very wide range of parametric models [ wang ( @xcite ) ] . the present paper goes beyond this and provides a formal identification result and a general nonparametric regression method that is consistent in the presence of berkson errors , without requiring the distribution of the measurement error to be known a priori . instead , the method relies on the availability of a so - called instrumental variable [ e.g. , see chapter 6 in @xcite ] to recover the relationship of interest . for instance , in the epidemiological study of the effect of particulate matter pollution on respiratory health we consider in this paper , suitable instruments could include ( i ) individual - level measurement of contaminant levels that can even be biased and error - contaminated or ( ii ) incidence rates of diseases other than the one of interest that are known to be affected by the contaminant in question . our estimation method essentially proceeds by representing each of the unknown functions in the model by a truncated series ( or a flexible functional form ) and by numerically solving for the parameter values that best fits the observable data . although such an approach is easy to suggest and implement , it is a challenging task to formally establish that such a method is guaranteed to work in general . first , there is no guarantee that the solution ( i.e. , parameter values that best match the distribution of the observable data ) is unique . second , estimation in the presence of a number of unknown parameters going to infinity with sample size is fraught with convergence questions . can the postulated series represent the solution asymptotically ? is the parameter space too large to obtain consistency ? is the noise associated with estimating an increasing number of parameters kept under control ? our solution to these problems is two - fold . first , we target the most difficult obstacle by formally establishing identification conditions under which the regression function and the distribution of all the unobserved variables of the model are uniquely determined by the distribution of the observable variables . a second important aspect of our solution to the berkson measurement error problem is to exploit the extensive and well - developed literature on nonparametric sieve estimation [ e.g. , @xcite , @xcite , @xcite ] to formally address the potential convergence issues that arise when nonparametric unknowns are represented via truncated series with a number of terms that increases with sample size . these theoretical findings are supported by a simulation study and the usefulness of the method is illustrated with an epidemiological application to the effect of particulate matter pollution on respiratory health . we consider a regression model of the general form @xmath6 where the function @xmath7 is the ( unknown ) relationship of interest between @xmath8 , the observed outcome variable and @xmath1 , the _ unobserved _ true regressor , while @xmath9 is a disturbance . information regarding @xmath1 is only available in the form of an observable proxy @xmath0 contaminated by an error @xmath5 . equation ( [ eqz ] ) assumes the availability of an instrument @xmath10 , related to @xmath1 via an unknown function @xmath11 and a disturbance @xmath12 . our goal is to estimate the function @xmath13 in ( [ eqy ] ) nonparametrically and without assuming that the distribution of the measurement error @xmath5 is known . [ as by - products , we will also obtain @xmath11 and the joint distribution of all the unobserved variables . ] to this effect , we require the following assumptions , which are very common in the literature focusing on nonlinear models with measurement error [ e.g. , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ] . [ condindep]the random variables @xmath0 , @xmath5 , @xmath14 , @xmath12 are mutually independent . note that assumption [ condindep ] implies the commonly - made `` surrogate assumption '' @xmath15 , as can be seen by the following sequence of equalities between conditional densities : @xmath16 . [ condloc]the random variables @xmath5 , @xmath9 , @xmath17 are centered ( i.e. , the model s restrictions preclude replacing @xmath18 by @xmath19 for some nonzero constant @xmath20 , and similarly for @xmath9 and @xmath12 ; this includes either zero mean , zero mode or zero median , e.g. ) . as our approach relies on the availability of an instrument @xmath10 to achieve identification , it is instructive to provide practical examples of suitable instruments in common settings . although the use of instrumental variables has historically been more prevalent in the econometrics measurement error literature [ @xcite ] , instruments are gathering increasing interest in the statistics literature , especially in the context of measurement error problems [ see chapter 6 entitled `` instrumental variables '' in @xcite and the numerous references therein ] . note that instrument equation ( [ eqz ] ) is entirely analogous to ( [ eqy ] ) , the equation generating the main dependent variable . hence , the instrument is nothing but another observable `` effect '' caused by @xmath1 via a general nonlinear relationship @xmath11 . let us consider a few examples , which were inspired by some of the case studies found in @xcite , @xcite and @xcite . epidemiological studies . in these studies , the dependent variable @xmath8 is typically a measure of the severity of a disease or condition , while the true regressor @xmath21 is someone s true but unobserved exposure to some contaminant . the average concentration @xmath0 of this contaminant in the region where the individual lives is , however , observed . the error on @xmath0 is berkson - type because individual - specific @xmath1 typically randomly fluctuate around the region average @xmath0 . in this setup , multiple plausible instruments are available : a measurement of contaminant concentration in the individual s house ( these would be error - contaminated by classical errors , since the concentration at a given time randomly fluctuates around the time - averaged concentration which would be relevant for the impact on health ) . thanks to the flexibility introduced by the function @xmath11 in ( [ eqz ] ) , these measurements can even be biased . they can therefore be made with a inexpensive method ( that can be noisy and not even well - calibrated ) , making it practical to use at the individual level . hence , it is possible to combine ( i ) accurate , but expensive , region averages that are not individual - specific ( @xmath0 ) and ( ii ) inexpensive , inaccurate individual - specific measurements ( @xmath10 ) to obtain consistent estimates . another plausible instrument could be a measure of the severity of another disease or condition that is _ known _ to be caused by the contaminant . the fact that it is _ caused by _ the contaminant , introduces an error structure which is consistent with equation ( [ eqz ] ) . other measurable effects due to the contaminant ( e.g. , the results of saliva or urine tests for the presence of contaminants ) could also serve as instruments . clearly these measurements are not units of exposure , but the function @xmath11 can account for this . experimental studies . researchers may wish to study how an effect @xmath8 ( e.g. , the production of some chemical ) is related to some imposed external conditions @xmath0 ( e.g. , oven or reactor temperature ) , but the true conditions @xmath21 experienced by the sample of interest may deviate randomly from the imposed conditions ( e.g. , temperature may not be completely uniform ) . in this case , an instrument @xmath10 could be ( i ) another `` effect '' ( e.g. , the amount of another chemical ) that is known to be caused by @xmath21 or ( ii ) a measurement of @xmath1 that is specific to the sample of interest but that may be very noisy or even biased ( e.g. , it could be an easier - to - take temperature measurement after the experiment is completed and the sample has partly cooled down ) . self - reported data . @xcite have argued that individuals reporting data ( e.g. , their food intake , or exercise habits ) are sometimes aware of the uncertainty in their estimates of @xmath1 and , as a result , try to report an average @xmath0 over all plausible estimates consistent with the information available to them , thus leading to berkson - type errors , because the individuals try to make their prediction error independent from their report . in this setting , an instrument @xmath10 could be another observable outcome variable @xmath10 that is also related to @xmath1 . we now formally state conditions under which the berkson measurement error model can be identified with the help of an instrument . let @xmath22 , @xmath23 , @xmath24 and @xmath25 denote the supports of the distributions of the random variables @xmath8 , @xmath0 , @xmath1 and @xmath10 , respectively . we consider @xmath26 and @xmath10 to be jointly continuously distributed ( with @xmath27 , @xmath28 , @xmath29 and @xmath30 with @xmath31 ) . accordingly , we assume the following . [ conddens]the random variables @xmath32 admit a bounded joint density with respect to the lebesgue measure on @xmath33 . all marginal and conditional densities are also defined and bounded . we use the notation @xmath34 and @xmath35 to denote the density of the random variable @xmath36 and the density of @xmath36 conditional on @xmath37 , respectively . lower case letters denote specific values of the corresponding upper case random variables . next , as in many treatments of errors - in - variables models [ @xcite , @xcite , @xcite , @xcite , schennach ( @xcite ) ] , we require various characteristic functions to be nonvanishing . we also place regularity constraints on the two regression functions of the model . [ condinv]for all @xmath38 , @xmath39 \neq0 $ ] and for all @xmath40 , @xmath41 \neq0 $ ] ( where @xmath42 ) . [ condnodup]@xmath43 and @xmath44 are one - to - one ( but not necessarily onto ) . [ condcont]@xmath45 is continuous . assumption [ condnodup ] is somewhat restrictive when @xmath1 has a dimension larger or equal to the ones of @xmath8 ( or @xmath10 ) . fortunately , it is often possible to eliminate this problem by re - defining @xmath8 ( and @xmath10 ) to be a vector containing auxiliary variables in addition to the outcome of interest , in order to allow for enough variation in @xmath8 ( and @xmath10 ) to satisfy assumption [ condnodup ] . each of these additional variables need not be part of the relationship of interest per se , but does need to be affected by @xmath1 is some way . in that sense , such auxiliary variables would also be a type of `` instrument . '' our main identification result can then be stated as follows . ( note that the theorem also holds upon conditioning on an observed variable @xmath46 , so that additional , correctly measured , regressors can be straightforwardly included . ) [ thid]under assumptions [ condindep][condcont ] , given the true observed conditional density @xmath47 , the solution @xmath48 to the functional equation @xmath49 for all @xmath50 , @xmath51 , @xmath52 is unique ( up to differences on sets of null probability measure ) . a similar uniqueness result holds for the solution @xmath53 to @xmath54\\[-8pt ] & & \qquad = f_{x } ( x ) \int f_{\delta z } \bigl ( z - h \bigl ( x^{\ast } \bigr ) \bigr ) f_{\delta y } \bigl ( y - g \bigl ( x^{\ast } \bigr ) \bigr ) f_{\delta x^{\ast } } \bigl ( x^{\ast}-x \bigr ) \,dx^{\ast}.\nonumber\end{aligned}\ ] ] establishing this result demands techniques radically different from existing treatment of berkson error models , such as the spectral decomposition of linear operators [ see @xcite for a review ] , which are emerging as powerful alternatives to the ubiquitous deconvolution techniques that are typically applied in classical measurement error problems . the proof can be found in the and can be outlined as follows . assumption [ condindep ] lets us obtain the following integral equation relating the joint densities of the observable variables to the joint densities of the unobservable variables : @xmath55 from which equation ( [ eqfyxz ] ) follows directly . uniqueness of the solution is then shown as follows . equation ( [ eqpreindep ] ) defines the following operator equivalence relationship : @xmath56 where we have introduced the following operators : @xmath57 ( z ) & = & \int f_{y , z|x } ( y , z|x ) r ( x ) \,dx,\nonumber\\ { } [ f_{z|x^{\ast}}r ] ( z ) & = & \int f_{z|x^{\ast } } \bigl ( z|x^{\ast } \bigr ) r \bigl ( x^{\ast } \bigr ) \,dx^{\ast } , \nonumber\\ { } [ f_{z|x}r ] ( z ) & = & \int f_{z|x } ( z|x ) r ( x ) \,dx,\\ { } [ d_{y;x^{\ast}}r ] \bigl ( x^{\ast } \bigr ) & = & f_{y|x^{\ast } } \bigl ( y|x^{\ast } \bigr ) r \bigl ( x^{\ast } \bigr ) , \nonumber\\ { } [ f_{x^{\ast}|x}r ] \bigl ( x^{\ast } \bigr ) & = & \int f_{x^{\ast } |x } \bigl ( x^{\ast}|x \bigr ) r ( x ) \,dx \nonumber\end{aligned}\ ] ] for some sufficiently regular but otherwise arbitrary function @xmath58 . note that , in the above definitions , @xmath59 is viewed as a parameter ( the operators do not act on it ) and that @xmath60 is the operator equivalent of a diagonal matrix . next , we note that the equivalence @xmath61 also holds [ e.g. , by integration of ( [ eqleqlll ] ) over all @xmath50 ] . we can then isolate @xmath63 @xmath64 and substitute the result into ( [ eqleqlll ] ) to yield , after rearrangements , @xmath65 where all inverses can be shown to exist over suitable domains under our assumptions . equation ( [ eqdiag ] ) states that the operator @xmath66 admits a spectral decomposition . the operator to be `` diagonalized '' is defined in terms of observable densities , while the resulting eigenvalues @xmath67 ( contained in @xmath60 ) and eigenfunctions @xmath68 ( contained in @xmath69 ) provide the unobserved densities of interest . a few more steps are required to ensure uniqueness of this decomposition , which we now briefly outline . one needs to ( i ) invoke a powerful uniqueness result regarding spectral decompositions [ theorem xv 4.5 in @xcite ] , ( ii ) exploit the fact that densities integrate to one to fix the scale of the eigenfunctions , ( iii ) handle degenerate eigenvalues and ( iv ) uniquely determine the ordering and indexing of the eigenvalues and eigenfunctions . this last , and perhaps most difficult , step , addresses the issue that both @xmath68 and @xmath70 , for some one - to - one function @xmath71 , are equally valid ways to state the eigenfunctions that nevertheless result in different operators @xmath69 . to resolve this ambiguity , we note that for any possible operator @xmath72 satisfying ( [ eqdiag ] ) , there exist a unique corresponding operator @xmath63 , via equation ( [ eqgivexsx ] ) . however , only one choice of @xmath69 leads to an operator @xmath63 whose kernel @xmath73 satisfies assumption [ condloc ] . hence , @xmath73 , @xmath74 and @xmath75 are identified , from which the functions @xmath76 , @xmath77 , @xmath78 , @xmath45 and @xmath79 can be recovered by exploiting the centering restrictions on @xmath5 , @xmath9 and @xmath12 . an operator approach has recently been proposed to address certain types of nonclassical measurement error problems [ @xcite ] , but under assumptions that rule out berkson - type measurement errors : it should be emphasized that , despite the use of operator decomposition techniques similar to the ones found in @xcite ( hereafter hs ) , it is impossible to simply use their results to identify the berkson measurement error model considered here , for a number of reasons . first , the key condition ( assumption 5 in hs ) that the distribution of the mismeasured regressor @xmath0 given the true regressor @xmath1 is `` centered '' around @xmath1 does not hold for berkson errors . consider the simple case where the berkson measurement error is normally distributed and so are the true and mismeasured regressors . the distribution of @xmath0 given @xmath80 is a normal centered at @xmath81 . hence , there is absolutely no reasonable measure of location ( mean , mode , median , etc . ) that would yield the appropriate centering at @xmath82 that is needed in assumption 5 of hs . in addition , one can not simply replace the assumption of centering of @xmath0 given @xmath83 ( as in hs ) by a centering of @xmath1 given @xmath0 ( as would be required for berkson errors ) and hope that theorem 1 in hs remains valid . hs exploit the fact that , in a conditional density , there is no jacobian term associated with a change of variable in a conditioning variable ( here @xmath1 ) . however , with berkson errors , the corresponding change of variable would not take place in the conditional variables , and a jacobian term would necessarily appear , which makes the approach used in hs fundamentally inapplicable to the berkson case . solving this problem involves ( i ) using a different operator decomposition than in hs and ( ii ) using a completely different approach for `` centering '' the mismeasured variable.=1 a referee suggested an alternative argument ( formalized in the ) that makes a more direct connection with theorem 1 in hs but under the additional assumption that @xmath10 and @xmath1 have the same dimension . such an assumption is rather restrictive because it will often result in the assumption that @xmath11 is one - to - one ( assumption [ condnodup ] ) being violated . for instance , if @xmath1 is scalar and we have access to two instruments @xmath84 and @xmath85 such that neither @xmath86 $ ] nor @xmath87 $ ] are strictly monotone , then @xmath11 is not one - to - one for either instrument used in isolation . however , the mapping @xmath88,e [ z_{2}|x^{\ast } ] ) $ ] will typically be one - to - one , except for really exceptional cases . hence , allowing for the dimensions of @xmath1 and @xmath10 to differ is important . nevertheless , even assuming away this problem , such an approach still requires a different technique for centering @xmath1 than the one used in hs . that said , both hs and the current paper rely on operator spectral decomposition as an alternative to conventional convolution / deconvolution techniques , and it appears likely that these new techniques will find applications in a number of other measurement error models . observe that our identification result is also useful in a parametric and semi - parametric context , as it provides the confidence that , under simple conditions , the model is identified . rank conditions that would need to be verified on a case - by - case basis in any given parametric model are automatically implied by our identification results in a wide class of models . also , although @xmath0 is allowed to be random throughout , considering @xmath0 to be fixed poses no particular difficulty , since equation ( [ eqfyxz ] ) provides a valid conditional likelihood function in that case . as discussed in @xcite , a number of extensions of the method are possible : ( i ) relaxing the independence between @xmath0 and @xmath18 to allow for some heteroskedasticity in the measurement error and ( ii ) combining classical and berkson errors , a possibility considered in , for example , @xcite , @xcite , @xcite and @xcite . it can also be shown that some extensions are not plausible , such as assuming that both the measurement equation ( [ eqxs ] ) and the instrument equation ( [ eqz ] ) have a berkson error structure [ @xcite ] . a natural way to obtain a nonparametric estimator of the model is to substitute truncated series approximations into ( [ eqfyxz ] ) or ( [ eqfyxz3 ] ) for each of the unknown functions and construct a log likelihood function to be maximized numerically with respect to all coefficients of the series [ e.g. , @xcite ] . such sieve - based estimators have recently found applications in a variety of measurement error problems [ e.g. , @xcite , @xcite , @xcite , @xcite , among others ] . below we first define our estimator before establishing its consistency . we represent the regression functions @xmath7 and @xmath89 as @xmath90 where @xmath91 is some sequence ( indexed by the truncation parameters @xmath92 ) of progressively larger sets of basis functions indexed by @xmath93 while @xmath94 is a vector of coefficients to be determined . the @xmath95 could be some power series , trigonometric series , orthogonal polynomials , wavelets or splines , for instance . the double indexing by @xmath96 and @xmath92 is useful to allow for splines , where changing the number of knots modifies all the basis functions . a similar expansion in terms of basis functions @xmath97 ( with truncation parameter @xmath98 ) is used for the density of each disturbance @xmath99 , @xmath100 where @xmath101 is a vector of coefficients to be determined , and @xmath102 is a user - specified `` baseline '' function . the `` baseline '' function is convenient to reduce the number of terms needed in the expansion , when the approximate general shape of the density is known . it is not strictly needed , however , and can be set to 1 . either way , the method is fully nonparametric . a convenient choice of basis [ see @xcite ] is to take @xmath102 to be a gaussian and @xmath103 for any @xmath98 . an important distinction with the functions @xmath7 and @xmath11 is that some constraints have to be imposed on the densities . one constraint is needed to ensure centering ( assumption [ condloc ] ) , @xmath104 where , for some user - specified function @xmath105 , we define @xmath106 for instance , to impose zero mean on the disturbance @xmath107 , let @xmath108 . to impose zero median , let @xmath109 , where @xmath110 denotes an indicator function , while to impose zero mode , let @xmath111 ( a delta function derivative , in a slight abuse of notation ) . another constraint is needed to ensure unit total probability : @xmath112 . note that both types of constraints exhibit the computationally convenient property of being linear in the unknown coefficients . given the above definitions , we can define an estimator of all unknown functions based on a sample @xmath113 and equation ( [ eqfyxz ] ) [ a corresponding estimator based on equation ( [ eqfyxz3 ] ) can be derived analogously ] . let @xmath114 denote the minimizer of the sample log likelihood @xmath115 where @xmath116 @xmath117 , subject to @xmath118 for @xmath99 and subject to technical regularity constraints to be defined below . estimators are then given by @xmath119\\[-8pt ] \hat{f}_{v } ( v ) & = & \hat{f}_{v}^ { ( k_{v } ) } \bigl ( v,\hat{\theta}_{v}^ { ( k_{v } ) } \bigr ) \qquad\mbox{for } v= \delta x^{\ast } , \delta y,\delta z. \nonumber\end{aligned}\ ] ] this type of estimator falls within the very general class of sieve nonparametric maximum likelihood estimators ( mle ) , whose asymptotic theory has received considerable attention over the last few decades [ e.g. , @xcite , @xcite , @xcite ] . here , we parallel the treatment of @xcite and @xcite to establish the consistency of the above procedure . although the consistency of sieve - type estimators has been previously established in very general settings under some high - level assumptions , our contribution is to provide very primitive sufficient conditions for consistency for the class of models considered here . we first need to define the set in which the densities of interest reside . the formal proof of consistency of the estimator requires this set to be compact , although this requirement appears to have little impact in practice . in essence , compactness is helpful to rule out very extreme but rare events associated with very poor estimates . it is a standard regularity condition ; see , for example , @xcite , @xcite , @xcite . a well - known type of infinite - dimensional but compact sets are those generated via boundedness and lipschitz constraints in an @xmath120 space . here , we use a weighted lipschitz constraint in order to allow for densities supported on an unbounded set , while still maintaining compactness ( our treatment can be straightforwardly adapted to cover the simpler case where the variables are supported on finite intervals ) . following @xcite , we enforce restrictions that avoid too rapid divergences in the log likelihood . [ defsetf]let @xmath121 . let @xmath37 be finite and strictly positive . let @xmath122 be strictly positive and bounded function that is decreasing in @xmath123 , symmetric about @xmath124 and such that @xmath125 . let @xmath126 $ ] such that @xmath127 . let @xmath128 and @xmath129 be strictly positive and bounded functions with @xmath128 decreasing in @xmath123 and @xmath130 . let @xmath131 . we also define suitable norms and sets for the regression functions . here , we need to allow for functions that diverge to infinity at controlled rates toward infinite values of their argument . in analogy with any existing global measure of expected error , we also use a norm that downweights errors in the tails , which is consistent with the fact that the tails of a nonparametric regression function are always estimated with more noise , since there are fewer datapoints there . [ defsetg]let @xmath132 by some given strictly positive , bounded and differentiable weighting function . for any function @xmath133 , let @xmath134 where @xmath135 . let @xmath136 and @xmath137 where @xmath138 is a given positive function that is increasing in @xmath123 and symmetric about @xmath124 . we can now state the regularity conditions needed . [ condiid]the observed data @xmath139 are independent and identically distributed across @xmath140 . [ condball]we have @xmath141 and @xmath142 . [ conddense]the set of functions representable as series ( [ expf ] ) and ( [ expg ] ) are , respectively , dense in @xmath143 ( in the norm @xmath144 ) and @xmath145 ( in the norm @xmath146 ) . denseness results for numerous types of series are readily available in the literature [ e.g. , @xcite , @xcite ] . although such results are sometimes phrased in a mean square - type norm rather than the sup norm used here , lemma [ lemnorm2 ] below [ proven in @xcite ] establishes that , within the sets @xmath143 and @xmath145 , denseness in a mean square norm implies denseness in the norms we use . [ lemnorm2]let @xmath147 be a sequence in @xmath143 . then @xmath148 implies @xmath149 ( for @xmath143 and @xmath144 as in definition [ defsetf ] ) . we also need standard boundedness and dominance conditions . [ conddom]for any @xmath150 , @xmath151 for @xmath152 and @xmath153 as in definitions [ defsetg ] and [ defsetf ] , respectively . [ condmomex ] there exists @xmath154 such that @xmath155 < \infty$ ] , where @xmath156 for @xmath157 and @xmath158 as in definitions [ defsetf ] and [ defsetg ] , respectively . we can then state our consistency result [ proven in @xcite ] : [ thconsis]under assumptions [ conddens][condmomex ] , if @xmath159 , for @xmath160 , the estimators given by ( [ eqnest ] ) evaluated at the minimizer of ( [ eqnopti ] ) subject to ( [ eqnlincons ] ) , @xmath161 and @xmath162 and satisfying assumption [ conddom ] are such that @xmath163 , @xmath164 , @xmath165 , @xmath166 , @xmath167 , where the stared quantities denote the true values [ i.e. , the unique solution to ( [ eqfyxz ] ) ] . the practical implementation of the above approach necessitates the selection of the number of terms @xmath98 in each of the approximating series . theorem [ thconsis ] allows for a data - driven selection of the @xmath98 , since @xmath98 is allowed to be random . to select the @xmath98 , one can employ the bootstrap cross - validation model selection method based on the kullback leibler ( kl ) criterion , shown by @xcite to be consistent even when the number of candidate models grows to infinity with sample size ( as it is here ) . in this method , a fraction @xmath168 of the sample is excluded at random and the remaining @xmath169 fraction is used to estimate the model parameters with given numbers @xmath170 of terms in the corresponding series . the likelihood ( or kl criterion ) is then evaluated using the excluded fraction @xmath168 at the value of the estimated parameters found in the previous step . the process is repeated many times with different random partitions of the sample into fractions @xmath168 and @xmath171 , to obtain an average kl criterion with a sufficiently small variance ( which can be estimated from the kl criterion of each random partitions ) . this procedure is carried out for various trial choices of @xmath170 and the choice that yields the largest likelihood is selected . this method is consistent asymptotically ( as sample size @xmath172 ) as @xmath173 and @xmath174 and under some mild technical regularity conditions stated in @xcite . our nonparametric approach nests parametric and semiparametric models . these subcases can be easily implemented by replacing some , or all , of the nonparametric series approximations by suitable parametric models . it is possible to obtain convergence rates and limiting distribution results , along the lines of @xcite or @xcite , although we do not do so here due to space limitations [ stating suitable regularity conditions , even in high - level form , is rather involved , as seen in the supplementary material of @xcite , which covers a related but different measurement error model ] . it is , however , important to point out one important property . sieve nonparametric mle is optimal in the following sense : under suitable regularity conditions , any sufficiently regular semiparametric functional of the nonparametric sieve mle estimates is asymptotically normal and root @xmath175 consistent and reaches the semiparametric efficiency bound for that functional ; see theorem 4 in @xcite . this notion of optimality is a natural nonparametric generalization of the well - known efficiency of parametric maximum likelihood . we now investigate the practical performance and feasibility of the proposed estimator via a simulation example purposely chosen to be a difficult case . the data is generated as follows . the distribution of @xmath0 is a uniform distribution over @xmath176 $ ] ( implying a standard deviation of @xmath177 ) . we consider a thick - tailed @xmath178 distribution with 6 degrees of freedom scaled by @xmath179 as the distribution of @xmath18 . the standard deviation of the error @xmath5 is almost identical to the one of the `` signal '' @xmath0 , thus making this estimation problem exceedingly difficult . the distribution of @xmath9 is a logistic scaled by @xmath180 while the distribution of @xmath17 is a @xmath178 distribution with @xmath181 degrees of freedom scaled by @xmath182 . the regression function has the form=1 @xmath183=0 which is only finitely many times differentiable , thus limiting the convergence rate of its series estimator in the measurement - error - robust estimator ( the naive estimator would be less affected since it would `` see '' a smoothed version of this function ) . the instrument equation has a specification that is strictly convex and therefore tends to exacerbate the bias in many nonparametric estimators , @xmath184 a total of @xmath185 independent samples , each containing @xmath186 observations , were generated as above and fed into our estimator . for estimation purposes , the functions @xmath7 and @xmath11 are both represented by polynomials while the densities of @xmath5 , @xmath9 and @xmath12 are represented by a gaussian multiplied by a polynomial [ following @xcite , who establish that these choices satisfy a suitable denseness condition ] . the gaussian is centered at the origin , but its width is left as a parameter to be estimated . note that the functional forms considered are not trivially nested within the space spanned by the truncated sieve approximation . this was an intentional choice aimed at properly accounting for the nonparametric nature of the problem ( in which the researcher never has the fortune of selecting a truncated sieve fitting the true model exactly ) . the integral in equation ( [ eqfyxz ] ) is evaluated numerically by discretizing the integral as a sum over the range @xmath187 $ ] in intervals of @xmath188 . naive least - squares estimators ignoring measurement error ( i.e. , least - squares regressions of @xmath8 on @xmath0 and of @xmath10 on @xmath0 ) were used as a starting point for the numerical sieve optimization of the @xmath79 and @xmath45 functions , while the variances of the corresponding residuals were used to construct an initial gaussian guess for the optimization of all the error distributions . the simplex method due to @xcite ( also known as `` amoeba '' ) was used to carry out the numerical optimization of the log likelihood ( [ eqnopti ] ) with respect to all the parameters @xmath189 for @xmath190 and @xmath191 for @xmath192 simultaneously . the constraints that the estimated densities and regression functions lie , respectively , in the sets @xmath143 and @xmath145 of the form given in definitions [ defsetf ] and [ defsetg ] are implied by bounds on the magnitude of the sieve coefficients @xmath193 and @xmath194 in ( [ expf ] ) and ( [ expg ] ) . such constraints are easy to impose within the simplex optimization method : parameter changes that would yield violations of the bounds are simply rejected ( effectively assigned an `` infinite '' value)the simplex optimization method easily accommodates such extreme behavior in the objective function , since it does not rely on derivatives . however , we found that these constraints are rarely binding in practice , unless the number of terms @xmath98 in the expansions is large [ @xcite reports a similar observation ] . such large values of @xmath98 tend to be naturally ruled out via our data - driven selection method of the number of terms . to select the number of terms in the approximating series for a given sample , we use the `` bootstrap cross - validation '' method described in section [ secest ] with a fraction @xmath195 and @xmath185 bootstrap replications . trial values of the number of free parameters ( not counting parameters uniquely determined by zero mean and unit area constraints ) in the series representing @xmath196 each span the set @xmath197 while for @xmath79 , @xmath45 each span the set @xmath198 . the optimal numbers of parameters ( kept constant during the replications ) were found to be @xmath199 ; @xmath200 ; @xmath201 ; @xmath202 ; @xmath203 . figure [ figberksim ] summarizes the result of these simulations , where a naive nonparametric series least - squares estimator ignoring measurement error ( i.e. , least - square regressions of @xmath8 on @xmath0 and of @xmath10 on @xmath0 ) with the same number of sieve terms is also shown for comparison . the reliability of the method can be appreciated by noting how closely the median of the replicated measurement - error - robust estimates matches the true model , while the naive estimator ignoring the presence of measurement error is considerably more biased , even missing the fact that the true regression function is nearly flat in the middle section and instead producing a very misleading linear shape despite the strong nonlinearity of the true model . in fact , unlike the proposed estimator , the naive estimator is so significantly biased that any type of hypothesis test based on it would exhibit completely misleading confidence levels : the true model curves ( for @xmath79 and @xmath45 ) almost always lies beyond the 95% or 5% percentiles of the estimator distribution . overall , the proposed measurement - error - robust estimator exhibits low variability and low bias at the reasonable sample size of @xmath186 . the bias is not exactly zero in a finite sample because our estimator is a nonlinear functional of sample averages and because the sieve approximation necessarily has a limited accuracy in a finite sample . nevertheless , the fact that our estimator performs so well in the presence of measurement error of such large magnitude is a strong indication of its practical usefulness . this behavior is not specific to this model we have tested the method in other simulation settings ; see @xcite . numerous studies have sought to quantify the effect of air pollution on respiratory health [ e.g. , @xcite ] . specifically , there is a growing concern regarding the effect of small particulate matter [ @xcite , @xcite ] . a key difficulty with such studies is that air quality monitors are not necessarily located near the subjects being affected by air pollution , implying that the main regressor of interest is mismeasured . our approach to this question relies on very comprehensive country - wide data collected by environment protection agency ( epa ) and the center for disease control ( cdc ) in the united states . pollution levels are taken from epas monitor values report criteria air pollutants database for year 2005 . epas data provides point measurements of the particulate matter levels ( we focus on so - called 95th percentile level of pm2.5 particles , those having less than 2.5 micrometers in diameter ) at various monitoring stations throughout the united states , from which we construct state - averaged pollution levels ( our @xmath0 variable , measured in @xmath204 g of particles per m@xmath205 ) . we do so because pollution data is only available for a small fraction of counties , and even where it is available , the nature of its measurement error is complex ( it could be a mixture of classical and berkson errors ) . by constructing state - level averages , we average out the randomness in monitor measurements while leaving the randomness in the individual exposure untouched , thus obtaining a valid berkson error - contaminated estimate of the pollution level experienced by individuals from each state , whether they live in a county with a monitoring station or not . each individual faces an exposure equal to the state average plus an unknown random noise due to his / her precise geographic whereabouts and lifestyle . health data is obtained from the publicly available `` cdc wonder '' database entitled `` mortality underlying cause of death '' for year 2005 . to measure respiratory health , we use data on causes of death , which offers the advantage that it is very comprehensive and accurate ( medical professionals are required to collect it and there is no reliance on voluntary surveys ) . one limit to the completeness of the data is that , for some counties , the data is `` suppressed '' ( for privacy reasons ) or labeled as `` unreliable '' by the cdc and were therefore omitted from our sample . our dependent variable of interest ( @xmath8 ) is the rate ( per 10,000 ) of death due to `` chronic lower respiratory diseases '' ( e.g. , asthma , bronchitis , emphysema ) , while our instrument ( @xmath10 ) is the rate ( per 10,000 ) of death resulting from `` lung diseases due to external agents '' ( e.g. , pneumoconiosis due to organic or inorganic dust , coalworker s pneumoconiosis ) . the rationale is to use , as an instrument , a variable that is clearly expected to be affected by pollution levels . this variable indirectly provides information regarding the true level of pollution , so that the effect of pollution ( if any ) on the variable of interest can be more accurately assessed . we employ county - level data on causes of death because they are readily available without concerns for patient privacy issues . moreover , the cdc provides age - corrected death rates , thus correcting for demographic differences between counties . we construct our sample by matching mortality data via counties and matching pollution data via states , resulting in 1305 observations over as many counties and covering all 51 states . a limitation of our approach is that it does not control for other possible confounding effects , for example , if the proportion of smokers differs between industrial and nonindustrial cities . however , such a limitation is common in studies of this kind [ as noted in @xcite ] . we use the same types of sieves and computational methods as in the simulation example and select the number of terms using the `` bootstrap cross - validation '' method described in section [ secest ] with a fraction @xmath195 and @xmath185 bootstrap replications . trial values of the number of free parameters in the series representing @xmath196 span the range @xmath206 while trial values of the number of terms in the series representing @xmath79 and @xmath45 span the range @xmath207 ( increasing any one of the @xmath98 beyond that range resulted in clearly worse performances ) . the optimal numbers of free parameters ( not counting parameters uniquely determined by zero mean and unit area constraints ) were found to be @xmath208 ; @xmath209 ; @xmath210 ; @xmath211 ; @xmath212 . pointwise 90% confidence bands around the nonparametric estimates were obtained using the standard bootstrap [ see , e.g. , @xcite for general conditions justifying its use ] with 100 replications . results are shown in figure [ figapp ] . a few observations are in order . first , our measurement error - robust estimator is perfectly able to detect a clear monotone relationship between @xmath8 and @xmath1 and between @xmath10 and @xmath1 with useful confidence bands , despite the use of a fully nonparametric approach . second , although the distribution of the measurement error is difficult to estimate ( as reflected by the wide confidence bands ) , the impact of this uncertainty on the main function of interest [ @xmath213 is fortunately very limited . the 90% confidence bands indicate that the presence of substantial measurement error is consistent with the data : the measurement error is of the order of 10 @xmath204g / m@xmath205 , whereas the observed @xmath0 roughly ranges from 10 to 40 @xmath204g / m@xmath205 . third , the distribution of @xmath9 exhibits nonnegligible asymmetry , thus illustrating the drawbacks of methods merely assuming normality of all the error terms . in contrast , the distributions of @xmath5 and @xmath12 are apparently very close to symmetric ( this is a conclusion of the formal model selection procedure , not an assumption ) . for comparison purposes , we also naively regress the dependent variables ( @xmath8 or @xmath10 ) on the mismeasured regressor @xmath0 using a conventional least squares ( thereby neglecting measurement error ) with a polynomial specification with the same number of terms as our berkson model . a first troubling observation from this exercise [ see figure [ figapp](b ) ] is that the naive estimate of @xmath214 is not monotone , although in the region where it is unexpectedly decreasing , the confidence bands do not rule out a constant response . second , it is perhaps counter - intuitive that the confidence bands for the naive estimator are sometimes larger than the corresponding bands for the measurement error - robust estimator . this is a consequence of the fact that correcting for berkson errors amounts to an operation akin to convolution ( rather than deconvolution , as in classical measurement errors ) . unlike deconvolution , convolution is a noise - reducing operation , effectively averaging observations of @xmath8 over a wide range of values of @xmath0 to yield an estimate the expected value of @xmath8 given a specific value of @xmath0 . this phenomenon is probably also responsible for the more reasonable ( i.e. , increasing ) behavior of the response for the measurement error - robust estimate . finally , the measurement error - robust regression function often lies at or beyond the 95% or 5% percentiles of the naive estimator distribution ; see figure [ figapp](b ) . this implies that the level of any statistical test would be severely biased . for instance , the confidence bands of the naive estimator would reject our best estimate of @xmath214 obtained with the measurement - error robust procedure . in summary , this application example serves to illustrate that ignoring berkson errors can be seriously misleading in nonlinear settings . not only is the shape of the estimated response considerably affected , but statistical inferences based on a measurement error - blind method would be seriously biased . this application example also shows that our fully nonparametric and measurement error - robust method works well at sample sizes typically available in real data sets , without assuming the knowledge of the distribution of the measurement error . let @xmath215 with @xmath216 @xmath217 for some @xmath218 denote the set of all bounded functions in @xmath219 endowed with the usual @xmath220 norm . also , whenever we state an equality between functions in @xmath215 , we mean that their difference is zero in the @xmath220 norm . we provide two proofs of theorem 1 . the first one , suggested by a referee , relies on the additional assumptions that ( i ) @xmath10 and @xmath1 have the same dimension and ( ii ) @xmath45 and its inverse are differentiable . assumption ( i ) makes assumption [ condnodup ] unlikely to hold , but enables a somewhat direct application of theorem 1 in @xcite . the second proof relaxes those assumptions . it borrows some of the operator techniques from @xcite , yet requires considerable changes in the approach we focus here on the aspects of the proof that differ . proof theorem [ thid ] ( simple special case ) let variables from @xcite be denoted by the corresponding uppercase letter with tildes and make the following assignments : @xmath221 . we now verify the 5 assumptions of theorem 1 in @xcite . to verify assumption 1 , we observe that the densities of @xmath222 and @xmath223 are related through:@xmath224 where the density @xmath225 exists by assumption [ conddens ] , and @xmath226 exists by assumption [ condnodup ] . the jacobian @xmath227 matrix is only defined if @xmath1 and @xmath10 ( and therefore @xmath228 ) have the same dimension and is finite and nonsingular under the assumption that @xmath45 and its inverse are differentiable . a similar argument can be used for marginals and conditional distributions . to verify assumption 2 , we note that our model can be written in terms of tilded variables as @xmath229 to verify assumption 2(i ) , we write @xmath230 where we have used , in turn , ( i ) the equality @xmath231 and the fact that changes of variables in the conditioning variables do not introduce jacobian terms , ( ii ) the fact that conditioning on @xmath232 is equivalent to conditioning on @xmath233 , ( iii ) assumption [ condindep ] , ( iv ) the relationship between @xmath9 and @xmath8 via ( [ eqtilde1 ] ) and ( v ) the equality @xmath234 . assumption 4 requires that @xmath236 for @xmath237 . this can be verified as follows : @xmath238 by invoking ( i ) the definition of @xmath9 , ( ii ) independence of @xmath9 from @xmath1 ( and therefore @xmath228 ) , ( iii ) the fact that @xmath239 implies @xmath240 since @xmath7 and @xmath11 are one - to - one by assumption [ condnodup ] and so is @xmath241 . theorem 1 in @xcite then allows us to conclude that the joint distribution of @xmath242 is identified . however , in order to identify the distribution of @xmath243 , we need to identify @xmath11 . to this effect , we note that , conditional on @xmath244 , the fluctuations in @xmath228 are entirely caused by fluctuations in @xmath18 by equation ( [ eqtilde2 ] ) . moreover , @xmath5 is independent from @xmath0 , hence @xmath245 where the left - hand side was previously identified and where the jacobian term is well defined by assumptions [ condnodup ] and the assumed differentiability of @xmath226 . the jacobian can be identified by integrating ( [ eqfj ] ) with respect to @xmath246 to yield @xmath247 . by varying @xmath248 while keeping @xmath249 fixed in equation ( [ eqfj ] ) , we can identify the density @xmath78 up to a shift of @xmath226 . assumption [ condloc ] , pins down what the shift should be , so that @xmath250 is identified for any given @xmath249 . since @xmath89 is one - to - one by assumption [ condnodup ] , @xmath251 uniquely determines @xmath11 . hence , the joint distribution of @xmath252 is identified . finally , noting that @xmath253 ( by assumption [ condindep ] ) , then establishes the identification of @xmath214 with the help of assumption [ condloc ] . the definition of marginal and conditional densities in combination with assumption [ condindep ] lead to the following sequence of equalities : @xmath254 or , equivalently , @xmath255 as in @xcite , this integral equation can be written more conveniently as an operator equivalence relation @xmath256 by introducing the operators defined in equation ( [ eqdefop ] ) , which are acting on an arbitrary @xmath257 [ or @xmath258 . similarly , one can show that @xmath259 and thus @xmath260 . by assumptions [ conddens ] , [ condindep ] , [ condinv ] , [ condnodup ] , [ condcont ] and lemma [ leminj ] below , we know that @xmath261 admits an inverse on the range of @xmath69 ( and therefore the range of @xmath262 ) , and we can write @xmath263 substituting ( [ eqop2i ] ) into ( [ eqop1 ] ) , we obtain @xmath264 by assumptions [ conddens ] , [ condindep ] , [ condinv ] , [ condnodup ] , [ condcont ] and lemma [ leminj ] below again , @xmath262 admits an inverse . moreover , by lemma 1 in @xcite , the domain of @xmath265 is dense in @xmath266 , and we can then write @xmath267 equation ( [ eqdiagpr ] ) states that the operator @xmath66 admits a spectral decomposition , where the eigenvalues are given by the @xmath268 for @xmath269 ( for a fixed @xmath59 ) defining the operator @xmath60 while the eigenfunctions are the functions @xmath270 for @xmath269 defining the kernel of the operator @xmath69 . as usual , the knowledge of a linear operator [ e.g. , @xmath262 ] only determines the value of its kernel [ e.g. , @xmath272 everywhere except on a set of null lebesgue measure . the resulting equivalence class exactly matches the usual equivalence class for probability densities with respect to the lebesgue measure , so identifiability of the model is not affected . the operator to be diagonalized is entirely defined in terms of observable densities while the decomposition provides the unobserved densities of interest . to ensure uniqueness of this decomposition , we employ four techniques . first , a powerful result from spectral analysis [ theorem xv 4.5 in @xcite ] ensures uniqueness up to some normalizations . second , the a priori arbitrary scale of the eigenfunctions is fixed by the requirement that densities must integrate to one . third , to avoid any ambiguity in the definition of the eigenfunctions when degenerate eigenvalues are present , we use assumption [ condnodup ] and the fact that the eigenfunctions [ which do not depend on @xmath59 , unlike the eigenvalues @xmath273 must be consistent across different values of the dependent variable @xmath59 . these three steps are described in detail in @xcite and are not repeated here . the fourth step [ which differs from the approach taken in @xcite ] is to rule out that the eigenvalues @xmath274 and eigenfunctions @xmath275 could be indexed by a different variable without affecting the operator @xmath66 . ( this issue is analogous to the nonunique ordering of the eigenvalues and eigenvectors in matrix diagonalization . ) suppose that the eigenfunctions can be indexed by another value , that is , they are given by latexmath:[$f_{z|\tilde{x}^{\ast } } ( \cdot variable related to @xmath82 through @xmath277 for some one - to - one function @xmath71 . is also measurable , for otherwise @xmath278 would not be a proper random variable . ] under this alternative indexing , all the assumptions of the original model must still hold with @xmath82 replaced by @xmath279 , so a relationship similar to ( [ eqfzxsidx ] ) would still have to hold , for the same observed @xmath280 @xmath281 or , in operator notation , @xmath282 . in order for @xmath283 to be a valid alternative density , it must satisfy the same assumptions ( and their implications ) as @xmath75 . in particular , the fact that @xmath69 is invertible ( established above via lemma [ leminj ] ) must also hold for @xmath284 . hence , for any alternative @xmath284 , there is a unique corresponding @xmath285 , given by @xmath286 . we can find a more explicit expression for @xmath287 as follows . first note that we trivially have that @xmath288 since @xmath289 and @xmath71 is one - to - one . by performing the change of variable @xmath290 in ( [ eqfzxsidx ] ) , we obtain @xmath291 where the measure @xmath204 is defined , via @xmath292 for any measurable set @xmath293 , where @xmath294 denotes the lebesgue measure and @xmath295 . from this we can conclude the equality between the two following measures : @xmath296 by comparison to equation ( [ eqaltfzxs ] ) and the uniqueness of the measure@xmath297 due to the injectivity of the @xmath284 operator , shown in lemma [ leminj ] in the general case where the domain of @xmath284 could include finite signed measures . we will now show that @xmath287 necessarily violates assumption [ condloc ] ( with @xmath5 replaced by @xmath298 ) , unless @xmath299 is the identity function . since @xmath300 with @xmath5 independent from @xmath0 , we have @xmath301 and by a similar reasoning @xmath302 with @xmath303 . equation ( [ eqds1 ] ) then becomes @xmath304 now , for a given @xmath248 , consider radom nikodym derivative of @xmath305 with respect to the lebesgue measure @xmath306 , which is , by definition ( almost everywhere ) equal to @xmath307 , a bounded function by assumption [ conddens ] . by equation ( [ eqmueq ] ) , the existence of the radom nikodym derivative of the left - hand side implies the existence of the same radom nikodym derivative on the right - hand side , and we can write @xmath308 almost everywhere . integrating both sides of the equation over all @xmath309 , we obtain ( after noting that points where the equality may fail have null measure and therefore do not contribute to the integral ) , @xmath310 , since densities integrate to @xmath311 , which implies that @xmath312 , that is , @xmath204 is also the lebesgue measure . it follows from ( [ eqallf ] ) that , almost everywhere @xmath313 in order for assumption [ condloc ] to hold for both @xmath314 and @xmath5 , we must have that @xmath315 , when viewed as a function of @xmath249 for any given @xmath248 , is centered at @xmath316 , and we must simultaneously have that @xmath317 , when viewed as a function of @xmath82 for any given @xmath248 , is centered at @xmath318 , that is , @xmath319 . the two statements are only compatible if @xmath320 . thus , there can not exist two distinct but observationally equivalent parametrization of the eigenvalues / eigenfunctions . hence we have shown , through equation ( [ eqdiagpr ] ) , that the unobserved functions @xmath268 and @xmath321 are uniquely determined ( up to an equivalence class of functions differing at most on a set of null lebesgue measure ) by the observed function @xmath322 . next , equation ( [ eqop2i ] ) implies that @xmath323 is uniquely determined as well . once @xmath268 and @xmath324 are known , the functions @xmath214 and @xmath325 can be identified by exploiting the centering restrictions on @xmath9 , @xmath5 and @xmath12 , for example , @xmath326 if @xmath9 is assumed to have zero mean . next , @xmath327 can be straightforwardly identified , for example , @xmath328 for any @xmath269 . similar arguments yield @xmath325 and @xmath329 from @xmath330 as well as @xmath331 from @xmath73 . it follows that equation ( [ eqfyxz ] ) has a unique solution . the second conclusion of the theorem then follows from the fact that both @xmath322 and @xmath332 are uniquely determined ( except perhaps on a set of null lebesgue measure ) from @xmath333 . the following lemma is closely related to proposition 2.4 in @xcite . it is different in terms of the spaces the operators can act on and more general in terms of the possible dimensionalities of the random variables involved . [ leminj]let @xmath334 and @xmath10 be generated by equations ( [ eqxs ] ) and ( [ eqz ] ) . let @xmath335 be the set of finite signed measures on a given set @xmath336 or @xmath25 [ and note that @xmath335 includes @xmath337 as a special case , in the sense that for any function in @xmath338 , there is a corresponding measure @xmath339 whose radom nikodym derivative with respect to the lebesgue measure is @xmath58 ] . under assumptions [ condindep ] , [ conddens ] , [ condinv ] , [ condnodup ] and [ condcont ] , the operators @xmath341 , @xmath342 and @xmath343 , defined in ( [ eqdefop ] ) , are injective mappings . first , one can verify that @xmath344 implies that @xmath345 and similarly for @xmath69 and @xmath262 , since the ( conditional ) densities involving variables @xmath346 and @xmath10 are bounded by assumption [ conddens ] and are absolutely integrable . we now verify injectivity of @xmath69 . by assumptions [ condindep ] , [ conddens ] and equation ( [ eqz ] ) , we have , for any @xmath347 , @xmath348 ( z ) = \int f_{z|x^{\ast } } \bigl ( z|x^{\ast } \bigr ) \,dr \bigl ( x^{\ast } \bigr ) = \int f_{\delta z } \bigl ( z - h \bigl ( x^{\ast } \bigr ) \bigr ) \,dr \bigl ( x^{\ast } \bigr).\ ] ] next , let @xmath349 denote the signed measure assigning , to any measurable set @xmath350 , the value@xmath351 and note that @xmath349 is a finite signed measure since @xmath352 is . then , we can express @xmath353 as @xmath354 ( z ) = \int f_{\delta z } \bigl ( z-\tilde{x}^{\ast } \bigr ) \,d\tilde{r } \bigl ( \tilde{x}^{\ast } \bigr),\ ] ] that is , a convolution between the probability measure of @xmath12 ( represented by its lebesgue density ) and the signed measure @xmath349 ; see chapter 5 in @xcite . by the convolution theorem for signed measures [ theorem 5.1(iii ) in @xcite ] , one can convert the convolution ( [ eqconvol1 ] ) into a product of fourier transforms , @xmath355 where @xmath356 ( z ) e^{\mathbf{i}\zeta z}\,dz$ ] , @xmath357 $ ] and @xmath358 . since @xmath359 , the characteristic function of @xmath12 , is nonvanishing by assumption [ condinv ] , we can isolate @xmath360 as @xmath361 since there is a one - to - one mapping between finite signed measures and their fourier transforms [ by theorem 5.1(i ) in @xcite ] , @xmath349 can be recovered as the unique signed measure whose fourier transform is @xmath362 . we now show that the signed measure @xmath349 uniquely determines the measure @xmath363 . let @xmath364 for any measurable @xmath365 , and note that @xmath366 is also measurable since @xmath45 is continuous by assumption [ condcont ] . then observe that by assumption [ condnodup ] , @xmath367 if and only if @xmath368 , and we have @xmath369 since @xmath370 is arbitrary , the knowledge of @xmath371 uniquely determines the value assigned to any measurable set by the signed measure @xmath363 . injectivity of @xmath63 is a special case of the above derivation ( with @xmath372 replaced by @xmath346 ) , in which @xmath45 is the identity function . finally , injectivity of @xmath262 is implied by the injectivity of @xmath69 and @xmath63 , since @xmath61 by assumption [ condindep ] and equations ( [ eqxs ] ) and ( [ eqz ] ) .

we consider the classical response of a strongly chaotic hamiltonian system . the spectrum of such a system consists of discrete complex ruelle - pollicott ( rp ) resonances which manifest themselves in the behavior of the correlation and response functions . we interpret the rp resonances as the eigenstates and eigenvalues of the fokker - planck operator obtained by adding an infinitesimal noise term to the first - order liouville operator . we demonstrate how the deterministic expression for the linear response is reproduced in the limit of vanishing noise . for the second - order response we establish an equivalence of the spectral decomposition with infinitesimal noise and the long - time asymptotic expansion for the deterministic case .

nlin0703014

vibrational molecular motion is an interesting and important example of complex dynamics . due to nonlinearity , nonadiabatic effects and interactions with the environment , the behavior of such systems is remarkably rich , and their theoretical description away from the equilibrium becomes a challenge . in many cases , e.g. when the frequencies of the vibrational modes do not exceed the temperature , a completely classical description is adequate . the dynamics far from equilibrium is typically irregular and can exhibit chaotic features . in the extreme case of strong chaos all trajectories in phase space are unstable , and the phase - space dynamics shows mixing features . spectroscopic methods represent powerful tools for obtaining detailed information on vibrational dynamics . the outcome of spectroscopic experiments can be naturally interpreted in terms of optical response functions , either in the time or in frequency domain . the multi - time ( nonlinear ) response functions that constitute the main objects in multi - dimensional time - domain spectroscopy are usually analyzed using their spectral decompositions , i.e. the fourier transforms with respect to the time intervals . these time intervals can be viewed as the time delays between the exciting pulses , provided the latter are short compared to the typical time scales of the system dynamics . periodic features of response functions are routinely interpreted as signatures of periodic motions . the central frequencies of diagonal and off - diagonal ( cross ) peaks are associated with some coherent vibrations , whereas the peak shapes contain information on the system - bath coupling . smooth behavior of response functions in the case of almost harmonic vibration of the primary system is achieved due to its coupling to a macroscopically large number of the bath motions . the bath is often considered in the harmonic approximation : such model is known in the spectroscopic literature as a multi - mode brownian oscillator . the situation is totally different in the case of chaotic motion of the primary system . the spectrum of a strongly chaotic system is known to consist of so - called ruelle - pollicott ( rp ) resonances that represent the eigenmodes of the perron - frobenius operator @xcite . the imaginary part of any particular resonance describes an oscillating feature in the system correlations , whereas its real part is responsible for the correlations decay . an individual rp resonance is not directly related to any particular periodic motion , although their positions can be expressed in terms of all periodic orbits in a very collective way via the dynamical @xmath0-function . these collective resonances in chaotic systems should be distinguished from the signatures of stable periodic motions . nonlinear response functions of quantum systems can be conveniently represented in terms of spectral decompositions via the system s stationary states . these spectral decompositions that are often referred to as bloembergen s expressions allow to relate resonances in the spectroscopic measurement data to the transitions between the stationary states . a natural question that arises in the context of approaching the classical limit is what would be the classical counterpart of the quantum liouville space spectral decomposition for the linear and nonlinear response functions ? in quantum mechanics the system state can be described by a density matrix , whereas the evolution is governed by the quantum liouville operator . one can try to reach the classical limit by replacing the quantum density matrix by the classical phase - space distribution using the wigner transformation . the quantum liouville operator should be naturally replaced by its classical counterpart . this conceptually straightforward approach , however , faces certain major difficulties . quantum liouville operators are second - order elliptic operators , whose eigenstates belong to certain natural hilbert spaces . in particular in the case of compact ( restricted ) coordinate spaces the corresponding spectra are discrete , and finding spectral decompositions does not face conceptual difficulties , at least on the physical level of rigor . the situation with the classical limit is much more involved since the classical liouville operator is a first - order non - elliptic operator that describes the propagation of phase - space distributions along classical trajectories . in particular any non - periodic classical trajectory generates a family of eigenstates of the classical liouville operator , concentrated on the trajectory with all possible eigenvalues . such a spectrum contains no useful information on the system relaxation . a meaningful spectrum that provides detailed information on the relaxation of a strongly chaotic system is known to be represented by the rp resonances . one of the ways to reproduce the rp resonances as eigenvalues of the classical liouville operator is based on the appropriate simultaneous definition of the functional space where the operator acts . these functional hilbert spaces , often referred to as rigged spaces , are very different from `` standard '' hilbert spaces involved in spectral decompositions of quantum operators . they are usually chosen on the case - to - case basis . to avoid these difficulties we follow a more physical approach introduced in ref . . the approach is based on introducing weak langevin noise , followed by considering the limit of vanishing noise . it has several major advantages . first of all , this describes a realistic situation since any system is at least weakly coupled to some environment , and in many cases the system - bath coupling can be described on the level of langevin noise . second , it allows to avoid dealing with a problem of choice of the appropriate hilbert space . the stochastic langevin processes can be described by adding a diffusion operator to the liouville operator @xmath1 , which results in the fokker - planck operator @xmath2 . the fokker - planck operator @xmath3 is a second - order elliptic operator , and its spectral decomposition is free of the aforementioned difficulties that arise in the case of classical liouville operator . the rp resonances are obtained from the spectrum of the fokker - planck operator by taking the limit @xmath4 . this allows to interpret introducing infinitesimal noise as a regularization procedure similar to coarse graining . the dependence of the eigenfunctions of the fokker - planck operator @xmath3 on the noise strength @xmath5 is nonanalytical . in the limit @xmath4 of the vanishing noise strength the smooth eigenfunctions turn into generalized functions ( distributions ) . the rigged hilbert spaces are reproduced from the `` standard '' hilbert spaces in the limit @xmath4 . they are spanned on the generalized eigenfunction obtained from the eigenfunctions of @xmath6 by applying the noiseless limit . the calculation of the linear response function in terms of the generalized functions does not pose a problem , since the expression involves an expansion of the smooth function over generalized functions . the calculation of nonlinear response functions is essentially more complicated because one needs to define expansions of the generalized functions over similar generalized functions . we avoid dealing with the latter problem by performing all calculations for weak , yet finite noise , followed by applying the limit @xmath4 to the final expressions . in our previous work @xcite we demonstrated the convergence of the second - order response function for strongly chaotic systems . as an example of the strong chaos , we considered free motion on a compact surface of constant negative curvature . the model has been studied since more than a century and served as a prototype for quantum chaos @xcite . the motion on a surface of genus @xmath7 is in particular known as a hadamard billiard . using the dynamical symmetry ( ds ) of the system we found analytical expressions for the response functions . the long - time asymptotic expansion of the second - order response function turned out to have the form of the double decomposition in the resonances that appear in the linear response . in the present work we find the decomposition of the linear and nonlinear response in ruelle - pollicott resonances for this classical strongly chaotic model by introducing weak langevin noise . the spectral decomposition in this case is conceptually straightforward . we further demonstrate that the decomposition coefficients converge in the noiseless limit @xmath4 , and the resulting spectral decompositions reproduce the asymptotic expansions for the purely classical response functions computed in our earlier work @xcite . to the best of our knowledge , this is the first calculation of the classical nonlinear response for a chaotic flow that uses spectral decomposition based on the noise regularization . our paper is organized as follows . in sec . [ section : model ] we summarize the geometry and dynamical symmetry of free motion on a compact surface of constant negative curvature . statistical description of the response is presented in section [ section : noise ] . we regularize the dynamics by adding noise to the liouville operator and derive general forms of the spectral decompositions in section [ section : decomposition - general ] . the eigenstates of the resulting fokker - planck operator are found in section [ section : eigenstates ] . explicit forms of spectral decompositions are obtained in section [ section : linear ] for the linear response and in section [ section : second - order ] for the second - order response . following @xcite we consider free motion on a @xmath8 compact surface @xmath9 of constant negative curvature ( gaussian curvature ) . this strongly chaotic system is described by the classical free - particle hamiltonian @xmath10 that depends on the absolute value @xmath0 of the momentum @xmath11 only . the curvature of the @xmath8 configuration space is expressed in terms of the metric tensor @xmath12 . the hamiltonian classical dynamics preserves the smooth compact @xmath13 manifold @xmath14 that corresponds to a fixed value of energy . points @xmath15 of the reduced phase space are described by two coordinates @xmath16 and the momentum direction angle @xmath17 . hereafter , we will use dimensionless units so that the mass @xmath18 and the curvature @xmath19 . despite the complexity of the flow due to its chaotic nature , its strong dynamical symmetry ( ds ) enables an analytical treatment @xcite . the geodesic flow @xmath20 , where @xmath21 denotes a point in phase space , is generated by the liouville operator @xmath22 hereafter , we identify any vector field with the corresponding first - order operator of the derivative in the vector field direction . we denote by @xmath23 the vector field that determines the phase space velocity : @xmath24 . we further introduce the second vector field @xmath25 in the tangent space , and finally set @xmath26 $ ] ( see e.g. refs . for the details ) . a simple local calculation yields @xmath27=-\sigma_1 $ ] and @xmath28=-k\sigma_{z}$ ] , which implies that in the constant negative curvature case the vector fields @xmath23 , @xmath29 and @xmath30 form the lie algebra @xmath31 . the group @xmath32 action in the reduced phase space @xmath14 is obtained by integrating the @xmath31 algebra action . ds with respect to the action of the group @xmath33 does not mean symmetry in a usual sense , i.e. that the system dynamics commutes with the group action , but rather reflects the fact that the vector field @xmath34 that determines the classical dynamics is represented by an element of the corresponding lie algebra @xmath31 , whereas the stable and unstable directions of our hyperbolic flow are determined by @xmath35 . the algebra generators @xmath36 , @xmath37 are anti - hermitian , i.e. satisfy the equalities @xmath38 with respect to the natural scalar product @xmath39 . in addition , the poisson bracket of two functions @xmath40 and @xmath41 reads @xmath42 the smooth action of @xmath43 in @xmath14 can be interpreted as that the space @xmath44 of smooth functions in @xmath14 constitutes a representation of @xmath45 , which turns out to be a unitary representation ( see refs . ) , and therefore can be decomposed into a direct sum of irreducible representations of @xmath45 . stated differently , any distribution in the reduced phase space @xmath14 can be decomposed in irreducible representations . ds implies that the distributions in different representations evolve independently . we focus on the principal series representations of @xmath32 since only these provide experimentally interesting contributions to the linear and second - order response @xcite . a principal series representation @xmath46 is labeled by an imaginary number @xmath47 , @xmath48 . eigenfunctions @xmath49 of the momentum rotation operator @xmath50 , hereafter also referred to as angular harmonics , form a convenient basis set in the irreducible representation @xmath46 . the functions @xmath51 do not depend on the momentum direction and can be viewed as eigenfunctions of the laplacian operator on our compact riemann surface @xmath9 @xcite . the laplacian eigenvalues provide a set of numbers @xmath52 according to the equation @xmath53 . the angular harmonics have the following properties : @xmath54 where we introduced the raising and lowering operators @xmath55 . the operators @xmath56 are anti - hermitian conjugated : @xmath57 . the description of unitary representations of @xmath45 @xcite allows us to identify the functions @xmath58 in reduced phase space with the corresponding functions on the circle @xmath59 for any @xmath52 . in the case of the principal series @xmath60 , @xmath48 , and @xmath59 form an orthogonal normalized basis set with the natural scalar product . therefore , the normalized functions @xmath58 are naturally associated with @xmath61 . on the circle the algebra generators are represented by ( see @xcite ) @xmath62 using the dynamical symmetry ( ds ) of the problem , one can calculate the response functions . two - point correlations that are related to the linear response via the fluctuation - dissipation theorem have been also considered in ref . . detailed calculations of the response functions are presented in our previous paper @xcite . since the zero harmonics @xmath51 are momentum - independent , the expansion of the dipole @xmath63 in irreducible representations has the form @xmath64 where only the principle series contributions are retained . typically the dipole is represented by a smooth distribution , and , therefore , only a small number @xmath65 of representations are relevant in the expansion . for the sake of the presentation clarity we focus on the case @xmath66 . generalizations to @xmath67 are straightforward and can be easily performed . strongly chaotic systems are characterized by complicated irregular dynamics @xcite . exponential divergence of initially close trajectories and mixing make the description in terms of individual trajectories complicated and inadequate . an alternative picture , based on liouville representation of classical mechanics , involves propagating phase space densities . suppose that the system initially occupies a certain region in the phase space which correspond to a range of initial phase space variables . in the course of the evolution of a chaotic system , the shape of the region becomes stretched along the unstable directions and contracted along the stable ones . the response functions characterize the system response to a time - dependent external driving field @xmath68 and constitute the basic outcome of spectroscopic measurements . the external field is coupled to the system via the polarization @xmath63 , also often referred to as the dipole . the function @xmath69 of the system coordinates is the classical counterpart of the polarization operator . the total classical hamiltonian of the system coupled to the driving field has the form @xmath70 the initial equilibrium phase space distribution @xmath71 starts to evolve once the external field is turned on . the phase space density @xmath72 at time @xmath73 depends on the values of the external field at preceding times . in most cases the observed signal at time @xmath73 is directly related to the same polarization @xmath63 , which describes the system - field coupling . within the classical mechanics formalism the observed polarization is given by @xmath74 . the evolution of the phase space distribution @xmath75 is governed by the classical liouville equation @xmath76 where the liouville operator @xmath34 acts as the poisson bracket with the hamiltonian according to eq . ( [ liouville - operator ] ) . the sign of the poisson bracket is defined by a convention that @xmath77 for canonically conjugate coordinates @xmath78 and momenta @xmath11 . note that the liouville operator is a first - order linear differential operator . as stated earlier , it can be identified with the vector field @xmath79 in the phase space . the liouville operator is anti - hermitian with respect to the conventional scalar product , so that @xmath80 . this makes the evolution operator @xmath81 unitary and leads to the conservation of phase space volume ( liouville theorem ) . one can solve eq . ( [ liouville - equation ] ) iteratively considering the right - hand side as a small perturbation and taking @xmath71 for the zero approximation . then the phase space density , and hence the response ( observed signal ) can be naturally represented in the form of the functional taylor expansion over the external field . the @xmath82-th order response function @xmath83 is defined as the coefficient ( more precisely , an integral kernel ) in front of the @xmath82-th power of the driving field . the response function @xmath83 depends on @xmath84 time moments , and if the unperturbed system is initially at equilibrium , it actually depends on @xmath82 time intervals . for the linear and second - order response functions we have the following expressions : @xmath85 here @xmath86 is the the phase space density evolution operator , often referred to as the perron - frobenius operator . we understand @xmath87 as the solution @xmath88 of the equation @xmath89 with the initial condition @xmath90 . since @xmath34 is a first - order differential operator , the solution can be readily found using the method of characteristics . in other words , the first - order operator @xmath34 allows to write @xmath91 . phase space trajectories found from the hamilton equations of motion are invariant with respect to the time reversal . this property is easily observed in the behavior of integrable systems . namely , any integrable dynamics can be represented by a set of quasiperiodic motions , which implies reversibility and recurrence in the values of physical observables . the behavior of chaotic systems is quite different . although described by the same formalism based on eqs . ( [ liouville - operator ] ) and ( [ liouville - equation ] ) that possess time - reversal symmetry , one observes obviously irreversible features such as relaxation phenomena . the apparent paradox and its solution are well known in statistical mechanics . in the course of evolution of a chaotic system , more and more fine features develop in the phase space distribution . the distribution width decreases exponentially along the stable directions . at the same time , physical quantities are represented by smooth functions in phase space . therefore , fine features of the distribution are actually irrelevant for the smooth observables . in particular , this results in the exponential damping of the linear and nonlinear response functions in a strongly chaotic system @xcite . only coarse grained properties of the phase space density remain relevant . for instance , in the long - time asymptotic , the strongly chaotic system may be found in the neighborhood of any point of the phase space with equal probability , which is reflected in the homogeneity of the stationary phase space density . a physically useful and meaningful definition of the evolution operator which relates initial and final distributions must rely on some kind of regularization that eliminates unnecessary distracting details . coarse graining can be introduced either directly in phase space or , alternatively , in the functional space of phase space distributions . sometimes coarse graining is inevitable , e.g. in computer simulations , where numerical errors determine the precision . the finest scale of classical dynamics is limited by the onset of quantum effects . one of well - known examples is the quasiclassical calculation of the entropy . quantum effects were also shown to remove unphysical power - law divergences in the nonlinear response functions of integrable systems @xcite . it has been demonstrated that the limit @xmath92 should be taken after calculating the long - time asymptotics . this property resembles the characteristic feature of our approach to the second - order response in a chaotic system , as presented below . besides , real physical systems are never isolated . the influence of an environment shows up as noise or random forces at the level of equations of motion , and as unobservable degrees of freedom and diffusive behavior of relevant variables in the statistical description . although in many cases interaction with the environment is sufficiently weak and can be neglected , it is often utilized as a convenient and physically meaningful way to perform calculations . the effect of the diffusion in the chaotic system is to regularize the long - time dynamics and to introduce irreversible leveling of gradients in the stable direction . we regularize the operator generating the density evolution by adding a small diffusive term in the form of the second - order differential operator : @xmath93 since @xmath94 generates rotations of the momentum , @xmath95 has the meaning of the angular diffusion coefficient . factor @xmath96 is chosen for reasons of convenience . we will be interested in the case @xmath97 . as we will see , the limit @xmath4 does exist in the spectral decomposition , therefore in the limit the factor does not play any role . the resulting fokker - planck operator @xmath98 defines the evolution of the phase space density . this statistical description for the density is equivalent to the description in terms of the langevin equations for the dynamical variables . such equations contain a random force which tends to change the momentum direction keeping the energy constant . this can be formally described by a stochastic liouville operator @xmath99 where @xmath100 is a random gaussian markovian process with the zero mean . the noise intensity is determined by @xmath5 , and the two - point correlation function reads @xmath101 the addition of the diffusion term with the second derivative defines the spectrum of the resulting operator in the space of smooth functions . the eigenfunctions become regular , differentiable functions . as we will see , they are still singular at @xmath102 , and in the limit @xmath4 they turn into generalized functions , which is natural for the case of a small parameter in front of the highest derivative . the mixing property of chaos makes the type of noise irrelevant . random force exerted on any of the phase space variables affects other variables if they are not fixed by conservation laws . thus , in the reduced phase space represented by the shell of constant energy the mixing leads to the fast and irregular randomization of the position variables @xmath78 . as a result , the liouville equation ( [ liouville - equation ] ) that describes the phase space density evolution in a system perturbed by an external field @xmath103 , is replaced by the fokker - planck equation : @xmath104 the evolution operator in the unperturbed system can be written symbolically as @xmath105 , in a full analogy with the noiseless case . similarly , the iterative solution of eq . ( [ fokker - planck - equation ] ) yields the phase space density in a form of an expansion in powers of @xmath103 . the response functions are obtained from eqs . ( [ first - order ] ) by replacing @xmath34 with @xmath106 : @xmath107 where angular brackets stand for the integral over the reduced phase space , and we defined the action of the operator @xmath108 on a function @xmath109 as the poisson bracket of @xmath110 and @xmath109 so that @xmath111 . in the well - known situation with quantum response , the response functions can be readily represented in the form of spectral decompositions , since the infinitesimal evolution is determined by a hermitian operator . the conjugation property with respect to a simple scalar product facilitates expansions in the basis of the eigenstates . we have a natural scalar product for functions in @xmath112 which allowed to implement the unitary presentations in @xmath13 space by functions on the circle . with respect to this scalar product @xmath1 is an anti - hermitian first - order differential operator . however , the fokker - planck operator @xmath98 is obtained by adding a second - order hermitian contribution to @xmath1 . this makes the resulting operator @xmath98 neither hermitian nor anti - hermitian . namely , its adjoint is @xmath113 it seems that there is no natural choice of the scalar product that would make our fokker - planck operator simply related to its adjoint . the expressions for the response functions include integrations over the reduced phase space . the convolution @xmath114 of two functions @xmath115 and @xmath116 can be recast in the form of the natural scalar product as @xmath117 . as shown by the construction of eigenmodes in section [ section : eigenstates ] , the operator @xmath98 is diagonalizable in @xmath46 , and the jordan - block structures that are possible in a general case do not appear in our spectral decompositions . therefore , any function in @xmath118 can be expanded in the eigenfunctions @xmath119 that obey the equation @xmath120 the expansion coefficients are linear functionals of @xmath121 and hence can be represented by scalar products of @xmath121 with certain functions denoted by @xmath122 : @xmath123 it follows from the expansion of @xmath119 and @xmath124 that the functions @xmath122 are the eigenfunctions of the operator @xmath125 which satisfy the following properties : @xmath126 we focus our detailed treatment on the simplest case @xmath66 of a single irreducible representation contributing to the dipole moment @xmath63 . this corresponds to a single contribution in the expansion ( [ f - expansion ] ) , and below we imply @xmath127 . a generalization to an arbitrary linear combination of @xmath67 such terms with different @xmath47 is straightforward for both response functions under consideration @xcite . we can successively apply the expansion procedure ( [ varphi - expansion ] ) and obtain any response or correlation function as a spectral decomposition . based on eq . ( [ first - order - general1 ] ) we get the following form for the linear response function : @xmath128 for the calculation of the second - order response starting from eq . ( [ second - order - general1 ] ) , we in addition introduce a double expansion in the basis of the angular harmonics that are defined by eqs . ( [ sigma - act - irred0 ] ) : @xmath129 the double expansion in the angular harmonics yields a geometric matrix element @xmath130 studied in our previous work @xcite . note that @xmath131 in the middle angular brackets contains a derivative @xmath132 acting on all functions of the momentum to the right of it . this complication occurs due to the absence of the integration over @xmath0 in the angular brackets which correspond to projections onto the basis set in @xmath46 . as stated earlier the fokker - planck operator ( [ fokker - planck - operator ] ) is neither hermitian nor anti - hermitian with respect to any natural scalar product and , therefore , not necessarily diagonalizable . in what follows , we show that it actually is : we apply the representation on the circle to find its eigenstates and demonstrate that they constitute a basis set . the eigenfunctions and eigenvalues of @xmath98 can be found by solving the @xmath133 eigenvalue problem @xmath134 on a circle with the fokker - planck operator [ see eqs . ( [ so(2,1)-circle ] ) and ( [ fokker - planck - operator ] ) ] @xmath135 the first derivative in the operator can be eliminated by redefining the functions @xmath136 in terms of the functions @xmath137 , the eigenvalue problem ( [ eq - general ] ) assumes the form of a stationary schrdinger equation @xmath138 with the effective hamiltonian @xmath139 the hamiltonian describes a quantum particle in a complex - valued potential on the circle . the operator in not hermitian because of the imaginary part of the potential . however , since the hamiltonian does not contain first derivatives , we can define a symmetric ( non - hermitian ) scalar product @xmath140 so that the hamiltonian is self - adjoint with respect to it : @xmath141 . eigenfunctions of @xmath142 corresponding to different eigenvalues are orthogonal and can be normalized : @xmath143 the hamiltonian @xmath142 also possesses certain symmetries that simplify the analysis . first , it is invariant with respect to the sign change of @xmath144 , and therefore all its eigenfunctions are either even or odd functions of @xmath144 . as we will see , only even eigenstates contribute to the spectral decompositions of response functions . therefore , we present the detailed expressions only for even eigenstates on the half - circle @xmath145 . since the potential contains a nonzero imaginary part , its eigenvalues can be complex . the second symmetry involves the complex conjugation of the hamiltonian : @xmath146 . consequently , the eigenfunctions corresponding to the complex conjugate eigenvalues are related by @xmath147 detailed analysis of the schrdinger equation in the weak noise case @xmath148 is presented in appendix [ appendix : eigenmodes ] . the small parameter @xmath5 allows to solve the schrdinger equation using the wkb method . the imaginary part of the potential energy is small compared to its real part . this supports the use of such terms as the `` minimum '' of the potential and the `` under the barrier '' wave function . since the potential minima become deeper for smaller @xmath5 , the eigenfunctions are concentrated near @xmath149 or @xmath150 . these states do not mix together since @xmath47 is imaginary . due to the compact nature of the circle , the spectrum of the fokker - planck operator is discrete . its real part is positive and unbounded from above . for such large energies that @xmath151 , the spectrum can be adequately approximated by @xmath152 ( with the eigenfunctions @xmath153 ) of a free particle on the circle . in the limit @xmath154 even low - energy eigenstates of the first set are concentrated near @xmath149 and have energies @xmath155 with nonnegative even numbers @xmath156 ( @xmath157 ) . this discrete equidistant spectrum of the fokker - planck operator in the noiseless limit @xmath4 is a quite fascinating property . infinitesimal noise regularizes the liouvillian dynamics to yield a physical spectrum in the space of smooth functions . the normalized eigenfunctions that correspond to the specified above eigenvalues are concentrated near @xmath149 : @xmath158 where @xmath159 are hermite polynomials . the eigenfunctions @xmath160 are singular at @xmath149 in the limit @xmath4 , as they should form a basis set of distributions with large gradients in the stable direction . in the limit @xmath4 the eigenvalues in eq . ( [ eigenvalues ] ) are determined by the single potential minimum at @xmath149 . the form ( [ harmonic - zero ] ) of the eigenfunctions is appropriate in the region where the potential may be approximated by a harmonic well . outside the region , at @xmath161 , the values of the eigenfunction are exponentially small and do not influence its normalization . a more rigorous treatment in appendix provides the wkb approximation for the even eigenfunctions @xmath162 on the whole circle . under the potential barrier , if @xmath163 and @xmath164 the approximate solution at the energy @xmath165 is found to be @xmath166 in the vicinity of @xmath150 , at @xmath167 , we obtain : @xmath168 where @xmath169 even low - lying eigenstates of the second set concentrated near @xmath150 have eigenenergies @xmath170 , and their eigenfunctions can be obtained by using eq . ( [ conjugation - symmetry ] ) which follows from the symmetry of the hamiltonian . ( [ harmonic - zero])-([nu1 - 0 ] ) provide a zero - order approximation for the eigenfunction of the hamiltonian ( [ hamiltonian - on - circle ] ) on the circle . in the limit @xmath97 one can distinguish two types of corrections to the specified above eigenvalues and eigenfunctions of low - lying states . the first one is due to the omitted terms in the potential in the vicinity of its minima . such corrections are important in the treatment of states with @xmath171 . corrections of the second type originate from the tunneling through the potential barriers and therefore are exponentially small @xmath172 . details of computing the eigenfunctions are presented in appendix [ appendix : eigenmodes ] . it turns out that a straightforward calculation of the higher - mode contributions to the spectral decomposition requires knowledge of eigenfunctions with increasing accuracy in @xmath5 . in the next two sections we present a more elegant approach for which the approximation given by eqs . ( [ harmonic - zero])-([nu1 - 0 ] ) is sufficient . spectral decompositions of the response functions in the eigenmodes of the evolution operator are obtained in eqs . ( [ first - order - general2 ] , [ second - order - general2 ] ) in the general form . the expressions are not really useful for the calculation , since it would involve several three - dimensional integrations that involve the functions not explicitly known . nevertheless , the calculation can be carried out by implementing the representation in terms of functions on the circle . we start with the spectral decomposition of the linear response , and employ the correspondence between the functions @xmath173 in @xmath112 and @xmath174 on the circle , to obtain the following expansion : @xmath175 using the representation @xmath176 and the orthonormality of the functions @xmath137 with respect to the symmetric scalar product ( [ scalar - product - real ] ) , we can specify a rule to find the projection of @xmath173 on @xmath177 : @xmath178 this implies that @xmath179 is implemented on the circle by the function @xmath180 to proceed with the calculation of the response functions , we introduce a notation @xmath181 for the coefficient of the expansion of @xmath182 in @xmath119 . the expansion of @xmath119 in angular harmonics @xmath182 contains coefficients @xmath183 the symmetry property ( [ conjugation - symmetry ] ) allows to relate them to @xmath184 : @xmath185 therefore , the convolution in the first angular brackets in eq . ( [ first - order - general2 ] ) is equal to @xmath186 . the action of the @xmath187 in the last angular brackets , represented by a poisson bracket ( [ poisson - bracket ] ) , creates the following convolution @xmath188 where we have used the representation of @xmath189 on the circle given by eq . ( [ so(2,1)-circle ] ) . bearing in mind that @xmath190 is the eigenfunction of hamiltonian ( [ hamiltonian - on - circle ] ) and integrating by parts , we find the following relation : @xmath191 we conclude that the linear response function can be recast in the form of the time derivative of the two - point correlation function , in agreement with the fluctuation - dissipation theorem ( fdt ) : @xmath192 where we introduced @xmath193 following the approach developed in ref . for the purely deterministic situation , we introduce the matrix elements of the evolution operator between @xmath82-th and zero harmonics : @xmath194 spectral decomposition @xmath195 appears in the response functions of the system with noise . as we show below , the noiseless limit @xmath4 of the series @xmath196 reproduces the matrix element @xmath197 of the deterministic evolution operator , calculated in ref . . these series and their coefficients @xmath198 play an important role in the calculation of the second - order response function . we are now in a position to proceed with an explicit calculation of the coefficients in the spectral decomposition of the linear response function . we focus on the first set of modes with the energies @xmath165 , whose eigenfunctions are concentrated in the neighborhood of @xmath149 for @xmath148 . the coefficients @xmath199 for the other set of modes , with the energies @xmath170 , can be then easily found by employing the symmetry , described by eq . ( [ conjugation - symmetry ] ) : @xmath200 which also reflects the fact that the response function is real . the first two terms in the spectral expansion originate from the lowest rp resonances with the energies @xmath201 and @xmath202 . the eigenfunction with @xmath201 , concentrated in the vicinity of @xmath149 , is given by @xmath203 . the first integral , @xmath204 , is calculated without much effort . this is done by noticing that the main contribution to the integral comes from @xmath205 . expanding @xmath206 in the exponential results in a gaussian integral : @xmath207 where the normalization factor @xmath208 is related to @xmath5 via eq . ( [ a0 ] ) . the situation , however , is more complicated for higher modes with @xmath171 . due to the orthogonality of the eigenfunctions @xmath209 and @xmath210 with @xmath171 , the integral vanishes if we employ the lowest - order approximation for the wavefunctions : @xmath211 this actually means that more accurate expressions are needed to find the first nonvanishing term . note that one should use better approximations for both the exponential @xmath212 and the wave function . the corrections to the latter are computed by applying the standard quantum mechanical perturbation theory to the schrdinger equation . the procedure is feasible for few low modes only , since the calculation for the higher modes requires higher orders of the perturbation theory . however , it is possible to overcome this difficulty and find an explicit expression for @xmath213 for all @xmath156 by using the following trick . we notice that the dominant contribution to the integral @xmath214 does not involve higher - order perturbative calculations described above . the integral is dominated by the region @xmath215 , where we can approximate @xmath216 and expand it in the hermite polynomials @xmath217 with @xmath218 . all terms except for the last one vanish due to orthogonality of the hermite polynomials . further approximations lead to negligible corrections in the limit @xmath219 . we proceed by introducing @xmath220 our goal is to calculate the integral @xmath221 . this will be achieved by expressing it via the known integral @xmath222 for @xmath97 . consider @xmath223 . making use of @xmath224 , we integrate in eq . ( [ p - definition ] ) by parts , and neglect higher - order terms in @xmath5 . this results in the recurrence relation @xmath225 which being applied @xmath226 times and followed by setting @xmath227 yields : @xmath228 and we arrive at @xmath229 the second integral @xmath230 in the linear response is expressed via @xmath231 [ see eqs . ( [ r - conjugate - integral ] ) and ( [ r - conjugate ] ) ] . making use of the complex conjugation symmetry specified in eq . ( [ conjugation - symmetry ] ) this can be recast as @xmath232 the obtained integral looks more complicated compared to @xmath213 , even for the lowest resonance with @xmath233 . we first point out the problem and present its straightforward solution for @xmath233 . afterwards we apply a more elegant method , similar to the one developed above , which provides the solution for all modes . before starting we emphasize that , due to the eigenfunction symmetry , the discussion can be limited to the interval @xmath145 . three different analytical approximate forms for the eigenfunction near @xmath149 , @xmath150 , and under the barrier are specified in eqs . ( [ harmonic - zero ] ) - ( [ nu1 - 0 ] ) . as opposed to the case of @xmath213 , the integrand in @xmath231 does not experience exponential dependence on @xmath144 , and the exponential part of its dependence on @xmath5 in the contributions from all three regions amounts to a common factor @xmath234 . in the region @xmath235 , where the approximate forms , given by eqs . ( [ under - barrier - positive - half ] ) and ( [ harmonic - pi ] ) match , the solution behaves @xmath236 . this region , where the two solutions match , provides an essential contribution to the integral . we can break up the integral into two parts by introducing an intermediate point @xmath237 that belongs to the matching region . since for small @xmath148 one can always choose @xmath238 , the approximate forms provide good approximations to the left and to the right of the intermediate point , respectively . we intend to show that the sum of two integrals is independent of @xmath239 . the contribution from the region nearby @xmath150 has the form : @xmath240 we chose @xmath241 so that @xmath242 and retained first two terms in the expansion of @xmath243 . the integral is calculated by expressing @xmath244 in terms of @xmath245 and @xmath246 , since all three functions are solutions of the same second - order differential equation ( see ref . ) . we also employ the solution symmetry and the recurrence relation @xmath247 for hermite functions . the integration with the under - the - barrier function can be safely extended to @xmath149 , since the deviation of the approximate integrand from the actual function in a short region of length is finite . then the integration is performed exactly in terms of the hypergeometric function : @xmath248 we further expand eq . ( [ other - minimum - integral ] ) in @xmath249 and eq . ( [ under - barrier - integral ] ) in @xmath250 . in the sum of two pieces of the integral ( [ r0-conjugate - integral ] ) , the strongest dependence on @xmath241 cancels out . for the lowest resonance with @xmath233 there are no other contributions comparable with the constant in eq . ( [ under - barrier - integral ] ) , which is not the case if @xmath171 . as @xmath156 increases the divergence of the under - barrier solution becomes stronger , and one should take into account all @xmath241-dependent terms with non - positive real parts of the exponents that appear in eqs . ( [ other - minimum - integral ] ) and ( [ under - barrier - integral ] ) . for large @xmath156 this requires higher order of the perturbative calculations . one might have argued that , since the corrections to the wave functions have the same structure as the initial approximations for them , all relevant @xmath241-dependent terms should cancel out , and the corrections to the constant are small in @xmath5 . however , we can avoid exact unpleasant calculations by applying the method developed above . this is achieved by relating @xmath231 to the integral that can be calculated using the principal approximation for the eigenfunction specified in eqs . ( [ harmonic - zero ] - [ nu1 - 0 ] ) . according to the definition ( [ p - definition ] ) of @xmath251 and due to the wavefunction symmetry ( [ conjugation - symmetry ] ) , we have @xmath252 . the recurrence relation ( [ p - recurrence ] ) yields @xmath253 next , we consider the integral @xmath254 and notice that for its calculation it is sufficient to take the under - barrier approximation ( [ under - barrier - positive - half ] ) for @xmath210 on the whole half - circle . the resulting integral can be easily calculated : @xmath255 the integral converges , and the differences between the approximate and exact integrands in the narrow regions of width @xmath256 near @xmath149 and @xmath150 are smaller than @xmath257 and @xmath258 , respectively . the calculation of the second integral is completed by combining eqs . ( [ second - int - recurrence ] ) and ( [ second - int - integral ] ) followed by expressing @xmath259 via @xmath260 , according to eq . ( [ a2 ] ) : @xmath261 the method is naturally also applicable to the case @xmath233 , which has been treated earlier , using a more straightforward approach , with much more effort involved . we complete the calculation by inserting eqs . ( [ r - result ] ) and ( [ r - conjugate - result ] ) into eq . ( [ cal - qn ] ) to find that @xmath262 and @xmath263 have finite limits at @xmath4 , namely @xmath264 we conclude the section by demonstrating that the spectral decomposition of the linear response in the noiseless limit @xmath154 reproduces the asymptotic expansion of its purely deterministic counterpart presented in ref . . to that end we show that in the noiseless limit @xmath262 and @xmath263 coincide with the coefficients in the expansion @xmath265 at @xmath266 , where @xmath197 is defined as a matrix element of the liouvillian deterministic evolution : @xmath267 the integral can be calculated using the representation on the circle by solving the @xmath133 liouville equation @xmath268 [ see refs . ] : @xmath269 the expression is substantially simplified in the relevant for the linear response case @xmath266 . expanding the gauss hypergeometric function @xmath270 in @xmath271 into the hypergeometric series of @xmath272 to obtain @xmath273 and @xmath274 , we see directly that they are indeed reproduced by the noiseless limit of the coefficients @xmath275 and @xmath263 given by eq . ( [ limit - cal - q0 ] ) , and in the limit @xmath4 , i.e. @xmath276 spectral decomposition of nonlinear response functions in the case of finite noise is conceptually straightforward . the general expression for @xmath277 , decomposed in the eigenmodes of the fokker - planck operator @xmath6 is given by eq . ( [ second - order - general2 ] ) . in this section we demonstrate that in the noiseless limit @xmath154 the expansion coefficients converge to the coefficients of the long - time asymptotic expansion of the purely deterministic @xmath102 response that have been derived in our previous work @xcite . for the sake of simplicity we focus on the @xmath66 case when only one irreducible representation contributes to the dipole , i.e. @xmath278 . the expression ( [ second - order - general2 ] ) for the second - order response contains four matrix elements calculated in the previous section . the fifth matrix element in the middle angular brackets requires a careful treatment since @xmath131 includes the operator @xmath279 that acts on all momentum - dependent functions to the right . we first perform integration over the reduced phase space , represented by the middle angular brackets , which results in a @xmath0-dependent expression that also includes derivatives . integrating over @xmath0 by parts , and employing the symmetries , we obtain the second - order response function in the form of the following spectral decomposition : @xmath280 { \cal q}^*_{\mu^*,n}{\cal q}_{\lambda , n+1 } \\ & % \nonumber -\left(n - s+\frac{1}{2}\right ) \biggl[\zeta\mu t_2+n-1\biggr ] { \cal q}^*_{\mu^*,n+1}{\cal q}_{\lambda , n } \biggr ) \,,\end{aligned}\ ] ] where @xmath198 is given by eq . ( [ cal - qn ] ) . the symmetric @xmath281 coefficients @xmath282 are purely geometrical factors that do not depend on the dynamics and are @xcite . they are all related to @xmath283 with the help of the recurrence relations @xmath284 and , therefore , can be expressed in terms of the laplacian operator eigenfunctions @xmath285 . the spectral decomposition [ second - order2 ] is an exact expression for any given @xmath5 . the diffusion coefficient @xmath5 enters the expression via the eigenvalues as well as via @xmath286 , expressed in terms of the eigenfunctions as specified in eq . ( [ cal - an - series ] ) . in what follows we determine the leading contributions in the noiseless limit @xmath4 . compared to the linear case , noise plays a more delicate role for the spectral decomposition of the second - order response function . it provides convergence of the series over angular harmonics in the expression for the response function that does not appear in the linear case [ compare eqs . ( [ linear - response1 ] ) and ( [ second - order2 ] ) ] . to retrieve the asymptotic behavior of @xmath198 for @xmath148 and @xmath287 we apply the method developed in section [ section : linear ] to derive the following recurrence relation : @xmath288 specifically we replace @xmath289 with @xmath290 in the integral representation [ eq . ( [ r - definition ] ) ] for @xmath291 followed by integrating by parts . ( [ recurrence - previous ] ) results in another exact recurrent relation @xmath292 since only symmetric eigenfunctions contribute to the spectral decomposition , we have @xmath293 , which allows to express all quantities @xmath198 via @xmath199 specified by eqs . ( [ cal - qn ] ) . we start with the limit @xmath154 for fixed @xmath82 . although the dependence of @xmath184 on @xmath5 is singular , the coefficients @xmath198 have well - defined limits at @xmath4 that are equal to the corresponding coefficients in the expansion ( [ an - expansion ] ) of the purely deterministic matrix elements @xmath197 . this can be established in the following way . viewing @xmath197 as a matrix element of the evolution operator between zero and @xmath82-th harmonic , we employ the identity @xmath294 and the relations between the neighboring harmonics that follow from eqs . ( [ so(2,1)-circle ] ) to express @xmath295 in terms of @xmath296 and @xmath297 . finally , the result is expanded in @xmath272 according to eq . ( [ an - expansion ] ) . for the components oscillating as @xmath298 we arrive at a relation : @xmath299 which that can be viewed as the noiseless limit of the recurrence relation ( [ recurrence ] ) . combined with the already established equivalence @xmath300 for the zero term , this implies the equivalence @xmath301 for any given @xmath82 . so far the equivalence has been established for the rp resonances with @xmath165 . the coefficients for the other set of rp resonances can be easily found by employing the symmetry of the recurrence relation ( [ recurrence ] ) combined with @xmath302 , which results in : @xmath303 in the limit @xmath4 . this establishes the equivalence @xmath304 for the other set of resonances . we are now in a position to demonstrate that the series in angular harmonics @xmath82 for any coefficient in the spectral decomposition converges for small @xmath305 and in the noiseless @xmath154 limit reproduces the corresponding coefficient in the long - time asymptotic series for the purely deterministic response function @xmath277 . explicit expressions for the coefficients @xmath306 and @xmath307 that enter the expansion ( [ an - expansion ] ) for @xmath197 become increasingly lengthy as @xmath156 grows . the coefficients can be obtained by expanding the deterministic evolution matrix element in eq . ( [ an - result ] ) in powers of @xmath272 . the simplest expressions can be found for the lowest modes with @xmath233 whose energies are @xmath308 . for example , for the mode with @xmath309 , the recurrence relation ( [ q - recurrence ] ) implies @xmath310 , where @xmath311 and @xmath312 is specified in the r.h.s . ( [ limit - cal - q0 ] ) taken at @xmath233 . to analyze the limit @xmath313 for fixed finite @xmath148 we view @xmath184 as the fourier coefficients of the smooth function @xmath314 . consequently , @xmath198 decay faster than any power of @xmath82 for @xmath313 , and indeed we find from the recurrence relations the intermediate asymptotic of @xmath198 at @xmath315 : @xmath316 for larger @xmath82 the recurrence relation implies @xmath317 , which also leads to the decay faster than any power law as @xmath82 increases . finally we compare eq . ( [ cal - q - asympt ] ) with the asymptotic @xmath318 form of the deterministic coefficients @xmath306 to derive @xmath319 which is valid in the range @xmath315 for small @xmath5 . ( [ q - regularized ] ) means that noise provides a homogeneous cut - off for all rp resonances and the series over angular harmonics for the spectral decomposition coefficients converges at @xmath320 where eq . ( [ q - regularized ] ) holds . it remains to be demonstrated that the resulting spectral decomposition in the @xmath154 limit of vanishing noise coincides with the asymptotic long - time expansion of @xmath321 in the deterministic case . the latter expansion has been derived in ref . . first of all , we observe that the eigenvalues of the perron - frobenius operator converge in the @xmath4 to the factors that appear in the expansion of @xmath197 . the same factors that appear in the spectral decomposition of linear response have been interpreted in ref . @xcite as rp resonances . while the linear response function may be represented in the form of a converging series , the expansion of @xmath322 appears to be more involved . if @xmath102 , for a pair of resonances , the series ( [ second - order2 ] ) over @xmath82 diverges , as clearly seen from the power - law growth in eq . ( [ cal - q - asympt ] ) . in the deterministic case , the time dependence of the second - order response is determined by the converging series which contains the matrix elements @xmath197 . the expansion should be formally performed after the series summation . the long - time expansion of @xmath321 is an asymptotic , rather than a converging expansion in @xmath323 and @xmath324 . we have developed a method , equivalent to regrouping , which allowed to approximate the infinite sum by a sum of a finite number of terms . due to the alternating character of the series , any smooth cut - off effective for larger term numbers did not influence the result . the noise @xmath325 actually introduces a smooth cutoff in the sum over @xmath82 in eq . ( [ second - order2 ] ) for a given pair of resonances @xmath326 . the suppressive exponential factor is present for arbitrarily small positive @xmath5 . the series over angular harmonics is almost alternating and converging . therefore , the sum of the series does not depend on the value of @xmath5 as long as it is small . it is intuitively clear that another form of regularizing noise would lead to the same results . for the second - order response function we have demonstrated the equivalence of the spectral decomposition in the limit of vanishing noise @xmath4 with the asymptotic expansion in the case of deterministic dynamics . in short , this follows from the equality @xmath327 for the evolution operator matrix elements and the property that the asymptotic expansion of @xmath321 originates from the expansion of @xmath197 and the subsequent proper summation of the apparently diverging series . in the present manuscript we have studied linear and second - order nonlinear response of a stochastic system obtained by adding langevin noise to a deterministic system whose dynamics is strongly chaotic . the deterministic system is represented by a free particle moving on a @xmath8 compact riemann surface @xmath9 with constant negative curvature ( geodesic flow ) . we chose the random langevin force to be orthogonal to the particle momentum , so that the noise does not change the energy , and random walk occurs on the energy shell represented by the @xmath13 reduced phase space @xmath14 . the stochastic dynamics has been analyzed in terms of the fokker - planck operator @xmath328 , where the second term stands for the deterministic component ( advection ) , whereas the first term describes the langevin noise in the form of diffusion in the momentum space with @xmath5 being the diffusion coefficient . the @xmath102 case that corresponds to purely deterministic dynamics has been studied in our earlier work @xcite , where we have employed strong dynamical symmetry ( ds ) in the system to find the analytical solution to the problem . the langevin noise added to the system has been chosen in a way that it does not break down the ds . similar to the deterministic case , where the space of reduced phase space distributions has been decomposed into simpler components [ irreducible representations of the ds group @xmath32 ] invariant with respect to the liouville operator @xmath1 , the same components form invariant subspaces for our fokker - planck operator @xmath6 . here we have considered noise as a regularization to construct spectral decompositions of the perron - frobenius operator and focused on the noiseless @xmath154 limit of the stochastic dynamics . we have employed the ds to analyze the eigenvalue problem for the fokker - planck operator @xmath3 whose eigenfunctions are nice smooth functions for finite values of @xmath5 . in each irreducible component the distributions can be represented by functions on the circle , whereas the fokker - planck operator is represented by a sum of the first - order liouville operator and a @xmath133 diffusion operator @xmath330 . the deterministic dynamics has the stable @xmath150 and unstable @xmath149 fixed points that represent the dynamical processes along the stable and unstable directions , respectively . mapping the original problem onto much simpler @xmath133 stochastic dynamics on the circle allowed for an explicit analysis of the eigenvalue problem . being interested in the noiseless limit we focused on small finite values @xmath148 and found the relevant eigenfunctions of @xmath6 analytically using the wkb method where the diffusion coefficient @xmath5 plays the role of the square @xmath331 of the planck constant . the fluctuation - dissipation theorem relates the linear response function to the two - point correlation function calculated earlier @xcite , the latter being interpreted as an expansion in rp resonances using the language of rigged hilbert spaces . the long - time asymptotic expression for the second - order response functions , has the form of a spectral decomposition over the same set of resonances @xcite . interpretation of the asymptotic expansion in the nonlinear case is more involved due to the following reasons . the eigenmodes that correspond to the rp resonances are represented by generalized , rather than smooth functions . in the linear case the initial smooth distribution should be decomposed in the rp modes . the signal is computed by convoluting the rp modes with the smooth dipole function . both operations are well - defined for generalized functions . in the nonlinear case the second interaction with the driving field involves applying a differential operator to a generalized function followed by projecting it onto another generalized function . the legitimacy of the latter operation is less obvious , and has been related to the cancellation of apparently dangerous terms @xcite . interpretation of nonlinear response in terms of rp resonances has been addressed in this manuscript by considering the noiseless limit of the langevin dynamics . in the nonzero noise case @xmath332 , the spectral decomposition of the response functions of any order is conceptually more or less straightforward . we have demonstrated explicitly for geodesic flows that in the @xmath154 limit the relevant eigenvalues of @xmath6 converge to the rp resonances , whereas the coefficients in the spectral decompositions of the response functions converge to the coefficients of the asymptotic series in the purely deterministic @xmath102 expressions derived in ref . . summarizing , rp resonances can be interpreted as the noiseless limit for the eigenvalues of the fokker - planck operator @xmath6 , and the spectral decompositions in this limit reproduce the long - time asymptotic series for the response functions . note that the dynamical @xmath0-function can be also reproduced as the limit @xmath333 . in our model the rp decomposition for the linear response is represented by a converging series , whereas the nonlinear response is given by a non - converging asymptotic series . the converging character of the spectral decomposition is lost when the limit @xmath154 is applied . computation of the expansion coefficients in the nonlinear case requires a delicate summation of almost sign - alternating series whose convergence is ensured thanks to the noise . applications of our results are not limited to interpretations of purely deterministic quantities . irreversibility that shows itself in the decaying correlations appears only when the deterministic chaotic dynamics is regularized by some kind of coarse graining . full `` physical '' mixing always requires some diffusion mechanism . the difference between stable and chaotic deterministic dynamics is that the diffusion - induced `` physical '' mixing is much more efficient in a chaotic system . we consider a small fraction of the phase space that represent the initial conditions . as the ball of initial conditions is stretched and folded back , the shape becomes elongated along unstable directions and contracted along stable ones , while the phase space volume remains constant . the diffusion time scales as a square of the blurring size . therefore , a purely diffusive relaxation in the stable system occurs on the time scale of @xmath334 . in a chaotic system the scale of the density inhomogeneity decreases with time exponentially . considering for simplicity a @xmath13 reduced phase space of a chaotic hamiltonian system , a ball of size @xmath335 of initial conditions becomes a fettuccine - like shape . the fettucine length grows as @xmath336 . the size of the system @xmath337 induces the folding of the unstable manifold . we estimate the number of folds as @xmath338 . then the characteristic `` physical '' mixing time @xmath339 in a chaotic system with weak diffusion is very short compared to the stable case . note that the `` physical '' mixing time @xmath340 can be measured in photon echo experiments , where it represents the characteristic time scale of the photon echo decay as a function of the delay between the exciting and the dynamics - reversing pulses . in this section we calculate the relevant eigenmodes of the fokker - planck operator in a given representation of @xmath32 labeled by @xmath47 . they are given by symmetric solutions of the schrdinger equation on a circle with the effective hamiltonian ( [ hamiltonian - on - circle ] ) : @xmath341 the hamiltonian is self - adjoint with respect to a natural symmetric scalar product . for our purposes we focus on the weak limit @xmath97 . the low lying states with @xmath342 are concentrated near the the potential minima , in the vicinity of @xmath149 and @xmath150 . the principal approximation for the solutions of eq . ( [ schreq ] ) in the regions @xmath343 or @xmath344 can be found by retaining up to second order terms in the expansion of the potential . for @xmath343 we arrive at the approximate equation : @xmath345 that reproduces the well - known schrdinger equation for a linear harmonic oscillator . one can easily identify the characteristic scales @xmath346 and @xmath347 of the energy and length , respectively . a general solution of eq . ( [ schreq - near - zero - app ] ) can be represented in terms of the hermite functions as @xmath348 with @xmath349 and @xmath350 being complex constants . the parameter @xmath156 is related to the energy @xmath351 by @xmath352 as opposed to the case of a harmonic oscillator , the general solution ( [ harmonic - zero - app ] ) contains both decaying and growing waves . their relative amplitude can be determined only by solving the equation on the whole circle . the functions @xmath353 and @xmath354 that are linearly independent for any parameter @xmath156 can be simply related to confluent hyperbolic functions and parabolic cylinder functions , the latter being the solutions of the original schrdinger equation with the harmonic potential . we prefer to deal with the hermite functions rather than with other special functions since the former reproduce hermite polynomials in the case of integer order . we begin with the construction of a symmetric solution near @xmath150 . the requirement of the solution to be invariant with respect to @xmath359 combined with the relations between hermite functions leads to an identity @xmath360 and after some transformations of the hermite functions we arrive at an explicit form of a symmetric solution for ( [ harmonic - pi - app ] ) : @xmath361 therefore , in the region @xmath344 the symmetric solution is determined by two unknown parameters @xmath362 and @xmath357 ; both can assume complex values . the general wkb solution of eq . ( [ schreq ] ) under the barrier and around @xmath363 is represented by a superposition of two waves : @xmath364 where @xmath365 and @xmath366 are defined by @xmath367 the quasiclassical expansion over @xmath5 for the wavefunction phase can be employed when the wavelength @xmath368 does not change too fast : @xmath369 this holds when we are not too close to the classical turning points where @xmath365 turns to zero . small value of @xmath5 ensures that the classical turning points lie close to @xmath149 or @xmath150 . the wkb solutions under the barriers are valid at least for @xmath370 and @xmath371 . the right - hand sides of these inequalities contain expressions for the classical turning points . if @xmath372 , the validity of the quasiclassical approximation ( [ under - barriers ] ) conditions is limited to the regions latexmath:[\ ] ] in eq . ( [ compare - asympt ] ) we have also used the symmetry @xmath384 , which implies that for positive @xmath385 the asymptotic form of @xmath386 ( [ harmonic - pi ] ) includes the only term that is proportional to @xmath358 . therefore , the matching near @xmath150 yields @xmath387 the value of @xmath357 is then found from eq . ( [ bovera - app ] ) . the component of @xmath386 in eq . ( [ harmonic - pi - app ] ) with @xmath388 in front of it decreases rapidly @xmath389 under the barrier as @xmath144 deviates from @xmath378 . its magnitude can be estimated as @xmath390 the procedure of solving the schrdinger equation near @xmath149 is completely similar : in the region @xmath391 both approximate solutions ( [ harmonic - zero - app ] ) and ( [ under - barriers ] ) are adequate and can be matched . to that end we expand @xmath366 in the interval @xmath392 , neglecting the terms that are small compared to one : @xmath393 and also employ the asymptotic forms of the hermite functions in the overlap region . with @xmath144 approaching @xmath394 the second term in eq . ( [ under - barriers ] ) decreases and hence matches the second term in eq . ( [ harmonic - zero - app ] ) , @xmath395 the component with @xmath382 in front of it matches under the barrier with the component that has @xmath349 in front of it and decreases as @xmath144 deviates from @xmath394 : @xmath396 first we assume that the ratio @xmath397 given by eq . ( [ bovera - app ] ) does not contain exponentially small term @xmath398 . ( [ b1-app])-([a2-app ] ) imply that the second component in eq . ( [ harmonic - zero - app ] ) scales @xmath399 near @xmath149 . its presence is related to the nonresonant tunneling along the circle from the potential minimum into itself . the argument in the suppressive exponential factor is evaluated as @xmath400 the tunneling is nonresonant , since the two wells near @xmath149 and @xmath150 are offset by @xmath47 which is not an integer number . performing the same procedure in the region with @xmath401 as it has been done for @xmath402 , we find the symmetric solution @xmath403 valid for @xmath343 . for arbitrary @xmath156 the function may have a cusp at @xmath149 . the solution is smooth if @xmath156 is a nonnegative even integer . then the hermite functions @xmath159 reduce to hermite polynomials of even order . the dependence @xmath404 on @xmath156 with @xmath405 of the eigenstate energy for the states concentrated near @xmath149 is given by eq . ( [ nu ] ) . exponentially small corrections @xmath406 , induced by the tunneling can be neglected . on the other hand , the power - like in @xmath5 corrections to the energies , play a crucial role for @xmath171 . the latter corrections appear because of the deviation of trigonometric functions from their harmonic approximations . however the method developed in section [ section : linear ] allows to avoid these calculations completely . complex conjugation of the hamiltonian that corresponds to the change of sign of @xmath47 , or to the shift @xmath407 implies the quantization condition @xmath408 with @xmath409 for the other set of eigenstates . they have energies @xmath410 and are concentrated near @xmath150 . the eigenfunctions satisfy the relations similar to the specified above . the eigenfunctions of the hamiltonian ( [ hamiltonian - on - circle ] ) found above significantly differ from those of the linear harmonic oscillator only far from the interval @xmath411 . outside the interval the exponentially decaying functions take negligibly small values so that the normalization factor has the form typical to the harmonic oscillator : @xmath412 the phase of the normalized eigenfunctions of hermitian operators can be chosen arbitrarily . for our hamiltonian and scalar product ( [ scalar - product - real ] ) this reduces to the freedom in the sign choice . finally , for illustrative purposes , we calculate the correction to the energy @xmath165 of the eigenstate that originates from the vicinity of @xmath149 ( note that the result is not used anywhere ) : @xmath413 one of the ways derive eq . ( [ do - not - need - it ] ) is to consider the next term in the wkb expansion under the barrier . the wkb wavefunction contains the state energy . one can simply require that the wavefunction does not contain logarithmic terms due to their absence near @xmath149 .

modal dependence logic ( ) was introduced recently by vnnen . it enhances the basic modal language by an operator . for propositional variables the atomic formula intuitively states that the value of is determined solely by those of . we show that model checking forformulae over kripke structures is-complete and further consider fragments ofobtained by restricting the set of allowed propositional and modal connectives . it turns out that several fragments , e.g. , the one without modalities or the one without propositional connectives , remain-complete . we also consider the restriction ofwhere the length of each single dependence atom is bounded by a number that is fixed for the whole logic . we show that the model checking problem for this boundedis still-complete while for some fragments , e.g. , the fragment with only , the complexity drops to . we additionally extendby allowing classical disjunction introduced by sevenster besides dependence disjunction and show that classical disjunction is always at least as computationally bad as bounded arity dependence atoms and in some cases even worse , e.g. , the fragment with nothing but the two disjunctions is-complete . furthermore we almost completely classifiy the computational complexity of the model checking problem for all restrictions of propositional and modal operators for both unbounded as well as boundedwith both classical as well as dependence disjunction . this is the second arxiv version of this paper . it extends the first version by the investigation of the classical disjunction . a shortened variant of the first arxiv version was presented at sofsem 2012 . * acm subject classifiers : * f.2.2 complexity of proof procedures ; f.4.1 modal logic ; d.2.4 model checking * keywords : * dependence logic , modal logic , model checking , computational complexity

1104.1034

dependence among values of variables occurs everywhere in computer science ( databases , software engineering , knowledge representation , ai ) but also the social sciences ( human history , stock markets , etc . ) . in his monograph @xcite in 2007 vnnen introduced functional dependence into the language of first - order logic . functional dependence of the value of @xmath9 from the values of @xmath10 means that there exists a determinating function @xmath11 with @xmath12 , i.e. , the value of @xmath9 is completely determined by the values of @xmath13 alone . we denote this form of dependence ( or determination ) by the _ dependence atom _ @xmath14 . to examine dependence between situations , plays , worlds , events or observations we consider collections of these , so called _ teams_. for example , a database can be interpreted as a team . in this case @xmath15 means that in every record the value of the attribute @xmath9 is determined by the values of the attributes @xmath13 . in modal logic a team is a set of worlds in a kripke structure . here @xmath14 means that in every world of the team the value of the atomic proposition @xmath9 is determined by the propositions @xmath10 , i.e. , there is a fixed boolean function @xmath16 that determines the value of @xmath9 from the values of @xmath13 for all worlds in the team . in first - order logic @xmath17 means the same for a function @xmath18 where @xmath19 is the universe of a first - order structure . _ dependence logic _ @xcite is then defined by simply adding dependence atoms to usual first - order logic and _ modal dependence logic _ ( @xmath0 ) @xcite is defined by introducing dependence atoms to modal logic . besides the inductive semantics ( which we will use here ) vnnen also gave two equivalent game - theoretic semantics for @xmath0@xcite . sevenster showed that for singleton sets of worlds there exists a translation from @xmath0to plain modal logic @xcite . sevenster also showed that the satisfiability problem for @xmath0is @xmath20@xmath21-complete @xcite and lohmann and vollmer continued the complexity analysis of the satisfiability problem for @xmath0by systematically restricting the set of allowed modal and propositional operators and completely classifying the complexity for all fragments of @xmath0definable in this way @xcite . sevenster @xcite also introduced classical disjunction ( which is _ classical _ in a more set theoretic way of looking at the semantics ; cf . @xcite ) into the language of @xmath0 . in the following we always think of the version that includes both classical disjunction ( here denoted by @xmath22 ) as well as dependence disjunction when we write @xmath0 . the method of systematically classifying the complexity of logic related problems by restricting the set of operators allowed in formulae goes back to lewis who used this method for the satisfiability problem of propositional logic @xcite . recently it was , for example , used by hemaspaandra et al . for the satisfiability problem of modal logic @xcite and by lohmann and vollmer for the satisfiability problem of @xmath0@xcite . the motivation for this approach is that by systematically examining all fragments of a logic one might find a fragment which allows for efficient algorithms but still has high enough expressivity to be useful in practice . on the other hand , this systematic approach usually leads to insights into the sources of hardness , i.e. , the exact components of the logic that make satisfiability , model checking etc . hard . in this paper we transfer the method from satisfiability @xcite to model checking and classify the model checking problem for almost all fragments of @xmath0definable by restricting the set of allowed modal ( @xmath23 , @xmath7 ) and propositional ( @xmath24 , @xmath25 , @xmath22 , @xmath26 ) operators to an arbitrary subset of all operators . the model checking problem asks whether a given formula is true in a given team of a given kripke structure . for plain modal logic this problem is solvable in @xmath8as shown by clarke et al . a detailed complexity classification for the model checking problem over fragments of modal logic was shown by beyersdorff et al . @xcite ( who investigate the temporal logic @xmath27which contains plain modal logic as a special case ) . in the case of @xmath0it turns out that model checking is @xmath6-complete in general and that this still holds for several seemingly quite weak fragments of @xmath0 , e.g. , the one without modalities or the one where nothing except dependence atoms and @xmath7 is allowed ( first and fourth line in [ results - unbounded - dep ] ) . strangely , this also holds for the case where only the both disjunctions @xmath28 and @xmath22 are allowed and not even dependence atoms occur ( third line in [ results - unbounded - dep ] ) . furthermore it seems natural to not only restrict modal and propositional operators but to also impose restrictions on dependence atoms . one such restriction is to limit the arity of dependence atoms , i.e. , the number @xmath29 of variables @xmath13 by which @xmath9 has to be determined to satisfy the formula @xmath14 , to a fixed upper bound @xmath30 ( the logic is then denoted by @xmath31 ) . for this restriction model checking remains @xmath6-complete in general but for the fragment with only the @xmath7 operator allowed this does not hold any more ( seventh line in table [ results - bounded - dep ] ) . in this case either @xmath24 ( fourth line in table [ results - bounded - dep ] ) or @xmath25 ( sixth line in table [ results - bounded - dep ] ) is needed to still get @xmath6-hardness . we classify the complexity of the model checking problem for fragments of @xmath0with unbounded as well as bounded arity dependence atoms . we are able to determine the tractability of each fragment except the one where formulae are built from atomic propositions and unbounded dependence atoms only by disjunction and negation ( sixth line in table [ results - unbounded - dep ] ) . in each of the other cases we either show @xmath6-completeness or show that the model checking problem admits an efficient ( polynomial time ) solution . @xmath32 @xmath33 operator present @xmath34 operator absent @xmath35 complexity independent of operator @xmath36 @xmath33 operator present @xmath34 operator absent @xmath35 complexity independent of operator in table [ results - unbounded - dep ] we list our complexity results for the cases with unbounded arity dependence atoms and in table [ results - bounded - dep ] for the cases with an a priori bound on the arity . in these tables a + means that the operator is allowed , a - means that the operator is forbidden and a * means that the operator does not affect the complexity of the problem . we will briefly present the syntax and semantics of @xmath0 . for a more in - depth introduction we refer to vnnen s definition of @xmath0@xcite and sevenster s model - theoretic and complexity analysis @xcite which also contains a self - contained introduction to @xmath0 . [ defmdlsyn](syntax of @xmath0 ) + let @xmath37 be an arbitrary set of atomic propositions and @xmath38 . then @xmath0is the set of all formulae built from the following rules : @xmath39 note that negation is only atomic , i.e. , it is only defined for atomic propositions and dependence atoms . we sometimes write @xmath40 ( resp . @xmath41 ) for @xmath42 ( resp . @xmath43 ) . for a dependence atom @xmath44 we define its _ arity _ as @xmath29 , i.e. , the arity of a dependence atom is the arity of the determinating function whose existence it asserts . in section [ sec : bounded - fragments ] we will investigate the model checking problem for the following logic . [ defmdlk](@xmath31 ) + @xmath31is the subset of @xmath0that contains all formulae which do not contain any dependence atoms whose arity is greater than @xmath45 . we will classify @xmath0for all fragments defined by sets of operators . [ defopfrag](@xmath46 ) + let @xmath47 . by @xmath46 ( resp . @xmath48 ) we denote the subset of @xmath0(resp . @xmath31 ) built from atomic propositions using only operators from @xmath49 . we sometimes write @xmath50 instead of @xmath51 . @xmath0formulae are interpreted over kripke structures . [ defkripke](kripke structure ) + an _ @xmath37-kripke structure _ is a tuple @xmath52 where @xmath53 is an arbitrary non - empty set of _ worlds _ , @xmath54 is the _ accessibility relation _ and @xmath55 is the _ labeling function_. [ defmdlsem](semantics of @xmath0 ) + in contrast to common modal logics , truth of a @xmath0formula is not defined with respect to a single world of a kripke structure but with respect to a set ( or _ team _ ) of worlds . let @xmath37 be a set of atomic propositions and @xmath56 . the _ truth _ of a formula @xmath57 in a team @xmath58 of an @xmath37-kripke structure @xmath52 is denoted by @xmath59 and is defined as follows : @xmath60 note that this semantics is a conservative extension of plain modal logic semantics , i.e. , it coincides with the latter for formulae which do neither contain dependence atoms nor classical disjunction . rationales for this semantics especially for the case of the negative dependence atom were given by vnnen @xcite . in the remaining we will classify the complexity of the model checking problem for fragments of @xmath0and @xmath31 . [ defmdlmc](@xmath61@xmath62 ) + let @xmath47 . then the model checking problem for @xmath46 ( resp . @xmath48 ) over kripke structures is defined as the canonical decision problem of the set @xmath63 we write @xmath61@xmath62for @xmath64 . before we begin with the classification we state a lemma showing that it does not matter whether we include @xmath65 , @xmath66 or @xmath26 in a sublogic @xmath46 of @xmath0since this does not affect the complexity of @xmath61@xmath67 . [ neg - dont - matter ] let @xmath49 be an arbitrary set of @xmath0operators , i.e. , @xmath68 . then we have that @xmath69 it suffices to show @xmath70 . so let @xmath71 a kripke structure , @xmath58 , @xmath72 and the variables of @xmath73 among @xmath13 . let @xmath74 be fresh propositional variables . then @xmath75 iff @xmath76 , where @xmath77 with @xmath78 defined by @xmath79 for all @xmath80 and @xmath81 , and @xmath82 defined by @xmath83 first we will show that the most general of our problems is in @xmath6and therefore all model checking problems investigated later are as well . [ mdlmc - in - np ] let @xmath49 be an arbitrary set of @xmath0operators . then @xmath61@xmath67 is in @xmath6 . and hence also @xmath84 is in @xmath85 for every @xmath86 . the following non - deterministic top - down algorithm checks the truth of the formula @xmath73 on the kripke structure @xmath87 in the evaluation set @xmath88 in polynomial time . .... $ % bool check($w$ , $ t$ , $ \varphi$ ) % $ case $ \varphi$ when $ \varphi = p$ foreach $ s \in t$ if not $ p \in \pi(s)$ then return false return true \needspace{5\baselineskip} when $ \varphi=\neg p$ foreach $ s \in t$ if $ p \in \pi(s)$ then return false return true \needspace{7\baselineskip} when $ \varphi = \,\dep{p_1,\ldots , p_n}$ foreach $ ( s , s')\in t\times t$ if $ \pi(s)\cap\{p_1,\dots , p_{n-1}\}\,=\,\pi(s')\cap\{p_1,\dots , p_{n-1}\}$ then // i.e. , $ s$ and $ s'$ agree on the values of the propositions $ p_1,\dots , p_{n-1}$ if ( $ q\in\pi(s)$ and not $ q\in \pi(s')$ ) or ( not $ q\in\pi(s)$ and $ q\in \pi(s')$ ) then return false return true \needspace{4\baselineskip} when $ \varphi = \ , \neg\dep{p_1,\ldots , p_n}$ if $ t=\emptyset$ return true return false \needspace{6\baselineskip} when $ \varphi = \psi\vee \theta$ guess two sets of states $ a,\,b \subseteq s$ if not $ a\cup b = t$ then return false return ( check($w , a,\psi$ ) and check($w , b,\theta$ ) ) \needspace{2\baselineskip} when $ \varphi = \psi\nor \theta$ return ( check($w , t,\psi$ ) or check($w , t,\theta$ ) ) \needspace{2\baselineskip} when $ \varphi= \psi\wedge \theta$ return ( check($w , t,\psi$ ) and check($w , t,\theta$ ) ) \needspace{8\baselineskip} when $ \varphi= \ax \psi$ $ t':=\emptyset$ foreach $ s ' \in s$ foreach $ s \in t$ if $ ( s , s ' ) \in r$ then $ t':= t ' \cup \{s'\}$ // $ t'$ is the set of all successors of all states in $ t$ return check($w , t',\psi$ ) \needspace{7\baselineskip} when $ \varphi= \ex \psi$ guess set of states $ t'\subseteq s$ foreach $ s\in t$ if there is no $ s'\in t'$ with $ ( s , s')\in r$ then return false // $ t'$ contains at least one successor of every state in t return check($w , t',\psi$ ) .... now we will see that the model checking problem is @xmath6-hard and that this still holds without modalities . [ wedge - vee - np - complete ] let @xmath89 . then @xmath61@xmath67 is @xmath6-complete . furthermore , @xmath90@xmath91 is @xmath6-complete for every @xmath30 . membership in @xmath6follows from proposition [ mdlmc - in - np ] . for the hardness proof we reduce from @xmath92 . for this purpose let @xmath93 be an arbitrary @xmath94 formula with variables @xmath95 . let @xmath87 be the kripke structure @xmath96 over the atomic propositions @xmath97 defined by @xmath98 let @xmath99 be the @xmath100 formula @xmath101 and let @xmath102 the evaluation set . we will show that @xmath103 iff @xmath104 . then it follows that @xmath105 and therefore @xmath106 is @xmath6-hard . now assume @xmath103 and @xmath107 an interpretation with @xmath108 . from the valuations @xmath109 of all @xmath110 we construct subteams @xmath111 such that for all @xmath112 it holds that @xmath113 with @xmath114 . the @xmath115 are constructed as follows @xmath116 since every clause in @xmath73 is satisfied by some valuation @xmath117 or @xmath118 we have that @xmath119 such that @xmath120 . on the other hand , assume that @xmath104 , therefore we have @xmath121 such that for all @xmath122 it holds that @xmath123 . therefore @xmath124 is constant for all elements @xmath125 . from this we can construct a valid interpretation @xmath107 for @xmath73 . for all @xmath126 let @xmath127 . for every @xmath128 we consider @xmath115 . if for every element @xmath125 it holds that @xmath129 then we have for all @xmath130 that @xmath110 is a literal in @xmath131 . in order to satisfy those @xmath131 we set @xmath132 . if for every element @xmath125 it holds that @xmath133 then we have for every @xmath130 that @xmath134 is a literal in @xmath131 . in order to satisfy those @xmath131 we set @xmath118 . + since for every @xmath135 there is a @xmath126 with @xmath125 we have an evaluation @xmath107 that satisfies every clause in @xmath73 . therefore we have @xmath108 . instead of not having modalities at all we can also allow nothing but the @xmath7 modality , i.e. , we disallow propositional connectives and the @xmath23 modality , and model checking is @xmath6-complete as well . [ diamond - np - complete ] let @xmath136 . then @xmath61@xmath67 is @xmath6-complete . membership in @xmath6follows from proposition [ mdlmc - in - np ] again . for hardness we again reduce from @xmath137 . let @xmath138 be an arbitrary @xmath139formula built from the variables @xmath140 . let @xmath87 be the kripke structure @xmath141 , over the atomic propositions @xmath142 , shown in figure [ figure : diamond - dep1 ] and formally defined by @xmath143 ( c1 ) @xmath144 ; ( x1 _ ) [ below left of = c1 , label = left:@xmath145 @xmath146 ; ( x1 ) [ below right of = c1 , label = right:@xmath147 @xmath148 ; ( c1 ) edge [ bend left ] node ( x1 ) ; let @xmath99 be the @xmath149 formula @xmath150 and let @xmath151 . we will show that @xmath103 iff @xmath104 . hence , @xmath152 and @xmath153 is @xmath6-hard . first suppose we have an interpretation @xmath107 that satisfies @xmath73 . from the valuations of @xmath107 we will construct a successor team @xmath154 of @xmath88 , i.e. , for all @xmath155 there is an @xmath156 s.t . @xmath157 with @xmath158 . @xmath154 is defined by : @xmath159 since @xmath107 satisfies every clause @xmath131 of @xmath73 we have that for every @xmath131 there is an @xmath110 with @xmath160 it follows that for every @xmath161 there is an @xmath162 such that @xmath157 . by construction of @xmath154 it is not possible to have both @xmath146 and @xmath148 in @xmath154 . hence for all elements @xmath163 it follows that @xmath164 and therefore @xmath165 . thus @xmath158 . on the other hand assume @xmath166 . then there is a successor set @xmath154 of @xmath88 s.t . for every @xmath161 there is an @xmath162 with @xmath157 and @xmath167 . we construct @xmath107 as follows : @xmath168 note that in the latter case it does not matter if 0 or 1 is chosen . since @xmath169 and for every @xmath126 it holds that @xmath170 we have that for every @xmath126 at most one of @xmath146 or @xmath148 is in @xmath154 . it follows that @xmath107 is well - defined . since for every @xmath171 there is an @xmath172 s.t . @xmath173 with @xmath174 , we have by contruction of @xmath87 that @xmath107 satisfies every clause @xmath131 of @xmath73 . from this follows @xmath103 . if we disallow @xmath7 but allow @xmath23 instead we have to also allow @xmath25 to get @xmath6-hardness . [ box - vee - np - complete ] let @xmath175 . then @xmath61@xmath67 is @xmath6-complete . also , @xmath90@xmath91 is @xmath6-complete for every @xmath30 . membership in @xmath6follows from proposition [ mdlmc - in - np ] again . to prove hardness , we will once again reduce @xmath92to this problem . let @xmath138 be an arbitrary @xmath94 formula over the variables @xmath140 . let @xmath87 be the structure @xmath141 , over the atomic propositions @xmath13 , shown in figures [ chain1 ] to [ chain6 ] and formally defined as follows : @xmath176 \multicolumn{3}{l}{r\ , \cap\bigcup\limits_{j\in\{1,\dots , n\}}\,\{(s_i , r_i^j),(s_i,\overline r_i^j)\}\ { \mathrel{\mathop:}=}}\\ \multicolumn{3}{l}{\qquad\left\{\begin{array}{lll } \{(s_i , r_i^1)\ } & \text{iff $ x_1 $ occurs in $ c_i$ ( positively or negatively ) } & \text{\emph{(fig.~\ref{chain1})}}\\ \{(s_i , r_i^1),(s_i , \overline r_i^1 ) \ } & \text{iff $ x_1 $ does not occur in $ c_i$ } & \text{\emph{(fig.~\ref{chain2})}}\\ \end{array}\right.}\\[4ex ] \multicolumn{3}{l}{r \,\cap \bigcup\limits_{k\in\{1,\dots , n\}}\,\{(r_i^j , r_i^{k}),(r_i^j,\overline r_i^{k}),(\overline r_i^j , r_i^{k}),(\overline r_i^j,\overline r_i^{k})\}\ { \mathrel{\mathop:}=}}\\ \multicolumn{3}{l}{\qquad\left\{\begin{array}{lll } \{(r_i^j , r_i^{j+1})\ } & \parbox[t]{15em}{iff $ x_{j}$ and $ x_{j+1}$ both occur in $ c_i$ } & \text{\emph{(fig.~\ref{chain3})}}\\ \{(r_i^j , r_i^{j+1}),(r_i^j , \overline r_i^{j+1 } ) \ } & \parbox[t]{15em}{iff $ x_{j}$ occurs in $ c_i$ but $ x_{j+1}$ does not occur in $ c_i$ } & \text{\emph{(fig.~\ref{chain4})}}\\[3ex ] \{(r_i^j , r_i^{j+1}),(\overline r_i^j , r_i^{j+1})\ } & \parbox[t]{15em}{iff $ x_j$ does not occur in $ c_i$ but $ x_{j+1}$ does occur in $ c_i$ } & \text{\emph{(fig.~\ref{chain5})}}\\[3ex ] \{(r_i^j , r_i^{j+1}),(\overline r_i^j , \overline r_i^{j+1 } ) \ } & \parbox[t]{15em}{iff neither $ x_j$ nor $ x_{j+1}$ occur in $ c_i$ } & \text{\emph{(fig.~\ref{chain6 } ) } } \end{array}\right.}\\[11ex ] \pi(s_i ) & { \mathrel{\mathop:}= } & \emptyset\\ \pi(r_i^j ) & { \mathrel{\mathop:}= } & \begin{cases}\{p_j\ } & \text{iff $ x_j$ occurs in $ c_i$ positively or not at all}\\ \emptyset & \text{iff $ x_j$ occurs in $ c_i$ negatively}\end{cases}\\ \pi(\overline r_i^j ) & { \mathrel{\mathop:}= } & \emptyset \end{array}\ ] ] ( c1 ) @xmath177 ; ( s1x1 ) [ below of = c1 ] @xmath178 ; ( c1 ) edge node ( s1x1 ) ; ( c1 ) @xmath177 ; ( s1x1 ) [ below right of = c1 ] @xmath178 ; ( s1x1 _ ) [ below left of = c1 ] @xmath179 ; ( c1 ) edge node ( s1x1 _ ) edge node ( s1x1 ) ; ( s1x2 ) @xmath180 ; ( s1x1 ) [ above of = s1x2 ] @xmath181 ; ( s1x1 ) edge node ( s1x2 ) ; ( s1x1 ) @xmath181 ; ( s1x2 _ ) [ below left of = s1x1 ] @xmath182 ; ( s1x2 ) [ below right of = s1x1 ] @xmath180 ; ( s1x1 ) edge node ( s1x2 ) ; ( s1x1 ) edge node ( s1x2 _ ) ; ( s1x2 ) @xmath180 ; ( s1x1 ) [ above right of = s1x2 ] @xmath181 ; ( s1x1 _ ) [ above left of = s1x2 ] @xmath183 ; ( s1x1 ) edge node ( s1x2 ) ; ( s1x1 _ ) edge node ( s1x2 ) ; ( s1x1 ) @xmath181 ; ( s1x1 _ ) [ left of = s1x1 ] @xmath183 ; ( s1x2 ) [ below of = s1x1 ] @xmath180 ; ( s1x2 _ ) [ below of = s1x1 _ ] @xmath184 ; ( s1x1 ) edge node ( s1x2 ) ; ( s1x1 _ ) edge node ( s1x2 _ ) ; let @xmath99 be the @xmath185 formula @xmath186 and let @xmath187 . then , as we will show , @xmath103 iff @xmath188 and therefore @xmath189 is @xmath6-complete . intuitively , the direction from left to right holds because the disjunction splits the team @xmath190 of all starting points of chains of length @xmath29 into @xmath29 subsets ( one for each variable ) in the following way : @xmath177 is in the subset that belongs to @xmath110 iff @xmath110 satisfies the clause @xmath131 under the variable valuation that satisfies @xmath73 . then the team that belongs to @xmath110 collectively satisfies the disjunct @xmath191 of @xmath99 . for the reverse direction the @xmath192 states are needed to ensure that a state @xmath177 can only satisfy a disjunct @xmath191 if there is a variable @xmath110 that occurs in clause @xmath131 ( positively or negatively ) and satisfies @xmath131 . more precisely , assume @xmath107 is a satisfying interpretation for @xmath73 . from @xmath107 we construct subteams @xmath111 with @xmath119 s.t . for all @xmath126 it holds that @xmath193 . @xmath115 is defined by @xmath194 for all @xmath195 . obviously , for all @xmath126 it holds that @xmath193 . now we will show that for all @xmath196 there is a @xmath195 such that @xmath125 . for this purpose let @xmath196 and suppose @xmath131 is satisfied by @xmath132 for a @xmath195 . then , by definition of @xmath87 , @xmath197 , hence @xmath198 and therefore @xmath125 . if , on the other hand , @xmath131 is satisfied by @xmath199 then we have that @xmath200 , hence @xmath201 and again it follows that @xmath125 . altogether we have that for all @xmath202 there is a @xmath126 such that @xmath125 . it follows that @xmath119 and therefore @xmath188 . on the other hand assume @xmath188 . therefore we have @xmath203 with @xmath193 for all @xmath195 . we define a valuation @xmath107 by @xmath204 since every @xmath177 is contained in a @xmath115 we know that for all @xmath196 there is a @xmath195 with @xmath205 . from this it follows that @xmath110 occurs in @xmath131 ( positively or negatively ) since otherwise , by definition of @xmath87 , both @xmath181 and @xmath192 would be reachable from @xmath177 . it also holds that @xmath206 or @xmath207 . in the former case we have that @xmath208 , hence , by definition of @xmath87 , @xmath110 is a literal in @xmath131 . by construction of @xmath107 it follows that @xmath131 is satisfied . in the latter case it holds that @xmath209 is a literal in @xmath131 . again , by construction of @xmath107 it follows that @xmath131 is satisfied . hence , @xmath103 . the following example demonstrates the construction from the previous proof . [ box - vee - np - complete - example ] let @xmath73 be the @xmath139formula @xmath210 the corresponding kripke structure @xmath87 shown in figure [ fig : wedge - vee - np - complete - example ] has _ levels _ 0 to 4 where the @xmath126th level ( corresponding to the variable @xmath110 in the formula @xmath73 ) is the set of nodes reachable via exactly @xmath126 transitions from the set of nodes @xmath211 , @xmath212 and @xmath213 ( corresponding to the clauses of @xmath73 ) . in this example all non connected states ( which do not play any role at all ) are not shown . the @xmath0formula corresponding to @xmath73 is @xmath214 ( c1 ) @xmath211 ; ( c1 _ ) [ right of = c1 ] @xmath211 ; ( c2 ) [ right of = c1 _ ] @xmath212 ; ( c2 _ ) [ right of = c2 ] @xmath211 ; ( c3 ) [ right of = c2 _ ] @xmath213 ; ( s1x1 ) [ below of = c1 , label = left : ] @xmath215 ; ( s2x1 _ ) [ below left of = c2 ] @xmath216 ; ( s2x1 ) [ below right of = c2 , label = left:@xmath217 @xmath218 ; ( s3x1 ) [ below of = c3 , label = left:@xmath217 @xmath219 ; ( s1x2 ) [ below of = s1x1 , label = left:@xmath220 @xmath221 ; ( s2x2 ) [ below left of = s2x1 , label = left:@xmath220 @xmath222 ; ( s3x2 ) [ below of = s3x1 ] @xmath223 ; ( s1x3 ) [ below of = s1x2 , label = left:@xmath224 @xmath225 ; ( s2x3 ) [ below of = s2x2 ] @xmath226 ; ( s3x3 _ ) [ below left of = s3x2 ] @xmath227 ; ( s3x3 ) [ below right of = s3x2 , label = left:@xmath224 @xmath228 ; ( s1x4 ) [ below right of = s1x3 , label = left:@xmath229 @xmath230 ; ( s1x4 _ ) [ below left of = s1x3 , label = left:@xmath231 @xmath232 ; ( s2x4 ) [ below of = s2x3 , label = left:@xmath229 @xmath233 ; ( s3x4 ) [ below of = s3x3 , label = left:@xmath229 @xmath234 ; ( s3x4 _ ) [ below of = s3x3 _ , label = left:@xmath231 @xmath235 ; ( c1 ) edge node ( s1x1 ) ( s1x1 ) edge node ( s1x2 ) ( s1x2 ) edge node ( s1x3 ) ( s1x3 ) edge node ( s1x4 _ ) edge node ( s1x4 ) ( c2 ) edge node ( s2x1 ) edge node ( s2x1 _ ) ( s2x1 ) edge node ( s2x2 ) ( s2x1 _ ) edge node ( s2x2 ) ( s2x2 ) edge node ( s2x3 ) ( s2x3 ) edge node ( s2x4 ) ( c3 ) edge node ( s3x1 ) ( s3x1 ) edge node ( s3x2 ) ( s3x2 ) edge node ( s3x3 ) ( s3x2 ) edge node ( s3x3 _ ) ( s3x3 _ ) edge node ( s3x4 _ ) ( s3x3 ) edge node ( s3x4 ) ; let @xmath236 with @xmath188 and for all @xmath237 let @xmath238 with @xmath123 and @xmath239 . by comparing the formulae @xmath240 with the chains in the kripke structure one can easily verify that @xmath241 i.e. , there can at most be one of @xmath211 and @xmath213 in @xmath242 since @xmath243 and @xmath212 can not be in @xmath242 since its direct successors @xmath244 do not agree on @xmath245 . in this case @xmath246 means that @xmath247 is satisfied by setting @xmath248 and the fact that @xmath249 corresponds to the fact that there is no way to satisfy @xmath250 via @xmath251 , because @xmath251 does not occur in @xmath250 . analogously , @xmath252 or @xmath253 , and @xmath254 and @xmath255 . now , e.g. , the valuation @xmath107 where @xmath251 , @xmath256 and @xmath257 evaluate to true and @xmath258 to false satisfies @xmath73 . from this valuation one can construct sets @xmath259 with @xmath260 such that @xmath123 for all @xmath261 by defining @xmath262 for all @xmath126 . this leads to @xmath263 , @xmath264 and @xmath265 . the gray colourings indicate which chains ( resp . clauses ) are satisfied on which levels ( resp . by which variables ) . @xmath99 ( resp . @xmath73 ) is satisfied because there is a gray coloured state in each chain . if we disallow both @xmath7 and @xmath25 the problem becomes tractable since the non - deterministic steps in the model checking algorithm are no longer needed . [ box - wedge - in - p ] let @xmath266 . then @xmath61@xmath67 is in @xmath8 . algorithm [ algo : check ] is a non - deterministic algorithm that checks the truth of an arbitrary @xmath0formula in a given structure in polynomial time . since @xmath267 it holds that @xmath268 . therefore the non - deterministic steps are never used and the algorithm is in fact deterministic in this case . note that this deterministic polynomial time algorithm is a top - down algorithm and therefore works in a fundamentally different way than the usual deterministic polynomial time bottom - up algorithm for plain modal logic . now we have seen that @xmath153 is tractable if @xmath269 and @xmath270 since these two operators are the only source of non - determinism . on the other hand , @xmath153 is @xmath6-complete if @xmath271 and either @xmath272 ( theorem [ diamond - np - complete ] ) or @xmath273 ( theorem [ box - vee - np - complete ] ) . the remaining question is what happens if only @xmath25 ( but not @xmath23 ) is allowed . unfortunately this case has to remain open for now . we will now show that @xmath274 is in @xmath8if we impose the following constraint on the dependence atoms in formulae given as part of problem instances : there is a constant @xmath275 such that in any input formula it holds for all dependence atoms of the form @xmath276 that @xmath277 . to prove this statement we will decompose it into two smaller propositions . first we show that even the whole @xmath278 fragment with unrestricted @xmath1 atoms is in @xmath8as long as it is guaranteed that in every input formula at least a specific number of dependence atoms depending on the size of the kripke structure occur . we will need the following obvious lemma stating that a dependence atom is always satisfied by a team containing at least half of all the worlds . [ split - in - two ] let @xmath52 be a kripke structure , @xmath279 ( @xmath280 ) an atomic formula and @xmath58 an arbitrary team . then there is a set @xmath281 such that @xmath282 and @xmath283 . let @xmath284 and @xmath285 . then @xmath286 and @xmath287 . therefore there is an @xmath288 such that @xmath289 . let @xmath290 . since @xmath9 is either labeled in every state of @xmath154 or in no one , it holds that @xmath291 . we will now formalize a notion of many dependence atoms in a formula . [ defmdlmcl ] for @xmath57 let @xmath292 be the number of positive dependence atoms in @xmath73 . let @xmath293 an arbitrary function and @xmath294 . then @xmath295@xmath296 ( resp . @xmath90@xmath297 ) is the problem @xmath61@xmath67 ( resp . @xmath90@xmath91 ) restricted to inputs @xmath298 that satisfy the condition @xmath299 . if we only allow @xmath25 and we are guaranteed that there are many dependence atoms in each input formula then model checking becomes trivial even for the case of unbounded dependence atoms . [ many - dep - atoms - trivial ] let @xmath300 . then @xmath301 is trivial , i.e. , for all kripke structures @xmath52 and all @xmath72 such that the number of positive dependence atoms in @xmath73 is greater than @xmath302 it holds for all @xmath58 that @xmath120 . let @xmath52 , @xmath72 , @xmath58 be an arbitrary instance with @xmath303 dependence atoms in @xmath73 . then either @xmath304 or @xmath305 where each @xmath306 is either a ( possibly negated ) atomic proposition or a negated dependence atom . * claim . * for all @xmath307 there is a set @xmath308 such that @xmath309 and @xmath310 . the main proposition follows immediately from case @xmath311 of this claim : from @xmath312 follows that @xmath313 and from @xmath314 follows that @xmath315 . _ inductive proof of the claim . _ for @xmath316 we can choose @xmath317 . for the inductive step let the claim be true for all @xmath318 . by lemma [ split - in - two ] there is a set @xmath319 such that @xmath320 and @xmath321 . let @xmath322 . since @xmath323 it follows by definition of the semantics of @xmath25 that @xmath324 . furthermore , @xmath325 note that @xmath301 is only trivial , i.e. , all instance structures satisfy all instance formulae , if we assume that only valid instances , i.e. , where the number of dependence atoms is guaranteed to be large enough , are given as input . however , if we have to verify this number the problem clearly remains in @xmath8 . now we consider the case in which we have very few dependence atoms ( which have bounded arity ) in each formula . we use the fact that there are only a few dependence atoms by searching through all possible determinating functions for the dependence atoms . note that in this case we do not need to restrict the set of allowed @xmath0operators as we have done above . [ few - dep - atoms - in - p ] let @xmath30 . then @xmath326 is in @xmath8 . from the semantics of @xmath327 it follows that @xmath328 is equivalent to @xmath329 where @xmath330 and @xmath331 both introduced by sevenster ( * ? ? ? * section 4.2 ) have the following semantics : @xmath332 now let @xmath52 , @xmath58 and @xmath333 be a problem instance . first , we count the number @xmath334 of dependence atoms in @xmath73 . if @xmath303 we reject the input instance . otherwise we replace every dependence atom by its translation according to [ eq : dep - as - function ] ( each time using a new function symbol ) . since the dependence atoms in @xmath73 are at most @xmath45-ary we have from the transformation that the introduced function variables @xmath335 are also at most @xmath45-ary . from this it follows that the upper bound for the number of interpretations of each of them is @xmath336 . for each possible tuple of interpretations @xmath337 for the function variables we obtain an @xmath338formula @xmath339 by replacing each existential quantifier @xmath340 by a boolean formula encoding of the interpretation @xmath341 ( for example by encoding the truth table of @xmath342 with a formula in disjunctive normal form ) . for each such tuple we model check @xmath339 . that is possible in polynomial time in @xmath343 as shown by clarke et al . @xcite . since the encoding of an arbitrary @xmath45-ary boolean function has length at most @xmath344 and @xmath45 is constant this is a polynomial in @xmath345 . furthermore , the number of tuples over which we have to iterate is bounded by @xmath346 with proposition [ many - dep - atoms - trivial ] and proposition [ few - dep - atoms - in - p ] we have shown the following theorem . [ vee - bounded - in - p ] let @xmath347 , @xmath30 . then @xmath90@xmath91 is in @xmath8 . given a kripke structure @xmath52 and a @xmath348 formula @xmath73 the algorithm counts the number @xmath349 of dependence atoms in @xmath73 . if @xmath350 the input is accepted ( because by proposition [ many - dep - atoms - trivial ] the formula is always fulfilled in this case ) . otherwise the algorithm from the proof of proposition [ few - dep - atoms - in - p ] is used . and there is another case where we can use the exhaustive determinating function search . [ no - wedge - no - vee ] let @xmath351 . then @xmath84 is in @xmath352 for every @xmath30 . . then there can be at most one dependence atom in @xmath73 because @xmath49 only contains unary operators . therefore we can once again use the algorithm from the proof of proposition [ few - dep - atoms - in - p ] . in theorem [ diamond - np - complete ] we saw that @xmath354 is @xmath6-complete . the previous theorem includes @xmath355 as a special case . hence , the question remains which are the minimal supersets @xmath49 of @xmath356 such that @xmath84 is @xmath6-complete . we will now see that adding either @xmath24 ( ) or @xmath25 ( ) is already enough to get @xmath6-completeness again . but note that in the case of @xmath25 we need @xmath357 while for @xmath316 the question remains open . [ diamond - wedge - bounded ] let @xmath358 . then @xmath84 is @xmath6-complete for every @xmath30 . membership in @xmath6follows from [ mdlmc - in - np ] . for hardness we once again reduce @xmath92to our problem . for this purpose let @xmath359 be an arbitrary @xmath139formula built from the variables @xmath360 . let @xmath87 be the kripke structure @xmath361 shown in [ figure : diamond - wedge - bounded ] and formally defined by @xmath362 ( c1 ) @xmath363 ; ( c2 ) [ right of = c1 ] @xmath364 ; ( c3 ) [ right of = c2 ] @xmath365 ; ( cx ) [ right of = c3 ] @xmath366 ; ( x1 ) [ below of = c1 ] @xmath367 ; ( tx1 ) [ right of = x1 ] @xmath368 ; ( x2 ) [ right of = tx1 ] @xmath369 ; ( tx2 ) [ right of = x2 ] @xmath370 ; ( x3 ) [ right of = tx2 ] @xmath371 ; ( xx ) [ right of = x3 ] @xmath366 ; ( tt1 ) at ( -1.5,-2 ) [ label=225:@xmath372 @xmath373 ; ( t1 ) at ( -3,-1 ) [ label=225:@xmath374 @xmath375 ; ( xx1 ) [ below of = x1 ] @xmath376 ; ( xx ) at ( -2.5,-3 ) @xmath377 ; ( txx1 ) [ below of = tx1 ] @xmath378 ; ( xx2 ) [ below of = x2 ] @xmath379 ; ( txx2 ) [ below of = tx2 ] @xmath380 ; ( xx3 ) [ below of = x3 ] @xmath381 ; ( xx ) [ right of = xx3 ] @xmath366 ; ( xxx1 ) [ below of = xx1 ] @xmath382 ; ( txxx1 ) [ below of = txx1 ] @xmath383 ; ( xxx2 ) [ below of = xx2,label = below:@xmath377 ] @xmath385 ; ( txxx2 ) [ below of = txx2 ] @xmath386 ; ( xxx3 ) [ below of = xx3 ] @xmath387 ; ( xxx ) [ right of = xxx3 ] @xmath366 ; ( c1 ) edge node ( x1 ) ; ( c1 ) edge node ( tx2 ) ; ( c2 ) edge node ( x1 ) ; ( c2 ) edge node ( x2 ) ; ( c2 ) edge node ( x3 ) ; ( c3 ) edge node ( tx1 ) ; ( c3 ) edge node ( x3 ) ; ( x1 ) edge node ( xx1 ) ; ( tx1 ) edge node ( txx1 ) ; ( x2 ) edge node ( xx2 ) ; ( tx2 ) edge node ( txx2 ) ; ( x3 ) edge node ( xx3 ) ; ( x1 ) edge node ( t1 ) ; ( x2 ) edge [ bend right=14 ] node ( t1 ) ; ( tx2 ) edge [ bend right=14 ] node ( t1 ) ; ( x3 ) edge [ bend right=14 ] node ( t1 ) ; ( tx1 ) edge [ bend left=10 ] node ( tt1 ) ; ( x2 ) edge [ bend left=10 ] node ( tt1 ) ; ( tx2 ) edge [ bend left=10 ] node ( tt1 ) ; ( x3 ) edge [ bend left=10 ] node ( tt1 ) ; ( xx1 ) edge node ( xxx1 ) ; ( txx1 ) edge node ( txxx1 ) ; ( xx2 ) edge node ( xxx2 ) ; ( txx2 ) edge node ( txxx2 ) ; ( xx3 ) edge node ( xxx3 ) ; the underlying @xmath139formula contains the clauses @xmath388 , @xmath389 and @xmath390 [ figure : diamond - wedge - bounded ] and let @xmath99 be the @xmath391 formula @xmath392 we again show that @xmath103 iff @xmath393 . first assume that @xmath103 and that @xmath107 is a satisfying valuation for the variables in @xmath73 . now let @xmath394 for all @xmath395 . then it holds that @xmath396 . furthermore , since @xmath107 satisfies @xmath73 it holds for all @xmath397 that there is a @xmath398 such that @xmath399 . hence , @xmath400 . for the reverse direction assume that @xmath401 . now let @xmath402 such that @xmath403 and for all @xmath397 there is a @xmath155 with @xmath404 . since @xmath405 there is no @xmath195 with @xmath406 and also @xmath407 . now let @xmath107 be the valuation of @xmath408 defined by @xmath409 since for each @xmath397 there is a @xmath195 such that either @xmath410 and @xmath406 or @xmath411 and @xmath407 it follows that for each clause @xmath131 of @xmath73 there is a @xmath195 such that @xmath110 satisfies @xmath131 under @xmath107 . [ diamond - vee - bounded ] let @xmath412 . then @xmath84 is @xmath6-complete for every @xmath357 . as above membership in @xmath6follows from proposition [ mdlmc - in - np ] and for hardness we reduce @xmath92to our problem . for this purpose let @xmath359 be an arbitrary @xmath139formula built from the variables @xmath413 let @xmath87 be the kripke structure @xmath361 shown in figure [ figure : diamond - vee - bounded ] and formally defined by @xmath414 \{p_j\ } & \text{iff } p_j\in c_i\\ \emptyset & \text{iff } \neg p_j \in c_i \end{array}\right . \end{array}\ ] ] ( c1 ) [ label = left:@xmath9 ] @xmath416 ; ( c2 ) [ right of = c1,label = left:@xmath9 ] @xmath417 ; ( c3 ) [ right of = c2 ] @xmath418 ; ( c12 ) [ below of = c1 ] @xmath419 ; ( c22 ) [ below of = c2,label = left:@xmath220 @xmath420 ; ( c32 ) [ below of = c3,label = left:@xmath9 ] @xmath421 ; ( c13 ) [ below of = c12,label = left:@xmath9 ] @xmath422 ; ( c23 ) [ below of = c22,label = below:@xmath377 ] @xmath423 ; ( c33 ) [ below of = c32,label = left:@xmath9 ] @xmath424 ; ( cx ) [ right of = c32 ] @xmath366 ; ( x2 ) [ below of = c23,label = left:@xmath9 ] @xmath425 ; ( x1 ) [ left of = x2,label = left:@xmath426 @xmath427 ; ( x3 ) [ right of = x2,label = left:@xmath9 ] @xmath428 ; ( xx2 ) [ below of = x2,label = left:@xmath429 @xmath430 ; ( xx3 ) [ below of = x3,label = left:@xmath9 ] @xmath431 ; ( xx ) [ right of = xx3 ] @xmath366 ; ( xxx3 ) [ below of = xx3,label = left:@xmath432 @xmath433 ; ( c1 ) edge node ( c12 ) ; ( c2 ) edge node ( c22 ) ; ( c3 ) edge node ( c32 ) ; ( c12 ) edge node ( c13 ) ; ( c22 ) edge node ( c23 ) ; ( c32 ) edge node ( c33 ) ; ( x2 ) edge node ( xx2 ) ; ( x3 ) edge node ( xx3 ) ; ( xx3 ) edge node ( xxx3 ) ; let @xmath99 be the @xmath0formula @xmath434 once again we show that @xmath103 iff @xmath435 . first assume that @xmath103 and that @xmath107 is a satisfying valuation for the variables in @xmath73 . now let @xmath436 for all @xmath395 . then it follows that @xmath437 and that @xmath438 for all @xmath395 . additionally , it holds that @xmath439 ( @xmath395 ) . together it follows that @xmath440 for all @xmath395 . this implies @xmath441 which is equivalent to @xmath442 for the reverse direction assume that @xmath188 with @xmath443 . let @xmath444 be subsets of @xmath88 with @xmath445 such that for all @xmath195 it holds that @xmath446 . then it follows that @xmath447 since the chain starting in @xmath427 consists of only one state . from @xmath448 and @xmath449 it follows that @xmath450 and hence ( again because of the length of the chain ) @xmath451 . inductively , it follows that @xmath452 for all @xmath395 . now , it follows from @xmath452 that for all @xmath196 with @xmath453 : @xmath454 ( because @xmath455 , @xmath456 ) . since @xmath446 , it then holds that @xmath457 . now let @xmath107 be the valuation of @xmath13 defined by @xmath458 since for each @xmath397 there is a @xmath195 such that @xmath453 it follows that for each clause @xmath131 of @xmath73 there is a @xmath195 such that @xmath459 satisfies @xmath131 under @xmath107 . first we show that classical disjunction can substitute zero - ary dependence atoms . [ nor replaces dep ] let @xmath460 . then @xmath461 follows immediately from the equivalence of @xmath462 and @xmath463 together with [ neg - dont - matter ] . the following surprising result shows that both kinds of disjunctions together are already enough to get @xmath6-completeness . [ nor vee ] let @xmath464 . then @xmath84 is @xmath6-complete for every @xmath30 . as above membership in @xmath6follows from [ mdlmc - in - np ] and for hardness we reduce @xmath92to our problem using a construction that bears some similarities with the one used in the proof of [ diamond - vee - bounded ] . for this purpose let @xmath359 be an arbitrary @xmath139formula built from the variables @xmath413 . let @xmath87 be the kripke structure @xmath361 shown in [ figure : nor vee ] and formally defined by @xmath465 ( c1 ) [ label=225:@xmath245 , @xmath466 @xmath467 ; ( c2 ) [ right of = c1,label=225:@xmath468 , @xmath469 @xmath470 ; ( c3 ) [ right of = c2,label=225:@xmath471 , @xmath229 @xmath472 ; ( cx ) [ right of = c3 ] @xmath366 ; the underlying @xmath139formula contains the clauses @xmath473 , @xmath474 and @xmath475 [ figure : nor vee ] let @xmath99 be the @xmath0formula @xmath476 once again we show that @xmath103 iff @xmath477 . first assume that @xmath103 and that @xmath107 is a satisfying valuation for @xmath73 . now let @xmath478 for all @xmath395 . then it follows that @xmath479 and that @xmath480 for all @xmath395 . together it follows that @xmath481 for the reverse direction assume that @xmath188 with @xmath482 . let @xmath483 be subsets of @xmath88 with @xmath445 such that for all @xmath195 it holds that @xmath484 . now let @xmath107 be the valuation of @xmath13 defined by @xmath485 since for each @xmath397 there is a @xmath195 such that @xmath486 it follows that for each clause @xmath131 of @xmath73 there is a @xmath195 such that @xmath459 satisfies @xmath131 under @xmath107 . now we show that [ no - wedge - no - vee ] still holds if we additionally allow classical disjunction . [ nor and unary ] let @xmath487 . then @xmath84 is in @xmath8for every @xmath30 . let @xmath72 . because of the distributivity of @xmath22 with all other @xmath0operators there is a formula @xmath99 equivalent to @xmath73 which is of the form @xmath488 with @xmath489 for all @xmath490 . note that there are only linearly many formulas @xmath491 because @xmath73 does not contain any binary operators aside from @xmath22 . further note that @xmath99 can be easily computed from @xmath73 in polynomial time . now it is easy to check for a given structure @xmath87 and team @xmath88 whether @xmath166 by simply checking whether @xmath492 ( which can be done in polynomial time by [ no - wedge - no - vee ] ) consecutively for all @xmath490 . in this paper we showed that @xmath61@xmath62is @xmath6-complete ( theorem [ wedge - vee - np - complete ] ) . furthermore we have systematically analyzed the complexity of model checking for fragments of @xmath0defined by restricting the set of modal and propositional operators . it turned out that there are several fragments which stay @xmath6-complete , e.g. , the fragment obtained by restricting the set of operators to only @xmath493 and @xmath327 ( theorem [ box - vee - np - complete ] ) or only @xmath7 and @xmath327 ( theorem [ diamond - np - complete ] ) . intuitively , in the former case the @xmath6-hardness arises from existentially guessing partitions of teams while evaluating disjunctions and in the latter from existentially guessing successor teams while evaluating @xmath7 operators . consequently , if we allow all operators except @xmath7 and @xmath25 the complexity drops to @xmath8(theorem [ box - wedge - in - p ] ) . for the fragment only containing @xmath25 and @xmath327 on the other hand we were not able to determine whether its model checking problem is tractable . our inability to prove either @xmath6-hardness or containment in @xmath8led us to restrict the arity of the dependence atoms . for the aforementioned fragment the complexity drops to @xmath352 in the case of bounded arity ( theorem [ diamond - vee - bounded ] ) . furthermore , some of the cases which are known to be @xmath6-complete for the unbounded case drop to @xmath8 in the bounded arity case as well ( theorem [ no - wedge - no - vee ] ) while others remain @xmath6-complete but require a new proof technique ( theorems [ diamond - wedge - bounded ] and [ diamond - vee - bounded ] ) . most noteworthy in this context are probably the results concerning the @xmath7 operator . with unbounded dependence atoms this operator alone suffices to get @xmath6-completeness whereas with bounded dependence atoms it needs the additional expressiveness of either @xmath24 or @xmath25 to get @xmath6-hardness . considering the classical disjunction operator @xmath22 , we showed that the complexity of @xmath494 is never higher than the complexity of @xmath495 , i.e. , @xmath22 is at least as bad as @xmath1 with respect to the complexity of model - checking ( in contrast to the complexity of satisfiability ; cf . @xcite ) . and in the case where only @xmath25 is allowed we even have a higher complexity with @xmath22 ( ) than with @xmath327 ( ) . the case of @xmath496 is also our probably most surprising result since the non - determinism of the @xmath25 operator turned out to be powerful enough to lead to @xmath6-completeness although neither conjunction nor dependence atoms ( which also , in a sense , contain some special kind of conjunction ) are allowed . interestingly , in none of our reductions to show @xmath6-hardness the @xmath0formula depends on anything else but the number of propositional variables of the input @xmath139formula . the structure of the input formula is always encoded by the kripke structure alone . so it seems that even for a fixed formula the model checking problem could still be hard . this , however , can not be the case since , by theorem [ few - dep - atoms - in - p ] , model checking for a fixed formula is always in @xmath8 . another open question , apart from the unclassified unbounded arity case , is related to a case with bounded arity dependence atoms . in [ diamond - vee - bounded ] it was only possible to prove @xmath6-hardness for arity at least one and it is not known what happens in the case where the arity is zero . additionally , it might be interesting to determine the exact complexity for the cases which are in @xmath8since we have not shown any lower bounds in these cases so far . johannes ebbing and peter lohmann . complexity of model checking for modal dependence logic . in _ proceedings sofsem 2012 : theory and practice of computer science _ , volume 7147 of _ lecture notes in computer science _ , pages 226237 . springer berlin / heidelberg , 2012 . peter lohmann and heribert vollmer . complexity results for modal dependence logic . in _ proceedings 19th conference on computer science logic _ , volume 6247 of _ lecture notes in computer science _ , pages 411425 . springer berlin / heidelberg , 2010 . jouko vnnen . modal dependence logic . in krzysztof r. apt and robert van rooij , editors , _ new perspectives on games and interaction _ , volume 4 of _ texts in logic and games _ , pages 237254 . amsterdam university press , 2008 .

in this paper we discuss the age and spatial distribution of young ( age gyr ) smc and lmc clusters using data from the magellanic cloud photometric surveys . luminosities are calculated for all age - dated clusters . ages of 324 and 1193 populous star clusters in the small and the large magellanic cloud have been determined fitting padova and geneva isochrone models to their resolved color - magnitude diagrams . the clusters cover an age range between 10 myr and 1 gyr in each galaxy . for the smc a constant distance modulus of = 18.90 and a metallicity of z = 0.004 were adopted . for the lmc , we used a constant distance modulus of = 18.50 and a metallicity of z = 0.008 . for both galaxies , we used a variable color excess to derive the cluster ages . we find two periods of enhanced cluster formation in both galaxies at 160 myr and 630 myr ( smc ) and at 125 myr and 800 myr ( lmc ) . we present the spatially resolved recent star formation history of both clouds based on young star clusters . the first peak may have been triggered by a close encounter between the smc and the lmc . in both galaxies the youngest clusters reside in the supergiant shells , giant shells , the inter - shell regions , and toward regions with a high h content , suggesting that their formation is related to expansion and shell - shell interaction . most of the clusters are older than the dynamical age of the supergiant shells . no evidence for cluster dissolution was found . computed v band luminosities show a trend for fainter magnitudes with increasing age as well as a trend for brighter magnitudes with increasing apparent cluster radii .

1004.1247

due to their proximity the magellanic clouds ( mcs ) offer an excellent opportunity to study their spatially resolved star formation ( sf ) histories . sf can be triggered by several mechanisms such as , e.g. , the self - induced gravitational collapse of a molecular cloud , tidal shocking , a turbulent interstellar medium , or cloud - cloud collisions ( e.g. , * ? ? ? the mcs and the milky way ( mw ) are interacting with each other . the formation of star clusters younger than @xmath31 gyr in the mcs was probably triggered by interactions of the galaxies with each other and with the mw ( e.g. , * ? ? ? star clusters may be produced through strong shock compressions induced by close encounters of their host galaxies , which causes enhanced star formation . conversely , the star formation rate decreases again once the galaxies recede from each other . repeated encounters then lead to episodic cluster formation . in the mcs , a correlation between young star clusters and putative close encounters with each other and mw has been suggested by , e.g. , ( * ? ? ? * g95 ) , ( * ? ? ? * pu00 ) , and ( * ? ? ? * c06 ) . strong tidal perturbations induced by the encounters could also have triggered the formation of clusters ( e.g. , * ? ? ? * ) in the mcs . possible orbits of the small magellanic cloud ( smc ) , the large magellanic cloud ( lmc ) , and the mw have been modeled by several authors ( e.g. , * ? ? ? they found that it was difficult to keep the clouds bound to each other for more than 1 gyr in the past . the lmc and the smc have been part of a triple system together with the milky way since at least 1 gyr ( e.g. , * ? ? ? * kallivayalil et al . 2006a / b ) . it is possible that the clouds are not a bound system and that they are making their first passage close to the mw . interestingly , the cluster formation histories of the lmc and smc show large differences . in the lmc , two main epochs of cluster formation ( e.g. , * ? ? ? * ) have been observed that are separated by an age gap of about 4 - 9 gyr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , in which no star clusters have formed . the two epochs of pronounced cluster formation occurred @xmath49 gyr ago and @xmath53 - 4 gyr ago . in the lmc , a few globular clusters are found that are as old as the oldest galactic globulars @xcite . during the past @xmath54 gyr , star clusters have been forming continuously until the present day . in contrast , the star clusters in the smc cover a wide range of ages and continued to form over at least the last @xmath510.5 gyr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to the present day . interestingly , in the smc the cluster formation history appears to have started with a delay since the smc formed its first and only globular cluster , ngc121 , 2 - 3 gyr later than the lmc or the mw ( * ? ? ? * and references therein ) . the lmc contains about @xmath54200 star clusters , while in the smc @xmath5770 star clusters have formed ( and survived ) . the cluster census is probably still incomplete , missing small and faint clusters that are yet to be detected . ongoing and prospective space - based observations may further increase the number of known star clusters . the most recent catalog cross - correlating all known objects of the lmc , smc , and the magellanic bridge region was published by @xcite ( b08 ) . however , the cluster sample still is highly incomplete as pointed out by the authors . only for a few clusters in b08 s catalog , ages have been determined . for young smc clusters , @xcite ( pu99 ) used isochrone fitting on data from the optical gravitational lensing experiment ( ogle ii ; * ? ? ? * ) to determine ages for 93 clusters . @xcite compiled ages for 306 binary cluster candidates in the lmc from a variety of literature sources ranging from multiwavelength integrated light studies to isochrone fitting to resolve clusters . @xcite ( rz05 ) made use of integrated colors and derived ages for 200 clusters . c06 determined ages of 164 associations and 311 star clusters based on data from the ogle using isochrone fitting . their sample is the largest available catalog with smc cluster ages . ages for young lmc clusters have been provided by g95 based on integrated colors ( 624 objects ) and by pu00 using isochrone fitting applied to ogle - ii data ( @xmath5600 clusters ) . luminosities have been published for 204 smc star clusters by rz05 measuring integrated colors from the magellanic clouds photometric survey ( mcps ) . @xcite ( b96 ) published integrated photometry of 624 lmc star clusters that was based on observations carried out at the 0.61-m telescope at ctio in chile and at the 2.15-m casleo telescope in argentina . in the present study we increase the number of age - dated young lmc and smc star clusters and calculate v - band luminosities . we aim at improving the understanding of the cluster age distribution of these two irregular galaxies and present spatial distribution maps of the star clusters in both galaxies . to achieve this goal , we make use of ground - based data of the magellanic clouds photometric surveys ( mcpss ) @xcite . in the next section the observations and data reduction are described . in @xmath6 [ sec : metunddistmod ] the distances , reddenings , and metallicities of both the smc and the lmc are given . in @xmath6 [ sec : agedist ] the clusters age distribution , spatial distribution , and dissolution effects are discussed and in @xmath6 [ sec : youngies_lum ] the correlation between the cluster luminosities and age / radius is derived . we made use of three catalogs in order to select clusters and to obtain color - magnitude diagrams ( cmds ) of the young smc and lmc star clusters ( age @xmath01 gyr ) . the first catalog was published by b08 ( see also * ? ? ? * ; * ? ? ? * ) and includes cluster positions , angular sizes , and object classes for 17,815 objects in the lmc , smc , and magellanic bridge . this catalog combines and cross - identifies objects measured on the eso / serc r and j sky survey schmidt films and other catalogs ( e.g. , * ? ? ? the other two catalogs are from the mcps presented by @xcite containing u , b , v , and i point - source photometry of the central 18 deg@xmath7 area of the smc ( 5,156,057 stars ) and of the central 64 deg@xmath7 area of the lmc ( 24,107,014 stars ) . the data were obtained using the las campanas 1-m swope telescope in chile . the v frame was used as reference and only stars detected in both the v and b frame were retained . in both catalogs , the limiting magnitude of the photometry is v@xmath524 mag . the photometry is highly incomplete below u=21.5 mag , b=23.5 mag , v=23 mag , and i=22 mag . this means that it is difficult to derive ages of star clusters older than @xmath51 gyr due to the limited photometric depth of the mcps , which does not resolve main - sequence turnoff - points ( mstos ) of intermediate - age and older clusters . each object in b08 s catalog is categorized by an object class . obvious star clusters are indicated by a c and emissionless associations by an a. ca and ac indicate intermediate classes between these two types for which a classification was not clear . nc refers to small hii regions with embedded star clusters , while cn are clusters that show traces of emission . dcn refers to clusters that are decoupled from nebulae . for our study , only objects including a c-classification were used ( smc : 765 objects ; lmc : 4089 objects ) , because these objects are well defined and not too extended so that cluster stars can be distinguished from the surrounding field stars . star clusters that have very small radii or that are very sparse could not be age - dated ( see @xmath6 [ sec : youngies_ages ] ) . our resulting catalogs contain ages of star clusters between @xmath510 myr and 1 gyr . clusters whose appearance in the cmds indicated an age older than 1 gyr were discarded from the sample , because the mstos were not resolved . because we only age - dated objects including a c in the b08 classification , our samples contain only objects with a minimum age of 10 myr , since very young objects are usually classified as associations or nebulae and thus were excluded from our sample . for many intermediate - age and old smc and lmc clusters accurate ages have been determined elsewhere in the literature mostly based on hst data ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the latest low - resolution spectroscopic measurements of smc star clusters by @xcite showed that the young clusters with ages between @xmath5 0.9 and 2 gyr have a spread in metallicity of 0.53 dex . the two youngest clusters in their sample , lindsay106 and lindsay108 , both with ages around 0.9 gyr , have mean metallicities of [ fe / h]=-0.88 and -1.05 , respectively . studies based on spectroscopy of six supergiants in the young cluster ngc330 ( age@xmath520 - 25 myr ; @xcite ) rendered a metallicity of [ fe / h ] = @xmath8 dex ( e.g. , * ? ? ? * ) , whereas @xcite found [ fe / h ] = @xmath9 from high - resolution sprectra of seven supergiants in this cluster . it has been suggested repeatedly that ngc 330 is more metal - poor than the surrounding field star population ( e.g. , * ? ? ? * ; * ? ? ? * ) . on the other hand , most field star studies agree on a value of [ fe / h ] = @xmath10 dex ( e.g. , * ? ? ? * ; * ? ? ? this metallicity corresponds to an isochrone model of z = 0.004 , which we used for the age determination . we do not consider a possible metallicity spread here . a distance modulus of ( m - m ) = 18.90 mag ( @xmath560 kpc , e.g. , * ? ? ? * ) was assumed for the smc . there are no direct determinations of the cluster distances along the line of sight , but it is assumed that the smc has a depth extent of up to 20 kpc @xcite . the clusters analyzed in our study lie in the central region of the smc main body . @xcite found a variation in the distance modulus of @xmath50.14 mag assuming an elongation of 4 kpc . the resulting error in log(age ) is less than 0.05 . @xcite studied 19 cepheid variables in the lmc and derived a mean present - day metallicity of [ fe / h ] = @xmath11 dex for this galaxy . this result is in agreement with previous determinations of the metal abundance within the young population of the lmc ( e.g. , * ? ? ? * ; * ? ? ? * ) also using cepheid variables . this metallicity matches best the isochrone models of z = 0.008 , which were used for the age determination . for the lmc , a distance modulus of ( m - m ) = 18.50 mag ( @xmath550 kpc ; e.g. , * ? ? ? * ) was assumed . there are indications that the lmc bar may be offset from the plane of the disk by about 2 kpc ( e.g. , * ? ? ? * ; * ? ? ? this leads to a variation in the distance modulus of @xmath50.08 mag . the resulting error in log(age ) is less than 0.05 . the cluster age distribution of the lmc and the smc was obtained by first matching the clusters central positions adopted from b08 with the mcps @xcite . from the apparent major and minor axis adopted by b08 we computed the mean apparent diameter @xmath12 for each object ( see for an example the smc cluster ngc 376 in fig . [ fig : radii376 ] ) . all stars within these diameters are considered cluster members . four cmds were plotted for each object : two v vs b - v cmds and two v vs v - i cmds showing the cluster and the surrounding field . ccccccccl smc0017 & 0.02 & 0.25 & 9.00 & 2 & 0:28:31 & -73:00:49 & @xmath13 & bs2 + smc0018 & 0.01 & 0.60 & 8.70 & 1 & 0:30:00 & -73:22:45 & @xmath14 & k9 , l13 + smc0023 & 0.01 & 0.60 & 9.10 & 3 & 0:32:41 & -72:34:53 & @xmath15 & l14 + smc0026 & 0.03 & 1.50 & 8.85 & 3 & 0:32:56 & -73:06:58 & @xmath16 & ngc152 , k10 , l15 , eso028sc24 + [ tab : smc_table ] ccccccccl lmc0020 & 0.02 & 0.75 & 8.80 & 1 & 4:37:51 & -69:01:45 & @xmath17 & sl8 , lw13 , kmhk21 + lmc0021 & 0.08 & 0.29 & 8.00 & 1 & 4:38:07 & -68:46:39 & @xmath18 & ngc1649 , eso055sc31 , kmhk22 + lmc0022 & 0.08 & 0.75 & 8.50 & 1 & 4:38:22 & -68:40:21 & @xmath19 & sl14 , lw21 , kmhk28 + lmc0030 & 0.02 & 0.55 & 8.30 & 2 & 4:40:28 & -69:38:57 & @xmath20 & sl15 , lw23 , kmhk29 + [ tab : lmc_table ] we fitted the cmds with two different isochrone models : padova isochrones @xcite and geneva isochrones @xcite . both the padova and the geneva isochrone grids have an age resolution of log(t)=0.05 . the dartmouth isochrone models have a youngest age of 250 myr @xcite , which is too old for many of the star clusters in our sample . therefore , we did not use these models . to fit the isochrones , a constant distance modulus and metallicity were used . the best - fit isochrone was then found by using different combinations of reddening and age . the age and reddening of each cluster are derived by visual inspection . human judgement is needed in particular in order to decide about the inclusion of luminous supergiants in the fit , since the apparent main - sequence turnoff of sparse , young clusters is subject to pronounced statistical fluctuations ( e.g. , * ? ? ? the geneva isochrone models with z=0.004 deviate from the padova isochrones for ages younger than log(age)=6.9 by about log(age)=@xmath21 . the geneva models with z=0.008 deviate by about log(age)=@xmath21 from the padova isochrones for ages younger than log(age)=8.00 . in our catalogs , we list the ages determined using the padova isochrones . many star clusters in the present samples are located in the main body of the smc and along the lmc bar , both of which are highly crowded . therefore , field star contamination is a severe effect and influences the age determination . the field population was sampled within an annulus between 1 and 2 arcmin around the cluster centers outside the apparent cluster radii ( for examples see lower panels of figs . [ fig : cmd376 ] and [ fig : cmd0899 ] ) . for clusters with larger apparent radii than 1 arcmin , the field population was sampled within an annulus between 2 and 3 arcmin . the selected stars lying within these annuli were plotted on top of the cluster cmds to illustrate the location of the smc field stars . the accuracy of the inferred ages depends on the number of cluster member stars and on the density of the surrounding field . many star clusters in the sample could not be age - dated , because they are too sparse or they have very small radii , which include too small a number of stars against the smc field background for a reliable isochrone fit . in the densest smc regions , the cluster membership determination on the basis of photometric information is difficult , and therefore the here derived ages are uncertain . in figures [ fig : cmd376 ] and [ fig : cmd0899 ] two examples are shown . for the smc cluster ngc376 we derive an age of @xmath530 myr and @xmath22 . the lmc cluster sl410 is a slightly older star cluster for which we determine an age @xmath5110 myr and @xmath23 . in the respective upper panels , stars that were selected within the cluster radii adopted from b08 are displayed . the magenta dots represent the field stars selected within a concentric 0.5 arcmin wide annulus located 0.1 armin away from the clusters apparent radii . the blue solid line shows the best - fitting geneva isochrone . the red solid line is the best - fitting padova isochrone . in the lower panels , field stars selected within an annulus between 1 and 2 arcmin radius around the cluster centers are shown . young cluster ages derived from isochrone fitting are not severely affected by the discreteness of isochrones due to the logarithmic age - steps of the applied isochrone models , which produce a fairly dense grid for small ages . the interstellar extinction , however , can have a severe impact on the determination of the cluster ages . we compared the reddenings found by isochrone fitting to the reddenings adopted from pu99 . the standard deviation is of the order of @xmath24=0.03 . in section [ sec : metunddistmod ] , the uncertainty in the present - day metallicity of the smc is mentioned . consequently , for some clusters , the applied isochrone metallicity of z=0.004 might not be the best choice . however , the attempt to fit the clusters with a different metallicity ( e.g. , z=0.008 ) provided fits of lower quality ( age difference of @xmath5log(age)=0.2 ) . therefore , the uncertainties of the age determinations are partly a function of interstellar extinction , partly a function of the age itself in the sense that older clusters are more difficult to identify , and partly a function of the cluster density . an uncertainty that can not be handled with our data occurs when the clusters are all inside large groups of associations having small separations along the line of sight or when they are very sparse . sparse objects usually do not have a well - defined main sequence turn - off . on one hand , supergiants with their short lifetimes ( typically @xmath520 myr ) are well - suited as age tracers if they show a concentration toward the cluster center . we therefore require our `` best '' isochrones to reproduce the location to the supergiants in the cmds and use that as a primary criterion for age determinations . the reddenings , on the other hand , are based on the perceived locus of the blue envelope of the main sequence . the determination of cluster membership for individual stars is a difficult matter ( based on only photometric information ) and further investigation with e.g. , the hubble space telescope or studies based on spectroscopy are necessary . a clear separation between clusters and their surroundings becomes particularly difficult when they are embedded in associations of a very similar age . for the smc and the lmc , ages of 324 clusters and 1193 clusters , respectively , have been determined in our study . the full tables will be made available electronically , but illustrative excerpts are shown in tables [ tab : smc_table ] and [ tab : lmc_table ] . column 1 gives the cluster identification number . the reddening parameter @xmath25 is listed in column 2 and the apparent radii in column 3 . the derived ages are shown in column 4 , while column 5 gives the degree of reliability of our age measurements . class 1 indicates having errors @xmath26 ; class 2 indicates objects having errors @xmath27 ; and class 3 indicates objects having errors @xmath28 , adopting the same notation as used by c06 . columns 6 and 7 give the right ascension and the declination , respectively . the calculated total v band magnitudes ( see @xmath6 [ sec : youngies_lum ] ) are listed in column 8 and finally , the cross - identifications with other catalogs are given in column 9 . c06 used data from ogle ii @xcite for the smc disk and data obtained at the eso 2.2 m telescope for the region around ngc269 that is located at the border of the supershell 37 a to derive ages of 461 smc clusters and associations . the ages were derived using padova isochrones @xcite . a distance modulus of ( m - m ) = 18.9 was assumed and the reddening was derived by main - sequence fitting . a mean smc metallicity of z = 0.008 was assumed ( in the present study we used z = 0.004 ) as found spectroscopically for young stars by @xcite . an upper age limit was set by the limiting magnitude of the available photometry indicating that clusters that have an msto fainter than v=20 mag in the ogle ii field could not be age - dated . c06 s study was restricted to clusters younger than 1 gyr . they derived ages within an age range of 4 myr and 1 gyr for clusters covering an area of 2.4 deg@xmath7 of the smc main body . 136 of their clusters are also included in our study . in the first panel of fig . [ fig : ageallvergleich ] cluster ages determined in our work are compared to ages presented by c06 . the ages derived in our work tend to be @xmath50.2 - 0.3 in log(age ) older than the ages by c06 , an offset we anticipated above . the main reason for the difference is probably the different metallicity of the applied isochrone models as we tested for a subset of cmds . for those clusters with the most significant age deviation large age uncertainties are stated both in their and our work ( e.g. bs271 , bs272 , b114 ) . the dispersion about the zero line ( red solid line ) for the sample is @xmath29 = 0.13 . rz05 derived ages of 204 smc star clusters using integrated colors , of which 112 clusters are included in our sample . the second panel of fig . [ fig : ageallvergleich ] shows the comparison between our ages and ages adopted by rz05 . the largest deviations are found at the ends of the upper and the lower age limits . a few clusters having ages younger than 25 myr and older than 1 gyr show an age deviation of up to 1 gyr . stochastic effects on the number of bright stars , uncertainties on the metallicity , dubious cluster membership , and on the adopted stellar models may contribute to the large uncertainties in the conclusion of the cluster age . rz05 used a metallicity of z = 0.004 . those clusters with the largest age differences are mostly very sparse objects or objects to which very young isochrones were fitted based on only 2 - 3 bright stars . large uncertainties for the oldest clusters are caused by large field contamination and unresolved main - sequence turnoff points . the dispersion about the zero line ( red solid line ) for the sample is @xmath29=0.3 . pu99 determined ages for 93 smc star clusters from the ogle catalog @xcite , of which 51 are included in our sample . they fitted isochrones from the library of @xcite applying a distance modulus of @xmath1=18.65 @xcite and a metallicity of z=0.004 . in their study , they derived cluster ages within an age range of 10 myr and 1 gyr . in the third panel of fig . [ fig : ageallvergleich ] we compare ages presented by pu99 to our study . the ages are in very good agreement and the dispersion about the zero line ( red solid line ) is @xmath29=0.17 . this result may seem surprising considering the difference of the applied distance moduli of 0.25 . however , fitting isochrones to the same cmd using both values for the distance moduli leads to age differences for these young clusters of no more than 0.1 - 0.15 in log(age ) . ages of about 600 lmc star clusters were determined using ogle ii data @xcite by pu00 . the ages were derived fitting isochrones of @xcite . a distance modulus of 18.24 mag and a metallicity of z = 0.008 were assumed . the reddenings were adopted from @xcite . clusters older than @xmath51.2 gyr could not be reliably age - dated , because the main - sequence turnoff points are located close to the limit of the ogle ii photometry ( [email protected] mag ) . 49 clusters are in common with our cluster sample . in the fourth panel of fig . [ fig : ageallvergleich ] we compare the derived ages found in this study with pu00 . the dispersion about the 1:1 agreement ( red solid line ) for the sample is @xmath29=0.15 . again the good agreement is surprising , considering the difference of the applied distance moduli of 0.26 . as for the smc , the age difference for these young clusters is no larger than 0.1 - 0.15 in log(age ) fitting two isochrones to the same cmd using both distance modulus values . g95 used integrated colors to determine the s parameter for all 624 lmc objects listed in the catalog of @xcite . the s parameter is an age indicator based on integrated u - b and b - v colours ( see also * ? ? ? unfortunately , a direct comparison of our data is not possible because the cluster ages of g95 are not tabulated . @xcite fitted isochrones to data from earlier versions of the mcps @xcite and ogle @xcite and transformed g95 s s parameter into ages . the authors found a good overall agreement in the age distribution . all in all , the scatter of @xmath50.2 in age derived by other groups and in this work is in good agreement with the typical age uncertainties derived in either study . uncertainties aside , the age distribution of star clusters is influenced by two effects @xcite . the first is cluster fading that occurs through stellar evolution when the clusters get fainter with time . therefore , the number of observed star clusters appears to decrease with increasing age for a given magnitude limit since sparse , faint clusters become increasingly harder to detect . a second effect acting upon gas - free clusters is caused by the combined effects of 2-body relaxation and external tidal stripping : this is the so - called secular evolution @xcite . a third effect altering the initial age distribution at the lower end is infant mortality caused by gas expulsion . in the mcs the phase of infant mortality lasts up to @xmath3040 myr @xcite . the cluster age distributions of the smc and the lmc are shown in fig . [ fig : agedist_smc ] and [ fig : agedist_lmc ] . in the upper panels of these figures only cluster ages determined in our study and in the lower panels the combination of this study with c06 ( smc ) and pu00 ( lmc ) samples are displayed . for the clusters in common with our work ages derived in this study were used . because we age - dated only objects that included a c in the b08 classification , our samples contain only a few objects younger than @xmath520 myr , since very young objects are usually classified as associations or nebulae and thus were excluded from our sample . the age distribution of the smc star clusters derived in our study shows a first peak of enhanced cluster formation around 160 myr and a second one around 630 myr . only clusters with an age uncertainty smaller than class 3 were used . c06 found two main episodes of enhanced cluster formation between 5 - 15 myr and at 90 myr . their cluster sample contains ages for 164 associations , which is the reason why their sample reaches down to ages of @xmath53 myr . in the combined sample three episodes of enhanced cluster formation are detectable , the youngest around 6.5 myr , followed by a second peak around 160 myr , and a third around 630 myr . the analysis in the age range of the oldest peak is limited by the depth of the photometry , while the youngest peak mostly comes from the associations adopted from c06 . the difference between the oldest peak found in our study and by c06 lies within the given uncertainties for the cluster ages ( @xmath29=0.3 ) . for the lmc , pu00 detected three periods of enhanced cluster formation at about 7 myr , 125 myr , and 800 myr , which are similar to the findings of g95 . no associations and nebulae were considered in the study of pu00 and therefore the age distribution of the combined lmc sample does not reach as young an age as the combined smc sample . the youngest clusters in pu00 s sample have ages of @xmath55 myr , but come with high age uncertainties as stated by the authors . the age determination of the oldest clusters in our sample with ages close to 1 gyr is highly uncertain ( class 3 ) . clusters with uncertainties smaller than class 3 were discarded and therefore the oldest peak is very weak . in the combined sample an increased number of clusters appears at @xmath59 myr and 630 myr , which is in very good agreement with g95 and p00 . the most enhanced peak occurs at about 125 myr . in both the smc and the lmc , we find evidence for periods of enhanced cluster formation , which appear to have occurred during the same periods . the peaks at @xmath5125 and 160 myr , respectively , are very pronounced in both galaxies and are probably correlated . the difference between the peaks is within the given uncertainties for the cluster ages ( difference in log(age)=0.1 ) . model calculations performed by e.g. , @xcite and @xcite showed that the mw , the lmc , and the smc have only interacted long enough to produce the magellanic stream . according to these models , the last close encounter between smc and lmc occurred about 100 - 200 myr ago . the smc star formation rate increases , if the lmc orbit leads to a close encounter with the smc and vice versa . the star formation rate decreases again when the lmc recedes from smc , thus leading to episodic cluster formation . hence , this increase in cluster formation some 100 - 200 myr ago may have been triggered by a tidal interaction with the neighboring galaxy ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . the youngest peaks , however , might have another origin . probably they are caused by associations , which did not yet dissolve . high velocity cloud - cloud collisions are another trigger mechanism of cluster formation @xcite . these collisions are particularly effective during galaxy interactions and mergers . high - speed motions may produce a high - pressure environment that in return can trigger turbulences or shocks @xcite . stellar winds and supernova explosions can also trigger star formation through compression by turbulent motions @xcite . the spatial distribution of the star clusters in fig . [ fig : smcageall ] clearly shows that the youngest clusters are located in the two large hi supershells 37 a and 304 a @xcite . these features are believed to have their origin in the evolution of an ob association in which the most massive stars become supernovae and/or where stellar winds lead to a runaway expansion and the formation of a supergiant shell ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? most of the shells and super - shells are associated with young objects @xcite . inside the shells , second generation star clusters may form due to supernova explosions in the first generation clusters . our ( or the c06 ) data do not cover the western side of supershell 37 a , but most of the remaining objects are age - dated . we confirm the findings of c06 about the discontinuity in the spatial distribution of younger clusters in the supershells 37 a and 304 a : that is , the youngest objects are found toward the eastern rim of the shell ( 37a ) where gas and dust are located @xcite and toward regions luminous in h@xmath2 . such hii regions visible in h@xmath2 indicate the presence of massive stars ( e.g. , ob - stars , supergiants , luminous blue variable stars ) , supernova remnants , or diffuse ionized nebulae . the youngest objects of the shell 304 a are located toward its northern and western rim ( toward 37 a ) and also in the central part . the compression of gas due to shell - shell interactions related to the expansion of the shells might have triggered the cluster formation at the opposing rims as well as in the inter - shell region . h@xmath2 is concentrated in the northern part of 304 a where we also find the youngest objects . in fig . [ fig : smcageslides ] four snapshots are shown displaying the spatial distribution of star clusters within different age ranges . in the first snapshot only clusters younger than 16 myr are shown . the clusters are distributed along the smc bar in an elongated but narrow area . the two supershells and the inter - shell region are clearly visible in this plot . clusters with ages between 16 myr and 63 myr are also concentrated along the bar , but there are outliers found toward the smc wing or slightly offset of the bar ( second snapshot ) . the supershell 304 a is highly populated by clusters of this age range . widely spread over the entire smc main - body are clusters with ages between 63 myr and 315 myr ( third snapshot ) . moreover , the northern part of the smc body is covered with clusters in this age range . finally , the last snapshot displays clusters older than 315 myr ( up to @xmath5 1 gyr ) . these clusters mainly populate the western part of the smc main body and only a few are found in the north or in the east . supershell 304 a contains fewer of these older objects than 37 a and they are mainly located at the western rim of the shell , while in 37 a they are widely distributed . generally , the cluster distribution in 304 a indicates a continuous cluster formation from a few myr to 1 gyr . according to the standard model of shell formation @xcite the older objects are supposed to be found in the center of the shell while younger objects are distributed around the edge . as mentioned above , the younger objects in 37 a and 304 a are clustered toward the eastern and the western rim of the shells , respectively . shell interaction due to collisions following shell expansion has probably triggered the cluster formation in the inter - shell region as well as at the two opposing rims . the reason might be that the shell is surrounded by an interstellar medium with differing densities . shell expansion into higher - density regions leads to slower expansion ; the corresponding strong gas compression leads to enhanced star formation in that region . on the other hand , if the shell expands into a region with lower density the star formation rate will be lower since there is less compression and less material to be swept up . we confirm c06 s finding that the location of the young smc clusters is correlated with hi intensities , which decrease with increasing age . @xcite found that young emission objects are positionally well correlated with co clouds , while emissionless objects of ages @xmath56 - 100 myr do not show any correlation to the co clouds . emission objects were excluded from our sample and therefore we can not investigate their possible association with co clouds . the youngest star clusters are found mostly in the lmc hi supergiant shells ( sgss ) 30 doradus , sgs 7 ( lmc 5 ) , sgs 11 ( lmc 4 ) , sgs 12 ( lmc 3 ) , and sgs 19 ( lmc 2 ) and in the 103 giant shells published by @xcite ( see also * ? ? ? * ; * ? ? ? the authors classified the shells into three categories ; the _ single sgss _ are simple expanding stalled shells ( solid ) , _ complex sgss _ consist of smaller interlocked shells and their rims often contain smaller shells ( sgs 4 , 9 , 10 , 12 , 17 , 18 , 20 ; dashed ) , and _ propagated sgss _ are surrounded by another sgs ( sgs 11 , 19 ; dash - dotted ) . in most of the complex sgss ( see fig . [ fig : lmcageall ] ) star clusters are located along the rims and lack the youngest clusters . the smaller shells along their rims might indicate that triggered star formation occurred recently . the propagated sgss may have formed by sequential star formation . de boer et al . ( 1998 ) proposed a scenario in which star formation in the lmc is triggered by the gas being compressed at the leading edge due to the bow - shocks as the lmc moves through the gaseous galactic halo around the milky way . the bow - shock compresses large areas which leads to star formation on a large scale . due to the lmc s rotation , the material at the leading edge will move away clockwise and distribute itself around the lmc s rim . in this scenario one would expect to find a progression in the age of star clusters along the direction of the rotation ( in fig . [ fig : lmcageall ] the galaxy rotates clockwise ) . the lmc disc shows solid - body rotation and has an approximate full rotation velocity for a small inclination of @xmath5150 kms@xmath31 at 1.5 kpc @xcite . the time of one full rotation period is @xmath5250 myr . if present the progression of age along the galaxy s rim should be detectable . the youngest shell structures lie in the proximity of 30 doradus , the largest star forming region of the lmc . 30 doradus lies close to the leading edge of the lmc . it is the largest star forming region in the galaxy and may very well be related to the bow shock . @xcite used cepheids and other supergiant stars to study the recent star formation history of the lmc . they found that the majority of objects younger than 30 myr are concentrated on the south - eastern border , while others are widely distributed across the entire disk which can not be explained with the bow - shock star formation model . a progression of age was found in several giant shells along the lmc rim moving from the south - east ( lmc 2 ) to the north ( ngc 1818 at @xmath2 = @xmath32 , @xmath33 = @xmath34 ) @xcite . 30 doradus has an age of 3 - 5 myr , lmc 4 of 9 - 16 myr , and ngc 1818 of 25 - 30 myr @xcite . the difference in age between these shells corresponds to their distance along the border of the disk divided by the galaxy s rotation velocity @xcite . we find no evidence for such a cluster age distribution for the lmc . the older clusters are mostly located in the bar region and a few are located along the rim . most clusters populating the lmc bar were adopted from the sample published by pu00 . ages of clusters older than @xmath51 gyr can not be derived due to the limited photometric depth of the mcpss , which does not resolve mstos of older clusters . overall , our spatial cluster age distribution does not support ram pressure as the primary agent of cluster formation in the lmc . as mentioned above , the highest concentration of young clusters can be found in the 30 doradus region ( ngc2070 ) in the southeastern part of the lmc at roughly @xmath2 = @xmath35 , @xmath33 = @xmath36 . in this large star forming complex many sgss are interlocking , where active star formation occurred simultaneously @xcite . a number of smaller shells have formed along the rims of the supershells , as expected for self - propagating star formation ( e.g. , * ? ? ? the hii regions are sometimes associated with the hi shells and the largest concentration can be found in the 30 doradus region as well as in sgss hosting the youngest star clusters . [ fig : lmcageslides ] displays six snapshots showing the spatial distribution of lmc star clusters within different age bins . the first snapshot shows only clusters younger than 20 myr . the star clusters are located in the 30 doradus region , lmc 4 , in the western part of the bar , and along the rim in the southwestern part of the galaxy . star clusters within the age range of 20 myr and 100 myr in the second and third snapshot are more widely distributed over the entire sgss and in their inter - shell regions as well as in the giant shells . there is no one - to - one correlation between the hi shells and the ionized gas traced out by the h@xmath2 regions @xcite . along the bar , we now see pronounced activity in the western region and in the center , while the eastern part of the bar is essentially quiescent . note also the particularly strong activity in the 30 doradus region and in and around sgs 11 ( lmc 4 ) . the lmc bar and the lmc rim are fully populated with star clusters within the age range of 100 myr and 250 myr ( fourth snapshot ) . star clusters with ages between 250 myr and 500 myr mostly populate the lmc bar , but many are also distributed along the western and northwestern lmc rim as well as in the northeast ( fifth snapshot ) . star clusters older than 500 myr ( up to 1 gyr ) are mostly distributed along the lmc rim , while the bar region is essentially devoid of clusters in this age range . also for the lmc , we find the older objects outside the shell regions and the younger objects are located along the rims and inter - cloud regions . the formation of young lmc clusters may have been triggered by shell expansion and interactions , which might be the reason why we do not find evidence for the standard model of shell formation ( mccray & kafatos 1987 , see also braun et al . 1997 , points et al . 1999 ) . in the following we will discuss the spatial distribution of the clusters of different ages inside the supershells sgs 3 ( lmc 1 ) , sgs 4 ( lmc 8) , sgs 9 ( lmc 9 ) , sgs 11 ( lmc 4 ) , sgs 7 ( lmc 5 ) , sgs 12 ( lmc 3 ) , and sgs 19 ( lmc 2 ) in more detail . _ sgs 3 _ ( lmc 1 ) is located in the northwestern corner of the lmc at @xmath2 = @xmath37 , @xmath33 = @xmath38 and has a dynamical age of @xmath57 myr @xcite . blue and red supergiants in the southern shell area were found to have ages less than 30 myr @xcite . young stellar objects are mostly distributed around the shell s southern periphery @xcite inside nearby hii regions . there is no cluster younger than 30 myr in our sample in sgs 3 ( lmc 1 ) and only a few clusters within the age range of 30 myr-100 myr are located in the southeastern region . clusters with ages between 30 myr and 500 myr are distributed over the entire projected area of shell . the clusters from our sample in this shell have a mean age of @xmath5150 myr . star formation in this shell mostly occurred in hii regions independent of the shell @xcite . one ob association stretches from the center of the shell to a bright southeastern hii region indicating a propagation of star formation . no indication of self - propagating star formation can be found in this study , which might be due to incompleteness of the sample . given this age structure , the formation of the shell is unrelated to the comparatively much older star clusters . _ sgs 4 _ ( lmc 8) lies in the southwestern corner of the lmc at @xmath2 = @xmath39 , @xmath33 = @xmath40 and has a dynamical age of @xmath56 myr @xcite . this shell is clearly visible in the spatial distribution of supergiants ( ages @xmath0 30 myr , grebel & brandner 1998 ) . ob associations were found in the north and the east of the shell . sgs 4 ( lmc 8) has a complex structure of hi lobes , which are connected @xcite . star clusters in this shell cover an age range between @xmath512 to 500 myr and have a mean age of 100 myr . the northern , the north - eastern , and the eastern lobes show expansion where also a few clusters younger than 50 myr are located ( see fig . [ fig : lmcageslides ] ) . the southern lobes show no distinct expansion @xcite and in this region the shell appears to be essentially quiescent . a few clusters in the age range between 50 myr and 100 myr are found in the center and the norther region of the shell . clusters in the age range betweeen 100 myr and 250 myr are located in the center , the east , and the northwest . the shell is mostly devoid of clusters older than 250 myr . only a small number of clusters is located along the rim regions in the east and the west . as for sgs 3 ( lmc 1 ) , no indication for self - propagating star formation is found in this study . _ sgs 9 _ ( lmc 9 ) is located in the southern part of the lmc at @xmath2 = @xmath41 , @xmath33 = @xmath42 . the shell is a collection of several giant h@xmath2 shells , but has no corresponding hi shell structure , which is why @xcite characterized the shell as _ this shell has almost never shown cluster formation activity within the period studied here . hi supershells are distributed from west to northeast along the periphery . the shell showing the most cluster formation activity is a region in the southeast , outside the boundary of the shell . in this region many clusters with ages between 50 myr and 250 myr are found , which might be correlated with the expanding h@xmath2 and hi region . the few clusters younger than 50 myr are located at the northeastern shell rim , which might be associated with hii regions . this shell is probably very old , because almost no young clusters are located in this shell and barely any supergiants and cepheids were found in the shell region @xcite . the star clusters from our sample in this shell have ages between @xmath520 myr to 1 gyr with a mean age of 170 myr . _ sgs 11 _ ( lmc 4 ) is the largest sgs in the lmc and is located at @xmath2 = @xmath43 , @xmath33 = @xmath44 . the shell is in collision with sgs 7 ( lmc 5 ) . the densest part of the shell lies at the northwestern border in the region between sgs 11 ( lmc 4 ) and sgs 7 ( lmc 5 ) . multiple ob associations exist in hii regions in the interaction zone , along the shell rim , and near the center of the shell @xcite . @xcite found the clusters at the rim of sgs 11 ( lmc 4 ) to be associated with co clouds . they are mostly noticable in the interaction region . stars inside the shell were found to be @xmath59 - 16 myr old , while those at the rim are @xmath36 myr old @xcite . the analysis of cepheids and other supergiants showed that star formation in this shell started about 25 myr ago in its center region and then slowly moved out toward the outer rims @xcite . sgs 11 ( lmc 4 ) appears to be the only shell that shows a weak indication of propagating star formation and has formed only within the last 10 myr . star clusters from our sample cover an age range between @xmath512 and 500 myr . the mean age lies at 60 myr . star clusters with ages less than 20 myr are found along the shell rim where also the hii regions were identified @xcite . star clusters with ages between @xmath550 - 250 myr are widely distributed across the entire shell . clusters older than 250 myr are mostly found at the southern rim of the shell . two are found in the center region and a single one at the northern rim . if the very young age of this shell is correct , the formation of clusters older than @xmath520 myr is not related to sgs 11 ( lmc 4 ) . no indication for self - propagating star formation is found in this study . _ sgs 7 _ ( lmc 5 ) is located to the northwest of sgs 11 ( lmc 4 ) at @xmath2 = @xmath45 , @xmath33 = @xmath46 and has a dynamical age of @xmath55 myr @xcite . the shell is in collision with sgs 11 ( lmc 4 ) , which caused a density enhancement on the eastern side of the shell . the cluster age distribution is similar to the one in sgs 11 ( lmc 4 ) . the star clusters from our sample in this shell have ages between 20 to 400 myr with a mean age of 130 myr . only five clusters from our sample younger than 100 myr are found in this shell and three of them are located at the eastern rim . no ob associations are found inside the shell , but along the interaction zone and along the rim @xcite . no ob association was detected in the shell interior . the shell has a relatively large co concentration @xcite . the highest co density is located in the interaction zone , superposed on ob associations and corresponding hii regions . two more high - density co regions are located in the southern region of the interaction zone and in the north . only very few objects younger than 20 myr were found in the present sample because of our object selection . the southern part of the shell and the surroundings are covered by star clusters with ages between 100 - 500 myr . the shell is devoid of objects older than 500 myr . like sgs 11 ( lmc 4 ) , this shell probably is very young . especially in the interaction zone , where sgs 7 ( lmc 5 ) and sgs 11 ( lmc 4 ) are colliding , a large quantity of blue and red supergiants are located indicating recent star formation @xcite . the old clusters in the south of the shell might have formed in a shell that is dissolved already or via a non - shell related mechanism . star formation in this shell is probably triggered by shell expansion , especially in the collision zone . _ sgs 12 _ ( lmc 3 ) is located at @xmath2 = @xmath47 , @xmath33 = @xmath48 and is in collision with sgs 19 ( lmc 2 ) . the large star forming region 30 doradus lies in the interaction zone . the sgs is divided into northern and southern lobes and shows significant hi and h@xmath2 structures @xcite . old ob associations ( age@xmath4912 myr ) exist in the interior of the shell , while young ob associations ( age@xmath312 myr ) are found within dense hii regions . the cluster mean age of this shell lies at @xmath575 myr and they cover an age range of 5 myr to 1 gyr . clusters younger than 20 myr are concentrating in the center region of the shell . clusters with ages between 20 - 250 myr are more widely distributed . the shell barely contains any clusters older than 250 myr . no signs of propagating star formation events are found . clusters might have formed via shell expansion , most efficiently in the collision zone . looking at the age distribution of blue and red supergiants ( age@xmath030 myr ) the intershell region is clearly visible as a region with a high concentration of these objects @xcite . the oldest supergiants may have triggered the subsequent compression of dense molecular clouds and resulting star bursts . _ sgs 19 _ ( lmc 2 ) is located to the southeast of 30 doradus at @xmath2 = @xmath50 , @xmath33 = @xmath51 and is the brightest sgs in the lmc in h@xmath2 @xcite . the shell is colliding with sgs 12 ( lmc 3 ) . the shell has very complex structures in hi and h@xmath2 and shows star forming regions on its western edge @xcite . the clusters from our sample cover an age range between 40 myr and 1 gyr . the cluster mean age lies at @xmath5250 myr . clusters younger than 50 myr are located in the center of the shell and toward 30 doradus . the shell hosts only a few clusters with ages between 50 myr and 250 myr and is almost devoid of older clusters . 29 clusters in the direction of lmc 2 were reported for which ages have been determined ( e.g. , * ? ? ? * ; * ? ? ? eleven of them are younger than 10 myr , seven have ages between 10 - 30 myr , one has an age of 30 - 70 myr , and one is significantly older with an age of 400 - 800 myr . the youngest clusters are associated with hii regions that are located along the periphery of lmc 2 @xcite . the analysis of supergiants revealed that this shell is very young ( 10 - 15 myr , grebel & brandner 1998 ) and in the spatial distribution of supergiants the interaction region between sgs 19 ( lmc 2 ) and sgs 12 ( lmc 3 ) is clearly visible . @xcite found that most of the co clouds in the lmc are distributed in dense parts of hi gas and over several hii regions . lmc co clouds are often located at the outer edge of sgss as for e.g. , sgs 11 ( lmc 4 ) @xcite . furthermore these authors found that clusters associated with the co clouds are younger than 5 myr , which we can not investigate due to the limited age range of the sample . due to the different morphology of the mcs compared to the mw and the absence of differential rotation of smc and lmc , low - concentration clusters can easily survive in these two galaxies , while the mw rotates differentially , which leads to shear and interaction with massive molecular clouds that can result in cluster dissolution . for the lmc @xcite found a cluster formation rate of 0.3 clusters myr@xmath31 5 gyr ago , which increased steadily to a present rate of @xmath525 clusters myr@xmath31 . for the smc such studies have not been published yet . the main distinct phases and corresponding typical timescales of cluster disruption in the mcs are : ( i. ) _ infant mortality _ ( @xmath510@xmath52 yr ) , ( ii . ) mass loss dominated by _ stellar evolution _ ( @xmath510@xmath53 yr ) , and ( iii . ) a phase dominated by _ tidal relaxation _ ( @xmath510@xmath54 yr ) in which the mass loss is driven by the clusters dynamical evolution and external influence of the tidal field of the host galaxy ( * ? ? ? * ; * ? ? ? * and references therein ) . additionally , tidal external perturbations speed up the process of disruption during all three phases , but these effects operate on longer timescales and mostly affect phase iii . most of the clusters in our sample have survived the first phase and are dominated by the second and third phase . the rate of infant mortality is highly dependent on the ambience of the host galaxy , but is largely mass - independent - at least for masses in excess of @xmath55 ( e.g. , * ? ? ? infant mortality is mainly driven by gas expulsion . this lifecycle of a cluster results from the expulsion of the intra cluster gas due to explosive expansion driven by stellar winds or supernova activity ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the end of cluster infant mortality depends on the crossing time of the cluster s gaseous progenitor and on the external tidal field @xcite . the denser the gaseous precursor of the star cluster , the shorter the cluster s crossing time and the quicker the cluster s violent relaxation . the stronger the external tidal field , the smaller the cluster s tidal radius and the faster unbound stars leave the cluster . because the external tidal fields of the mcs are not very strong , most star clusters in the mcs surviving the phase of infant mortality will have returned to an equilibrium state after @xmath540 - 50 myr @xcite . therefore , some of the clusters in the present sample might dissolve and not survive the first phase of cluster dissruption , while most clusters have ages of 40 myr and higher , and therefore are in the phase where stellar evolution has become a major agent of change , whereas the gas has already been expelled . the second ( mass - dependent ) lifecycle of a cluster includes the so - called secular evolution ( gas free evolution ) . this phase is driven by 2-body relaxation and the morphology of the host galaxy . therefore , the disruption time of a cluster is dependent on the internal cluster conditions , such as the initial mass , density , and velocity dispersion , as well as on external conditions , such as the orbit in the galaxy and tidal heating ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? [ fig : dissolution ] shows the age - radius relation of the combined smc and lmc cluster sample with ages between @xmath55 myr and 1 gyr . smc and lmc cluster ages are plotted against their apparent overall radii ( eq . 1 ) from the combined samples derived in this study and in c06 ( smc ) and pu00 ( lmc ) . the youngest objects in the smc have radii up to @xmath54 arcmin , but these objects are clusters adopted from c06 and are mostly classified as associations . moreover , until an age of @xmath540 myr star clusters in the mcs can still undergo infant mortality @xcite and therefore we have to treat this subsample with caution . the radii decrease until an age of @xmath525 myr and then remain relatively constant until 1 gyr . because the youngest objects in the lmc were not age - dated no extended objects can be seen at the young age end of the distribution . the cluster radii in the lmc show the same behavior as the smc clusters and are constant with time during the analyzed period . the cluster disruption time for the smc and the lmc derived in the literature is of the order of @xmath56 myr ( e.g. , * ? ? ? * ; * ? ? ? we have shown @xcite that clusters with ages older than 1 gyr have a trend for larger core radii with increasing age . this trend was first found by @xcite and is also visible for lmc , fornax , and sagittarius clusters @xcite . we note , however , that the radii in fig . [ fig : dissolution ] are _ apparent overall _ radii ( see section [ sec : youngies_ages ] ) so that we can not draw any firm conclusions on possible dissolution effects based on the observed sizes alone . rz05 published luminosities of 204 smc star clusters measuring integrated colors from the mcps . b96 published integrated photometry of 624 lmc star clusters that was based on observations carried out at the ctio in chile and at the casleo in argentina . for the age - dated clusters in our samples , we computed luminosities by summing the flux of each star within the individual cluster radii ( eq . 1 ) . to correct for field stars , the luminosity of a concentric annulus between 2 and 1 arcmin was calculated for each cluster , area corrected , and subtracted from the cluster luminosity . 112 clusters of our sample are in common with rz05 s sample and 217 clusters are in common with b96 s sample . the comparison between the v band luminosity derived in our study with rz05 and b96 ( fig . [ fig : vrafmevergl ] ) shows that the v band magnitudes are overall in good agreement . the lmc v band magnitudes @xmath012 mag published by b96 are systematically fainter by a small amount than the ones computed in this study . on the other hand , at the faint end the v band magnitudes ( [email protected] mag ) by b96 appear to be slightly brighter . the dispersion about the 1:1 relation ( red solid lines ) for all clusters in common is @xmath57 = 0.28 ( smc ) and 0.22 ( lmc ) . the younger the star clusters the brighter they are because young clusters still contain very massive hot stars ( e.g , supergiants ) , which contribute most of the light . the older a cluster gets the more of these massive stars are in their end stage of evolution and no longer contribute to the cluster light . therefore , clusters become fainter and redder with increasing age . this effect can be seen in the upper panels of figs . [ fig : smcraf_vager ] ( smc ) and [ fig : lmcraf_vager ] ( lmc ) , in which cluster age is plotted versus luminosity . the red dots represent the mean of the age distribution in magnitude bins of 0.3 mag . we overplotted three galev models @xcite for the total cluster masses of @xmath58 ( blue dash - dotted line ) , @xmath59 ( blue dotted line ) , and @xmath60 ( blue solid line ) using a salpeter initial mass function . galev is a model for computing the spectral evolution of single stellar populations and galaxies . the galev models give an overview of the cluster mass range . the lower plots of figs . [ fig : smcraf_vager ] and [ fig : lmcraf_vager ] show that star clusters of the smc and lmc become brighter with increasing radius , which is expected assuming a larger number of stars within the cluster radius , contributing to the total cluster luminosity . we have presented ages and luminosities of 324 and 1193 populous smc and lmc star clusters , respectively . an age range of @xmath59 myr to 1 gyr was covered based on isochrone fitting to resolved color - magnitude diagrams in both galaxies . using only cluster ages derived in this study , we find two maxima of enhanced cluster formation for both galaxies which appear to be correlated . in the smc , the peaks are found at @xmath5160 myr and @xmath5630 myr , and in the lmc at @xmath5125 myr and @xmath5800 myr . model calculations predict that the last close encounter between lmc and smc occurred around 100 - 200 myr ago . during a close encounter , the star formation is expected to be enhanced . therefore , the first peaks in the cluster age distributions could have been triggered by this tidal interaction . extending our samples with cluster ages derived by c06 we find a third pronounced period of enhanced cluster formation in the smc at around 8 myr . we find the same in the lmc combining our sample with the one of pu00 . these peaks are only visible if we extend our sample with objects classified as associations , objects which did not or could not reach higher ages because they dissolve too quickly . the youngest objects in both galaxies are associated with super giant shells , giant shells , the inter - shell region , and with hii regions . their formation is probably related to shell expansion and shell interaction . in the spatial distribution of the clusters younger than @xmath516 myr the two smc shells are clearly visible . the older objects are widely spread across the entire smc main body , but show a concentration in the western part of the galaxy . in the lmc , the youngest objects are concentrated in 30 doradus , sgs 11 ( lmc 4 ) , and in the giant shells located in the western part and in the bar region . the older lmc clusters are mostly distributed along the bar and along the rim . one can see nicely how star cluster formation propagated along the lmc bar . we find no indication for propagating star cluster formation in the sgss in either lmc or smc . most of the lmc star clusters are older than the dynamical ages of the sgss and therefore may have formed in shells , which already have dissolved and can not be detected at the present day . no obvious dissolution effects were found for mcs star clusters younger than @xmath51 gyr . it is quite difficult to ascertain a real absence of cluster dissolution using this study . two biases may play a major role : 1 . infant mortality can not be accounted for , because very young star clusters and ob - associations are not included in our sample ; and 2 . cluster dissolution processes for clusters older than @xmath51 gyr , because we did not age - date clusters in this age range . within the time period considered here - 10 myr to 1 gyr - we do not find evidence for cluster dissolution . in both galaxies , the clusters become fainter with increasing age . the very massive hot stars , which are still present in the young star clusters and contribute most of the light , become fainter and redder with increasing age and so do the star clusters . this trend can be seen in both the lmc and the smc . the total cluster luminosity increases with increasing radius due to a larger number of stars within the cluster radius . 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 . andreas koch acknowledges support by an stfc postdoctoral fellowship .

the high mass neutral quantum states envisaged by theories of physics beyond the standard model can at the hadron colliders reveal themselves through their decay into a pair of photons . once such a peak in the diphoton invariant mass distribution is discovered , the determination of its spin through the distinctive photon angular distributions is needed in order to identify the associated nonstandard dynamics . we here discuss the discrimination of the spin-2 randall - sundrum graviton excitation against the hypothesis of a spin-0 exchange giving the same number of events under the peak , by means of the angular analysis applied to resonant diphoton events expected to be observed at the lhc . the spin-0 hypothesis is modelled by an effective interaction of a high mass gauge singlet scalar particle interacting with the standard model fields . the basic observable of our analysis is the symmetrically integrated angular asymmetry , calculated for both graviton and scalar-channel exchanges to next - to - leading order in qcd . desy 11 - 139

1108.3764

diphoton final states represent a very important testing ground for the standard model ( sm ) , for example they may be one of the main discovery channels for the higgs boson searches at the cern lhc . moreover , similar to the case of dileptons , the inclusive production of two - photon high mass resonance states at the lhc : @xmath2 is considered as a powerful , clean test of new physics ( np ) , would an excess of @xmath3 events be observed with respect to the prediction from the sm cross section . one np scenario of particular importance is the case of the spin-2 kaluza - klein ( kk ) graviton excitations predicted by the randall - sundrum ( rs ) model of gravity in one warped spatial extra dimension @xcite . this model suggests a rich phenomenology that includes the production of diphoton resonances , to be explored at collider energies , see , for example , refs . @xcite . the existence of such graviton excitations can be signalled by the occurrence of peaks in the invariant mass distribution of the photon pairs and , indeed , the lowest lying predicted diphoton peak has recently been searched for in experiments at the @xmath4 fermilab tevatron collider @xcite , and at the 7 tev @xmath5 lhc collider with time - integrated luminosity of the order of 40 @xmath6 @xcite . in these experiments , exclusion mass limits on the lightest rs resonance of the tev order have been set , and graviton mass scales larger than 1 tev will certainly be in the kinematical reach of lhc . assuming that a diphoton peak at an invariant mass value @xmath7 is observed , its association to a specific np scenario would be possible only if we are able to discard other competitor models , potential sources of the peak itself with same @xmath7 and same number of events . basically , for any nonstandard model one can define , on the basis of the foreseeable statistics and uncertainties , a _ discovery reach _ on the relevant heavy resonance @xmath8 as the upper limit of the range in @xmath7 where , in a specific domain of the model parameters called `` signature space '' , the peak is expected to give a signal observable over the sm prediction to a prescribed confidence level . instead , the _ identification reach _ on the model is the upper limit of the range in @xmath7 where it can be identified as the source of the peak , once discovered , or , equivalently , the other competitor models can be excluded for all values of their respective parameters . of course , for many models , identification should be possible only in a subdomain of their signature space . the determination of the spin of an observed resonance clearly represents an important selection among different classes of nonstandard interactions . in the case of the inclusive diphoton production ( [ proc ] ) , the tool to directly test the spin-2 of the rs graviton resonance or , equivalently , exclude the hypothesis of a spin-0 scalar particle exchange , would be provided by the distinctive angular distributions in the angle @xmath9 between the incident quark or gluon and the final photon in the diphoton center - of - mass frame . this is similar to dilepton production , the difference being that in this case the hypotheses of both the spin-0 and the spin-1 exchanges must simultaneously be excluded . the spin-2 test of the lowest - lying rs graviton in lepton - pair collider events , through the direct comparison of the angular distributions for the various spin hypotheses , was earlier discussed in several papers , see , e.g. , refs . @xcite , and experimental angular analysis were attempted at the tevatron in ref . a potential difficulty of the direct - fit angular analysis at the lhc is that generally , due to the symmetry of the proton - proton initial configuration , the determination on an event - by - event basis of the direction of the initial parton , hence of the sign of @xmath10 , is in principle not fully unambiguous , so that cuts in phase space must be applied in this regard . the spin-2 rs graviton analysis of lhc dilepton events proposed in ref . @xcite , makes use of a `` center - edge '' angular asymmetry @xmath0 where the above mentioned ambiguity should not be present @xcite . essentially , in this observable the dilepton events are weighted according to the @xmath10 differential distributions , and the asymmetry is defined between cross sections symmetrically integrated over `` center '' and `` edge '' angular intervals . recently , asymmetries conceptually analogous to @xmath0 , have been applied to heavy quantum states spin identification in refs . @xcite , and a comparison of the performances of different methods for heavy resonances identification has been presented in ref . angular analyses for different spin - mediated drell - yan processes have been applied to a variety of np models in ref . @xcite . here , we propose the application of @xmath11 to the angular analysis of the diphoton production process ( [ proc ] ) at the lhc . as remarked previously , the selection of the spin-2 rs graviton amounts in practice to exclude the hypothesis of a spin-0 particle exchange with same mass @xmath7 and producing the same number of diphoton events . ideally , one advantage of the diphoton channel over dileptons can be represented by the doubled statistics expected in the former case @xcite . also , the automatic exclusion of the spin-1 hypothesis @xcite , should in any case allow a simplification of the analysis from the phenomenological point of view . finally , the consideration of process ( [ proc ] ) , in addition to dilepton production , is needed for an exhaustive test of model @xcite . for our analysis we have used the calculations of the required differential cross sections to next - to - leading order ( nlo ) in qcd , and this is essential at a hadron collider as the theoretical uncertainties get reduced when higher order corrections are included . furthermore , as a result of new interactions in a np model , there will be additional subprocesses that contribute at leading order ( lo ) itself ( e.g. , @xmath12 in the rs model ) and hence the signal can receive enhanced contributions due to the nlo corrections . specifically , in sec . ii we review the definitions of the basic cross sections involved in the asymmetry @xmath0 ; sec . iii will be devoted to the relevant properties and the characteristic angular distributions for the rs graviton and for the competitor scalar particle exchanges in process ( [ proc ] ) , for which we will adopt the model recently proposed in ref.@xcite . in sec . iv we discuss the nlo qcd effects to the diphoton production rates and to the angular distributions , for both kinds of spin exchange . v contains an outline of the @xmath0-based angular analysis and the consequent numerical results for rs identification , in the lhc center - of - mass running configurations @xmath13 tev and @xmath14 tev . finally , sec . vi contains some conclusive remarks . the total cross section for a heavy resonance discovery in the events ( [ proc ] ) at a diphoton invariant mass @xmath15 can be expressed as @xmath16 where the rapidity of the individual photon @xmath17 and @xmath18 is chosen such that @xmath19 . resonance spin - diagnosis uses the comparison between the characteristic photon differential distributions for the two hypotheses for the resonance bump , @xmath20 spin-2 kaluza - klein ( kk ) modes of the rs model graviton and @xmath21 a massive scalar : @xmath22 in eqs . ( [ totcr ] ) and ( [ diffcr ] ) , cuts on the phase space accounting for detector acceptance are implicit , and @xmath23 is an invariant mass bin around @xmath7 , which should somehow reflect the detector energy resolution and be sufficiently large as to include the resonance width . in the calculations worked out in the sequel , we use for this mass window the expression @xcite : @xmath24 actually , eq.([deltam ] ) was derived in connection to the atlas and cms experiments on dilepton production , but we assume it also for the calculations of diphoton production of interest here . obviously , for a resonance sufficiently narrow , the integral over @xmath25 should be practically insensitive to the size of @xmath23 , whereas it should be essentially proportional to @xmath23 for a flat background such as the sm . besides @xmath26 mentioned above , the assumed typical cuts on harder ( softer ) photons are @xmath27 gev , and the statistics will be estimated by taking a photon reconstruction efficiency @xmath28 . moreover , to evaluate eqs . ( [ totcr ] ) and ( [ diffcr ] ) , the partonic cross sections will be convoluted with the cteq6l and cteq6 m parton distributions sets for lo and nlo cross sections , respectively , with @xmath29 gev @xcite . in particular , for resonance discovery , process ( [ proc ] ) must be observed with a number of events well - above the background from sm processes . specifically , denoting by @xmath30 and @xmath31 the numbers of signal and sm events in the @xmath32 invariant mass window , the statistical significance of a 5-@xmath33 signal would be ensured by the criterion that @xmath30 should be larger than max(@xmath34,@xmath35 ) . the @xmath36-evenly - integrated center - edge angular asymmetry @xmath0 is defined as : @xmath37 \frac{{\rm d } \sigma}{{\rm d } z}\ , { \rm d } z.\ ] ] in eq . ( [ ace ] ) , @xmath38 defines the separation between the `` center '' ( @xmath39 ) and the `` edge '' ( @xmath40 ) angular regions and is _ a priori _ arbitrary to some extent . in previous applications , see for example refs . @xcite , the `` optimal '' numerical value turned out to be @xmath41 , and we shall keep this value of @xmath42 here as well . one can notice that by definition @xmath0 is symmetric under @xmath43 , hence it is insensitive to the sign of @xmath36 . moreover , as being a ratio of integrated cross sections , an advantage of @xmath0 is that it should be less sensitive to theoretical systematic uncertainties , such as the uncertainties from different sets of parton distributions and from the particular choice of factorization and renormalization scales . we here sketch the models we are interested in , together with their features relevant to the resonance spin and the distinctive angular distributions for process ( [ proc ] ) . this model , originally proposed as a solution to the gauge hierarchy problem @xmath44 , consists of two 3-branes , and one compactified warped extra spatial dimension @xmath45 with exponential warp factor @xmath46 @xcite . here , @xmath47 is the 5d curvature , assumed to be of the order of @xmath48 . the two branes are placed at orbifold fixed points , @xmath49 with positive tension called the planck brane and the second brane at @xmath50 with negative tension called the tev brane . the basic , simplifying , hypothesis is that the sm fields are localized on the tev brane , whereas gravity originates on the planck brane but is allowed to propagate everywhere in the 5d space . the consequence of this setup is the existence of kk modes of the graviton that can be exchanged in the interactions among sm particles in tev brane . owing to the exponentially suppressing warp factor , mass scales , in passing from the planck brane to the tev brane , can get the size of the tev order . moreover , a specific mass spectrum of such kk resonances is predicted in terms of an effective mass scale defined as @xmath51 , that for @xmath52 happens to be of the tev order ( here , @xmath53 with @xmath54 the newton constant ) . these resonances , represented by spin-2 fields @xmath55 , can in process ( [ proc ] ) show up as ( narrow ) peaks in @xmath56 , through the interaction @xmath57 here : @xmath58 is the energy - momentum tensor of the sm ; the characteristic mass spectrum is @xmath59 ( with @xmath60 the roots of the bessel function @xmath61 ) ; and the resonance widths are @xmath62 , with @xmath63 a calculable constant depending on the number of open decay channels , of the order of @xmath64 . the model can therefore conveniently be parametrized in terms of @xmath65 , the mass of the lowest graviton excitation , and of the `` universal '' dimensionless graviton coupling @xmath66 . theoretically , the expected `` natural '' ranges for these parameters , avoiding additional mass hierarchies , are : @xmath67 and @xmath68 tev @xcite . the 95% cl experimental lower bounds on @xmath69 from previous analysis vary , essentially , from @xmath70 to @xmath71 tev as @xmath72 ranges from @xmath73 to @xmath64 @xcite . quite recently , preliminary results from rs graviton searches in dilepton inclusive production at the 7 tev lhc with luminosity 1.2 @xmath74 , indicate 95% cl lower limits on @xmath69 of 0.7 tev for @xmath75 up to 1.6 - 1.7 tev for @xmath76 @xcite . in hadronic collisions , in qcd at lo , photon pairs can be produced _ via _ the quark antiquark annihilation @xmath77 , and the gluon gluon fusion @xmath78 . the relevant diagrams at this order , for the sm and the rs graviton exchange , are represented in figs.[qq ] and [ gg ] . actually , the sm box diagram in fig . [ gg ] is of higher order in @xmath79 and , as discussed in the next section , for the values of the @xmath32 invariant mass @xmath25 in the tev range of interest here , its contribution turns out to be negligible , as earlier noticed also in ref . @xcite . , where the dashed line represents either a rs graviton kk mode @xmath80 or a scalar exchange @xmath81 . the crossed diagrams are not displayed.,width=377 ] , including the exchange of either a rs graviton kk mode @xmath80 or a scalar @xmath81 which is denoted by the dashed line . the crossed diagrams are not displayed.,width=377 ] using feynman rules for graviton exchange @xcite , the @xmath36-even angular dependencies needed in ( [ ace ] ) can be written as @xcite : @xmath82}\left(1+z^2\right ) \bigr]\ , \label{dsigmaqq}\end{aligned}\ ] ] where the subprocess mandelstam variable is the diphoton invariant mass , @xmath83 , and @xmath84 is the electromagnetic coupling constant with @xmath85 the quark electric charge . moreover , in eq . ( [ dsigmaqq ] ) , @xmath86 represents the sum of kk graviton propagators with masses @xmath87 and widths @xmath88 : @xmath89 in practice , from the phenomenology , just the lowest graviton mass @xmath65 and , perhaps , the next one @xmath90 at most , can be expected to fall within the discovery reach of lhc . the cross section for the @xmath91 subprocess _ via _ the rs graviton excitation exchange is @xmath92 notice that a factor @xmath93 is embodied in eqs . ( [ dsigmaqq ] ) and ( [ dsigmagg ] ) to account for the identical final state photons . in principle , in order to discriminate the graviton spin-2 angular distribution from the spin-0 hypothesis , we might limit ourselves to make a comparison of eqs . ( [ dsigmaqq ] ) and ( [ dsigmagg ] ) with the results of a generic flat ( in @xmath36 ) distribution numerically tuned to the same number of events around @xmath69 . however , for the graviton exchange we shall use a description supplemented by the cross sections calculated to nlo in qcd and , to consistently fully exploit the qcd dynamics also for the spin-0 scenario , we need an explicit model for a scalar particle exchange with definite couplings to the sm fields , in particular to photon pairs . we consider the simple model of a scalar particle @xmath81 , singlet under the sm gauge group and with mass @xmath94 of the tev order , proposed in ref . the trilinear couplings of @xmath81 with gluons , electroweak gauge bosons and fermions , are in this model : @xmath95 in eq . ( [ scalar ] ) , @xmath96 is a high mass scale , of the tev order of magnitude , and @xmath72 s are dimensionless coefficients that are assumed to be of order unity , reminiscent of a strong novel interaction . in our subsequent analysis we take , following ref . @xcite , @xmath97 tev . as for the @xmath72 s , we shall leave their numerical values free to vary in a range of the order of ( or less than ) unity , but constrained so that the scalar particle width @xmath98 could be comparable to ( or included in ) the diphoton mass window @xmath23 of eq . ( [ deltam ] ) . the leading order diphoton production process is in this model dominated by the @xmath1-channel exchange @xmath99 . as it is of order @xmath100 at lo , it will be sensitive to the choice of this coupling constant . the feynman diagrams in the scalar model will be similar to those in the rs model , as shown in figs . [ qq ] and fig . [ gg ] with the kk mode replaced by the scalar . the corresponding differential cross section reads , including a factor @xmath93 for identical final particles : @xmath101 in eq . ( [ dsigmaggs ] ) , @xmath102 is the scalar propagator , @xmath103 and the expression of the total width @xmath98 in terms of the @xmath72 s and @xmath96 introduced in eq . ( [ scalar ] ) can be obtained from @xcite , by summing the partial widths reported there into @xmath104 ( dominant , as is the cross section , proportional to @xmath105 ) , @xmath32 , @xmath106 , @xmath107 , and @xmath108 . in fig . [ fig - narrow - scalar ] we represent , as an example , the domains in @xmath105 _ vs _ all other parameters @xmath109 ( @xmath110 ) assumed equal to each other , allowed by the constraint @xmath111 for different values of @xmath112 . the dashed lines in the parameter plane @xmath113 with equal @xmath109 ( @xmath110 ) , correspond to the equation @xmath114 for @xmath115 tev , 1.5 tev and 3 tev , @xmath116 tev . the allowed area is at the left of these lines.,width=302 ] as the coupling of the scalar particle to quarks is proportional to the quark mass much smaller than @xmath96 , one can neglect at lo the subprocess @xmath117 , and the cross section for @xmath118 will be expressed simply by the gluon - initiated process plus the sm contribution appearing in eq . ( [ dsigmaqq ] ) . thus , at this lo , there is in the considered spin-0 model no interference between the sm quark - initiated contribution and the scalar - exchange amplitude . however , quark contributions can not be neglected at the nlo qcd order and , in particular , a ( small ) interference of the @xmath104 sm box diagram with the @xmath104 scalar exchange will occur , as it will be specified in the next section . preliminary to the numerical analysis of @xmath0 , in the next section we briefly describe the estimate of the next - to - leading order qcd effects . the diphotons produced in a hadronic collision could originate from the hard partonic interaction ( direct photon ) or at least one photon could be produced in the hadronisation of a parton ( fragmentation photon ) . at higher orders in qcd , there would be final state collinear singularities of qed origin as a result of the emission of a photon from a quark . these singularities can be factored out and absorbed into the fragmentation functions . the fragmentation functions are additional non perturbative inputs that are not well - understood . an alternate approach to isolate direct photons is the smooth cone isolation criterion @xcite , which ensures that the fragmentation contributions are suppressed without affecting the cancellation of the conventional qcd singularities . we start from the extra dimension scenario , where the nlo corrections in qcd were considered for the phenomenologically interesting process like the dilepton @xcite and diphoton production @xcite at hadron colliders . the essential feynman rules for kk modes coupling to ghosts and kk mode propagator in @xmath119-dimensions needed for the nlo computation have been introduced in @xcite . for the diphoton computation a semi - analytical two - cutoff phase space slicing method @xcite to deal with various singularities of infrared ( ir ) and collinear origin that appear at nlo in qcd was used . after cancellation and mass factorisation of these singularities , the remaining finite part is numerically integrated over the phase space by using monte - carlo techniques . as stated in previous sections , diphotons can in the sm be produced to lo in qcd , in the quark anti - quark annihilation subprocess @xmath120 . photon pairs produced _ via _ the gluon fusion subprocess through a quark box diagram @xmath121 , though of the order @xmath100 , have cross sections comparable to those of the @xmath120 subprocess for the low diphoton invariant mass . in the light higgs boson searches , this subprocess plays an important role , due to the large gluon flux at small fractional momentum @xmath122 , and is formally treated as a lo contribution although it is of @xmath123 , hence is in reality a next - to - next - to leading order contribution . however , it falls off rapidly with increasing diphoton invariant mass and in the mass range of interest for the tev scale gravity models it need not be included at lo . it has been demonstrated in @xcite that the contribution of this subprocess in the sm is few orders of magnitude smaller than that of the @xmath124 sub process for diphoton invariant mass @xmath125 gev . ) in ref . @xcite . ] . the dashed line could be either @xmath20 or @xmath81 depending on the model . for the scalar exchange @xmath81 the last diagram involving a four point vertex does not exist.,width=491 ] , where the dashed line corresponds to a spin-2 rs kk mode @xmath80 or a scalar exchange @xmath81 . in the case of a scalar exchange the last diagram involving a four point vertex does not exist.,width=491 ] the nlo qcd corrections to the diphoton production process would involve real emission diagrams corresponding to the following @xmath126 subprocesses : ( a ) @xmath127 ( fig . [ qqreal ] ) , ( b ) @xmath128 ( fig . [ ggreal ] ) and ( c ) @xmath129 ( the @xmath130 diagrams can be obtained from the @xmath124 diagram by appropriate inter change of initial and final state particles ) . in the virtual part , @xmath131 corrections at one loop arise as a result of the interference between the one loop graphs at @xmath131 of @xmath132 and the born graphs at @xmath133 of @xmath132 . some of the typical np one loop feynman diagrams to @xmath131 are shown in fig . [ virtual ] . the @xmath124 channel gets such contributions from both the sm and the graviton exchange , while in the @xmath104 channel the sm contribution already begins at @xmath131 . the sm @xmath134 subprocess amplitude can interfere with the gluon initiated lo gravity exchange subprocess , giving a @xmath131 contribution fig . [ gg ] . in the @xmath135 channel there is no virtual contribution to this order in either the sm or the np models of interest here . and @xmath104 channel in the np models.,width=491 ] both the virtual and real corrections have been evaluated with 5 quark flavors and in the limit of vanishing quark masses . the @xmath136-point tensor integrals appearing from integration over loop momenta were simplified using the passarino - veltman reduction @xcite , the computation was done using dimensional regularization in @xmath119-dimensions , in the @xmath137 scheme . the virtual contribution here does not contain uv singularities , this can be attributed to the facts that _ i ) _ the electromagnetic current is conserved and does not receive any qcd corrections , and _ ii ) _ that the graviton couples to the energy - momentum tensor of the sm fields which is a conserved quantity and therefore is not renormalized . the poles in @xmath138 arise from loop integrals and correspond to the soft and collinear divergences , configurations where a virtual gluon momentum goes to zero give soft singularities while the collinear singularities arise when two massless partons become collinear to each other . the 3-body phase space of the real emission diagrams has regions which are soft and collinear divergent . the phase space can accordingly be separated into soft @xmath139 and hard regions and , furthermore , the hard region can be separated into hard collinear @xmath140 and hard non - collinear @xmath141 parts as follows : @xmath142 the small cut - off parameters @xmath143 and @xmath144 set arbitrary boundaries for the soft ( gluon energy @xmath145 ) and collinear ( @xmath146 ) regions , respectively . here , for the @xmath126 process with momenta @xmath147 , the mandelstam variables are defined as @xmath148 , @xmath149 , with @xmath150 . in these mutually exclusive soft and hard collinear regions , the 3-body cross section can be factored into the born , @xmath151 , cross section and the remaining phase space integral can easily be evaluated in @xmath119-dimensions to get the expansion of the soft and collinear singularities in powers of @xmath138 . all positive powers of the small cut - off parameters @xmath143 and @xmath144 are set to zero , only logarithms of the cut - off parameters are retained . adding the virtual and real contributions , all double and single poles of ir origin are automatically cancelled between the virtual and the first two terms of eq . ( [ phs1 ] ) . the remaining collinear poles are absorbed into the parton distribution functions . the hard non - collinear part in eq . ( [ phs1 ] ) corresponds to the 3-body phase space , which by construction is finite , and can be evaluated in 4-dimensions ( @xmath152 ) . the sum of real and virtual contributions is now free of singularities and can hence be evaluated numerically using a monte - carlo method . it can be further seen that the explicit logarithmic dependence on @xmath143 and @xmath144 in the 2-body phase space ( @xmath153 ) is cancelled by the implicit dependence of the 3-body hard non - collinear part @xmath154 on these parameters , after the numerical integration is carried out . the sum of the 2-body and 3-body parts in eq . ( [ phs1 ] ) would be independent of the slicing parameters @xmath143 and @xmath144 , which is explicitly verified before the code is used for the analysis . now these codes @xcite can be used to study the full quantitative implication of the nlo qcd corrections to the various distributions of interest in the extra dimension searches . turning to the scalar - exchange model , the matrix elements corresponding to the interference between the sm box diagram and the lo @xmath104 initiated tree level diagram is given by @xmath155 { 1 \over [ n^2 - 1 ] } ~\left({c_3~g_s^2 } \over \lambda \right ) ~ \left ( { ( c_1+c_2)~e^2 \over \lambda } \right)~ \nonumber \\ { } & & \times \bigg\ { \left ( { \hat{t}-\hat{u } \over 2}\right ) \left[\hbox{ln}\left({-\hat{t } \over \hat{s}}\right ) - \hbox{ln}\left({-\hat{u } \over \hat{s}}\right ) \right ] \nonumber \\ { } & & -\left ( { { \hat{u}^2+\hat{t}^2 } \over 2s } \right ) \left [ \hbox{li}_2\left({-\hat{u } \over \hat{t}}\right ) + \hbox{li}_2\left({-\hat{t } \over \hat{u}}\right ) -2~\zeta(2 ) \right ] \bigg\},\end{aligned}\ ] ] where @xmath156 , @xmath157 , @xmath158 are the usual mandelstam variables , @xmath159 is the colour degrees of freedom and @xmath160 is the riemann zeta function . in the scalar model , the uv divergences in the virtual corrections to the @xmath104 initiated diphoton production process are removed by renormalization . the remaining finite contribution is given by @xmath161 where @xmath162 is the born contribution given in eq . ( [ dsigmaggs ] ) . the remaining soft and collinear finite contributions coming from the real corrections to the lo scalar model diagram are proportional to the born cross section . as they originate from pure qcd , they are independent of the hard scattering process , hence they will be the same as those for the @xmath1-channel diphoton production process in the rs model and are given in ref . @xcite . at the nlo in qcd , the following three subprocesses contribute to the scalar model cross section : _ i ) _ @xmath163 _ ii ) _ @xmath164 and _ iii ) _ @xmath165 , all followed by @xmath166 . the feynman diagrams for the @xmath167 and @xmath104 channels are given figs . [ qqreal ] and [ ggreal ] , respectively , wherein the dashed line now represents the scalar @xmath81 and the four point diagrams will be absent for the scalar case . the couplings of @xmath81 to the light quarks are proportional to the masses and hence are negligible so that the main contribution to the @xmath167 channel is from the @xmath81 coupling to the gluons . again the @xmath168 channel is related to the @xmath167 channel by suitable change of initial and final state . [ fig - ang - distr ] shows , as an example , for the diphoton resonance mass values @xmath169 and 3 tev , the angular distributions at the 14 tev lhc , at lo and nlo , in the cases of the sm , the rs graviton exchange with coupling constant @xmath75 , and the scalar particle exchange with couplings ( @xmath170 ) and ( @xmath171 ) , @xmath110 . angular distributions of photons for process @xmath172 at lo and nlo qcd in sm , rs model with graviton coupling constant @xmath75 , and scalar resonance model with @xmath173 , @xmath174 , @xmath110 ( left and right top panels ) and @xmath173 , @xmath175 ( left and right bottom panels ) . @xmath169 tev and 3 tev at the 14 tev lhc . , title="fig:",width=340 ] angular distributions of photons for process @xmath172 at lo and nlo qcd in sm , rs model with graviton coupling constant @xmath75 , and scalar resonance model with @xmath173 , @xmath174 , @xmath110 ( left and right top panels ) and @xmath173 , @xmath175 ( left and right bottom panels ) . @xmath169 tev and 3 tev at the 14 tev lhc . , title="fig:",width=340 ] angular distributions of photons for process @xmath172 at lo and nlo qcd in sm , rs model with graviton coupling constant @xmath75 , and scalar resonance model with @xmath173 , @xmath174 , @xmath110 ( left and right top panels ) and @xmath173 , @xmath175 ( left and right bottom panels ) . @xmath169 tev and 3 tev at the 14 tev lhc . , title="fig:",width=340 ] angular distributions of photons for process @xmath172 at lo and nlo qcd in sm , rs model with graviton coupling constant @xmath75 , and scalar resonance model with @xmath173 , @xmath174 , @xmath110 ( left and right top panels ) and @xmath173 , @xmath175 ( left and right bottom panels ) . @xmath169 tev and 3 tev at the 14 tev lhc . , title="fig:",width=340 ] k - factor _ vs _ @xmath7 for 14 tev ( left panel ) and 7 tev ( right panel ) lhc . , title="fig:",width=389 ] k - factor _ vs _ @xmath7 for 14 tev ( left panel ) and 7 tev ( right panel ) lhc . , title="fig:",width=389 ] k - factor _ vs _ @xmath176 for the 7 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 7 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] k - factor _ vs _ @xmath176 for the 14 tev lhc . , title="fig:",width=351 ] the k - factors represent quantitatively the magnitude of the nlo qcd corrections , and are defined as the ratio of the nlo cross section to the corresponding lo one as follows : @xmath177 here , the subscript @xmath178 refers to the new physics contribution ( extra dimension or scalar exchange ) and its interference with the sm to lo or nlo as the case may be , while the subscript @xmath179 refers to only the sm contribution . the ratio could be of total cross sections as is the case of fig . [ fig - kfac - mr ] , or of angular distributions as is the case of figs . [ fig - kfac - z-7tev ] and [ fig - kfac - z-14tev ] . including higher order qcd corrections to an observable at the hadron collider reduces the dependence on the factorization and renormalization scales , further since the k - factor itself could be large , it is essential to include these corrections . the dependence of the k - factors on the coupling constant @xmath72 in the rs case , exhibited in the lower panel of fig . [ fig - kfac - mr ] , can be understood by the fact that for low values of @xmath180 the np interference with the sm could be of the same order as the pure np part . for values of @xmath181 , the sm contribution is negligibly smaller than the pure np part and can be neglected both in the numerator and in the denominator of eq . ( [ kfactor ] ) , which in this case is determined by the np solely so that the dependence from @xmath72 cancels . it is interesting to notice from the upper panel of fig . [ fig - kfac - mr ] that the k - factors in the scalar case are larger than those in the rs case : the nlo corrections have enhanced the cross sections but , as shown in figs . [ fig - kfac - z-7tev ] and [ fig - kfac - z-14tev ] , they did not noticeably change the shape of the angular distributions from the flat behavior of the pure scalar particle exchange cross section in eq . ( [ dsigmaggs ] ) . also , one may remark that in the example discussed here the rs graviton couples to quarks and gluons with equal coupling strength whereas , in the model of ref . @xcite we have adopted , the scalar particle couples mainly to gluons ( couplings to quarks are identically zero in the limit of vanishing quark masses ) . furthermore , the production of a scalar particle in the gluon - gluon fusion subprocess is _ qualitatively _ equivalent to the higgs boson production in the limit of infinite top quark mass : it is well - known that k - factors ( due to nlo qcd corrections ) for the higgs production process at hadron colliders are very high and can easily be greater than 2.0 in the light higgs mass region @xcite . hence , a similar pattern of k - factors can be expected in the case of a scalar production ( followed by decay to photons ) in the model considered here , see figs . [ fig - kfac - mr ] , [ fig - kfac - z-7tev ] , and [ fig - kfac - z-14tev ] . the @xmath0-based angular analysis will essentially proceed as follows . the first step will be the determination of graviton - scalar `` confusion regions '' , namely , of the subdomains in respective discovery signature spaces of coupling constants and masses where , for @xmath182 , the two models predict equal numbers of resonance signal events , @xmath30 , hence are not directly distinguishable on a statistical basis . in such confusion regions , one then can try to discriminate the models from one another by means of the different values of the @xmath0 asymmetry generated by the respective photon angular distributions . the representation of @xmath0 _ vs _ @xmath42 predicted at nlo for the two models with diphoton resonance masses of 2 and 4 tev , at the 14 tev lhc , is shown in fig . [ fig - ace - zst ] . nlo @xmath0 asymmetry vs @xmath183 for the inclusive process ( [ proc ] ) at the 14 tev lhc for the rs model with graviton coupling constant @xmath75 @xmath184 , and the scalar resonance model with @xmath173 , @xmath174 ( @xmath173 , @xmath175 ) . @xmath185 tev : top - left panel ( top - right panel ) . bottom panels : same as for top panels , but with @xmath186 tev.,title="fig:",width=340 ] nlo @xmath0 asymmetry vs @xmath183 for the inclusive process ( [ proc ] ) at the 14 tev lhc for the rs model with graviton coupling constant @xmath75 @xmath184 , and the scalar resonance model with @xmath173 , @xmath174 ( @xmath173 , @xmath175 ) . @xmath185 tev : top - left panel ( top - right panel ) . bottom panels : same as for top panels , but with @xmath186 tev.,title="fig:",width=340 ] nlo @xmath0 asymmetry vs @xmath183 for the inclusive process ( [ proc ] ) at the 14 tev lhc for the rs model with graviton coupling constant @xmath75 @xmath184 , and the scalar resonance model with @xmath173 , @xmath174 ( @xmath173 , @xmath175 ) . @xmath185 tev : top - left panel ( top - right panel ) . bottom panels : same as for top panels , but with @xmath186 tev.,title="fig:",width=340 ] nlo @xmath0 asymmetry vs @xmath183 for the inclusive process ( [ proc ] ) at the 14 tev lhc for the rs model with graviton coupling constant @xmath75 @xmath184 , and the scalar resonance model with @xmath173 , @xmath174 ( @xmath173 , @xmath175 ) . @xmath185 tev : top - left panel ( top - right panel ) . bottom panels : same as for top panels , but with @xmath186 tev.,title="fig:",width=340 ] one can start from the assumption that an observed peak at @xmath15 is due to the lightest spin-2 rs graviton ( thus , that @xmath187 ) , and define a `` distance '' from the scalar - exchange model hypothesis as : @xmath188 an indication of the domain in the ( @xmath189 ) rs parameter plane , where the competitor spin-0 hypothesis giving same number of resonant events for @xmath190 can be excluded by the starting rs graviton hypothesis , can be obtained from a simple - minded @xmath191-like numerical procedure , similar to that used in ref . the comparison of the deviations ( [ deltags ] ) to the statistical uncertainty @xmath192 pertinent to the rs model , suggests the following criterion for spin-0 exclusion : @xmath193 eq . ( [ chisquare ] ) shows the definition of the @xmath191 , and @xmath194 specifies a desired scalar - exchange exclusion confidence level ( for example , @xmath195 for 95% cl ) . with @xmath196 calculated in terms of @xmath7 and model coupling constants , this condition will define the domain in the confusion - regions of model parameters where the rs spin-2 hypothesis can be discriminated from the scalar exchange . with @xmath197 much smaller than unity for values of @xmath42 around 0.5 , we have to a good approximation : @xmath198 where @xmath199 will be the minimum number of rs resonance events needed to satisfy the criterion ( [ chisquare ] ) , hence to exclude the spin-0 exchange model with same @xmath182 in the confusion - region of the parameters . the knowledge of @xmath199 determines , in turn , the rs resonance identification subdomain in the ( @xmath189 ) parameter plane . number of resonance ( signal ) events @xmath30 _ vs _ @xmath200 ( @xmath201 ) at the lhc with @xmath202 tev and @xmath203 for the process @xmath204 at lo ( left panel ) and nlo ( right panel ) qcd . the yellow area corresponds to the kk graviton signature space for @xmath205 ; the signature space for the scalar @xmath81 resonance is represented by the two specific cases , @xmath170 and @xmath173 , @xmath175 . , title="fig:",width=302 ] number of resonance ( signal ) events @xmath30 _ vs _ @xmath200 ( @xmath201 ) at the lhc with @xmath202 tev and @xmath203 for the process @xmath204 at lo ( left panel ) and nlo ( right panel ) qcd . the yellow area corresponds to the kk graviton signature space for @xmath205 ; the signature space for the scalar @xmath81 resonance is represented by the two specific cases , @xmath170 and @xmath173 , @xmath175 . , title="fig:",width=302 ] similar to fig . [ fig - sig-14tev ] , but for nlo qcd at @xmath2067 tev and @xmath2071 fb@xmath208 ( left panel ) and @xmath20710 fb@xmath208 ( right panel).,title="fig:",width=302 ] similar to fig . [ fig - sig-14tev ] , but for nlo qcd at @xmath2067 tev and @xmath2071 fb@xmath208 ( left panel ) and @xmath20710 fb@xmath208 ( right panel).,title="fig:",width=302 ] we apply this procedure to the design lhc running conditions @xmath13 tev and time - integrated luminosity @xmath209 @xmath74 . figure [ fig - sig-14tev ] shows the rs graviton signature domain for @xmath210 at lo ( left panel ) and to nlo ( right panel ) in qcd . the red line labelled as `` discovery '' indicates the minimum number of events statistically needed for the rs kk resonance discovery in process ( [ proc ] ) at the 5-@xmath33 level , as anticipated in sec . the domain between the lines labelled as @xmath75 and @xmath76 represents the number of events for rs kk graviton production followed by decay into a photon pair , theoretically evaluated as described in the previous section , for different values of @xmath69 . the scalar resonance signature space is also included in this figure , at the same lo and nlo , for simplicity by the representative lines labelled as @xmath170 and @xmath211 ( @xmath110 ) , respectively . one can see that , for these values of the scalar coupling constants , there is a finite confusion region where , at given mass @xmath182 , the numbers of predicted resonance signal events can be equal . actually , such a confusion region might easily be extended to almost completely overlap with the full rs signature space by partially weakening the condition @xmath212 . this condition is , anyway , to be understood in a qualitative sense , so that more numerical freedom to the scalar coupling constants of eq . ( [ scalar ] ) might be allowed . indeed , if the width @xmath98 turned out to be larger than @xmath23 in eq . ( [ deltam ] ) , the analysis proposed here should still be viable , and could in this case discriminate a narrow kk graviton vs the scalar resonances , both by the angular observable @xmath0 and by the size of the widths themselves . also , we can remark that , as relying just on specific angular distributions , the kind of analysis proposed here should be applicable more generally , to the identification of the rs graviton excitation from different scalar exchange models than studied here . the line labelled as `` i d '' in fig . [ fig - sig-14tev ] essentially represents the solution of eq . ( [ chisquare ] ) relating @xmath69 to @xmath72 , i.e. , it is the minimum number of events needed for a discovered rs graviton resonance to be identified at 95% cl against the scalar particle exchange hypothesis , according to the @xmath0-based angular analysis . the differences between the left and right panels clearly show the need to account for the large nlo qcd effects in the theoretical description of the resonant diphoton inclusive production at lhc . in practice , as one can read from the right , next - to - leading order panel in fig . [ fig - sig-14tev ] , if an rs graviton excitation will be discovered in diphoton events at the 14 tev lhc with 100 @xmath74 luminosity , its spin may be identified for @xmath69 up to , roughly , 23 tev for @xmath213 if the number of observed rs resonance signal events will be larger or equal to those indicated by the line `` i d '' : namely , @xmath214 . of course , these indications follow from the criterion ( [ chisquare ] ) outlined above , hence rely on statistical arguments and theoretical calculations of the relevant cross sections , in particular one may notice in this regard that the sm background turns out to be completely negligible with respect to the signal . the detailed assessment of the `` experimental '' backgrounds to the resonance discovery in process ( [ proc ] ) , and of the related systematic uncertainties , is out the scope of this paper , our purpose here is to just compare ( and discriminate ) two different theoretical explanations for same resonance mass events , once observed , on the basis of nlo calculations in qcd . the grey area in the right panel of fig . [ fig - sig-14tev ] represents the 95% cl exclusion , where the rs resonance should not be observable , if we account for the lower limits on @xmath69 vs @xmath72 derived from the dilepton production analyses recently presented in refs . @xcite and @xcite . thus , the range in @xmath7 of interest would start , in view of the new lhc results , from @xmath215 tev for @xmath76 and @xmath216 tev for @xmath75 . also drawn in fig . [ fig - sig-14tev ] is the line @xmath217 tev : the `` theoretical '' condition @xmath218 tev mentioned in sec . , if enforced literally , would dramatically constrain the rs discovery domain in the plane @xmath219 to the events located above this line . however , also this condition should be understood in a qualitative sense , as is the case , in principle , of the assumed range of values for @xmath72 . to make a contact with the current lhc running conditions , we repeat the rs identification procedure outlined above for the 7 tev case , with time - integrated luminosities @xmath220 and 10 @xmath74 . the analogue of fig . [ fig - sig-14tev ] , but this time for the nlo calculations only , is represented in fig . [ fig - sig-7tev ] , with the same significance of the symbols . the interpretation of the left and right panels in this figure is also completely analogous . for example , if a rs resonance were discovered in diphoton events at 1 @xmath74 , its spin-2 might be identified , to 95% cl , up to @xmath221 tev for @xmath76 , provided the collected signal were about @xmath222 events ; and for 10 @xmath74 , identification would be possible up to @xmath223 tev for @xmath76 with a collected signal of , say , about @xmath224 events . however , the situation is drastically modified by the 95% cl exclusion limits from the dilepton analysis of refs . @xcite at 1.2 @xmath74 , reported in sec . iii and represented by the grey areas in both panels of fig . [ fig - sig-7tev ] . the experimental limit @xmath225 tev for @xmath76 is not quite inconsistent with the left panel of fig . [ fig - sig-7tev ] , which shows that for these values the theoretically predicted statistics for rs events falls below that needed for 5-@xmath33 discovery . the exclusion range starts from the `` low '' values @xmath75 and @xmath216 tev , so that the left panel of this figure shows that , at the present stage , in principle there may be a little corner left available for discovery , roughly @xmath69 between 0.7 and 1.3 tev and small @xmath226 . on the other hand , this panel clearly indicates that there is no room for rs graviton identification at 1 @xmath74 luminosity , at least with the angular analysis presented here . moving to the 10 @xmath74 case , the right panel of fig . [ fig - sig-7tev ] shows that , with the current lhc limits , discovery might still be possible up to about @xmath227 tev with @xmath76 , but the identification would be allowed only for @xmath228 tev with @xmath72 not larger than , say , @xmath229 . in the previous sections , we have discussed the features of different - spin @xmath1-channel exchanges in the inclusive diphoton production process ( [ proc ] ) at the lhc . specifically , we have considered the hypothesis of the spin-2 rs graviton excitation exchange as the source of an eventually discovered peak in the diphoton invariant mass , and compared it to the interpretation of the same peak as a spin-0 scalar exchange exemplified by model @xcite . the aim has been of determining quantitatively the domain in the rs graviton mass @xmath69 and coupling constant @xmath66 , where the former hypothesis can be identified against the latter ( the so - called identification reach ) . clearly , to this purpose , in the situation of equal number of peak events from the two models , the information from the distinctive photon angular distributions is needed . the relevant cross sections , differential and angular integrated , at next - to - leading order in qcd have been introduced for both the rs and the scalar - exchange model . the comparison with the leading order calculations shows that the nlo effects are substantial , and non - negligible for this analysis . as shown in sec . iv , in the tev resonance mass range of interest here and at the considered lhc energies , k - factors turn out to be large , of the order of 1.5 or so for the rs exchange , and even larger , of the order of 3 or so , for the scalar exchange model . moreover , while k - factors exhibit an angular dependence in the case of the rs model , they do not noticeably alter the flat shape of the angular distribution for the pure scalar resonance exchange derived at leading order in qcd . the angular analysis to discriminate the rs from the scalar model has been based on the center - edge asymmetry @xmath0 , also estimated at nlo in qcd , for both the 14 tev and the 7 tev lhc . the numerical results for the predicted total number of resonant events and the minimum number of events for rs identification , obtained by the simple statistical arguments outlined in the previous section , are presented in figs . [ fig - sig-14tev ] and [ fig - sig-7tev ] , and summarized in table [ table : dis - id ] . as regards the 7 tev lhc , the theoretical results obtained above show that , taking into account recent experimental limits on the rs graviton mass , for luminosity around 10 @xmath74 there may still be the possibility to identify the rs graviton excitation , in a rather limited range of @xmath69 and small @xmath72 . higher luminosity , and/or larger lhc energy , would be required to extend this region by the @xmath0 based angular analysis . finally , one may observe that the results regarding the rs graviton identification obtained in the previous section show that the inclusive diphoton process ( [ proc ] ) at the lhc is complementary to the drell - yan dilepton production , in the sense that only the scalar exchange needs be considered as a competitor , alternative , hypothesis for the source of resonance events . we have performed the @xmath0 angular analysis of the diphoton production cross sections for both hypotheses at nlo in qcd , and estimated numerical results for discovery and identification . the @xmath0 method has so far been applied to the dilepton channel at the lo in qcd only . therefore , it should be interesting , for a fully exhaustive comparison of the results achievable from the two channels , to extend the @xmath0-based angular analysis at nlo also to the dilepton production process . this research has been partially supported by funds of the university of trieste and by the abdus salam ictp under the tril and the associates . pm acknowledges the ictp associateship , which lasted until the initial stage of this collaboration . the work of mck and vr has been partially supported by the funds of regional center for accelerator based particle physics ( recapp ) , department of atomic energy , govt . of india . we would also like to thank the cluster computing facility at harish - chandra research institute , where part of the computational work for this study was carried out . a conversation with i. a. golutvin is gratefully acknowledged . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . h. davoudiasl , j. l. hewett and t. g. rizzo , phys . d * 63 * , 075004 ( 2001 ) [ arxiv : hep - ph/0006041 ] ; phys . * 84 * , 2080 ( 2000 ) [ arxiv : hep - ph/9909255 ] . h. davoudiasl and t. g. rizzo , jhep * 0811 * , 013 ( 2008 ) [ arxiv:0809.4440 [ hep - ph ] ] . a. v. kisselev , jhep * 0809 * , 039 ( 2008 ) [ arxiv:0804.3941 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 83 * , 011102 ( 2011 ) . [ arxiv:1012.2795 [ hep - ex ] ] . b. c. allanach , k. odagiri , m. a. parker and b. r. webber , jhep * 0009 * , 019 ( 2000 ) . [ arxiv : hep - ph/0006114 ] . b. c. allanach , k. odagiri , m. j. palmer , m. a. parker , a. sabetfakhri and b. r. webber , jhep * 0212 * , 039 ( 2002 ) [ arxiv : hep - ph/0211205 ] . r. cousins , j. mumford , j. tucker and v. valuev , jhep * 0511 * , 046 ( 2005 ) . a. abulencia _ et al . _ [ cdf collaboration ] , phys . * 95 * , 252001 ( 2005 ) [ arxiv : hep - ex/0507104 ] . p. osland , a. a. pankov , n. paver and a. v. tsytrinov , phys . d * 78 * , 035008 ( 2008 ) [ arxiv:0805.2734 [ hep - ph ] ] . e. w. dvergsnes , p. osland , a. a. pankov and n. paver , int . j. mod . a * 20 * , 2232 ( 2005 ) [ arxiv : hep - ph/0410402 ] . e. w. dvergsnes , p. osland , a. a. pankov and n. paver , phys . d * 69 * , 115001 ( 2004 ) [ arxiv : hep - ph/0401199 ] . r. diener , s. godfrey and t. a. w. martin , phys . d * 80 * , 075014 ( 2009 ) [ arxiv:0909.2022 [ hep - ph ] ] . u. de sanctis , m. fabbrichesi and a. tonero , arxiv:1103.1973 [ hep - ph ] . r. kelley , l. randall and b. shuve , jhep * 1102 * , 014 ( 2011 ) [ arxiv:1011.0728 [ hep - ph ] ] . c. w. chiang , n. d. christensen , g. j. ding and t. han , arxiv:1107.5830 [ hep - ph ] . t. han , j. d. lykken and r. j. zhang , phys . d * 59 * , 105006 ( 1999 ) [ arxiv : hep - ph/9811350 ] . l. d. landau , dokl . sssr * 60 * , 207 ( 1948 ) . r. barbieri and r. torre , phys . b * 695 * , 259 ( 2011 ) [ arxiv:1008.5302 [ hep - ph ] ] . d. feldman , z. liu and p. nath , jhep * 0611 * , 007 ( 2006 ) [ arxiv : hep - ph/0606294 ] . j. pumplin , d. r. stump , j. huston , h. l. lai , p. m. nadolsky and w. k. tung , jhep * 0207 * , 012 ( 2002 ) [ arxiv : hep - ph/0201195 ] . s. chatrchyan _ et al . _ [ cms collaboration ] , jhep * 1105 * , 093 ( 2011 ) [ arxiv:1103.0981 [ hep - ex ] ] . t. berger - hrynova [ atlas collaboration ] , presentation at the _ international europhysics conference on high energy physics ( eps - hep 2011 ) _ , 21 - 27 july 2011 , grenoble , france ; + j. abdallah [ atlas collaboration ] , presentation at the _ 19th international conference on particles and nuclei ( panic 11 ) _ , 24 - 29 july 2011 , cambridge , ma , usa ; + see also : atlas collaboration , cern - ph-2011 - 123 , arxiv:1108.1582 [ hep - ex ] . j. tucker [ cms collaboration ] , presentation at the _ international europhysics conference on high energy physics ( eps - hep 2011 ) _ , 21 - 27 july 2011 , grenoble , france ; + see also : cms collaboration , cms physics analysis summary , cms pas exo-11 - 09 . o. j. p. eboli , t. han , m. b. magro and p. g. mercadante , phys . rev . d * 61 * , 094007 ( 2000 ) [ arxiv : hep - ph/9908358 ] . g. f. giudice , r. rattazzi and j. d. wells , nucl . b * 544 * , 3 ( 1999 ) [ arxiv : hep - ph/9811291 ] . k. m. cheung and g. l. landsberg , phys . d * 62 * , 076003 ( 2000 ) [ arxiv : hep - ph/9909218 ] ; + k. m. cheung , phys . d * 61 * , 015005 ( 2000 ) [ arxiv : hep - ph/9904266 ] . k. sridhar , jhep * 0105 * , 066 ( 2001 ) [ arxiv : hep - ph/0103055 ] . g. f. giudice , t. plehn and a. strumia , nucl . b * 706 * , 455 ( 2005 ) [ arxiv : hep - ph/0408320 ] . s. frixione , phys . b * 429 * , 369 ( 1998 ) [ arxiv : hep - ph/9801442 ] . p. mathews , v. ravindran , k. sridhar , w. l. van neerven , nucl . * b713 * , 333 - 377 ( 2005 ) [ hep - ph/0411018 ] . m. c. kumar , p. mathews , v. ravindran and a. tripathi , nucl . b * 818 * , 28 ( 2009 ) [ arxiv:0902.4894 [ hep - ph ] ] . m. c. kumar , p. mathews , v. ravindran and a. tripathi , phys . b * 672 * , 45 ( 2009 ) [ arxiv:0811.1670 [ hep - ph ] ] . b. w. harris and j. f. owens , phys . d * 65 * , 094032 ( 2002 ) [ arxiv : hep - ph/0102128 ] . j. m. campbell , r. k. ellis and c. williams , jhep * 1107 * , 018 ( 2011 ) [ arxiv:1105.0020 [ hep - ph ] ] . g. passarino and m. j. g. veltman , nucl . b * 160 * ( 1979 ) 151 . d. graudenz , m. spira and p. m. zerwas , phys . * 70 * ( 1993 ) 1372 . a. djouadi , m. spira and p. m. zerwas , phys . b * 264 * ( 1991 ) 440 . s. dawson , nucl . b * 359 * ( 1991 ) 283 .

sparse matrix - vector multiplication ( spmv ) is a central building block for scientific software and graph applications . recently , heterogeneous processors composed of different types of cores attracted much attention because of their flexible core configuration and high energy efficiency . in this paper , we propose a compressed sparse row ( csr ) format based spmv algorithm utilizing both types of cores in a cpu - gpu heterogeneous processor . we first speculatively execute segmented sum operations on the gpu part of a heterogeneous processor and generate a possibly incorrect results . then the cpu part of the same chip is triggered to re - arrange the predicted partial sums for a correct resulting vector . on three heterogeneous processors from intel , amd and nvidia , using 20 sparse matrices as a benchmark suite , the experimental results show that our method obtains significant performance improvement over the best existing csr - based spmv algorithms . sparse matrices , sparse matrix - vector multiplication , compressed sparse row , speculative execution , segmented sum , heterogeneous processor

1504.06474

sparse matrix - vector multiplication ( spmv ) is perhaps the most widely - used non - trivial sparse basic linear algebra subprogram ( blas ) in computational science and modeling . the operation multiplies a sparse matrix @xmath0 of size @xmath1 by a dense vector @xmath2 of size @xmath3 and gives a dense vector @xmath4 of size @xmath5 . despite its simplicity at the semantic level , an efficient spmv implementation is generally hard , because @xmath0 s sparsity structure can be very irregular and unpredictable . compared to cpus , co - processors ( e.g. , gpus and xeon phi ) promise much higher peak floating - point performance and memory bandwidth . thus a lot of research has focused on accelerating spmv on co - processors . one straightforward way on utilizing co - processors is to develop all - new sparse matrix formats ( e.g. , hyb @xcite , cocktail @xcite , jad @xcite , esb @xcite , bccoo @xcite and brc @xcite ) for specific hardware architectures . the experimental results showed that these formats can provide performance improvement for various spmv benchmarks . however , the completely new formats bring several new problems . the first one is backward - compatibility . when the input data are stored in basic formats such as compressed sparse row ( csr ) , a format conversion is required for using the new format based spmv . in practice , fusing a completely new format into well - established toolkits ( e.g. , petsc @xcite ) for scientific software is not a trivial task @xcite because of the format conversion . moreover , kumbhar @xcite pointed out that once an application ( in particular a non - linear solver ) needs repeated format conversion after a fixed small number of iterations , the new formats may degrade overall performance . furthermore , langr and tvrdk @xcite demonstrated that isolated spmv performance is insufficient to evaluate a new format . thus more evaluation criteria , such as format conversion cost and memory footprint , must be taken into consideration . secondly , when the spmv operation is used with other sparse building blocks ( e.g. , sparse matrix - matrix multiplication @xcite ) that require basic storage formats , using the all - new formats is less feasible . to leverage the spmv performance and the practicality , liu and vinter proposed the csr5 format @xcite to extend the basic csr format . the experimental results showed that the csr5 format delivers excellent spmv performance , but merely needs very short format conversion time ( a few spmv operations ) and very small extra memory footprint ( around 2% of the csr data ) . because the csr5 format shares data with the csr format , the csr - based sparse blas routines can efficiently work with the csr5 format . however , when a solver only needs a few iterations , the csr5 may not deliver speedups , compared to using the basic csr data . thereofore , improving performance of spmv using the most widely supported csr format has also gained plenty of attention @xcite . most of the related work @xcite has focused on improving row block method for the csr - based spmv . however , these newly proposed approaches are not highly efficient . the main reason is that co - processors are designed for load balanced high throughput computation , which is not naturally offered by the row pointer information of the csr format . on the other hand , using segmented sum method for the csr - based spmv has been proposed by @xcite and been implemented in libraries cudpp @xcite and modern gpu @xcite for nvidia gpus . unlike the row block methods , the segmented sum algorithms evenly partition an input matrix @xmath0 for nearly perfect load balancing , and thus may be suitable for a co - processor implementation . but unfortunately , this method can not recognize empty rows and requires more costly global operations . these extra overheads may offset performance gain of load balanced segmented sum and degrade overall spmv efficiency . recently , heterogeneous processors ( which are also known as heterogeneous chip multiprocessors ) have been designed @xcite and implemented @xcite . compared to homogeneous processors such as cpus or gpus , heterogeneous processors can deliver improved overall performance and power efficiency @xcite , while sufficient heterogeneous parallelisms exist . the main characteristics of heterogeneous processors include unified shared memory and fast communication among different types of cores ( e.g. , cpu cores and gpu cores ) . practically , heterogeneous system architecture ( hsa ) @xcite , opencl @xcite and cuda @xcite have supplied toolkits for programming heterogeneous processors . our work described in this paper particularly focuses on accelerating csr - based spmv on cpu - gpu heterogeneous processors . the main idea of our spmv algorithm is first speculatively executing spmv on a heterogeneous processor s gpu cores targeting high throughput computation , and then locally re - arranging resulting vectors by the cpu cores of the same chip for low - latency memory access . to achieve load balanced first step computation and to utilize both cpu and gpu cores , we improved the conventional segmented sum method by generating auxiliary information ( e.g. , segment descriptor ) at runtime and recognizing empty rows on - the - fly . compared to the row block methods for the csr - based spmv , our method delivers load balanced computation to achieve higher throughput . compared with the classic segmented sum method for the csr - based spmv , our approach decreases the overhead of global synchronization and removes pre- and post - processing regarding empty rows . this paper makes the following contributions : * we propose a fast csr - based spmv algorithm that efficiently utilizes different types of cores in emerging cpu - gpu heterogeneous processors . * we develop an speculative segmented sum algorithm by generating auxiliary information on - the - fly and eliminating costly pre- and post - processing on empty rows . * we evaluate our csr - based spmv algorithm on a widely - adopted benchmark suite and achieve stable spmv performance independent of the sparsity structure of input matrix . on a benchmark suite composed of 20 matrices with diverse sparsity structures , our approach greatly outperforms the row block methods for the csr - based spmv running on gpu cores of heterogeneous processors . on an intel heterogeneous processor , the experimental results show that our method obtains up to 6.90x and on average 2.57x speedup over an opencl implementation of the csr - vector algorithm in cusp running on its gpu cores . on an amd heterogeneous processor , our approach delivers up to 16.07x ( 14.43x ) and on average 5.61x ( 4.47x ) speedup over the fastest single ( double ) precision csr - based spmv algorithm from paralution and an opencl version of cusp running on its gpu cores . on an nvidia heterogeneous processor , our approach delivers up to 5.91x ( 6.20x ) and on average 2.69x ( 2.53x ) speedup over the fastest single ( double ) precision csr - based spmv algorithm from cusparse and cusp running on its gpu cores . the paper is organized as follows . we first introduce background knowledge about the csr format , the csr - based spmv algorithms and heterogeneous processors in section 2 . then we describe our csr - based spmv algorithm in section 3 . moreover , we give and analyze our experimental results in section 4 . we review the related approaches in section 5 . the csr format of a sparse matrix consists of three separate arrays : ( 1 ) row pointer array of size @xmath6 , where @xmath5 is the number of rows of the matrix , ( 2 ) column index array of size @xmath7 , where @xmath7 is the number of nonzero entries of the matrix , and ( 3 ) value array of size @xmath7 . hence the overall space complexity of the csr format is @xmath8 . below we show a sparse matrix @xmath0 of size @xmath9 and its csr representation . @xmath10 @xmath11 \\ column\ ; index & = \begin{bmatrix}0 , 2 , 5,\;\ ; 0 , 1 , 2,\;\ ; 2 , 4,\;\ ; 4,\;\ ; 2 , 3 , 4\end{bmatrix } \\% \big [ [ \ ; 0\ ; 2\ ; 5\ ; ] [ \ ; 0\ ; 1\ ; 2\ ; 3\ ; 4\ ; 5\ ; ] [ \ ; 2\;\;\ ; 4\;\;\ , ] [ \ ; 4\;\;\ , ] [ \;\ , 2\;\;\ , 3\;\;\,\ , 4\;\,\ , ] \big ] \\ value & = \begin{bmatrix}a , b , c,\;\ ; d , e , f,\;\ ; g , h,\;\ ; i,\;\;\ ; j , k , l\end{bmatrix}. \nonumber% \big [ [ \ ; 1\ ; 2\ ; 3\ ; ] [ \ ; 4\ ; 5\ ; 6\ ; 7\ ; 8\ ; 9\ ; ] [ \ ; 10\;11\ ; ] [ \ ; 12\ ; ] [ \ ; 13\ ; 14\ ; 15\ ; ] \big ] \\ % segment\ ; descriptor & = [ 1 , 0 , 0,\;\ ; 1 , 0 , 0,\;\ ; 1 , 0,\;\ ; 1,\;\ ; 1 , 0 , 0 ] . \big [ [ \ ; 1\ ; 0\ ; 0\ ; ] [ \ ; 1\ ; 0\ ; 0\ ; 0\ ; 0\ ; 0\ ; ] [ \ ; 1\;0\ ; ] [ \ ; 1\ ; ] [ \ ; 1\ ; 0\ ; 0\ ; ] \big ] \nonumber % row\ ; pointer & = \big[\;\;\;0\;\;\;\ ; 3\;\;\;\ ; 9\;\;\;\ ; 11\;\;\;\ ; 11\;\;\;\ ; 12\;\;\;\ ; 15\;\;\;\ ; \big ] \\ % column\ ; index & = \big [ [ \ ; 0\ ; 2\ ; 5\ ; ] [ \ ; 0\ ; 1\ ; 2\ ; 3\ ; 4\ ; 5\ ; ] [ \ ; 2\;\;\ ; 4\;\;\ , ] [ \ ; 4\;\;\ , ] [ \;\ , 2\;\;\ , 3\;\;\,\ , 4\;\,\ , ] \big ] \\ % value & = \big [ [ \ ; 1\ ; 2\ ; 3\ ; ] [ \ ; 4\ ; 5\ ; 6\ ; 7\ ; 8\ ; 9\ ; ] [ \ ; 10\;11\ ; ] [ \ ; 12\ ; ] [ \ ; 13\ ; 14\ ; 15\ ; ] \big ] \\ % value & = \big [ [ \ ; 1\ ; 0\ ; 0\ ; ] [ \ ; 1\ ; 0\ ; 0\ ; 0\ ; 0\ ; 0\ ; ] [ \ ; 1\;0\ ; ] [ \ ; 1\ ; ] [ \ ; 1\ ; 0\ ; 0\ ; ] \big ] \nonumber \end{split}\ ] ] assume we have an input dense vector @xmath12 we can obtain a dense vector @xmath4 by multiplying the sparse matrix @xmath0 by the vector @xmath2 : @xmath13 the straightforward way to multiply @xmath0 with @xmath2 on a multicore processor is assigning a row block ( i.e. , multiple rows ) to each core . since the row pointer array records offset information of column index and value of nonzero entries , each core can easily position data in @xmath0 and @xmath2 . then generating corresponding entries of @xmath4 merely needs some simple multiply and add operations . for example , we assume that using a six - core processor for the above spmv operation and each core is responsible for one row . we can notice that the cores calculating the first , the second and the sixth rows are busier than the other cores . meanwhile , the core doing the fourth row is actually idle while the other cores are working . therefore , the row block method can not naturally handle load balance on multicore processors . on co - processors composed of a large amount of lightweight single instruction , multiple data ( simd ) units , the problem can heavily degrade performance of spmv operation . even though many strategies , such as vectorization @xcite , data streaming @xcite , memory coalescing @xcite , static or dynamic binning @xcite , dynamic parallelism @xcite and dynamic row distribution @xcite , have been proposed for the row block method , it is still impossible to achieve nearly perfect load balancing in general sense , simply since row sizes are irregular and unpredictable . the other method of computing the csr - based spmv is utilizing a segmented sum algorithm . this method first generates a segment descriptor of size @xmath7 . the descriptor marks the first nonzero entry of each non - empty row as 1 ( or equivalently ` true ` ) and the other nonzero entries as 0 ( or equivalently ` false ` ) . using the above 6-by-6 sparse matrix as an example , we have @xmath14 . \nonumber \end{split}\ ] ] then an element - wise product array of size @xmath7 is allocated and filled by calculating @xmath15 = x[column\ ; index[i]]\times value[i ] , i\in[0,nnz ) . \nonumber \end{split}\ ] ] the third step conducts a segmented sum operation on the product array by using segment information stored in the segment descriptor . finally , the sum in each segment is stored to a contiguous location in @xmath4 . we can see that the segmented sum method can achieve nearly perfect load balance in the nonzero entry space . however , this method has two obvious drawbacks : ( 1 ) since the segment descriptor is binary , this method is unable to recognize empty rows , thus a pre - processing ( squeezing out possible empty rows ) is required for calculating a `` clean '' row pointer array , and a post - processing ( adding zeros to proper locations ) is needed for a correct resulting vector @xmath4 , and ( 2 ) this method requires more expensive global synchronizations and global memory access than the row block method ( which needs only one single kernel launch ) . therefore , in practice , the segmented sum method is not necessarily faster than the row block methods . compared to homogeneous chip multiprocessors such as cpus and gpus , the heterogeneous processors are able to combine different types of cores into one chip . thus heterogeneous processors offer more flexibilities in architecture design space . because of mature cpu and gpu architectures and applications , cpu - gpu integrated heterogeneous processor with multiple instruction set architectures ( isas ) is the most widely adopted choice . representatives of this model include amd accelerated processing units ( apus ) @xcite , intel multi - cpu and gpu system - on - a - chip ( soc ) devices @xcite , nvidia echelon heterogeneous gpu architecture @xcite , and many mobile processors ( e.g. , nvidia tegra @xcite and qualcomm snapdragon @xcite ) . figure [ spmv.parco.fig.hcmps ] shows two block diagrams of heterogeneous processors used as experimental testbed in this paper . in general , a heterogeneous processor consists of four major parts : ( 1 ) a group of cpu cores with hardware - controlled caches , ( 2 ) a group of gpu cores with shared command processors , software - controlled scratchpad memory and hardware - controlled caches , ( 3 ) shared memory management unit , and ( 4 ) shared global dynamic random - access memory ( dram ) . the last level cache of the two types of cores can be separate as shown in figure [ spmv.parco.fig.hcmps](a ) or shared as shown in figure [ spmv.parco.fig.hcmps](b ) . the cpu cores have higher single - thread performance due to out - of - order execution , branch prediction and large amounts of caches . the gpu cores execute massively parallel lightweight threads on simd units for higher aggregate throughput . the two types of compute units have completely different isas and separate cache sub - systems . in this paper , our experiments run on three different platforms ( shown in table [ spmv.parco.tab.testbed ] ) , the platforms from amd and nvidia are based on the design of figure [ spmv.parco.fig.hcmps](a ) ; the intel platform uses the design of figure [ spmv.parco.fig.hcmps](b ) . note that in the current amd apu architecture , although the two types of cores have separate last level caches , the gpu cores are able to snoop the last level cache on the cpu side . compared to loosely - coupled cpu - gpu heterogeneous systems , the two types of cores in a heterogeneous processor share one single unified address space instead of using separate address spaces ( i.e. , system memory space and gpu device memory space ) . one obvious benefit is avoiding data transfer through connection interfaces ( e.g. , pcie link ) , which is one of the most well known bottlenecks of co - processor computing @xcite . additionally , gpu cores can access more memory by paging memory to and from disk . further , the consistent pageable shared virtual memory can be fully or partially coherent , meaning that much more efficient cpu - gpu interactions are possible due to eliminated heavyweight synchronization ( i.e. , flushing and gpu cache invalidation ) . currently , programming on the unified address space and low - overhead kernel launch are supported by hsa @xcite , opencl @xcite and cuda @xcite . we first evenly decompose nonzero entries of the input matrix to multiple small tiles for load balanced data parallelism . here we define a tile as a 2d array of size @xmath16 . the width @xmath17 is the size of a thread - bunch , which is the minimum simd execution unit in a given vector processor . it is also known as wavefront in amd gpus or warp in nvidia gpus . the height @xmath18 is the workload ( i.e. , the number of nonzero entries to be processed ) of a thread . a tile is a basic work unit in matrix - based segmented sum method @xcite , which is used as a building block in our spmv algorithm . actually , the term `` tile '' is equivalent to the term `` matrix '' used in original description of the segmented scan algorithms @xcite . here we use `` tile '' to avoid confusion between a work unit of matrix shape and a sparse matrix in spmv . since a thread - bunch can be relatively too small ( e.g. , as low as 8 in current intel gpus ) to amortize scheduling cost , we combine multiple thread - bunches into one thread - group ( i.e. , work - group in opencl terminology or thread block in cuda terminology ) for possibly higher throughput . we define @xmath19 to denote the number of thread - bunches in one thread - group . additionally , we let each thread - bunch compute @xmath20 contiguous tiles . thus higher on - chip resource reuse and faster global synchronization are expected . therefore , we can calculate that each thread - group deals with @xmath21 nonzero entries . thus the whole nonzero entry space of size @xmath7 can be evenly assigned to @xmath22 thread - groups . figure [ spmv.parco.fig.datadecomposation ] shows an example of the data decomposition . in this example , we set @xmath23 , @xmath24 , @xmath25 , and @xmath26 . thus each thread - group is responsible for 32 nonzero entries . then @xmath27 thread - groups are dispatched . our csr - based spmv is based on fundamental segmented sum algorithm , which guarantees load balanced computation in the nonzero entry space . while utilizing segmented sum as a building block in our spmv algorithm , we have three main performance considerations : ( 1 ) the segment descriptor needs to be generated in on - chip memory at runtime to reduce overhead of global memory access , ( 2 ) empty rows must be recognized and processed without calling specific pre- and post - processing functions , and ( 3 ) taking advantages of both types of cores in a heterogeneous processor . hence we improve the basic segmented sum method to meet the above performance requirements . the algorithm framework includes two main stages : ( 1 ) speculative execution stage , and ( 2 ) checking prediction stage . the first stage speculatively executes spmv operation and generates a possibly incorrect resulting vector @xmath4 . here the term `` incorrect '' means that the layout of entries in @xmath4 can be incorrect , but the entries are guaranteed to be numerically identified . then in the second stage we check whether or not the speculative execution is successful . if the prediction is wrong , a data re - arrangement will be launched for getting a completely correct @xmath4 . we first give an example of our algorithm and use it in the following algorithm description . figure [ spmv.parco.fig.example ] plots this example . the input sparse matrix includes 12 rows ( 2 of them are empty ) and 48 nonzero entries . we set @xmath19 to 1 , @xmath20 to 2 , @xmath17 to 4 and @xmath18 to 6 . this setting means that one thread - group is composed of one thread - bunch of size 4 ; each thread - bunch runs 2 iteration steps . before gpu kernel launch , three containers , _ synchronizer _ , _ dirty_counter _ and _ speculator _ , are pre - allocated in dram for global synchronization and speculative execution . algorithm [ spmv.parco.alg.spmv ] in appendix a lists pseudo code of the first stage . the * speculative execution stage * includes the following steps : ( 1 ) positioning a range of row indices for nonzero entries in a given tile , ( 2 ) calculating segment descriptor based on the range , ( 3 ) conducting segmented sum on the tile , ( 4 ) saving partial sums to the computed index range in vector @xmath4 . this stage also has some input - triggered operations such as labeling a tile with empty rows . first , each thread - bunch executes binary search of @xmath28 tile boundaries on the csr row pointer array . then we obtain corresponding row indices and store them in a scratchpad array _ tile offset _ of size @xmath28 . the results of the binary search are starting and ending row indices of the nonzero entries in each tile . thus each tile knows the locations to store generated partial sums . lines 37 of algorithm [ spmv.parco.alg.spmv ] give a code expression of this step . in our example shown in figure [ spmv.parco.fig.example ] , the 2 tiles of size 24 have 3 boundaries \{0 , 24 , 48}. the results of binary search of \{0 , 24 , 48 } on the csr row pointer array are \{0 , 4 , 12}. note that the binary search needs to return the rightmost match , if multiple slots have the same value . then each thread - bunch executes an iteration of @xmath20 steps . lines 859 of algorithm [ spmv.parco.alg.spmv ] give code expression of this step . each iteration deals with one tile . by calculating offset between the left boundary of a tile and the covered row indices , a local segment descriptor is generated ( lines 1421 in algorithm [ spmv.parco.alg.spmv ] ) . for example , the left boundary of the second tile is 24 and its row index range is 412 . we need to compute offset between 24 and the row pointer \{19 , 27 , 29 , 29 , 34 , 37 , 37 , 44 , 48}. then we obtain a group of offsets \{-5 , 3 , 5 , 5 , 10 , 13 , 13 , 20 , 24}. after removing duplicate values and overflowed values on the left and the right sides , the effective part \{3 , 5 , 10 , 13 , 20 } in fact implies local segment descriptor for the current tile . we can easily convert it to a binary expression \{0 , 0 , 0 , 1 , 0 , 1 , 0 , ... , 0 , 0 , 1 , 0 , 0 , 0 } through a scatter operation in on - chip scratchpad memory . moreover , since each tile is an independent work unit , the first bit of its segment descriptor should be ` true ` . thus the final expression becomes \{1 , 0 , 0 , 1 , 0 , 1 , 0 , ... , 0 , 0 , 1 , 0 , 0 , 0}. in figure [ spmv.parco.fig.example ] , the filled and empty circles are heads ( i.e. , 1s or ` true`s ) and body ( i.e. , 0s or ` false`s ) of segments , respectively . while generating the segment descriptor , each thread detects whether or not its right neighbor wants to write to the same slot . if yes ( like the duplicate offset information \ { ... , 5 , 5 , ... } and \ { ... , 13 , 13 , ... } in the above example ) , we can make sure that this tile contains at least one empty row , since an empty row is expressed as two contiguous indices of the same value in the csr row pointer array . then we mark this tile as `` dirty '' ( line 19 in algorithm [ spmv.parco.alg.spmv ] ) . further , the _ dirty counter _ array stored in dram is incremented by atomic operation , and this tile s offset is recorded in the _ speculator _ array ( lines 5358 in algorithm [ spmv.parco.alg.spmv ] ) . in our example , _ dirty counter _ is 1 and _ speculator _ array has a pair of offsets \{@xmath294 , 12@xmath30 } ( ( shown with angular brackets in figure [ spmv.parco.fig.example ] ) . then we calculate and save element - wise products in scratchpad memory , based on its nonzero entries column indices , values and corresponding values in the vector @xmath2 . lines 2226 of algorithm [ spmv.parco.alg.spmv ] show code expression of this step . when finished , we transmit the sum of the last segment to an intermediate space for the next iteration ( lines 2731 in algorithm [ spmv.parco.alg.spmv ] ) . in our example , the first tile s last value @xmath31 is transmitted to the next tile . then we execute the matrix - based segmented sum ( lines 3233 ) on the tile . because the segmented sum algorithm used here is very similar to the method described in @xcite , we refer the reader to @xcite and several pervious gpu segmented sum algorithms @xcite for details . but note that compared to @xcite , our method makes one difference : we store partial sums in a compact pattern ( i.e. , values are arranged in order from the first location in the thread work space ) , but not save them to locations of corresponding segment heads . for this reason , we need to record the starting position and the number of partial sums . then we can use an ordinary exclusive scan operation ( lines 3435 ) for obtaining contiguous indices of the partials sums in @xmath4 . in figure [ spmv.parco.fig.example ] , we can see that the partial sums ( expressed as filled hexagons ) are aggregated in the compact fashion . note that empty hexagons are intermediate partial sums , which are already added to the correct position of segment heads . finally , we store the partial sums to known locations in the resulting vector . lines 3652 of algorithm [ spmv.parco.alg.spmv ] show code expression . as an exception , the sum result of the first segment in a thread - bunch is stored to the _ synchronizer _ array ( lines 4043 ) , since the first row of each thread - bunch may cross multiple thread - bunch . this is a well known issue while conducting basic primitives , such as reduction and prefix - sum scan , using more than one thread - group that can not communicate with each other . in fact , atomic add operation can be utilized to avoid the global synchronization . but we choose not to use relatively slow global atomic operations and let a cpu core to later on finish the global synchronization . lines 6268 of algorithm [ spmv.parco.alg.spmv ] show the corresponding code expression . since the problem size ( i.e. , @xmath32 ) can be too small to saturate a gpu core , a cpu core is in fact faster for accessing short arrays linearly stored in dram . taking the first tile in figure [ spmv.parco.fig.example ] as an example , its first partial sum is @xmath33 , which is stored with its global index 0 to the _ synchronizer_. after that , the value @xmath33 is added to position 0 of @xmath4 . when the above steps are complete , the resulting vector is numerically identified , except that some values generated by dirty tiles are not in their correct locations . in figure [ spmv.parco.fig.example ] , we can see that after synchronization , vector @xmath4 is already numerically identified to its final form , but entries @xmath34 , @xmath35 , @xmath36 and @xmath37 generated by the second tile are located in wrong slots . the * checking prediction stage * first checks value of the _ dirty counter _ array . if it is zero , the previous prediction is correct and the result of the first stage is the final result ; if it is not zero , the predicted entries generated by dirty tiles are scattered to their correct positions in the resulting vector . in this procedure , the csr row pointer array is required to be read for getting correct row distribution information . again , we use a cpu core for the irregular linear memory access , which is more suitable for cache sub - systems in cpus . in our example , entries @xmath34 , @xmath35 , @xmath36 and @xmath37 are moved to their correct positions . then the spmv operation is done . @xmath38.start$ ] @xmath39.stop$ ] @xmath40.index ] \gets y[synchronizer[i].index ] + synchronizer[i].value$ ] our csr - based spmv algorithm pre - allocates three auxiliary arrays , _ synchronizer _ , _ dirty counter _ and _ speculator _ , in dram . the space complexity of _ synchronizer _ is @xmath32 , equivalent to the number of thread - bunches . the size of _ dirty counter _ is constant 1 . the _ speculator _ array needs a size of @xmath41 , equivalent to the number of tiles . since @xmath18 and @xmath17 are typically set to relatively large values , the auxiliary arrays merely slightly increase overall space requirement . for each thread - bunch , we executes @xmath28 binary searches in the row pointer array of size @xmath6 . thus @xmath42 is work complexity of this part . on the whole , generating segment descriptor needs @xmath43 time . collecting element - wise products needs @xmath44 time . for each tile , segmented sum needs @xmath45 time . thus all segmented sum operations need @xmath46 time . saving entries to @xmath4 needs @xmath43 time . synchronization takes @xmath47 time . possible re - arrangement needs @xmath43 time in the worst case . thus overall work complexity of our csr - based spmv algorithm is @xmath48 . based on the above analysis , we can see that when the input matrix is fixed , the cost of our spmv algorithm only depends on two parameters : @xmath17 and @xmath18 . in our algorithm implementation , @xmath17 is set to simd length of the used processor . choosing @xmath18 needs to consider the capacity of on - chip scratchpad memory . the other two parameters @xmath19 and @xmath20 are empirically chosen . table [ spmv.parco.tab.parameters ] shows the selected parameters . note that double precision is not currently supported in intel opencl implementation for its gpus . > m1.5 cm > m1.25 cm > m1.25 cm > m1.25 cm > m1.25 cm > m1.25 cm processor & intel & & precision & 32-bit single & 32-bit single & 64-bit double & 32-bit single & 64-bit double @xmath17 & 8 & 64 & 64 & 32 & 32 @xmath18 & 16 & 16 & 8 & 8 & 4 @xmath19 & 4 & 2 & 2 & 5 & 5 @xmath20 & 6 & 2 & 5 & 7 & 7 we implement the first stage of our algorithm in opencl for the intel and amd platforms ( and cuda for the nvidia platform ) for gpu execution and the second stage in standard c language running on the cpu part . since our algorithm needs cpu and gpu share some arrays , we allocate all arrays in shared virtual memory supported by opencl for the best performance . on the nvidia platform , we use unified memory in cuda sdk . we use three heterogeneous processors , intel core i3 - 5010u , amd a10 - 7850k apu and nvidia tegra k1 , for evaluating spmv algorithms . table [ spmv.parco.tab.testbed ] shows specifications of the three processors . all of them are composed of multiple cpu cores and gpu cores . the two types of cores in the intel heterogeneous processor share a 3 mb last level cache . in contrast , gpu cores in the amd heterogeneous processor can snoop the l2 cache of size 4 mb on the cpu side . unlike those , the cache systems of the cpu part and the gpu part in the nvidia tegra processor are completely separate . note that currently the intel gpu can run opencl program only on microsoft windows operating system . also note that we use kb , mb and gb to denote @xmath49 , @xmath50 and @xmath51 bytes , respectively ; and use gflop to denote @xmath52 flops . m1.3 cm > m1.3 cm > m1.3 cm > m1.3 cm > m1.3 cm > m1.3 cm processor & & & core type & x86 cpu & gpu & x86 cpu & gpu & arm cpu & gpucodename & broadwell & hd 5500 & steamroller & gcn & cortex a15 & keplercores @ clock ( ghz ) & 2 @ 2.1 & 3 @ 0.9 & 4 @ 3.7 & 8 @ 0.72 & 4 @ 2.3 & 1 @ 0.85 sp flops / cycle & 2@xmath5332 & 3@xmath53128 & 4@xmath538 & 8@xmath53128 & 4@xmath538 & 1@xmath53384sp peak ( gflop / s ) & 134.4 & 345.6 & 118.4 & 737.3 & 73.6 & 327.2dp flops / cycle & 2@xmath5316 & 3@xmath5332 & 4@xmath534 & 8@xmath538 & 2@xmath532 & 1@xmath5316dp peak ( gflop / s ) & 67.2 & 86.4 & 59.2 & 46.1 & 18.4 & 13.6l1 data cache & 4@xmath5332 kb & 3@xmath534 kb & 4@xmath5316 kb & 8@xmath5316 kb & 4@xmath5332 kb & 1@xmath5316 kbl2 cache & 4@xmath53256 kb & 3@xmath5324 kb & 2@xmath532 mb & unreleased & 2 mb & 128 kbl3 cache & n / a & 384 kb & n / a & n / a & n / a & n / ascratchpad & n / a & 3@xmath5364 kb & n / a & 8@xmath5364 kb & n / a & 1@xmath5348 kb shared last level cache & & & dram & & & dram capacity & & & dram bandwidth & & & operating system & & & gpu driver version & & & compiler & & & toolkit version & & & to evaluate our method , we choose 20 unstructured matrices from the university of florida sparse matrix collection @xcite . table [ spmv.parco.tab.benchmark ] lists main information of the evaluated sparse matrices . the first 14 matrices of the benchmark suite have been widely used in previous work @xcite . the last 6 matrices are chosen as representatives of irregular matrices extracted from graph applications , such as circuit simulation and optimization problems . the first 10 matrices are relatively regular , due to short distance between the average value and the maximum value of @xmath7/row . the other matrices are relatively irregular . in this context , ` regular ' is used for a sparse matrix including rows of roughly the same size . in contrast , an ` irregular matrix ' can have some very long rows and many very short rows . for example , matrices generated from power - law graphs can have a few rows with @xmath54 nonzero entries and many rows with @xmath55 nonzero entries . .overview of evaluated sparse matrices [ cols="^,^,^,^,^,^ " , ] to analyze efficiency of the proposed spmv algorithm , we also benchmark parallel csr - based spmv using some other libraries or methods on cpus and gpus . on cpus , we execute three csr - based spmv approaches : ( 1 ) openmp - accelerated basic row block method , ( 2 ) poski library @xcite using oski @xcite as a building block , and ( 3 ) intel mkl v11.2 update 2 in intel parallel studio xe 2015 update 2 . the three approaches are running on all cpu cores of the used heterogeneous processors . for the intel cpu , we report results from mkl , since it always delivers the best performance and the poski is not supported by the used microsoft windows operating system . for the amd cpu , we report the best results of the three libraries , since none of the three libraries outperforms all the others . for the arm cpu included in the nvidia tegra k1 platform , we only report results from openmp , since the current poski and intel mkl implementations do not support the arm architecture . moreover , single - threaded nave implementation on cpu is included in our benchmark as well . on gpus , we benchmark variants of the csr - scalar and the csr - vector algorithms proposed in @xcite . the opencl version of the csr - scalar method is extracted from parallution v1.0.0 @xcite and evaluated on the amd platform . the opencl implementation of the csr - vector method is extracted from semantically equivalent cuda code in the cusp library v0.4.0 and executed on both the intel and the amd platforms . on the nvidia platform , we run the csr - based spmv from vendor - supplied cusparse v6.0 and cusp v0.4.0 libraries . for all tests , we run spmv 200 times and record averages . the implicit data transfer ( i.e. , matrices and vectors data copy from their sources to opencl shared virtual memory or cuda unified memory ) is not included in our evaluation , since spmv operation is normally one building block of more complex applications . all participating methods conduct general spmv , meaning that symmetry is not considered although some input matrices are symmetric . the throughput ( flops per second ) is calculated by @xmath56 the bandwidth ( bytes per second ) is calculated by @xmath57 -6pt -6pt -6pt -6pt -6pt -6pt -6pt -6pt figures [ spmv.parco.fig.spthroughput ] and [ spmv.parco.fig.dpthroughput ] show throughput of single precision and double precision spmv of the tested csr - based approaches , respectively . in figure [ spmv.parco.fig.spthroughput ] , we can see that on the intel heterogeneous processor , our approach obtains up to 6.90x and on average 2.57x speedup over the csr - vector method running on the used gpu . although the speedup mainly comes from irregular matrices , our method generally does not obviously lose performance on regular matrices . further , compared to cpu cores running mkl , both gpu spmv algorithms are slower . for our algorithm , the main reason is that the integrated gpu implements scratchpad memory in its l3 cache , which has one order of magnitude higher latency compared to fast scratchpad in nvidia or amd gpus . our algorithm in fact heavily uses scratchpad memory for storing and reusing segment descriptor , element - wise products and other shared data by threads . thus even though the gpu part of the intel heterogeneous processor has higher single precision theoretical peak performance than its cpu part , the delivered spmv throughput is lower than expected . for the csr - vector method , the low performance has another reason : small thread - bunch of size 8 dramatically increases loop overhead @xcite , which is one of the well known bottlenecks @xcite of gpu programming . in figures [ spmv.parco.fig.spthroughput ] and [ spmv.parco.fig.dpthroughput ] , we can see that on the amd heterogeneous processor , our method delivers up to 71.90x ( 94.05x ) and on average 22.17x ( 22.88x ) speedup over the single ( double ) precision csr - scalar method running on the used gpu . compared to the gpu csr - vector method , our algorithm achieves up to 16.07x ( 14.43x ) and on average 5.61x ( 4.47x ) speedup . the csr - scalar and the csr - vector methods give very low throughput while running the last 6 irregular matrices , because of the problem of load imbalance . further , we find that the intel heterogeneous processor s gpu is actually faster than the amd gpu while running the last 6 matrices . the reason is that the shorter thread - bunch ( 8 in intel gpu vs. 64 in amd gpu ) brings a positive influence for saving simd idle cost by executing a much shorter vector width for dramatically imbalanced row distribution . on the other hand , for several very regular matrices with short rows , e.g. , _ epidemiology _ , the csr - scalar method offers the best performance because of almost perfect load balance and execution of short rows without loop cost . for most regular matrices , our method delivers comparable performance over the best cpu algorithm . in figures [ spmv.parco.fig.spthroughput ] and [ spmv.parco.fig.dpthroughput ] , we can see that on the nvidia platform , our method delivers up to 5.91x ( 6.20x ) and on average 2.69x ( 2.53x ) speedup over the single ( double ) precision spmv in the cusp library running on the used gpu . compared to cusparse , our method has higher speedups . since the both libraries use csr - vector algorithm , those speedups are within expectations . consider the tegra k1 platform only contains one single gpu core , the problem of load imbalance on this device is not as heavy as on the above amd platform . as a result , the speedups are not as high as those from the amd processor . here our method delivers on average 1.41x ( 1.42x ) speedup over openmp - accelerated spmv on the quad - core arm cpu , while using single ( double ) precision benchmark . figure [ spmv.parco.fig.bandwidth ] shows bandwidth utilization of our algorithm proposed in this paper . we can see that the regular matrices can use bandwidth more efficiently compared to the irregular ones . considering the throughput speedups listed above , our method can obtain higher bandwidth utilization than the other csr - based spmv algorithms running on gpus . we further conduct experiments to exploit how selected parameters influence overall performance . figure [ spmv.parco.fig.parameterw ] shows dependency of the overall performance ( harmonic means of the 20 benchmarks ) on the parameters , while we fix all the parameters except for parameter @xmath18 ( i.e. , workload per thread ) . we can see that in general the overall performance goes up as parameter @xmath18 increases . this trend matches the algorithm complexity analysis described in section 3.3 . however , when @xmath18 is larger than a certain value , the overall performance degrades . the reason is that device occupancy may decrease while more on - chip scratchpad memory is allocated for @xmath58 work space of each thread - bunch . figure [ spmv.parco.fig.parameters ] shows the trend of the overall performance while we change parameter @xmath20 ( i.e. , the number of iterations of each thread - bunch ) and fix all the other parameters . we can see that if we assign more work to each thread - bunch , a better performance can be expected . the performance improvement mainly comes from higher on - chip resource reuse . in recent years , some new formats have been designed for spmv operation on various processor architectures . because of less off - chip memory access and better on - chip memory localization , block - based formats or libraries , such as oski @xcite , poski @xcite , csb @xcite , bellpack @xcite , bccoo / bccoo+ @xcite , brc @xcite and rsb @xcite , attracted the most attention . however , block - based formats heavily rely on sparsity structure , meaning that the input matrix is required to have a block structure to meet potential block layout . therefore , block - based formats are mainly suitable for some matrices generated from scientific computation problems , but may not fit irregular matrices generated from graph applications . our method proposed in this paper is insensitive to the sparsity structure of input matrix , thus a generally better performance is achieved . a lot of research has focused on improving row block method csr - based spmv . williams et al . @xcite proposed multiple optimization techniques for spmv on multi - core cpus and cell b.e . nishtala et al . @xcite designed a high - level data partitioning method for spmv to achieve better cache locality on multicore cpus . pichel et al . @xcite evaluated how reordering techniques influence performance of spmv on gpus . baskaran and bordawekar @xcite improved off - chip and on - chip memory access patterns of spmv on gpus . reguly and giles @xcite improved thread cooperation for better gpu cache utilization . ashari et al . @xcite utilized static reordering and the dynamic parallelism scheme offered by nvidia gpus for fast spmv operation . greathouse et al . @xcite grouped contiguous rows for better runtime load balancing on gpus . lightspmv [ liu : lightspmv ] proposed to dynamically distribute matrix rows over warps in order for more balanced csr - based spmv without the requirement of generating auxiliary data structures , and implemented this approach using atomic operations and warp shuffle functions as the fundamental building blocks . however , again , the row block methods can not achieve good performance for input matrix with dramatically imbalanced row distribution . in contrast , our method is independent with the sparsity structure of input matrix . using segmented sum as a building block is potentially a better generic method for the csr - based spmv . an early segmented sum method gpu spmv was introduced by sengupta et al . @xcite and garland @xcite and implemented in the cudpp library @xcite . but the cost of segmented sum and global memory access degrade overall spmv performance . zhang @xcite improved backward segmented scan for a better cache efficiency and implemented the csr - based spmv on multicore cpus . recently , nvidia s modern gpu library @xcite implemented an improved reduction method , which has been used as a back - end of cudpp . however , its performance still suffered by pre- and post - processing empty rows in global memory space . our method , in contrast , uses scratchpad memory more efficiently and utilizes the two types of cores in a heterogeneous processor for better workload distribution . compared with our recent work csr5 @xcite , a format designed for cross - platform spmv on cpus , gpus and xeon phi , the spmv approach presented in this paper does not need to process any format conversion or generate any auxiliary data for the input csr matrix . consider the format conversion from the csr to the csr5 merely needs the cost of a few spmv operations , the csr5-based spmv and the csr - based spmv can find their own application scenarios , such as solvers with different number of iterations . we proposed an efficient method for spmv on heterogeneous processors using the csr storage format . on three mainstream platforms from intel , amd and nvidia , our method greatly outperforms row block method csr - based spmv algorithms running on gpus . the performance gain mainly comes from our newly developed speculative segmented sum strategy that efficiently utilizes different types of cores in a heterogeneous processor . in this work , we assign different task to the cpu part and the gpu part in one heterogeneous processor . however , the heaviest workload ( stage 1 in our method ) currently only runs on gpu cores , while the cpu cores may be idle . obviously , it is possible to schedule tasks in the first stage on both cpu cores and gpu cores simultaneously for potentially higher throughput . however , a heterogeneity - aware scheduling strategy is beyond the scope of the spmv algorithm focused in this paper . we refer the reader to @xcite for recent progress on utilizing both cpu cores and gpu cores in a heterogeneous environment . the authors would like to thank james avery for his valuable feedback . the authors also thank the anonymous reviewers for their insightful suggestions and comments on this paper . 10 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 n. bell , m. garland , implementing sparse matrix - vector multiplication on troughput - oriented processors , in : proceedings of the conference on high performance computing networking , storage and analysis , sc 09 , 2009 , pp . 18:118:11 . su , k. keutzer , clspmv : a cross - platform opencl spmv framework on gpus , in : proceedings of the 26th acm international conference on supercomputing , ics 12 , 2012 , pp . 353364 . r. li , y. saad , gpu - accelerated preconditioned iterative linear solvers , the journal of supercomputing 63 ( 2 ) ( 2013 ) 443466 . x. liu , m. smelyanskiy , e. chow , p. dubey , efficient sparse matrix - vector multiplication on x86-based many - core processors , in : proceedings of the 27th international acm conference on international conference on supercomputing , ics 13 , 2013 , pp . 273282 . s. yan , c. li , y. zhang , h. zhou , yaspmv : yet another spmv framework on gpus , in : proceedings of the 19th acm sigplan symposium on principles and practice of parallel programming , ppopp 14 , 2014 , pp . 107118 . a. ashari , n. sedaghati , j. eisenlohr , p. sadayappan , an efficient two - dimensional blocking strategy for sparse matrix - vector multiplication on gpus , in : proceedings of the 28th acm international conference on supercomputing , ics 14 , 2014 , pp . 273282 . s. balay , s. abhyankar , m. f. adams , j. brown , p. brune , k. buschelman , v. eijkhout , w. d. gropp , d. kaushik , m. g. knepley , l. c. mcinnes , k. rupp , b. f. smith , h. zhang , petsc users manual , tech . anl-95/11 - revision 3.5 , argonne national laboratory ( 2014 ) . v. minden , b. smith , m. knepley , preliminary implementation of petsc using gpus , in : d. a. yuen , l. wang , x. chi , l. johnsson , w. ge , y. shi ( eds . ) , gpu solutions to multi - scale problems in science and engineering , lecture notes in earth system sciences , springer berlin heidelberg , 2013 , pp . 131140 . p. kumbhar , performance of petsc gpu implementation with sparse matrix storage schemes , master s thesis , the university of edinburgh ( aug 2011 ) . d. langr , p. tvrdk , evaluation criteria for sparse matrix storage formats , ieee transactions on parallel and distributed systems [ in press ] . w. liu , b. vinter , an efficient gpu general sparse matrix - matrix multiplication for irregular data , in : proceedings of the 2014 ieee 28th international parallel and distributed processing symposium , ipdps 14 , 2014 , pp . 370381 . w. liu , b. vinter , csr5 : an efficient storage format for cross - platform sparse matrix - vector multiplication , in : proceedings of the 29th acm international conference on supercomputing , ics 15 , 2015 , pp . 339350 . s. williams , l. oliker , r. vuduc , j. shalf , k. yelick , j. demmel , optimization of sparse matrix - vector multiplication on emerging multicore platforms , parallel computing 35 ( 3 ) ( 2009 ) 178194 . j. l. greathouse , m. daga , efficient sparse matrix - vector multiplication on gpus using the csr storage format , in : proceedings of the international conference for high performance computing , networking , storage and analysis , sc 14 , 2014 , pp . 769780 . a. ashari , n. sedaghati , j. eisenlohr , s. parthasarathy , p. sadayappan , fast sparse matrix - vector multiplication on gpus for graph applications , in : proceedings of the international conference for high performance computing , networking , storage and analysis , sc 14 , 2014 , pp . 781792 . g. e. blelloch , m. a. heroux , m. zagha , segmented operations for sparse matrix computation on vector multiprocessors , tech . cmu - cs-93 - 173 , carnegie mellon university ( aug 1993 ) . m. harris , j. d. owens , s. sengupta , cudpp documentation , nvidia , 2nd edition ( aug 2014 ) . s. baxter , modern gpu , nvidia ( 2013 ) . y. liu , b. schmidt , lightspmv : faster csr - based sparse matrix - vector multiplication on cuda - enabled gpus , in : application - specific systems , architectures and processors ( asap ) , 2015 ieee 26th international conference on , 2015 . s. sengupta , m. harris , y. zhang , j. d. owens , scan primitives for gpu computing , in : proceedings of the 22nd acm siggraph / eurographics symposium on graphics hardware , gh 07 , 2007 , pp . 97106 . m. garland , sparse matrix computations on manycore gpu s , in : proceedings of the 45th annual design automation conference , dac 08 , 2008 , pp . r. kumar , d. tullsen , n. jouppi , p. ranganathan , heterogeneous chip multiprocessors , computer 38 ( 11 ) ( 2005 ) 3238 . s. keckler , w. dally , b. khailany , m. garland , d. glasco , gpus and the future of parallel computing , micro , ieee 31 ( 5 ) ( 2011 ) 717 . a. branover , d. foley , m. steinman , amd fusion apu : llano , ieee micro 32 ( 2 ) ( 2012 ) 2837 . amd , white paper : compute cores ( jan 2014 ) . s. damaraju , v. george , s. jahagirdar , t. khondker , r. milstrey , s. sarkar , s. siers , i. stolero , a. subbiah , a 22 nm ia multi - cpu and gpu system - on - chip , in : solid - state circuits conference digest of technical papers ( isscc ) , 2012 ieee international , 2012 , pp . nvidia , nvidia tegra k1 a new era in mobile computing , 1st edition ( jan 2014 ) . qualcomm , qualcomm snapdragon 800 product brief ( aug 2013 ) . e. chung , p. milder , j. hoe , k. mai , single - chip heterogeneous computing : does the future include custom logic , fpgas , and gpgpus ? , in : microarchitecture ( micro ) , 2010 43rd annual ieee / acm international symposium on , 2010 , pp . 225236 . hsa foundation , hsa programmer s reference manual : hsail virtual isa and programming model , compiler writer s guide , and object format ( brig ) , 0th edition ( may 2013 ) . a. munshi , the opencl specification , khronos opencl working group , 2nd edition ( mar 2014 ) . d. negrut , r. serban , a. li , a. seidl , unified memory in cuda 6 : a brief overview and related data access / transfer issues , tech . rep . tr-2014 - 09 , university of wisconsin madison ( jun 2014 ) . y. s. deng , b. d. wang , s. mu , taming irregular eda applications on gpus , in : proceedings of the 2009 international conference on computer - aided design , iccad 09 , 2009 , pp . 539546 . c. gregg , k. hazelwood , where is the data ? why you can not debate cpu vs. gpu performance without the answer , in : performance analysis of systems and software ( ispass ) , 2011 ieee international symposium on , 2011 , pp . 134144 . y. dotsenko , n. k. govindaraju , p .- sloan , c. boyd , j. manferdelli , fast scan algorithms on graphics processors , in : proceedings of the 22nd annual international conference on supercomputing , ics 08 , 2008 , pp . 205213 . t. a. davis , y. hu , the university of florida sparse matrix collection , acm trans . 38 ( 1 ) ( 2011 ) 1:11:25 . byun , r. lin , j. w. demmel , k. a. yelick , poski : parallel optimized sparse kernel interface library user s guide , university of california , berkeley , 1st edition ( apr 2012 ) . r. vuduc , j. w. demmel , k. a. yelick , oski : a library of automatically tuned sparse matrix kernels , journal of physics : conference series 16 ( 1 ) ( 2005 ) 521530 . d. lukarski , n. trost , paralution - user manual , tech . 1.0.0 , paralution labs ug ( haftungsbeschrnkt ) co. kg ( feb 2015 ) . m. m. baskaran , u. bondhugula , s. krishnamoorthy , j. ramanujam , a. rountev , p. sadayappan , a compiler framework for optimization of affine loop nests for gpgpus , in : proceedings of the 22nd annual international conference on supercomputing , ics 08 , 2008 , pp . 225234 . j. fang , a. varbanescu , h. sips , a comprehensive performance comparison of cuda and opencl , in : parallel processing ( icpp ) , 2011 international conference on , 2011 , pp . 216225 . r. vuduc , h .- j . moon , fast sparse matrix - vector multiplication by exploiting variable block structure , in : high performance computing and communications , vol . 3726 of lecture notes in computer science , springer berlin heidelberg , 2005 , pp . 807816 . r. w. vuduc , automatic performance tuning of sparse matrix kernels , ph.d . thesis , university of california , berkeley ( dec 2003 ) . a. bulu , j. t. fineman , m. frigo , j. r. gilbert , c. e. leiserson , parallel sparse matrix - vector and matrix - transpose - vector multiplication using compressed sparse blocks , in : proceedings of the twenty - first annual symposium on parallelism in algorithms and architectures , spaa 09 , 2009 , pp . 233244 . a. bulu , s. williams , l. oliker , j. demmel , reduced - bandwidth multithreaded algorithms for sparse matrix - vector multiplication , in : parallel distributed processing symposium ( ipdps ) , 2011 ieee international , 2011 , pp . 721733 . j. w. choi , a. singh , r. w. vuduc , model - driven autotuning of sparse matrix - vector multiply on gpus , in : proceedings of the 15th acm sigplan symposium on principles and practice of parallel programming , ppopp 10 , 2010 , pp . 115126 . m. martone , efficient multithreaded untransposed , transposed or symmetric sparse matrix - vector multiplication with the recursive sparse blocks format , parallel computing 40 ( 7 ) ( 2014 ) 251 270 . r. nishtala , r. vuduc , j. demmel , k. yelick , when cache blocking of sparse matrix vector multiply works and why , applicable algebra in engineering , communication and computing 18 ( 3 ) ( 2007 ) 297311 . j. c. pichel , f. f. rivera , m. fernndez , a. rodrguez , optimization of sparse matrix - vector multiplication using reordering techniques on gpus , microprocessors and microsystems 36 ( 2 ) ( 2012 ) 65 77 . m. m. baskaran , r. bordawekar , optimizing sparse matrix - vector multiplication on gpus , tech . rc24704 , ibm ( 2008 ) . i. reguly , m. giles , efficient sparse matrix - vector multiplication on cache - based gpus , in : innovative parallel computing ( inpar ) , 2012 , 2012 , pp . 112 . n. zhang , a novel parallel scan for multicore processors and its application in sparse matrix - vector multiplication , parallel and distributed systems , ieee transactions on 23 ( 3 ) ( 2012 ) 397404 . j. lee , m. samadi , y. park , s. mahlke , transparent cpu - gpu collaboration for data - parallel kernels on heterogeneous systems , in : proceedings of the 22nd international conference on parallel architectures and compilation techniques , pact 13 , 2013 , pp . 245256 . r. kaleem , r. barik , t. shpeisman , b. t. lewis , c. hu , k. pingali , adaptive heterogeneous scheduling for integrated gpus , in : proceedings of the 23rd international conference on parallel architectures and compilation , pact 14 , 2014 , pp . 151162 . j. shen , j. fang , a. l. varbanescu , h. sips , an application - centric evaluation of opencl on multi - core cpus , parallel computing 39 ( 12 ) ( 2013 ) 834850 . j. shen , a. l. varbanescu , p. zou , y. lu , h. sips , improving performance by matching imbalanced workloads with heterogeneous platforms , in : proceedings of the 28th acm international conference on supercomputing , ics 14 , 2014 , pp . 241250 . @xmath59 ; _ //positioning row indices of tiles in the thread - bunch _ ` boundary[i ] ` @xmath60 ` tile_offset[i]`@xmath61}}$ ] _ //iterative steps in a thread - bunch _ @xmath62}$ ] @xmath63}$ ] @xmath64 _ //calculating segment descriptor _ @xmath65 - boundary[$i$ ] ] } \gets \texttt{true}$ ] @xmath66 _ //collecting element - wise products _ @xmath67+$j$]]}$ ] @xmath68 } \gets x\_value\times \texttt{value[boundary[$i$]+$j$]}$ ] _ //transmitting a value from the previous tile _ @xmath69 } \gets \texttt{product[0 ] } + transmitter$ ] @xmath70 } \gets \texttt{true}$ ]

observations of narrow radio - emitting filaments near the galactic center have been interpreted in previous studies as evidence of a pervasive vertical ( i.e. to the galactic plane ) milligauss magnetic field in the central pc of the galaxy . a simple cylindrically symmetric model for the equilibrium in this central region is proposed in which horizontal ( i.e. to the galactic plane ) magnetic fields embedded in an annular band of partially ionized molecular material of radius pc are wrapped around vertical magnetic fields threading low - density hot plasma . the central vertical magnetic field , which has a pressure that significantly exceeds the thermal pressure of the medium , is confined by the weight of the molecular material . the stability of this equilibrium is studied indirectly by analyzing a uniformly rotating cylinder of infinite extent along the axis in cylindrical coordinates , with low - density plasma and an axial magnetic field at pc , high - density plasma and an azimuthal field at pc , and a gravitational acceleration directed in the direction . simple profiles are assumed for the density , pressure , and field strength , with the sound speed and alfvn speeds within the dense plasma . the density profile and gravity tend to destabilize the plasma , but the plasma tends to be stabilized by rotation and magnetic tension since the interface between the high and low - density plasmas can not be perturbed without bending either the horizontal or vertical field . normal modes proportional to with and are studied . such modes neither bend nor compress the axial field at pc but allow compressions of the dense plasma along the azimuthal magnetic field that enhance the destabilizing role of gravity . it is shown analytically that when is small and the dense plasma is supported against gravity primarily by rotation , the necessary and sufficient condition for stability to modes is , where is the effective gravity , is the uniform angular velocity , and is the sound speed in the dense plasma . since the effective gravity is determined by the degree to which magnetic ( and to a lesser degree pressure ) forces support the dense plasma , the stability criterion gives an upper limit on the strength of the axial magnetic field , which is mg for galactic - center parameters .

astro-ph0009184

the central @xmath1 pc of the galaxy host a phenomenon that is unique within the galaxy : narrow , `` vertical '' ( i.e. @xmath0 to the galactic plane ) filaments of radio emission that are typically a few tens of parsecs long and only a fraction of a parsec wide ( morris 1996 ) . polarization measurements indicate that the magnetic fields in the radio arc and northern thread are parallel to these filaments ( tsuboi et al . 1986 , reich 1994 , tsuboi et al . 1995 , lang et al . 1999b ) , and this alignment is presumed to be typical of the other filaments as well . in virtually all cases the filaments appear to be in contact with a molecular cloud at some point along their length . the absence of bending of these filaments by the randomly moving clouds has been interpreted as evidence of a milligauss lower limit to the field strength in the filaments ( yusef - zadeh & morris 1987a this point is discussed further in section [ sec : review ] ) . the apparent impossibility of forming or confining narrow filaments with such tremendous magnetic pressures well in excess of the thermal pressure of the medium suggests that if a mg vertical field is present , it pervades the central @xmath1 pc of the galaxy , and the filaments are simply those flux tubes that possess large populations of relativistic electrons ( morris & serabyn 1996 ) . a strong pervasive magnetic field at the galactic center would be a natural consequence of radial inflow in the galactic disk ( sofue & fujimoto 1987 , morris & serabyn 1996 ) , and magnetic - pressure forces would over time cause the volume - filling magnetic flux tubes to rise buoyantly away from the galactic mid - plane , forcing field lines that penetrate the mid - plane into a vertical orientation ( chandran et al . it is not clear , however , that such a scenario is inevitable , as the rate of radial inflow in the galactic disk is not well known and since sufficient turbulent resistivity can inhibit the central concentration of the field ( lubow , papaloizou , & pringle 1994 ) . alternative explanations of the filaments have been explored by a number of authors ( chudnovsky et al . 1986 , benford 1988 , heyvaerts , norman , & pudritz 1988 , lesch & reich 1992 , rosso & pelletier 1993 , shore & larosa 1999 ) . if a pervasive mg field is present , it would play a dominant role in the dynamics of the interstellar medium near the galactic center , and would suggest that the galaxy was born with a magnetic field @xmath24 g ( chandran et al 2000this point is discussed further in section [ sec : discussion ] ) . one of the most pressing questions regarding the pervasive - mg - field hypothesis is whether such a strong magnetic field could be stably confined . the volume - dominant phase of the central region is hot plasma at a temperature @xmath25 k which fills an elliptical region @xmath26 pc ( fwhm ) with a major axis that is tilted @xmath27 with respect to the galactic plane ( yamauchi et al . the density of this plasma was estimated to be @xmath28 by yamauchi et al . ( 1990 ) and @xmath29 by koyama et al . the mass - dominant phase of the central region is molecular gas with @xmath30 , @xmath31 k , a filling factor in excess of 0.1 , and a vertical thickness of @xmath32 pc ( bally et al . 1988 , morris & serabyn 1996 ) . if the magnetic field strength is 1 mg , then the magnetic pressure @xmath33 is @xmath34 , which far exceeds the thermal pressure of both the hot plasma , @xmath35 , and the molecular material , @xmath36 . the thermal pressure of the ambient medium is thus unable to confine mg fields , whether they are ubiquitous or in the form of isolated narrow flux tubes . the only force capable of confining a pervasive vertical mg field is the weight of the molecular material . interestingly , the polarization of dust emission indicates that the molecular material is threaded by horizontal ( i.e. , @xmath2 to the galactic plane ) magnetic fields ( morris & serabyn 1996 ) . because this horizontal field is perpendicular to the vertical magnetic field in the hot plasma , it is difficult for the two types of material to interpenetrate , which suggests that the weight of the molecular material may be able to stably prevent a strong vertical field from expanding . although the molecular clouds are clumpy , it is expected that the horizontal magnetic field completely encloses the central region . in this paper , the equilibrium of figure [ fig : equil ] is proposed as a starting point for investigating the confinement of vertical fields at the galactic center . to make a compact analytic treatment possible , the stability of this equilibrium is explored under a number of additional simplifying approximations . it is assumed that the fractional ionization within the molecular clouds is sufficiently large and the frequencies of any instabilities sufficiently small compared to the neutral - ion collision frequency that ambipolar diffusion can be ignored . the partially ionized clouds then behave as a single - species fluid to which the magnetic field is frozen , or , in other words , as a dense plasma described by the equations of magnetohydrodynamics ( mhd ) . the equilibrium is taken to be infinite and invariant along the spin axis the @xmath3 axis in cylindrical coordinates @xmath4 as depicted in figure [ fig : equil_picture ] . a constant angular velocity @xmath21 is assumed , which removes the magneto - rotational instability from the problem ( balbus & hawley 1998 ) . an axial magnetic field and low - density plasma are assumed for @xmath5 pc , and an azimuthal magnetic field and dense plasma for @xmath37 pc . the equilibrium density @xmath38 , pressure @xmath39 , and field strength @xmath40 are assumed to be proportional to powers of @xmath41 within the dense plasma , and @xmath42 is taken to be constant and small , which implies that the sound speed @xmath22 and alfvn speed @xmath43 are proportional to @xmath41 . the gravitational acceleration @xmath44 is taken to be proportional to @xmath41 and to be directed in the @xmath45 direction . if @xmath44 arises from stars , this amounts to assuming a constant mass density of stars throughout the region occupied by the plasma . it is also assumed that the plasma is supported primarily by rotation , so that @xmath46 in the dense plasma , where @xmath47 . although some of these assumptions are extreme , a number of the key ingredients are retained : compressibility , the coriolis force , the suspension of dense plasma `` above '' low - density plasma , and the orthogonal orientations of the magnetic fields in the low- and high - density plasmas . the analysis then focuses on modes which perturb the sharp boundary between the high- and low - density plasmas . such perturbations are to some extent stabilized by magnetic tension , since short - wavelength disturbances involve significant field - line bending of either the axial or azimuthal magnetic fields . uniform rotation also plays a stabilizing role ( e.g. , talwar 1960 , chandrasekhar 1961 , gilman 1970 , acheson & gibbons 1978 ) . normal modes of the form @xmath48 with @xmath18 and @xmath15 are studied analtyically . such modes avoid bending or compressing the axial magnetic field at @xmath49 pc and also allow compressions of the dense plasma along the azimuthal magnetic field that enhance the destabilizing effects of gravity . it is shown in sections [ sec : diffeq ] through [ sec : bc ] that the necessary and sufficient condition for stability to @xmath18 modes is @xmath50 when @xmath51 , where @xmath52 is the effective gravity , and @xmath22 and @xmath43 are evaluated in the dense plasma . a physical explanation of equation ( [ eq : stabcrit0 ] ) is given in section [ sec : stability ] . since the effective gravity is proportional to the partial support of the plasma by the magnetic field , equation ( [ eq : stabcrit0 ] ) can be re - written as an upper limit on the strength of the vertical field [ equation ( [ eq : stabcrit3 ] ) ] . for galactic - center parameters , the upper limit is @xmath53 mg . if equation ( [ eq : stabcrit0 ] ) is the approximate criterion for the stability of the boundary between the high and low - density plasmas in systems with differential rotation , a finite extent along the @xmath3 axis , and more general radial profiles say by evaluating the terms in equation ( [ eq : stabcrit0 ] ) near the inner edge of the dense plasma then a pervasive mg vertical magnetic field can be confined at the galactic center . equation ( [ eq : stabcrit0 ] ) would then also be relevant to other accreting systems with strong poloidal magnetic fields . global instabilities with @xmath54 , local quasi - interchange modes , and instabilities of the two - dimensional plasma of figure [ fig : equil ] are discussed in section [ sec : local ] . in section [ sec : review ] , observational constraints on the field - strength in the galactic - center filaments are reviewed . the implications of a pervasive mg field at the galactic center for the origin of the galactic magnetic field are discussed in section [ sec : discussion ] . lcc + symbol & equation where introduced & meaning + + + @xmath9 , @xmath10 , @xmath55 , @xmath56 & ( [ eq : cont ] ) , ( [ eq : mom ] ) & mass density , flow velocity , pressure , and magnetic field ; ( 0 + & & indicates equilibrium value , 1 indicates perturbed quantity ) + @xmath44 , @xmath57 & ( [ eq : mom ] ) , ( [ eq : g ] ) & true gravity , effective gravity + @xmath21 & ( [ eq : u0 ] ) & uniform angular velocity + @xmath22 , @xmath43 , @xmath58 & ( [ eq : a ] ) , ( [ eq : va ] ) , ( [ eq : vrot ] ) & sound speed , alfvn speed , rotation speed + @xmath59 & ( [ eq : a ] ) & adiabatic index + @xmath60 , @xmath61 , @xmath62 & ( [ eq : u1 ] ) , ( [ eq : l3 ] ) & azimuthal , axial , and radial mode numbers + @xmath63 , @xmath64 & ( [ eq : u1 ] ) , ( [ eq : omega ] ) & frequency in inertial frame , frequency in rotating frame + @xmath65 , @xmath66 , @xmath67 , @xmath68 , @xmath69 & ( [ eq : h ] ) , ( [ eq : f ] ) , ( [ eq : h ] ) , ( [ eq : gamma ] ) , ( [ eq : pn ] ) & convenient abbreviations + @xmath70 , @xmath71 , @xmath72 , @xmath73 & ( [ eq : ds ] ) , ( [ eq : da ] ) , ( [ eq : de ] ) , ( [ eq : dr ] ) & dispersion - relation polynomials for the sound wave , + & & alfvn wave , epicyclic - alfvn mode , and rayleigh - taylor mode + @xmath74 & ( [ eq : beta ] ) & ratio of thermal to magnetic pressure + @xmath75 & ( [ eq : x ] ) & normalized radius + @xmath76 & ( [ eq : r0s ] ) & exponent in constant-@xmath74 equilibrium ( @xmath77 ) + @xmath78 & ( [ eq : gva ] ) , ( [ eq : c ] ) & order - unity factor in constant-@xmath74 equilibrium + @xmath79 & ( [ eq : xb ] ) & exponent of @xmath41 in power - law solutions for @xmath80 + @xmath81 & ( [ eq : xi ] ) & lagrangian displacement + @xmath82 & ( [ eq : zeta ] ) & measure of the wavenumber + @xmath83 & ( [ eq : zetacrit ] ) & maximum value of @xmath82 corresponding to an unstable mode + @xmath84 , @xmath85 , @xmath86 & ( [ eq : fgp ] ) , ( [ eq : fcp ] ) , ( [ eq : fpfg ] ) & perturbed gravitational , coriolis , and pressure forces + @xmath87 & ( [ eq : stabcrit3 ] ) & strength of the pervasive vertical field at the galactic center + + the plasma is taken to obey the equations of ideal mhd and an adiabatic equation of state , @xmath88 @xmath89 @xmath90 @xmath91 where @xmath9 , @xmath10 , @xmath55 , and @xmath56 are the density , pressure , velocity , and magnetic field , @xmath44 is the gravitational acceleration , @xmath92 are cylindrical coordinates , and @xmath59 is the adiabatic index . an equilibrium is assumed in which all quantities are functions of @xmath41 alone . equilibrium quantities are denoted with the subscript 0 ; @xmath93 a uniform angular velocity is assumed , @xmath94 and the effective gravity @xmath57 is defined by the equation @xmath95 by virtue of equation ( [ eq : mom ] ) , @xmath96 the sound and alfvn speeds are defined , respectively , by the equations @xmath97 and @xmath98 small - amplitude disturbances to the equilibrium , denoted by the subscript 1 , can be decomposed into normal modes of the form @xmath99 with analogous equations for @xmath100 , @xmath101 , and @xmath102 . the equations governing these disturbances are obtained in the standard way , by substituting @xmath103 , @xmath104 , etc into equations ( [ eq : cont ] ) through ( [ eq : state ] ) and discarding all terms containing the product of more than one of the perturbed quantities . the analysis is confined to the case @xmath105 the frequency @xmath64 of the disturbance in the frame rotating with the plasma is given by @xmath106 from equations ( [ eq : cont ] ) and ( [ eq : state ] ) , the @xmath107 and @xmath108 components of equation ( [ eq : ind ] ) , and the @xmath108 component of equation ( [ eq : mom ] ) , one obtains @xmath109 @xmath110 @xmath111 @xmath112 - \frac{i\omega \rho_0}{d_{\rm s } } \frac { d u_{1r}}{dr } , \label{eq : rho1}\ ] ] and @xmath113 - \frac{i\omega \rho_0 a^2}{d_{\rm s } } \frac { d u_{1r}}{dr } , \label{eq : p1}\ ] ] where @xmath114 and @xmath115 upon substituting equations ( [ eq : b1r ] ) through ( [ eq : p1 ] ) into the @xmath116 component of equation ( [ eq : mom ] ) , one obtains @xmath117 + rfu_{1r } = 0 , \label{eq : diffeq}\ ] ] where @xmath118 @xmath119 and where @xmath120 for isothermal equilibria and @xmath121 , equation ( [ eq : diffeq ] ) reduces to the @xmath122 limit of equation ( 2.13 ) of acheson & gibbons ( 1978 ) . because @xmath123 , @xmath124 and @xmath125 decouple from the other variables and exhibit purely stable behavior : the @xmath126 component of equation ( [ eq : ind ] ) gives @xmath127 and the @xmath126 component of equation ( [ eq : mom ] ) then gives @xmath128 it is now assumed that the equilibrium has constant @xmath74 , where @xmath129 and that the equilibrium quantities in the dense plasma are proportional to powers of @xmath41 : @xmath130 @xmath131 @xmath132 where @xmath133 and @xmath134 , @xmath135 , and @xmath136 , are constants . it is also assumed that @xmath137 where @xmath138 is a constant . substituting equations ( [ eq : b0s ] ) through ( [ eq : gs0 ] ) into equation ( [ eq : hydrostat ] ) , one finds that @xmath139 @xmath140 @xmath141 and @xmath142 where @xmath143/2 . \label{eq : c}\ ] ] it is assumed that @xmath78 and @xmath76 are of order unity . using equations ( [ eq : b0s ] ) through ( [ eq : c ] ) in equation ( [ eq : f ] ) , one finds that @xmath144 where @xmath145 since @xmath71 , @xmath70 , @xmath65 , and @xmath67 are constant within the dense plasma , equation ( [ eq : diffeq ] ) has simple power - law solutions , @xmath146 where @xmath147 and @xmath148 the dense plasma is assumed to have an inner edge at @xmath149 , inside of which there is perfectly conducting plasma of vanishingly small density : @xmath150 three different cases are considered for the behavior at large @xmath41 : ( 1 ) the dense plasma has an outer edge at @xmath151 , outside of which is non - magnetized plasma of vanishingly small density , ( 2 ) there is an outer edge at @xmath151 , outside of which is low - density plasma with an axial magnetic field , and ( 3 ) the constant-@xmath74 equilibrium profile is taken to extend to infinitely large @xmath41 . all three cases lead to the essentially the same dispersion relation and stability criterion . in this section , cases ( 1 ) and ( 2 ) are treated ; case ( 3 ) is discussed in appendix [ ap : inf ] . for uniform rotation , the displacement @xmath152 of a fluid element relative to where it would have been in the equilibrium flow is given by @xmath153 the boundary conditions at each of the boundaries between the high and low density plasmas are @xmath154 and @xmath155 where @xmath156 is the unit normal to the perturbed boundary , and @xmath157 denotes the jump in @xmath158 across the perturbed boundary . the lagrangian perturbation of a quantity @xmath158 , denoted @xmath159 , is related to the eulerian perturbation , @xmath160 , by the equation @xmath161 . equation ( [ eq : pressb ] ) indicates that the lagrangian total - pressure perturbation ( magnetic plus thermal ) is constant across the sharp boundary . since @xmath162 @xmath22 and/or @xmath43 must be arbitrarily large in the low - density plasma in order to maintain pressure balance in the equilibrium state . it is assumed that @xmath64 is fixed by the dynamics inside the dense plasma , so that high - frequency alfvn and sound waves in the low - density plasma are not treated . for the comparatively low - frequency modes that are considered , the low - density plasma instantly rearranges itself to avoid the local compressions that would lead to thermal or magnetic - pressure perturbations . within the low - density plasma , the lagrangian total - pressure perturbation thus vanishes , as shown in appendix [ ap : axialfield ] . equation ( [ eq : pressb ] ) then implies that the lagrangian total - pressure perturbation in the dense plasma also vanishes at the sharp boundaries : @xmath163 using equations ( [ eq : b1 t ] ) and ( [ eq : p1 ] ) , one can rewrite equation ( [ eq : lagpr ] ) as @xmath164 where @xmath165 and @xmath166 the general solution to equation ( [ eq : diffeq ] ) for the constant-@xmath74 equilibrium is a linear combination of the solutions given in equation ( [ eq : xb ] ) , @xmath167 where @xmath168 and @xmath169 applying equation ( [ eq : bc ] ) at @xmath170 and at @xmath171 yields the matrix equation @xmath172 \end{array } \right ) \left(\begin{array}{c } c_1 \\ c_2 \end{array } \right ) = \left(\begin{array}{c } 0 \\ 0 \end{array } \right ) . \label{eq : meq}\ ] ] non - trivial solutions require that @xmath173 one type of normal mode results from setting @xmath174 , or , equivalently , @xmath175 equations ( [ eq : b1b ] ) , ( [ eq : b2b ] ) and ( [ eq : l3 ] ) yield the dispersion relation @xmath176 where @xmath177 ^ 2 + \left(\frac{q}{2 } - 1\right)^2 . \label{eq : pn}\ ] ] the lagrangian total - pressure perturbation of these modes vanishes at @xmath149 , @xmath151 , and at @xmath178 discrete values of @xmath41 between @xmath134 and @xmath179 . note that @xmath180 in equation ( [ eq : l3 ] ) does not correspond to an eigenmode , since it implies @xmath181 , @xmath182 from equation ( [ eq : meq ] ) , and thus @xmath183 . a second type of normal mode results from having either @xmath184 or @xmath185 in either case , one sets @xmath186 and substitutes into equation ( [ eq : quadratic ] ) to obtain @xmath187 @xmath188 . \label{eq : gdisp1}\ ] ] the lagrangian total - pressure perturbation of these modes vanishes everywhere they are the gravest modes of the system for any given @xmath60 . the right - hand side of equation ( [ eq : gdisp1 ] ) is proportional to @xmath70 , but the @xmath189 roots are spurious ; in deriving equation ( [ eq : diffeq ] ) it was assumed that @xmath190 , and it can be verified that @xmath191 does not lead to normal - mode solutions satisfying the boundary conditions . equations ( [ eq : ldisp1 ] ) and ( [ eq : gdisp1 ] ) can be rewritten in the form @xmath192 where @xmath193 @xmath194 @xmath195 and @xmath196 the value of @xmath197 is different for equation ( [ eq : ldisp1 ] ) and equation ( [ eq : gdisp1 ] ) . for equation ( [ eq : ldisp1 ] ) @xmath198 @xmath199 , \label{eq : rl}\ ] ] whereas for equation ( [ eq : gdisp1 ] ) @xmath200 @xmath201 + \frac{4mg\omega \omega^3}{r g } - \frac{m^2 \omega^2(4\omega^2 a^2 + g^2)}{g r^2}. \label{eq : rg}\ ] ] it is now assumed that the dense plasma satisfies the low-@xmath74 condition @xmath202 and is supported against gravity primarily by rotation , which implies that @xmath203 where @xmath204 equations ( [ eq : lowbeta ] ) and ( [ eq : rotsup ] ) imply that @xmath197 is negligible compared to the dominant terms in @xmath205 for all values of @xmath64 , @xmath60 , and @xmath69 . the approximate roots of equation ( [ eq : gform ] ) are then obtained by setting @xmath206 when @xmath207 , @xmath208 when @xmath209 , @xmath210 the errors in equations ( [ eq : ae ] ) and ( [ eq : gs ] ) vanish as @xmath211 and @xmath212 . comparisons between equation ( [ eq : gs ] ) and numerical solutions of equations ( [ eq : ldisp1 ] ) and ( [ eq : gdisp1 ] ) are given in figure [ fig : comparison ] . when @xmath213 and @xmath214 are small ( just how small will be seen below in the numerical solutions ) , both the approximate roots given in equations ( [ eq : ae ] ) and ( [ eq : gs ] ) and the exact roots of equations ( [ eq : ldisp1 ] ) and ( [ eq : gdisp1 ] ) are real for all @xmath60 and @xmath62 if and only if @xmath215 [ to be precise , the exact stability criterion is slightly different from equation ( [ eq : stabcrit ] ) due to small corrections associated with nonzero @xmath213 and @xmath214 . ] it can be seen directly from equation ( [ eq : gs ] ) that the approximate roots are real for all @xmath60 and @xmath62 if and only if equation ( [ eq : stabcrit ] ) is satisfied ; that the exact roots are also real can be seen from the following argument . let @xmath216 the four roots of the exact dispersion relation @xmath217 are real when the plot of @xmath218 passes through zero four times , as in figure [ fig : crossings ] . when @xmath219 and @xmath220 , @xmath221 always has two local minima at @xmath222 at which @xmath223 . also , @xmath224 as @xmath225 . the question of stability turns upon whether @xmath221 is non - negative for some value(s ) of @xmath226 in the interval @xmath227 . when @xmath228 and @xmath229 and equation ( [ eq : stabcrit ] ) is satisfied , the local maximum of @xmath230 at @xmath231 is positive . moreover , @xmath232 is negligible compared to the value of @xmath230 at @xmath233 ( unless the plasma is extremely close to marginal stability ) . hence , the value of @xmath221 is also positive at @xmath234 , and the four exact roots are real . on the other hand , when equation ( [ eq : stabcrit ] ) is not satisfied , the local maximum of @xmath230 at @xmath235 is negative , @xmath230 is negative for all @xmath64 in the interval @xmath236 , @xmath237 for all @xmath64 in the interval @xmath227 , @xmath238 is negative for all @xmath64 in the interval @xmath239 , and two of the exact roots have nonzero imaginary parts . the @xmath240 roots of equation ( [ eq : gdisp1 ] ) are a special case , since equation ( [ eq : gs ] ) gives a double root at @xmath241 . in this case , the local maximum of @xmath230 occurs at @xmath242 , and @xmath243 at the local maximum . it is thus not clear whether the local maximum of @xmath238 has a positive or negative value . in appendix [ ap : m1 ] , a new approximate factorization of equation ( [ eq : gdisp1 ] ) is given for @xmath240 and it is shown that the @xmath240 roots are always stable when @xmath219 and @xmath244 , whether or not equation ( [ eq : stabcrit ] ) is satisfied . figure [ fig : stabcrit ] summarizes a stability analysis based upon numerical solutions of equations ( [ eq : ldisp1 ] ) and ( [ eq : gdisp1 ] ) for values of @xmath213 and @xmath245 ranging from 0.01 to 1 . a combination of parameters @xmath246 and @xmath247 is said to be stable if the imaginary part of each of the numerically determined roots is less than @xmath248 times the real part of the same root for all values of @xmath60 from 1 to 100 , and in the case of the roots to equation ( [ eq : ldisp1 ] ) , for all values of @xmath62 from 1 to 500 , where @xmath249 is taken to be 5 , and @xmath76 is taken to be 5 . the roots are determined using an eigenvalue method ( press et al 1992 ) at double precision . equation ( [ eq : stabcrit ] ) describes the stability of the exact roots quite accurately provided @xmath250 . for larger values of @xmath245 , equation ( [ eq : stabcrit ] ) underestimates stability . for any given values of @xmath22 , @xmath21 , and @xmath57 , the most unstable mode has zero lagrangian total - pressure perturbation everywhere and @xmath251 . upon setting @xmath252 , where @xmath253 is the imaginary part of @xmath64 , one can show using equation ( [ eq : gs ] ) that the maximum growth rate @xmath254 occurs at @xmath255 and is given by @xmath256 where @xmath158 is a function that increases monotonically from 0 to 1 as its argument @xmath257 is decreased from @xmath258 ( marginal stability ) towards 0 . because of equation ( [ eq : gva ] ) , @xmath259 when the plasma is not very near marginal stability . when @xmath260 , modes with @xmath251 as well as modes with @xmath261 are unstable only when @xmath262 , where @xmath263 in the absence of background flow , the stability of an equilibrium to a small - amplitude displacement @xmath264 is determined by whether @xmath264 leads to an increase or decrease in @xmath265 , where @xmath265 is the sum of the perturbed magnetic , internal , and gravitational potential energies of the plasma ( bernstein et al 1958 ) . if @xmath265 is positive , then the mode is stable if the displacement increases with time initially , the initial bulk - motion kinetic energy present in the mode is converted into these other types of energy , and the plasma proceeds to oscillate stably . if @xmath265 is negative , then the magnetic , internal , and gravitational energies of the plasma can be tapped to drive an instability . in an equilibrium with flow , the kinetic energy of the flow is altered by plasma displacements , and stability is no longer determined solely by the increments to the magnetic , internal , and gravitational potential energies . the perturbed kinetic energy depends upon both the displacement @xmath264 and its time derivative @xmath266 , which gives rise to the phenomenon of overstability , in which both the real and imaginary parts of @xmath64 are nonzero , and which makes the stability analysis more complicated than in the static case ( frieman & rotenberg 1960 ) . in this section , the physical basis of the stability criterion is explained using the properties of the eigenmodes . for any set of plasma parameters , the most unstable @xmath18 modes are those described by equation ( [ eq : gdisp1 ] ) [ as opposed to those described by equation ( [ eq : ldisp1 ] ) ] , for which the lagrangian total - pressure perturbation vanishes everywhere . it is these modes that will be discussed in this section . the dispersion relation given by equation ( [ eq : gdisp1 ] ) has two branches , described approximately by equations ( [ eq : ae ] ) and ( [ eq : gs ] ) . equation ( [ eq : ae ] ) is an epicyclic - alfvnic branch . when @xmath267 , one finds that @xmath268 , @xmath269 , and @xmath270 . these circularly polarized epicyclic modes , which involve primarily a balance between inertia and the coriolis force , are stable for all plasma parameters : when a fluid element in the dense plasma of figure [ fig : equil_picture ] tries to sink towards @xmath271 , the coriolis force pushes it in the @xmath272 direction ( for positive @xmath21 ) ; as it moves in the @xmath272 direction , the coriolis force pushes it back towards larger @xmath41 . for @xmath273 , the modes are stable alfvn waves , with @xmath274 , @xmath275 , and @xmath276 . the dominant force across field lines for these modes is magnetic tension . equation ( [ eq : gs ] ) is the rayleigh - taylor / sound - wave branch of the dispersion relation . it is shown in appendix [ ap : eigenmodes ] that the eigenmodes corresponding to equation ( [ eq : gs ] ) satisfy latexmath:[$|u_{1\theta}| \gg little field - line bending . for @xmath267 , @xmath278 . in the frame rotating with the plasma , the momentum equation at first order in perturbation amplitude can be written @xmath279 , where only the coriolis force @xmath85 has been written explicitly on the right - hand side . since @xmath280 and @xmath281 is not along @xmath126 , some force must approximately cancel the coriolis force . for @xmath282 , the near cancellation is provided by the gravitational force @xmath84 . upon defining @xmath283 and @xmath284 , and using equation ( [ eq : b1r ] ) to eliminate @xmath285 , one obtains @xmath286 it is shown in appendix [ ap : eigenmodes ] that @xmath287 ( that is , the density perturbation is dominated by the compression associated with @xmath288 ) , so that @xmath289 while @xmath290 the phase velocity of a low-@xmath60 mode in the @xmath108 direction , @xmath291 , is set by the requirement that @xmath292 and @xmath84 approximately cancel , which yields @xmath293 , or @xmath294 . this phase velocity is in the eastwards direction when @xmath295i.e . , in the direction of rotation ( @xmath296 ) . because of the additional forces in the plasma , which are much smaller than either @xmath84 or @xmath85 , the exact phase velocity will be either a little larger than @xmath297 , in which case @xmath298 , or a little smaller than @xmath297 , in which case @xmath299 . it will prove useful to consider the radial momentum equation in somewhat greater detail . it is shown in appendix [ ap : eigenmodes ] that @xmath300 for the eigenmodes of equation ( [ eq : gs ] ) with @xmath282 . this is related to the fact that the lagrangian perturbation of the magnetic pressure , which is @xmath301 , is cancelled by the lagrangian perturbation of the much smaller thermal pressure . ( although @xmath302 is large , it is not large enough to enable @xmath303 to be @xmath304 . ) with the use of the relation @xmath300 , the vanishing of the lagrangian total - pressure perturbation , equations ( [ eq : b1r ] ) and ( [ eq : b1 t ] ) , and the equation @xmath305 ( frieman & rotenberg 1960 ) , the radial component of the momentum equation can be written @xmath306 the @xmath307 term is the radial component of what will be called the net gravitational force , @xmath308 . the @xmath309 term is the effective magnetic tension , which acts to restore a radially displaced fluid element in the dense plasma to its initial position . in order for a mode with @xmath310 to propagate , the radial components of the effective tension and the net gravitational force must approximately cancel ; otherwise , solving for @xmath64 yields @xmath311 , which is @xmath312 for @xmath267 . because gravity and the coriolis force nearly cancel , the low-@xmath60 eigenmodes of equation ( [ eq : gs ] ) do not enjoy the full stabilizing effect that the coriolis force has on epicyclic modes , nor the full destabilizing effect that gravity has on rayleigh - taylor modes in a static plasma . the physics behind the stability criterion can be understood by considering a mode with @xmath267 and @xmath313 slightly less than @xmath297 , which implies that @xmath314 is slightly bigger than @xmath315 and that the net gravitational force acts in the same direction as gravity . if @xmath316 , then the dense plasma reacts in a way that is similar to how it would react if @xmath317 were 0 : the maximum density enhancements occur at those points towards which the parallel component of the net gravitational force converges the troughs of the perturbed magnetic field lines , as in figure [ fig : eigen1 ] . the induced pressure force opposes the parallel component of the net gravitational force . in the radial direction , the net gravitational force and the effective magnetic tension oppose each other and can therefore approximately cancel , as required for a mode with @xmath318 . on the other hand , if @xmath319 , then sound waves can not propagate information about the net gravity upstream of the wave . at any instant in time , fluid elements at the troughs of the perturbed field lines can not tell that the net gravitational force is trying to compress them , and the plasma reacts inertially to the parallel component of the net gravitational force . plasma is accelerated towards positive @xmath108 at those points where @xmath320 , and the resulting azimuthal compression causes the density enhancements to occur at the crests of the field lines when @xmath64 is real , as in figure [ fig : eigen1 ] . [ this point can also be seen with the aid of equation ( [ eq : aparallel_mom ] ) in appendix [ ap : eigenmodes ] . ] in this case , the net gravitational force in the radial direction acts in the same direction as the effective magnetic tension , providing much more radial force than is consistent with @xmath294 , and no stable mode at approximately that frequency exists . overstable modes , however , do exist . one way of thinking about this is the following . suppose a low-@xmath60 mode were to start propagating with a phase velocity greater than @xmath22 and peak density enhancements at the crests of waves ( as would be the case for real @xmath64 ) . the radial component of the net gravitational force and magnetic tension would then provide much more force than required to return the displaced plasma to its initial position . a fluid element displaced a distance @xmath321 towards @xmath322 would in response pass through its initial position and then move a distance @xmath323 towards @xmath45 , leading to an oscillation of growing amplitude . if @xmath324 , where @xmath253 and @xmath325 are the imaginary and real parts of @xmath64 , respectively , then the phase relation between @xmath101 and @xmath80 is not as depicted in figure [ fig : eigen1 ] . when @xmath326 , @xmath327 are then effectively sine waves of equal amplitude that are nearly @xmath328 out of phase , and their sum is about 90 degrees out of phase with each of them . the locations at which @xmath101 obtains its peak magnitude are then approximately the steepest points on the perturbed magnetic field lines , in phase with @xmath285 and @xmath80 . the peak magnitude of the radial component of the net gravitational force is then about 90 degrees out of phase with @xmath101 and coincides with the troughs and crests of field lines . the radial component of the net gravity is then approximately cancelled by the effective magnetic tension . when @xmath329 , modes are stabilized when @xmath82 is increased above @xmath83 because pressure begins to dominate over gravity in the dynamics along the magnetic field lines , changing the modes from rayleigh - taylor instabilities into something akin to sound waves . it is shown in appendix [ ap : eigenmodes ] that @xmath330 when @xmath331 and @xmath332 , where @xmath333 is the component of the pressure force along the perturbed magnetic field lines in the dense plasma . it is also shown that as @xmath82 is increased towards @xmath83 , @xmath334 . for @xmath335 , @xmath336 for all @xmath60 with one exception : @xmath337 on the branch of the dispersion relation with @xmath338 . this exception corresponds to the segment of the plot of @xmath339 in the upper - left panel of figure [ fig : comparison ] just to the right of @xmath83 where the curve bifurcates , between the bifurcation point and where the plot of @xmath339 intersects @xmath340 ; the plot of @xmath339 bifurcates at @xmath341 , and the exception corresponds to the smaller value of @xmath64 at @xmath342 . to obtain further insight into the stability criterion , it is helpful to consider some limiting cases . of course , if there is neither rotation nor magnetic field , the equilibrium of figure [ fig : equil_picture ] is unstable to rayleigh - taylor modes as in the classic calculation of a dense fluid suspended above low - density fluid in slab geometry ( rayleigh 1883 ) . if the orthogonal magnetic fields of figure [ fig : equil_picture ] are present but the plasma is nt rotating , the situation is similar to the stationary plane - parallel problem solved by gratton , gratton , & gonzalez ( 1988 ) and gonzalez & gratton ( 1990 ) . these authors discussed a number of cases , including that of a plasma at @xmath343 suspended above vacuum at @xmath344 with gravity pointing in the @xmath345 direction , all equilibrium quantities being functions of @xmath346 alone , the magnetic field in the plasma pointing in the @xmath75 direction , and the magnetic field in the vacuum pointing in the @xmath3 direction . they found that the orthogonal magnetic fields act to stabilize the plasma , since the interface at @xmath347 can not be perturbed without bending either the magnetic field in the plasma or the magnetic field in the vacuum . compressibility , however , acts to destabilize the plasma ; if the plasma is strongly compressed along the magnetic field , the gravitational force acting on regions of enhanced density can overcome magnetic tension . when the wave vector @xmath348 of the perturbation in the @xmath349 plane is oriented along the magnetic field in the plasma at @xmath343 , magnetic tension on its own is not sufficient to stabilize small - scale modes : as @xmath350 , the maximum unstable wavenumber approaches @xmath351 . since thermal pressure limits the compressibility of the plasma along the magnetic field , magnetic tension and pressure together stabilize modes with @xmath352 , where @xmath353 . it can be shown that the condition @xmath354 corresponds to pressure dominating over gravity in the dynamics along the perturbed magnetic field lines , in analogy to the stabilization of the modes described by equation ( [ eq : gs ] ) when @xmath355 . on the other hand , sufficiently large - wavelength modes ( @xmath356 ) are always unstable in the magnetized slab , whereas all @xmath18 modes are stable in the rotating magnetized cylinder when @xmath357 . if there is rotation but no magnetic field , the plasma has a set of normal modes analogous to those described by equation ( [ eq : gdisp1 ] ) ( i.e. , @xmath18 and zero lagrangian pressure perturbation everywhere ) , with frequencies @xmath358 and @xmath359 these modes are circularly polarized ( @xmath360 ) , stable at long wavelengths ( @xmath361 ) , unstable when @xmath362 in equation ( [ eq : oh1 ] ) , and unstable when @xmath363 in equation ( [ eq : oh2 ] ) . as @xmath364 , the growth time of unstable modes ( @xmath365 ) is much shorter than the rotation period , and rotation can be neglected . the introduction of an azimuthal magnetic field into the dense plasma has a stabilizing influence insofar as magnetic tension eliminates these large-@xmath366 instabilities with @xmath367 . the magnetic field in the equilibrium of figure [ fig : equil_picture ] , however , can also be destabilizing : as the field strength is increased , the plasma becomes increasingly supported by magnetic pressure , thereby increasing the effective gravity @xmath57 . with the use of equation ( [ eq : gva ] ) , the stability criterion can be rewritten as an upper limit on the alfvn speed : @xmath368 ( where @xmath22 , @xmath43 , and @xmath58 are evaluated in the dense plasma ) . in the low-@xmath74 limit , pressure balance requires that the magnetic - field strength @xmath87 in the low - density plasma at @xmath369 be approximately equal to the field strength in the dense plasma near @xmath149 , @xmath135 . assuming a dense plasma of pure hydrogen with @xmath370 , @xmath371 , @xmath372 , @xmath373 km / s [ from the galactic rotation curve at @xmath374 pc ( ruzmaikin & shukurov 1981 ) ] , and @xmath375 k and @xmath376 near @xmath149 , one can rewrite equation ( [ eq : stabcrit2 ] ) as @xmath377 although it is tempting to infer from equation ( [ eq : stabcrit3 ] ) that mg vertical magnetic fields at the galactic center can be stably confined by a ring of horizontal magnetic fields threading molecular material , further study is needed to extend the present analysis to more realistic equilibria . sections [ sec : diffeq ] through [ sec : stability ] treated the global stability of the equilibrium pictured in figure [ fig : equil_picture ] to @xmath18 modes . this section addresses three other classes of instabilities : ( 1 ) local interchange and non - axisymmetric quasi - interchange modes in a uniformly rotating plasma with azimuthal magnetic field , ( 2 ) global instabilities in the 1d equilibrium of figure [ fig : equil_picture ] with @xmath54 , and ( 3 ) quasi - interchange and magneto - rotational instabilities in a two - dimensional differentially rotating equilibrium such as the one pictured in figure [ fig : equil ] . local modes in a plasma with an azimuthal magnetic field that rotates uniformly about the @xmath3 axis have been treated by a number of authors . for the constant-@xmath74 equilibrium profile of section [ sec : eq ] , local axisymmetric interchange instabilities are suppressed primarily because of the strongly stabilizing specific - angular - momentum profile when @xmath378 ( newcomb 1962 , acheson & gibbons 1978 , rogers & sonnerup 1986 , ferriere et al 1999 ) . such modes are further stabilized since the specific entropy and magnetic flux per unit mass of an azimuthal flux loop @xmath379 are taken to increase with @xmath41 in the constant-@xmath74 equilibrium of section [ sec : bc ] . the results of ferriere et al ( 1999 ) on a wide class of one - dimensional equilibria can be used to show that local non - axisymmetric quasi - interchange modes in a uniformly rotating plasma with azimuthal magnetic field are stable when two conditions are met assuming @xmath380 : ( 1 ) the `` magnetic brunt - visl '' frequency @xmath381 must be real , and ( 2 ) @xmath19 ( to lowest order in @xmath213 and @xmath382 . for the constant-@xmath74 equilibrium of section [ sec : bc ] , @xmath383 and the first condition is met . the second condition is the same as equation ( [ eq : stabcrit ] ) , and thus when the @xmath18 modes are stable , so are the quasi - interchange modes . the quasi - interchange modes have @xmath294 , and are stable ( provided @xmath381 is real ) when their phase velocity along the magnetic field is less than the sound speed , just as for @xmath18 modes with @xmath384 . global modes in the equilibrium of figure [ fig : equil_picture ] with @xmath54 and @xmath385 are analogous to the external kink instability in static plasmas . an @xmath386 mode is illustrated in figure [ fig : kink ] [ figure 9.23 of freidberg ( 1987 ) ] . because the strength of the azimuthal field increases where the field lines are compressed , the lorentz force of the azimuthal field acts to increase the perturbation . on the other hand , the magnetic tension of the axial magnetic field opposes the perturbation . in the static reversed field pinch , in which @xmath387 , the azimuthal magnetic field wins out over the axial field and the mode is unstable when the axial wavelength @xmath388 is greater than roughly the radius of the plasma cross section ( friedberg 1987 ) . how this instability mechanism affects the stability criterion of the rotating equilibrium of figure [ fig : equil_picture ] is left for future work . newcomb ( 1961 ) showed that the convective stability criterion for a compressible stratified magnetized slab when @xmath389 is less stringent than the stability criterion in the limit @xmath390 , where @xmath391 is the component of the wave vector along the magnetic field . this is because when @xmath392 , newcomb s type 2 modes , which involve motion primarily along the magnetic field , become pure translations which are neutrally stable ( @xmath393 ) . the @xmath394 stability criterion is thus the one that applies to newcomb s type 1 modes , which involve motion primarily across field lines . on the other hand , when @xmath395 is small but nonzero , the type 2 modes allow destabilizing compressions of the plasma along the magnetic field and can be unstable even when type 1 modes are stable . the distinction between @xmath394 and @xmath396 in newcomb s work is not directly analogous to the distinction between @xmath397 and @xmath398 in the equilibrium of figure [ fig : equil_picture ] , in that @xmath391 corresponds to @xmath60 since it is the dense plasma that dominates the dynamics of the unstable modes . the two - dimensional equilibrium of figure [ fig : equil ] is subject to instabilities not present in the one - dimensional infinite - cylinder equilibrium . if the vertical ( along @xmath3 ) stratification does not satisfy the schwarzchild criterion for a static non - magnetized gas , then the plasma is unstable to local non - axisymmetric quasi - interchange modes with @xmath399 and @xmath400 ( terquem & papaloizou 1996 ) . the displacements in such modes are essentially parallel to the spin axis , and thus the coriolis force vanishes , preventing the type of rotational stabilization that occurs for quasi - interchange modes in the one - dimensional infinite - cylinder equilibrium . if there is differential rotation with @xmath401 , the plasma is subject to the magnetorotational instability ( mri ) ( ogilvie & pringle 1996 , terquem & papaloizou 1996 , balbus & hawley 1998 ) . global three - dimensional numerical simulations suggest that turbulence excited by the mri will be present quite generally in astrophysical disks ( hawley , balbus , & stone 2001 ) . it is thus interesting to ask how the stability of the sharp boundary in figure [ fig : equil ] is affected by differential rotation in a quiescent equilibrium flow as well as by differential rotation with turbulence . these questions are left for future work . lang et al . ( 1999b ) found that if the magnetic energy density in the northern thread is equal to the relativistic - particle energy density , and if the relativistic protons and electrons have equal energies , then the field strength is @xmath402 g . comparable equipartition field strengths have been found for the radio arc ( tsuboi et al.1986 , yusef - zadeh & morris 1987b , reich 1994 ) . however , the equipartition estimate may lead to an underestimate of the field strength , particularly since the geometry of the filaments allows relativistic particles to stream directly out of the galactic - center region . all of the radio filaments that have been studied in detail near the galactic center appear to be interacting with at least one randomly moving molecular cloud , and yet most are undeformed at the sites of interaction ( morris & serabyn 1996 ) . if the magnetic field were dynamically insignificant , one would expect the turbulent motions of the clouds to cause turbulent bulk motions in the hot plasma that would deform the frozen - in magnetic field lines . the lack of deformation gives a lower limit on the field strength in the filaments that depends upon how many independently moving clouds ( or clumps within a cloud ) interact with a single filament . if a vertical filament interacts with a single molecular cloud at a single point along its length , then the lower limit on its field strength is fairly small . since the vertical field @xmath87 threads the very tenuous hot plasma , there is very little mass to anchor the ends of the vertical field line above and below the galactic disk . any deformation in the field line in the galactic plane by an interaction with a molecular cloud is carried away along the field line at the alfvn speed of the hot medium , @xmath403 . in order for the visible deformations in the filaments to be small , the alfvn speed in the hot medium must be much larger than the turbulent speed of the the clouds @xmath404 with respect to the ends of the filament . the angle at which a field line is bent will be @xmath405 , as depicted in figure [ fig : deformation ] . the condition @xmath406 can be written @xmath407 where @xmath408 is the number density of the hot plasma in @xmath409 . if @xmath410 km / s , and @xmath411 , then @xmath412 g . as pointed out by yusef - zadeh & morris ( 1987a ) and morris & serabyn ( 1996 ) , if a filament interacts with clouds at more than one point along its length , as happens in the radio arc ( morris 2001 ) , then the lower limit on the field strength may be considerably greater than in equation ( [ eq : b1 ] ) . for certain cloud shapes , a filament could conceivably avoid the appearance of deformation by rapidly slipping around the moving clouds . for some cloud shapes , however , this would not be possible , and to avoid deformation a filament would have to dynamically oppose the force applied by the clouds . if this requires that the magnetic pressure in the filament exceed the ram pressure in the clouds , then @xmath413 mg ( yusef - zadeh & morris 1987a ) . if a single cloud consists of independently moving clumps , then a similar argument would apply to a filament interacting with a single cloud ( morris & serabyn 1996 ) . because of the simplifying assumptions made in this paper , further work is needed to determine the extent to which the stability criterion given by equation ( [ eq : stabcrit ] ) applies to more realistic equilibria , and to extend the analysis to modes with @xmath414 . if equation ( [ eq : stabcrit ] ) does describe , even approximately , the stability of the boundary between the high and low - density plasmas in systems with differential rotation , a finite extent along the @xmath3 axis , and more general radial profiles say by evaluating the terms in equation ( [ eq : stabcrit ] ) near the inner edge of the dense plasma then it appears that a pervasive mg vertical magnetic field can be confined at the galactic center . equation ( [ eq : stabcrit ] ) would then also be relevant to other accreting systems with strong central poloidal magnetic fields . a pervasive mg vertical magnetic field in the inner 150 pc of the galaxy would have important implications for the origin of the galactic magnetic field . if the vertical field originated from radial inflow over the lifetime of the galaxy , then the strength of the average vertical magnetic field in the galactic disk at the time of galaxy formation , @xmath415 , can be estimated by magnetic - flux conservation . if the average mass accretion rate into the galactic - center region is 0.3 @xmath416/yr ( morris & serabyn 1996 ) , and if the surface density of gas in the disk has remained constant over the lifetime of the galaxy , then all the gas that was within roughly 10 kpc of the galactic center at the time of galaxy formation has been accreted into the galactic - center region . if the magnetic field was frozen in to the interstellar plasma , then the vertical magnetic flux that initially threaded the central 10 kpc of the disk , @xmath417 , is now concentrated near the galactic center . if there is a pervasive vertical magnetic field at the galactic center today , it is unlikely that it contains regions of oppositely directed vertical fields , since the parallel geometry of field lines would permit rapid destruction of the oppositely directed fields by turbulent diffusion and reconnection . the current magnetic flux through the galactic - center region corresponding to such a pervasive field , @xmath418 , can thus be estimated by multiplying field strength times area . if there is a mg vertical field throughout the central 150 pc of the galaxy , then @xmath419 . setting @xmath420 yields @xmath421 g ( chandran et al 2000 ) , which would be consistent with a turbulent dynamo operating in the protogalaxy ( kulsrud et al . 1997 ) . in this case , the boundary conditions are that the lagrangian total - pressure perturbation vanishes at @xmath149 and that the behavior as @xmath423 is such that the kinetic energy of the mode is finite . as in section [ sec : bc ] , the boundary conditions can be satisfied in two ways . the first is for both @xmath424 and @xmath425 to be non - zero . the boundary condition at @xmath149 is then given by the first row of equation ( [ eq : meq ] ) , and a finite kinetic energy requires that the radicals in equations ( [ eq : b1b ] ) and ( [ eq : b2b ] ) be purely imaginary ; if they are real or complex , then the @xmath426 term causes the kinetic energy to diverge , where @xmath427 is taken to be the root with a positive real part . the resulting dispersion relation is the same as equation ( [ eq : ldisp1 ] ) , but with @xmath69 replaced by @xmath428 ^ 2 + n$ ] , where @xmath429 is any positive real number . the second way is to have @xmath430 with @xmath431 , which leads to equation ( [ eq : gdisp1 ] ) and corresponds to a mode in which the lagrangian total - pressure perturbation vanishes everywhere . the differential equation for @xmath123 normal modes in a uniformly rotating plasma with axial magnetic field [ @xmath432 is @xmath433 + r k u_{1r } , \label{eq : diffeqa}\ ] ] where @xmath434 @xmath435 + \frac{\rho_0 } { r } - \frac{\rho_0}{\omega^2}\frac{d}{dr}\left(\frac{g}{r}\right ) - \frac{m^2 \rho_0}{\omega^2 r d_{\rm ms } } , \label{eq : k}\ ] ] and @xmath436 when @xmath437 ( i.e. , @xmath438 ) , equation ( [ eq : diffeqa ] ) reduces to the constant-@xmath21 limit of equation ( 3a ) of spies ( 1978 ) . in the limit @xmath439 , with @xmath38 constant , equation ( [ eq : diffeqa ] ) reduces to @xmath440 + m^2 u_{1r } = 0 , \label{eq : diffeq2a}\ ] ] with solutions @xmath441 with @xmath442 writing the lagrangian pressure perturbation as @xmath443 and the lagrangian perturbation of @xmath444 as @xmath445 , one finds that @xmath446 , \label{eq : lagpressa}\ ] ] which vanishes as @xmath447 for fixed @xmath64 , where @xmath64 is set by the dynamics inside the dense plasma . for the @xmath240 roots of equation ( [ eq : gdisp1 ] ) , the dispersion relation can be written as @xmath448 where @xmath449 , and @xmath450 \right\}. \label{eq : m=1}\ ] ] the value of @xmath197 can be recalculated and is different from the values in equations ( [ eq : rl ] ) and ( [ eq : rg ] ) . as in the case @xmath451 , when @xmath219 and @xmath452 , @xmath453 has local minima at @xmath454 at which @xmath455 , and stability again turns upon whether @xmath456 becomes non - negative in the interval @xmath457 . the value of @xmath458 has a local maximum at @xmath459 at which @xmath458 is always positive . it can be shown that @xmath460 at @xmath461 , and thus @xmath462 is always positive at @xmath463 , which guarantees that the four exact roots are always real . this appendix treats the eigenmodes corresponding to equation ( [ eq : gdisp1 ] ) , for which the lagrangian total - pressure perturbation vanishes everywhere . all of the results of this section assume that @xmath464 and a number of the equations to be presented assume that the plasma is not very near marginal stability , i.e. , @xmath465 upon substituting equation ( [ eq : gs ] ) into equation ( [ eq : ds ] ) , one finds that @xmath466 ^ 2 , \label{eq : ds2}\ ] ] where the @xmath467 sign has the same meaning as in equation ( [ eq : gs ] ) , that the error in the equation vanishes as @xmath468 and @xmath469 . when @xmath64 is real , @xmath470 , @xmath471 is real , and @xmath472 . this implies that all stable modes on the rayleigh - taylor branch of the dispersion relation [ equation ( [ eq : gs ] ) ] travel more slowly than sound waves . upon substituting equation ( [ eq : gs ] ) into equation ( [ eq : h ] ) , one finds that @xmath473 given equation ( [ eq : nnms ] ) , the @xmath474 term on the right - hand side is negligible compared to the other term for @xmath475 . it can be shown that @xmath476 for @xmath475 . when @xmath477 and equation ( [ eq : nnms ] ) is satisfied , it can be shown that @xmath478 , @xmath479 and , for unstable modes with @xmath480 , @xmath481 the value of @xmath303 in equation ( [ eq : b3 ] ) is @xmath482 when @xmath483 and increases monotonically with increasing @xmath82 , reaching a maximum value of @xmath484 at @xmath485 , the largest value of @xmath82 for which modes are still unstable [ equation ( [ eq : zetacrit ] ) ] . when @xmath282 , @xmath300 for stable modes as well . given equations ( [ eq : nnms ] ) and ( [ eq : mcond ] ) , it can be shown that @xmath486 and @xmath487 equation ( [ eq : hds ] ) then implies that @xmath488 equations ( [ eq : b2 ] ) and ( [ eq : u1t2 ] ) can be used to show that @xmath489 since @xmath490 , the relation ( frieman & rotenberg 1960 ) @xmath491 can be used to show that @xmath492 similarly , the relation @xmath493 ( frieman & rotenberg 1960 ) can be used to show that @xmath494 if one takes the dot product of equation ( [ eq : mom ] ) with @xmath495 , uses equation ( [ eq : ind ] ) to eliminate @xmath496 and equation ( [ eq : p1fr ] ) to eliminate @xmath102 , one finds that @xmath497 the term on the left - hand side represents inertia in the frame of reference rotating with the plasma , the terms @xmath498 , @xmath499 , and @xmath500 are the components of the coriolis , gravitational , and pressure forces along the perturbed magnetic field line in roughly the @xmath272 direction . one can show that for modes with complex @xmath64 , @xmath501 which is less than 1 for @xmath262 , but which approaches 1 as @xmath502 ( as @xmath82 is increased towards @xmath83 , @xmath503 is increased towards @xmath504 ) . the stabilization of modes as @xmath82 is increased to @xmath83 thus coincides with pressure becoming equal to gravity in the dynamics along field lines . with the use of equation ( [ eq : divu ] ) , equation ( [ eq : parallel_mom ] ) can be rewritten as @xmath505 equations ( [ eq : rho12 ] ) and ( [ eq : aparallel_mom ] ) show that the sign of @xmath70 is critical for determining the phase of @xmath101 relative to @xmath80 . let @xmath506 be the sum of the components of the gravitational and coriolis forces along the magnetic field . as explained in section [ sec : stability ] , if @xmath64 is real and @xmath507 , the peaks in the perturbed density will occur at the points of convergence of @xmath508 . if @xmath64 is real and @xmath509 , the peaks in the perturbed density will occur at the points of divergence of @xmath508 . acheson , d. j. , and gibbons , m. p. 1978 , j. fluid mech .

we study the model coupled to topological gravity as a candidate to describing string theory at the self - dual radius . we define the model by analytical continuation of topological recursion relations to . we show that at genus zero the recursion relations yield the ward identities for tachyon correlators on the sphere . a scheme for computing correlation functions of gravitational descendants is proposed and applied for the computation of several correlators . it is suggested that the latter correspond to correlators of discrete states of the string . in a similar manner to the models , we show that there exist topological recursion relations for the correlators in the theory that consist of only one and two splittings of the riemann surface . using a postulated regularized contact , we prove that the genus one recursion relations for tachyon correlators coincide with the ward identities on the torus . we argue that the structure of these recursion relations coincides with that of the ward identities for any genus . citex[#1]#2@filesw auxout tempcntb@ne h@ldciteacite for citeb:=#2h@ld # 1 = by 60 = # 1bsphack@filesw gtempaauxout gtempa @nobreak esphack eqnlabel#1 eqnlabel vacuum # 1 # 1 # 1 # 1@y to # 1 -1000*#1 * = 12 # 1 . ( [ # 1 ] ) # 1#2 u#1^#1 # 1_#1 # 1#1mu # 1-#1mu [ theorem]claim # 1,#2,*#1 * ( 19#2 ) # 1,#2,#3,_#1 _ * # 2 * ( 19#3 ) # 1,#2,_ann . physics _ * # 1 * ( 19#2 ) # 1,#2,_comm . math . phys . _ * # 1 * ( 19#2 ) # 1,#2,_class . quantum grav . _ * # 1 * ( 19#2 ) # 1,#2,_class . quantum grav . _ * # 1 * ( 19#2 ) # 1,#2,_int . j. mod . phys . _ * a#1 * ( 19#2 ) # 1,#2,_j . math . phys . _ * # 1 * ( 19#2 ) # 1,#2,_gen . rel . grav . _ * # 1 * ( 19#2 ) # 1,#2,_mod . phys . lett . _ * a#1 * ( 19#2 ) # 1,#2,_nuovo cim . _ * # 1 * ( 19#2 ) # 1,#2,_nucl . phys . _ * b#1 * ( 19#2 ) # 1,#2,_phys . lett . _ * # 1b * ( 19#2 ) # 1,#2,_phys . lett . _ * # 1a * ( 19#2 ) # 1,#2,_phys . rev . _ * # 1 * ( 19#2 ) # 1,#2,_phys . rev . _ * d#1 * ( 19#2 ) # 1,#2,_phys . rev . lett . _ * # 1 * ( 19#2 ) # 1,#2,_phys . rept . _ * # 1c * ( 19#2 ) # 1,#2,_prog . theor . phys . _ * # 1 * ( 19#2 ) # 1,#2,_prog . theor . phys . suppl . _ * # 1 * ( 19#2 ) # 1,#2,_rev . mod . phys . _ * # 1 * ( 19#2 ) # 1,#2,#3[#4,#5]_yad . fiz . _ * # 1 * ( 19#2 ) # 3 [ _ sov . j. nucl . phys . _ * # 4 * ( 19#2 ) # 5 ] # 1,#2,#3[#4,#5]_zh . exp . theor . fiz . _ * # 1 * ( 19#2 ) # 3 [ _ sov . phys . jetp _ * # 4 * ( 19#2 ) # 5 ] # 1 # 1 # 1 # 1 # 1 # 1 # 1| # 1 # 1#1 | # 1#1 # 1#1 : : 6 pt 22 cm -.5 cm 0 in 0 in = 11 addtoresetequationsection addtoresetfootnotesection addtoresetfootnotesubsection = 12 taup-2170 - 94 + hep - th / yymmddd 1 cm * model as a topological description of string theory * 1.8 cm yoav lavi , yaron oz _ and _ jacob sonnenschein 0.2 cm _ school of physics and astronomy + raymond and beverly sackler faculty of exact sciences + tel - aviv university + ramat aviv , tel - aviv 69978 , israel . _ 1 cm

hep-th9406056

non - critical string theory in general and the @xmath6 string models in particular attracted much attention in recent years . they provide a framework for studying fundamental questions of string theory and quantum gravity such as non - perturbative structure . the most challenging of these class of models is the @xmath5 string . it differs from the @xmath7 string models by that it has a space - time description and a propagating massless degree of freedom , the tachyon . furthermore , brst analysis in the continuum description of the theory revealed the existence of special states at discrete values of momentum @xcite , one of which was argued to correspond to a discrete graviton describing the two - dimensional black hole of @xcite . @xmath7 non - critical string models constructed by coupling @xmath7 conformal matter to two - dimensional gravity , have been studied using a variety of methods reflecting their integrable and topological structures . their topological phase is realized by @xmath8 minimal topological matter coupled to topological gravity @xcite , from which they can be reached by appropriate perturbations . intersection theory interpretation of the correlators has been given in @xcite . a @xmath9 kdv integrable hierarchy underlies the @xmath8 topological models and the corresponding @xmath10 non - critical strings @xcite . this together with the string equation provides a complete description of the theories . the generating functions for correlators of the theories are @xmath11functions of the corresponding integrable hierarchies @xcite . matrix integral representations of the generating functions have been suggested by @xcite . for the case @xmath12 , that is pure topological gravity , they reduce to the kontsevich integral @xcite . however , while the latter integral has a nice interpretation in terms of cell decomposition of the moduli space of riemann surfaces , the geometrical meanings of the general integrals are less clear . nevertheless , these matrix integrals provide a link between the integrable and the topological structures . a manifestation of the integrable and topological structures is also provided by a set of ward identities given by @xmath13 constraints on the partition function of the theory . this is argued to be equivalent to specifying the integrable hierarchy together with a string equation @xcite and to a set of topological recursion relations @xcite as we discuss in the sequel . another relation between the integrable and the topological structures of the theories has been discovered using the topological landau - ginzburg formulation @xcite . the integrable hierarchy appears in the landau - ginzburg description in its lax formulation . the landau - ginzburg superpotential is identified with the lax operator in the dispersionless limit . there exist two types of topological recursion relations for correlators of @xmath8 theories . the basic strategy behind both is analysis of contributions to the correlators from the boundary of the moduli space . one set of recursion relations has been derived , first for pure topological gravity in @xcite and later generalized to topological matter coupled to topological gravity @xcite . the second set has been derived for pure topological gravity in @xcite . in the latter the concept of contact algebra was introduced , which together with the requirement of the invariance of the correlators under the interchange of operators provides a complete solution of the theory . the , thus derived , topological recursion relations coincide with the virasoro constraints on the partition function of the theory . the uniqueness of the contact algebra underlying the theory has been proved in @xcite . the generalization of the contact algebra to @xmath8 models coupled to topological gravity was suggested in @xcite , and was argued to correspond to the @xmath13 constraints on the partition function . a key role in this topological procedure is played by multicontacts whose importance was realized in @xcite . correlators of general @xmath8 models on the sphere , as well as those of @xmath14 model on higher genus , were consistently calculated using the topological scheme . the computation of genus @xmath15 correlators for @xmath16 requires a consistent regularization scheme which is still lacking . however , as shown in @xcite , one can overcome this problem since the recursion relations can be recast in a form that involves only one and two contacts . the topological description of the @xmath5 string is conjectured to correspond to @xmath8 model coupled to topological gravity , analytically continued to @xmath3 or equivalently , to @xmath17 twisted minimal model coupled to topological gravity . indeed the cohomology of the latter , realized an @xmath18 coset at level @xmath19 , coincides with that of the @xmath5 string @xcite . interpretation of intersection theory calculations of correlators at @xmath19 as correlators in @xmath5 string theory is in agreement with matrix model results . first , the partition functions were observed to be identical @xcite . second , the four - point correlator computed in @xcite , when analytically to @xmath19 was shown to be the tachyons four - point function @xcite . third , @xmath20 amplitudes as well as five - point function at various kinematic regions agree @xcite . the integrable hierarchy underlying the @xmath5 string is the toda - lattice hierarchy @xcite . the generating function for tachyon correlators is a @xmath11function of the hierarchy . there exist a matrix integral representation of the latter @xcite , generalizing that of the @xmath7 case . recently the string equations were constructed @xcite . the topological landau - ginzburg description of the theory is constructed as the @xmath21 model at @xmath19 @xcite . a relation between the integrable and the topological structures is established by that the landau - ginzburg superpotential is the baker - akhiezer wave function of the toda lattice hierarchy @xcite . this parallels the identification of the superpotential with the lax operator in the @xmath7 cases . there exist @xmath4 ward identities for tachyon correlators in the @xmath5 string @xcite . these relations determine completely the tachyon dynamics . the @xmath4 algebra of constraints on the partition function is probably the @xmath5 analog of the @xmath13 algebra of constraints in the @xmath7 cases . these ward identities can be derived from the toda lattice and the string equations @xcite . on the other hand they coincide with period integrals in the topological landau - ginzburg formulation of the theory @xcite . these provide another link between the topological and the integrable structures of the @xmath5 string . as described , the integrable and topological structures of the @xmath5 case parallel those of the @xmath7 cases . one of the main differences , however , is the lack of topological recursion relations in the former . it is straightforward to verify that witten s topological recursion relations are no longer correct for the @xmath5 string , and should be modified . however , the required modification is still unknown . our aim in this paper is to generalize the second set of topological recursion relations as proposed in @xcite for the @xmath7 models , and provide a topological procedure for computation of correlators in @xmath5 . the idea will be to analytically continue the @xmath8 recursion relations to @xmath3 . we will show explicitly for genus zero and one that the analytically continued topological recursion relations coincide with the @xmath4 ward identities for tachyon correlators , and will argue that their structures coincide for any genus . we will further use the topological recursion relations to compute correlators of @xmath3 gravitational descendants which are suggested to correspond to discrete states of the @xmath5 string . similarly to the @xmath7 cases , we show that the topological recursion relations can be recast in a form that involves only one and two splittings of the riemann surface . this provides us with another set of ward identities for the theory . the paper is organized as follows . in section two we review the @xmath8 topological models and the topological procedure to compute their correlators . in section three the @xmath4 ward identities of @xmath1 string theory at the self - dual radius are described . in section four we pose the rules for the analytic continuation of the topological recursion relations from @xmath2 to @xmath3 . we use the analytically continued scheme to compute up to five point tachyon correlators at genus zero . the results are in agreement with @xmath5 matrix model calculations . the equivalence between the @xmath0 recursion relations and the @xmath4 ward identities at genus zero is proven in section five . in section six we propose a topological scheme for computation of correlators of @xmath3 gravitational descendants , which we interpret as discrete states and compute several correlators . topological recursion relations in terms of one and two splittings are written in section seven . section eight is devoted to the study of higher genus recursion relations . we define a regularized contact and use it to prove the equivalence between the recursion relations and the @xmath4 ward identities on the torus . two and three point tachyon correlators on the torus are computed via the topological procedure . section nine is devoted to discussion and conclusions . in appendix a we include an example of computation of three - point tachyon correlator on the sphere as well as two - point tachyon correlator on the torus using the new recursion relations introduced in section seven . @xmath8 models form a special sub sector of the @xmath10 minimal models . having zero physical fields , they are not well defined conformal field theories . however , they make sense as topological field theories , the so called topological minimal models @xcite . the observables consist of @xmath22 primary fields @xmath23 . when coupled to topological gravity , a family of gravitational descendants @xmath24 is associated with each primary field @xmath25 , where @xmath26 takes positive integer values . for @xmath12 we get topological gravity with @xmath27 as the puncture operator . @xmath28 correspond in this case to the mumford , morita and miller cohomology classes on the compactified moduli space @xmath29 . an integer ghost number is attributed to the fields : gh(p_k , ) = ( k-1)q + ( - 1)[gh ] from the viewpoint of the integrable structure the ghost number basically corresponds to the power of the kp lax operator associated with the field @xcite , while from the viewpoint of the topological structure it corresponds to the degree of the form on the moduli space associated with it @xcite . the ghost number conservation law for the genus @xmath15 correlator @xmath30 reads _ i=1^s gh(p_k_i,_i ) = 2(g-1)(1+q)[colaw ] the ghost number conservation follows from the requirement of having a residue in the kdv computational scheme , and from the demand that the total form associated with the correlator be a top form on the moduli space in the corresponding intersection theory . an equivalent description , which is naturally analytically continued to the @xmath3 case , is the following : @xmath31 where @xmath32 . @xmath33 are in @xmath34 correspondence with @xmath24 : = n mod q , k = . the conservation law for the correlator @xmath35 is _ 1=1^s n_i = ( s + 2 g -2)(q+1)[cgh ] a topological procedure to calculate correlation functions of the model which is equivalent to the @xmath13 constraints has been developed in @xcite . in contrast to the @xmath13 ward identities which are complicated and are not known in general , the topological procedure consists of simple topological rules . the idea behind the scheme is that the correlators can be determined by contacts between the operators and between them and the degenerations of the riemann surface . this procedure yields topological recursion relations generalizing those proposed by @xcite for topological gravity , i.e. the @xmath36 model . let us briefly review it . the first thing to notice is that the metric on the space of physical fields , defined by the genus zero two - point function , vanishes , since _ ij p_ip_j_0 = |i|_i+j,0 [ metric ] while the physical fields of the theory are @xmath33 with @xmath37 being positive integer charges . in order to overcome this difficulty , auxiliary unphysical fields with negative charges are introduced . these fields appear in the metric and decouple from higher point functions . note that the definition of the metric ( [ metric ] ) is differs from the standard one @xcite : @xmath38 with @xmath39 being the puncture operator and @xmath40 are primary operators . the metric ( [ metric ] ) is defined on the space of both primaries and descendants . define $ ] _ n |w_g@xmath65|w_0@xmath66_n^(0)@xmath67w_1+@xmath68(1,q=-1)@xmath69q>1@xmath702d@xmath71q>1@xmath72p_n@xmath73n@xmath74(x)@xmath75q>1@xmath76q=-1@xmath77q=-1@xmath78-p_-q-1@xmath79t_0@xmath80p_n@xmath81q=-1@xmath82@xmath83t_n@xmath84@xmath85t_-n@xmath862d@xmath87p_np_n_1p_n_2@xmath88(1,q)@xmath89n_i@xmath90p_n@xmath91q=-1@xmath92p_-n@xmath93-n@xmath94t_-n@xmath95p_np_n_1p_n_2p_n_3@xmath96n+1@xmath97p_n + n_i-(q+1)_i j = 1 ^ 3p_n_j @xmath98(1,q)@xmath99q>1@xmath100(n + n_i - ( q+1))@xmath101w_q@xmath102((k + k_i-1)+ ( + _ i -1)q)@xmath103q = - 1@xmath104(-n - n_i)@xmath105@xmath106c=1@xmath107q = -1@xmath108p_n_i=1 ^ 4p_n_i@xmath109q = -1@xmath110q = -1@xmath111q=-1@xmath112q > 1@xmath113q=-1@xmath1142d@xmath115(1,q=-1)@xmath116w_1+@xmath1172d@xmath118p_n_i = 1^m p_n_i@xmath119p_n@xmath120p_n@xmath121s , s_1 .. s_n@xmath122s s_1 .. s_n = ( 1 ... m)@xmath123n@xmath124m@xmath125q=-1@xmath1261m@xmath127t_n_i = 1^m t_n_i @xmath128\stop \label{wwi } \eeqar a general term in ( \ref{wwi } ) is , up to a combinatorial factor , of the form \beq res[\bar{w}_0^{n+1-p}\prod_{i \in s}\pa_{n_i}\bar{w}_0 \prod_{j_1 \in s_1}\pa_{n_{j_1}}\bar{w}_0 ... \prod_{j_n \in s_n}\pa_{n_{j_n}}\bar{w}_0 ] \comma \label{gt } \eeq where the sets $ ] s , s_1 .. s_n@xmath122s s_1 .. s_n = ( 1 ... m)@xmath129p@xmath130w_1+@xmath131(_i = 1^m n_i)@xmath132_i = 1^m n_i = - n@xmath133(1,q=-1)@xmath134(1,q)@xmath135w_q@xmath1362d@xmath137q > 1@xmath1382d@xmath139q=-1@xmath140j , m = 0,,1 .. @xmath660 m j@xmath141y^+_j ,- m@xmath142@xmath143k > @xmath144q=-1@xmath145(1,q=-1)@xmath146(1,q)@xmath147q=-1@xmath148^+_j , m@xmath149q>1@xmath150 consider first the two - point function @xmath151 . using ( [ cgh ] ) and ( [ metric ] ) we expect that the only non - vanishing two - point functions are @xmath152 . however , since we do not derive this result from the degeneration equation we can not exclude the possibility that other two - point functions that satisfy momentum conservation do not vanish . the choice of the non - vanishing two - point functions with gravitational descendants is basically part of the definition of the @xmath0 theory . different choices do not affect the correlators of the tachyons as they do not appear in their degeneration equation . they may show up , however , in the @xmath0 degeneration equation for gravitational descendants . since we consider the latter as an analytical continuation of the @xmath2 equation in which those correlators do not appear , they will actually have no effect on the higher point functions of descendants . the simplest correlator for which we have to use the degeneration equation is the three - point function . thus consider the correlator @xmath153 , where , for instance , @xmath154 and @xmath155 are primaries while @xmath33 is a descendant with @xmath32 . such a correlator corresponds in @xmath3 to @xmath156 with @xmath157 . there are two different ways to compute this correlator using the degeneration equation ( [ 3pte ] ) since one can choose the marked operator to be either the descendant or one of the primary fields . evidently , the result must be independent of the choice . in the @xmath2 degeneration equation scheme an operator has as many contacts as his primary index . using ( [ 3pt ] ) we see that the value of the correlator is @xmath158 when @xmath33 is the marked operator and @xmath159 when one of the primary fields is chosen as the marked operator . the result is explicitly dependent on the choice and consistency requires that the correlator vanishes . indeed the ghost number conservation law ( [ cgh ] ) reads in this case kq++ n_1 + n_2 = q + 1 and thus @xmath26 must be zero . the same problem with consistency occurs if we take two different descendants as marked operators . thus there is no non - vanishing three point function with a descendant in @xmath8 models . in order to get a consistent scheme at @xmath3 via analytical continuation of the @xmath160 degeneration equation we must require that three point functions with discrete states vanish , although momentum conservation does not forbid such correlators . thus we conclude that ^+_j_1,m_1y^+_j_2,m_2y^+_j_3,m_3= 0 [ ds3pt ] unless all the three fields are tachyons . consider now the four point function @xmath161 , where @xmath33 is a descendant and the rest are primaries . the ghost number conservation fixes @xmath162 , that is @xmath33 is the first descendant of the primary field @xmath163 . using ( [ e4pt ] ) we get : p_q+p_n_1p_n_2p_n_3= [ ds4pt ] unlike the correlator of four primaries , the correlator ( [ ds4pt ] ) does not include @xmath164 terms and thus does not depend on the choice of the kinematic region . the constraint @xmath162 on the descendant as a consequence of the ghost number conservation law is also necessary for consistency . if the constraint is not satisfied , that is @xmath165 , the value of the correlator depends on the choice of the marked operator . in such a case , when the descendant is chosen as the marked operator we get for the correlator @xmath166 while if one of the primaries , for instance @xmath154 , is taken to be the marked operator we get @xmath167 . consistency of the analytical continuation to @xmath3 requires that the only non - vanishing four point functions with discrete states are ^+_j , j-1t_n_1t_n_2t_n_3= 2j-1 [ ds4ptt ] where we factored out the @xmath168function enforcing the momentum conservation @xmath169 and used the normalization ( [ dsgd ] ) . in our picture @xmath170 corresponds to @xmath171 which is the dilaton operator , i.e. the first descendant of the puncture . thus , for @xmath172 equation ( [ ds4ptt ] ) is a special case of the dilaton equation on the sphere . let us derive the general genus @xmath15 dilaton equation . thus , consider the genus @xmath15 correlator @xmath173 . the degeneration equation reads : ( -q)p_q+1_i=1^sp_n_i_g + _ i=1^s_ij=1^sp_n_j_g p_-kp_n_i_0 = 0 [ dileq ] using ( [ cont ] ) and ( [ cgh ] ) we get p_q+1_i=1^sp_n_i_g = ( s+2g-2)_i=1^sp_n_i_g [ qdil ] at @xmath3 we have ^+_1,0 _ i=1^sy^+_j_i , m_i_g = ( 2 - 2g - s)_i=1^sy^+_j_i , m_i_g [ deq ] equation ( [ deq ] ) is expected from the topological viewpoint since the dilaton measures the euler characteristic of the riemann surface with @xmath174 punctures , that is @xmath175 . the correlator ( [ ds4pt ] ) belongs to a family of correlators that do not include @xmath164 terms , namely @xmath176 such that @xmath177 . they read _ i=1^np_n_i= 1q^n-2 ( n-3)!_i=1^n k_i ! _ i=1^n_j_i=0^k_i ( j_iq+_i ) [ sptype ] note that , as expected from the picture described in the previous section , the correlator for the normalized operators @xmath178 vanishes if @xmath179 . as each descendant @xmath180 in a correlator seems to contribute a multiplicative factor _ j=0^k(jq+ ) it implies the vanishing of the correlator at @xmath3 in theses cases . at @xmath3 ( [ sptype ] ) corresponds to _ i=1^ny^+_j_i , m_i= ( -1)^n ( n-3)!_i=1^n where @xmath181 . the first correlator with discrete states that exhibits dependence on the choice of kinematic region is the five point . the ghost number selection rule allows various possibilities . we can have in the correlator one descendant field @xmath33 with @xmath182 or two descendants with @xmath183 . two of these possibilities satisfy the condition @xmath184 and thus belong to the family of correlators that do not have @xmath164 terms as described above and are given by ( [ sptype ] ) . the third possibility is the correlator @xmath185 where @xmath162 and @xmath186 are primaries . we can compute this correlator using the degeneration equation ( [ 5pt ] ) in two ways , by choosing either the descendant or one of the primaries as the marked operator . taking the descendant field as the marked operator yields & & p_n_i=1 ^ 4p_n_i= _ i=1 ^ 4n_ip_+ n_i -1 _ ij=1 ^ 4p_n_j- + & & _ i , j=1 ; ij^4n_in_j ( + n_i + n_j - q - 2 ) p_+ n_i + n_j - q - 2 _ i , jk=1 ^ 4p_n_k+ + & & _ i=1 ^ 4n_i [ ds5pt ] it is straightforward to verify that one gets the same result if one of the primary fields is taken as the marked operator . analytical continuation of ( [ ds5pt ] ) to @xmath3 proceeds in the same manner as that of the tachyon correlators , namely @xmath187 . we get & & ^+_j , j-1_i=1 ^ 4t_n_i= -_i=1 ^ 4t_2(j-1)+n_i _ ij=1 ^ 4 t_n_j- + & & _ i , j = 1 ; ij^4(2j-1)(n_i + n_j + 2j-2 ) ( - n_i - n_j -2j+2 ) + & & t_2(j - 1 ) + n_i + n_j_i , jk=1 ^ 4t_n_k- ( 2j-1)(2j - 3 ) [ cds5pt ] let us turn now to the general case . consider the correlator @xmath188 where @xmath189 . from the ghost number conservation law one gets the constraint _ i=1^n k_i n-3 [ constr ] this constraint ( [ constr ] ) is also necessary for the consistency of the whole calculational scheme . consider , for instance , the case where @xmath190 and all the other @xmath191 zero . if we take @xmath154 as the marked operator we get by induction : _ i=2^nn_i [ npts ] on the other hand , if we take one of the primary fields for instance @xmath155 we get _ i=1^np_n_i= _ i 2^nn_i [ onpt ] and it clearly differs from ( [ npts ] ) . thus , although the constraint ( [ constr ] ) does not follow from momentum conservation in the @xmath3 model we must impose it in the analytical continuation scheme in order to have a consistent framework . an immediate consequence of the constraint is that there is no non - vanishing @xmath192 point function with more than @xmath193 descendants . evidently , the constraint is necessary for consistency but is it sufficient ? . for the @xmath160 models the degeneration equation is equivalent to the @xmath194 constraints and thus to the kdv calculational scheme and is therefore consistent . we strongly believe that the analytic continuation process to @xmath3 does not spoil this property . it is indeed the case for all the computations that we made but we do not have a complete proof for that . finally , let us consider the ward identity for the momentum one tachyon @xmath195 . topological and integrable evidence imply that it corresponds to the puncture operator in the topological phase of @xmath1 string theory @xcite . consider the genus @xmath15 correlator @xmath196 the degeneration equation at @xmath160 reads ( -q)p_1_i=1^sp_n_i_g + _ i=1^s_ij=1^sp_n_j_g p_-kp_n_i_0 = 0 or p_1_i=1^sp_n_i_g = _ i=1^sn_i(n_i+1-(q+1 ) ) p_n_i+1-(q+1)_ij=1^sp_n_j_g [ qpeq ] thus , the puncture operator shifts the momentum of a gravitational descendant and does not affect a primary . at @xmath3 we have & & t_1_i s_1y^+_j_i , m_i _ js_2y^+_j_j ,- j_j_k s_3y^+_j_k , j_k_g = + & & _ ls_1(j_j - m_l-1)^+_j_l-,m_l+ _ lis_1y^+_j_i , m_i _ js_2y^+_j_j ,- j_j _ ks_3y^+_j_k , j_k_g - + & & _ ls_2(2j_l-1)^+_j_l-,-j_l+ _ is_1y^+_j_i , m_i _ l js_2y^+_j_j ,- j_j _ ks_3y^+_j_k , j_k_g [ t1wi ] with @xmath197 . considering ( [ t1wi ] ) as the puncture equation in the topological phase of @xmath1 string theory implies that indeed the negative momentum tachyons and all the discrete states correspond to gravitational descendants while non - negative momentum tachyons correspond to gravitational primaries . note , however , that we can not take a negative tachyon to be the marked operator within the degeneration equation framework while considering him as a gravitational descendant of the zero momentum tachyon , since it will have zero number of contacts . on the other hand we saw that we can take it to be the marked operator with a formal negative number of contacts . the latter is probably related to the fact that gravitational descendants can be described via matter degrees of freedom @xcite . in @xcite it was shown that correlation functions of any @xmath198 model can be computed using recursion relations derived not from the @xmath13 constraints but rather from the @xmath199 constraints on the partition function . in terms of the topological procedure this means that only terms with one and two splittings of the riemann surface are needed in order to reproduce the results of the @xmath8 degeneration equation . it is natural to ask whether a similar statement can be made about the topological procedure for the @xmath3 model . as we shall see in this section , this is indeed the case . this will provide us with new recursion relations for tachyon correlation functions , as well as a support for the consistency of the @xmath0 degeneration equation for gravitational descendants . let us first briefly summarize the method of @xcite for the @xmath2 minimal topological models . the idea is to write down an algorithm for computing genus @xmath15 correlators @xmath200 based only on one and two splittings . consider the following relations @xmath201 where @xmath202 and @xmath203 and @xmath204 are some coefficients . @xmath205 and @xmath206 are each a sum of correlators which follow from the contacts of @xmath207 and @xmath208 respectively . the particular choice of the left hand sides ( l.h.s ) of equations ( [ mishw3 ] ) is made so that the following properties are obeyed : ( i ) the degeneration equation of only one and two splittings is required ; ( ii ) @xmath209 ; ( iii ) all the terms in @xmath205 and @xmath206 are either on genus @xmath210 or on genus @xmath15 and contain less operators then @xmath211 or contain @xmath211 operators with @xmath212 . the second property was proven by invoking certain novel polynomial identities which are generalizations of abel s identity @xcite while the third one was shown to hold using induction procedure . by subtracting the two equations ( [ mishw3 ] ) we derive the following recursion relation @xmath213 we would like to emphasize that since @xmath205 and @xmath206 include only lower correlators " , ( [ mishcc ] ) is a recursion algorithm for determining the original correlator . does this procedure hold also for the @xmath3 description of the @xmath5 string model ? . it seems that since in the derivation of ( [ mishcc ] ) we have used only the degeneration equation together with the consistency of the calculational scheme , it should hold also for the @xmath0 model . let us examine things more carefully . unlike the @xmath2 cases , at @xmath3,@xmath203 which is given by @xcite c_1= ( 2 q ) ^_1 - 2 _ i=1^_1 - 2 [ ( k_1+i)q+2 ] vanishes for certain values of @xmath214 and @xmath215 : @xmath216 , @xmath217 and @xmath218 , @xmath219 . note , however , that @xmath204 is different from zero . the difference @xmath220 is independent of @xmath221 and is given by @xcite c_1-c_2=2^_1 - 2 _ i=1^_1 - 2 ( k_1+i ) this obviously implies that even for the cases where @xmath222 the recursion relation ( [ mishcc ] ) still holds : @xmath223 . we have thus found that for the whole range of values of @xmath224 one can express the correlators in terms of lower correlators " using the degeneration equation with only one and two splittings , thus obtaining a new set of ward identities " . the main advantage of these new ward identities " for @xmath2 is that they provide a practical algorithm to compute correlators on higher genus riemann surfaces . let us briefly explain why the topological procedure of @xcite fails in these cases . when inserting ( [ i d ] ) in a degeneration of the sphere , no more than one field of ( [ i d ] ) contributes to the summation due to the ghost number conservation . on higher genus surfaces things change . when pinching a handle we insert both fields of ( [ i d ] ) on the same surface , thus the ghost number is conserved for an infinite number of fields . the @xmath164 functions that are inserted in order to avoid contributions from unphysical operators restrict the sum and prevent infinities . two point functions , however , do not include these @xmath164 functions and therefore divergences due to the infinite sum occur . for @xmath12 the regularization for these infinities is known , and for @xmath225 no new infinities occur . thus , the new set of ward identities " provide a regularized method for calculating correlators on higher genus riemann surfaces . for the case of interest @xmath3 this algorithm enables us to compute correlators of gravitational descendants on @xmath226 surfaces . furthermore one may use it to deduce the postulated regularization " which we use in the next section to retrieve the genus one tachyon correlators derived from the @xmath4 ward identities . in the appendix an explicit evaluation of @xmath227 on the sphere as well as @xmath228 on the torus are presented we have seen that at genus zero , that the topological recursion relations for tachyon correlators implied by the degeneration equation analytically continued to the @xmath0 theory coincide with the @xmath4 matrix model ward identities of the @xmath5 string . in this section we will show that the equivalence between these two sets of ward identities persists beyond genus zero . since the topological recursion relations for genus greater than zero receive contributions from pinching of handles , infinities are encountered due to the propagation of infinite number of fields in the degeneration and should be properly regularized . we will prove explicitly the equivalence for the genus one case , and discuss the equivalence for genus @xmath15 surfaces . let us begin by computing the two and three point tachyon correlators on the torus using the degeneration equation and compare to the @xmath4 ward identities . consider first the correlator for two tachyons on the torus @xmath229 . the degeneration equation reads : & & p_np_-n_1 + n_1p_-ip_-n_0 + _ i 0 ( n-1)p_-n_0 + & & + _ i 0 ( n-1)(n-2)_0 p_-jp_-n_0 = 0 [ g12pt ] the first two terms come from splitting and the rest from pinching of the handle as shown in fig.4 . note that in the latter terms the marked operator contacts two operators inserted around the pinching point . this implies that one has to insert @xmath230 additional splitted spheres . 0.5 cm the second term vanishes since it is proportional to @xmath231 , while the last two terms need to be regularized . we will now postulate a regularization for the contact operation which we justify , as we will show , by that it yields the correct results for general tachyon correlators on the torus . we also checked for several cases that it reproduces the correct results for minimal models . the regularization reads : & & reg[_i 0 _ 0 ] + & & reg[_i 0 p_n_m+1_0 ] ( 1 + n_m+1)(2 + n_m+1)p_-n_m+1 p_n_m+1 _ 0 + & & reg[_i 0 _ j = m+1^s p_n_j_0 ] ( 1+_j = m+1^s n_j)(2 + _ j = m+1^s n_j ) + & & ( _ j = m+1^s n_j ) p_n+ _ j=1^m n_j_j = m+1^sp_n_j_0 [ reg ] using the regularized contact ( [ reg ] ) in equation ( [ g12pt ] ) we get t_nt_-n_1 = ( n-1)(n+1)(n-2 ) [ nnone ] which is the correct result at the self - dual radius @xcite . as in the genus zero case we have a sign difference between the tachyon correlators derived from the @xmath3 degeneration equation and those that were derived from the matrix model . the second example in the appendix demonstrates how using the one and two splitting method one finds an identical result to that derived from the postulated regularization . in order to see the reasoning behind the definition of the regularized contact consider the @xmath4 ward identities for the two - point function on the torus . using ( [ wiw ] ) we get & & t_nt_-n_1 = res[|w_0^n-1_-n|w_0 |w_1 + |w_0^n_-n|w_1 + & & -(n-1 ) |w_0^n-2_-n|w_0^ - ( n-1)(n-2)|w_0^n-3_-n|w_0|w_0^ ] [ wt2pt ] the second term in ( [ wt2pt ] ) vanishes , as can be seen from ( [ wexp ] ) . the other terms of ( [ wt2pt ] ) coincide with those of ( [ g12pt ] ) if we use ( [ wpart ] ) and the definition of the regularized contact ( [ reg ] ) . consider next the three point function of tachyons on the torus @xmath232 , where @xmath233 . the degeneration equation reads : & & p_np_n_1p_n_2_1 + ( n p_n_2_1p_-ip_n_1_0 + ( n_1n_2 ) ) + & & + ( ( n-1)(n-2)p_n_2_0 p_-jp_n_1_0 + ( n_1n_2 ) ) + + & & ( n-1)(n-2)(n-3)_0 p_-jp_n_1_0 p_-kp_n_2_0 = 0 [ tr3pt ] where we have written only the non - vanishing terms . the first three terms in ( [ tr3pt ] ) arise from splitting and the rest from pinching . we can use now the regularization ( [ reg ] ) together with the appropriate @xmath164 terms and get , up to a sign , the right three point function for tachyons on the torus @xcite t_nt_n_1t_n_2_1 = ( n-1)(n-2)(n_1 ^ 2+n_2 ^ 2-n-2)it is more instructive , however , to compare the degeneration equation ( [ tr3pt ] ) with the @xmath4 ward identities for the three point function . using ( [ wiw ] ) we get : & & t_nt_n_1t_n_2_1 = res[|w_0^n-1(_n_1|w_0 _ n_2|w_1 + ( n_1n_2 ) ) - ( n-1)(n-2 ) |w_0^n-3 + & & ( _ n_1|w_0_n_2|w_0^ + ( n_1n_2 ) ) -(n-1)(n-2)(n-3 ) ( [ tr3pt ] ) , we have written only the non - vanishing terms . comparing ( [ tr3pt ] ) and ( [ wt3pt ] ) we see that by associating the pinching of a handle with @xmath234 , namely & & _ i 0_0 |w_0^ + & & _ i 0p_k_0 _ k |w [ wpp ] and using ( [ wexp ] ) , ( [ wpart ] ) the two equations are identical . the generalization of ( [ wpp ] ) is clear , _ i 0 _ ksp_n_k _ 0 _ ks_n_k|w_0^ our aim is to prove for the torus that the topological recursion relations of the @xmath0 theory are identical to the @xmath4 ward identities of @xmath1 string theory . consider the correlator @xmath235 . using the ward identities we have & & t_n_i = 1^m t_n_i _ n_1 ... _n_m t_n(t=0 ) = _ _n_m res(|w_0^n|w_1 ) - + & & ( n-1 ) _ _n_m res(|w^n-2_0 |w_0 ) [ npt ] we will distinguish between two types of terms , those that include a genus one correlator and arise topologically from splittings and those that include only genus zero correlators and arise topologically from pinching the handle of the torus . the former are derived from the first term in ( [ npt ] ) and the latter from the second term . consider first the terms coming from the process of splitting . they read : & & res[_i=1^m [ _ n_1^(0 ) .. _ n_i^(1) .. _n_m^(0)|w^n - m + 1_0 + + & & ( _ n_1(_n_2^(0 ) .. _ n_i^(1) .. _n_m^(0 ) ) + _ n_1^(0)_n_2(_n_3^(0 ) .. _ n_i^(1) .. _n_m^(0 ) ) + + & & + .. + _ n_1^(0 ) .. _ n_i^(1 ) .. _ n_m-1 _ n_m^(0))|w^n - m+2_0 + + & & + ... + ( _ n_1 ... _n_m - k-1 ( _ n_m - k^(0) .. _n_i^(1 ) .. _ n_m^(0 ) ) + ... ) |w^n - k_0 + ... ] + & & + ( n-1)(_n_1 .. _n_m-3(_n_m-2^(1 ) _ n_m-1^(0)_n_m^(0 ) ) + ... ) |w^n-2_0 + + & & + ( _ n_1 .. _n_m-2 ( _ n_m-1^(1)_n_m^(0 ) ) + ... ) |w^n-1_0 + _ _n_m-1 _ n_m^(1 ) |w^n_0][twwi ] this has the same structure as ( [ wwi ] ) with one of the @xmath236 being @xmath237 and the rest being @xmath238 . a general term in ( [ twwi ] ) takes the form of ( [ gt ] ) with one of the @xmath239 replaced by @xmath240 . using ( [ part ] ) we get & & t_-n_i = 1^m t_n_i_1 = _ i=1^m [ n_it_n_it_-n_i_1 - + & & ( _ i j = 1^m ( n + _ i , j l = 1^m n_l)(-n - _ i , j l = 1^m n_l ) t_n + _ i , j l = 1^m n_l t_n_it_n_j_1 - + & & _ i j , k = 1^m ( n + _ j , k l = 1^m n_l)(-n - _ j , k l = 1^m n_l)n_i t_n + _ j , k l = 1^m n_l t_n_jt_n_k_0 t_n_it_-n_i_1 ) - + & & ... - _ i i_1 .. ,i_k = 1^m ( n_i_1+ .. + n_i_k ) ( n_i_1+ .. + n_i_k)n_i + & & t_n + n_i_1 + .. + n_i_k_i_1 .. i_k j=1^m t_n_j _ 0 t_n_it_-n_i_1 - + & & - ... - ( n-1 ) _ i j , k = 1^m ( n + n_j + n_k)(-n - n_j - n_k)n_i t_n + n_j + n_k_j , k l=1^m t_n_l_0 + & & t_n_it_-n_i_1 ... -(n + n_i)(-n - n_i)n_i t_n + n_i_ij = 1^m t_n_j_0 t_n_it_-n_i_1][twwi ] consider now the correlator @xmath241 , where @xmath33 is a primary operator . taking @xmath33 as the marked operator , the degeneration equation reads : @xmath242 a general term in ( [ tqcorrel ] ) takes the form of ( [ gtp ] ) with one of the correlators is on the torus and the rest on the sphere . passing to @xmath3 it is easy to see that to each term in ( [ twwi ] ) correspond an identical term in ( [ tqcorrel ] ) and vice versa . let us turn now to the contributions from surfaces with pinched handles . the degeneration equation yields & & ( n-1)(n-2)_i 0_j k = 1^mp_n_k_0 p_n_jp_-n_j_0 + + & & ( n-1)(n-2)(n-3)_i 0 _ j , k l = 1^mp_n_l_0 p_n_jp_-n_j_0 p_n_kp_-n_k_0 + & & + ... [ pinch ] with a general term depicted in fig.5 0.5 cm _ i0 _ i sp_n_i_0 p_-i_1_j_1 s_1p_n_j_1_0 ... p_-i_n-2_j_n-2 s_n-2p_n_j_n-2_0 [ gttp ] where the sets @xmath243 are disjoint , possibly empty , and satisfy @xmath244 . the corresponding terms in the @xmath4 ward identities are & & res[(n-1)(n-2)_n_j|w_0_j k = 1^m_n_k |w_0^ + + & & ( n-1)(n-2)(n-3)_n_j|w_0_n_k |w_0_j , k l = 1^m_n_l |w_0^ + ... ] [ wpinch ] with a general term of the form res[_i s_n_i|w_0^ _ j_1 s_1_n_j_1|w_0 ... _ j_n-2 s_n-2_n_j_n-2|w_0 ] [ gwt ] using ( [ reg ] ) and ( [ wpp ] ) we establish the equivalence of equations ( [ pinch ] ) and ( [ wpinch ] ) . consider a general genus @xmath15 correlator @xmath245 in @xmath0 theory . the degeneration equation for this correlator receives contributions , as depicted in fig.6 , from splittings , for instance p_n+ _ i sn_i_j sp_n_j_g p_-n_i_k - n_i_lp_n_i_kp_n_i_l_0 .. [ split ] with @xmath246 , from pinching a dividing cycle such as p_n + _ i sn_i_j sp_n_j_g - g_1 p_-n_i_k - n_i_lp_n_i_kp_n_i_l_g_2 .. [ dcycle ] and from pinching @xmath247 non - trivial homology cycles _ j_1 ... j_h 0 _ j sp_n_j_g - h p_-n_i_k - n_i_lp_n_i_kp_n_i_l_0 .. [ hcycle ] 0.5 cm consider now a genus @xmath15 tachyons correlator @xmath248 . from the @xmath4 ward identities we expect terms with a similar structure to ( [ split ] ) , ( [ dcycle ] ) and ( [ hcycle ] ) . the term that corresponds to the splitting ( [ split ] ) is res[_i s_n_i|w_g _ n_i_k_n_i_l|w_0 ... ] [ wsplit ] the term that corresponds to the pinching of a dividing cycle ( [ dcycle ] ) is res[_i s_n_i|w_g - g_1 _ n_i_k_n_i_l|w_g_2 ... ] [ wcycle ] and the term that corresponds to the pinching of a non - trivial cycle ( [ hcycle ] ) res[_i s_n_i|w_g - h^(2h ) _ n_i_k_n_i_l|w_0 ... ] [ whcycle ] evidently , the contribution from reduction of the genus by pinching a non - trivial homology cycle needs to be properly regularized in the framework of the degeneration equation in order to be identified with ( [ whcycle ] ) . the general structures of the degeneration equation and the @xmath4 ward identities for any genus seem to be equivalent , thus it is plausible to conjecture that they coincide . however more work is needed in order to prove this in details . we studied in this paper the @xmath0 model coupled to topological gravity . this model is supposed to describe @xmath1 string theory at the self - dual radius , that is at its topological phase . we defined the model by analytical continuation of the @xmath2 degeneration equation to @xmath3 . we have shown that at genus zero the @xmath3 degeneration equation yields the genus zero @xmath4 ward identities for tachyon correlators . the positive momentum tachyons were identified as primary fields while the negative ones as their analytically continued @xmath2 auxiliary analogs . by defining the discrete states of @xmath1 string theory as the gravitational descendants of @xmath0 model we proposed a scheme for computing their correlators and computed some of them . unfortunately we could not compare the results since these correlators have not been computed by other means . we derived the puncture equation as well as the dilaton equation , with the puncture operator being the momentum one tachyon , and the dilaton operator being , as usual , its first descendant . the puncture equation supports the identification of the negative momentum tachyons as well as the discrete states as gravitational descendants . we showed that in a similar manner to the @xmath2 models , there exist recursion relations for the correlators that consist of only one and two splittings . unlike the @xmath4 ward identities which consist only of tachyon fields , the new recursion relations involve @xmath0 gravitational descendants . this provides us with other means to compute tachyon correlators and supports the consistency of the whole computational scheme . however , an open problem is to fully prove the consistency of the @xmath0 degeneration equation for gravitational descendants . a related question is what is the algebra underlying the @xmath0 degeneration equation . for @xmath2 the algebra is @xmath13 , while for @xmath3 on the space of tachyon times the algebra is @xmath4 , but the algebra on the full phase space that includes also @xmath3 gravitational descendants is not known . it is plausible to conjecture that it is also a @xmath4 algebra . as we discussed in section 7 , on @xmath226 riemann surfaces one encounters infinities in the degeneration equation framework that should be regularized . a proper regularization is known for the @xmath249 models , that is for one and two splittings of the riemann surface . thus , when using the recursion relations ( [ mishcc ] ) analytically continued to @xmath3 , we can calculate any genus @xmath15 correlator of primaries and descendants . however , the nature of this procedure is numerical and it does not easily yield closed and explicit expressions for correlators . one example of this type is described in the appendix . we therefore took another route . we postulated a regularized contact and used it to prove that the @xmath3 degeneration equation for tachyon correlators coincides with the @xmath4 ward identities on the torus . derivation of the regularized contact from the recursion relations ( [ mishcc ] ) is important and may lead to a regularized degeneration equation for @xmath3 gravitational descendants . this is currently in study . as we showed , the structure of the @xmath3 degeneration equation for tachyon correlators seems to parallel that of the @xmath4 ward identities for any genus . it is important to show that this is indeed correct in details , since it will provide us with a topological interpretation of the @xmath4 ward identities and thus with another link between the topological and integrable structures of @xmath1 string theory . there are several open problems that require further study . first , there are specific questions related to the present work . ( i ) is the analytical continuation of the @xmath2 models to @xmath3 unique ? ( ii ) can one prove that the @xmath3 model , as described in this paper , indeed describes @xmath1 string theory at the self - dual radius without the need to compare all correlation functions calculated in the present approach to those derived via other methods . second , there are questions related to more general implications + 1 . the relation between the degeneration equation and intersection theory for @xmath2 as well as for @xmath3 . can we consistently analytically continue the degeneration equation to other domains of definition of @xmath221 such as complex values , and are there corresponding physical systems . is it possible to write topological recursion relations based on contacts and degenerations for any topological string theory ? . computations done in @xcite seem to fall into this category . + 4 . finally , other generalizations defined by the assignment of different charges , e.g. with non - abelian structure , and different anomalous conservation laws may lead to topological recursion relations describing interesting theories . consider first the correlator @xmath250 which corresponds to the tachyons three point function on the sphere . the degeneration equation for this correlator consists of at least four splittings , which is the case when we choose @xmath251 as the marked operator . its numerical value at @xmath3 , before normalization , is 240 as given by ( [ 3pt ] ) . following ( [ mishw3 ] ) , let us consider the correlator @xmath252 , which vanishes by ( [ sptype ] ) . choosing @xmath253 to be the marked operator we have a degeneration equation with one and two splittings & & 0 p_2,2 p_2p_2p_6p_-10_0 = 2 q[4p_2 p_1,3 p_6 p_-10_0 + 6p_1,7 p_2 p_2 p_-10_0 + & & - 10p_1,-9 p_2 p_2 p_6_0 ] -2 q^2[4p_4 p_6 p_-10_0 + 12p_2 p_8p_-10_0 + & & - 40p_2 p_6 p_-8_0 -60p_2 p_2 p_-4_0 ] + & & + 2 q[p_2 p_2 p_-4_0p_4 p_6 p_-10_0 + 2p_2 p_6p_-8_0p_2 p_8 p_-10_0 ] [ ex1 ] where we used ( [ cont ] ) and ( [ 2pt ] ) . note that we do not have @xmath164 terms in ( [ ex1 ] ) . the reason is that the operator @xmath253 can perform a contact with two physical operators and the result is still physical : = p_n+m since the analytical continuation is defined such that we allow in the correlators operators @xmath254 with @xmath37 negative only at the final @xmath3 recursion relations , there are no @xmath164 terms in a five - point function with @xmath253 as the marked operator . note in contrast that a five - point function with @xmath255 as the marked operator includes @xmath164 terms ( [ ds5pt ] ) . the r.h.s of equation ( [ ex1 ] ) consist of the required correlator plus other terms , all of which can be computed via one and two splittings . note that the required correlator for primaries is expressed via correlators of primaries and descendants . using the degeneration equation for these terms we get @xmath256 as expected . consider next the correlator @xmath257 which corresponds to the tachyons two - point function on the torus . the degeneration equation for this correlator consists of four splittings , and its value is @xmath258 . following ( [ mishw3 ] ) , let us consider the correlator @xmath259 which vanishes by the arguments of section six . taking @xmath253 as the marked operator we get & & 0 p_2,2 p_2p_2p_-4>_1 = 2 q [ 4 < p_2 p_1,3 p_-4_1 -4p_1,-3 p_2 p_2 _ 1 ] + & & -2 q^2 [ 4p_4 p_-4_1 ] + 2 q p_2 p_2 p_-4_0 p_4 p_-4_1 + & & -1 q^2 [ -4 p_2 p_2 p_-2 p_-2_0 - 8 p_2 p_2 p_-1 p_-3_0 ] [ 2pex ] using @xmath260 as the marked operator for the computation of the first two correlators on the r.h.s of ( [ 2pex ] ) we have & & p_4 p_-4 _ 1 = 4 p_2 p_-1 p_-1_0 -8 p_1,3 p_-2_1 + & & + 1 2 p_2 p_2 p_-2 p_-2_0 + p_2 p_2 p_-1 p_-3_0 [ 2pex1 ] in order to evaluate ( [ 2pex1 ] ) we have to compute the correlator @xmath261 , which we calculate by considering the correlator @xmath262 . we get @xmath263 , and using ( [ 2pex1 ] ) we arrive at the required result p_4p_-4 _ 1 = -20 b. lian and g. zuckerman , `` new selection rules and physical states in 2d gravity , '' 266,91,21 . s. mukherji , s. mukhi , and a. sen , `` null vectors and extra states in c = 1 string theory , '' 266,91,337 . p. bouwknegt , j. mccarthy , and k. pilch , `` brst analysis of physical states for 2d gravity coupled to @xmath6 matter , '' 145,92,541 . e. witten , `` ground ring of two - dimensional string theory , '' hep - th/9108004 , 373,92,187 . klebanov and a.m. polyakov , `` interaction of discrete states in two - dimensional string theory , '' hep - th/9109032 , 6,91,3273 . e. witten , `` on string theory and black holes , '' 44,91,314 . k. li , `` topological gravity with minimal matter , '' 354,91,711 . k. li , `` recursion relations in topological gravity with minimal matter , '' 354,91,725 . e. witten , `` algebraic geometry associated with matrix models of two dimensional gravity , '' iassns - hep-91/74 . e. witten , `` the n matrix model and gauged wzw models , '' 371,92,191 . m. r. douglas , `` strings in less than one dimension and the generalized kdv hierarchies , '' 238,90,176 . r. dijkgraaf , e. verlinde , and h. verlinde , `` loop equations and virasoro constraints in nonperturbative 2d quantum gravity , '' 348,91,435 . m. fukuma , h. kawai , and r. nakayama , `` continuum schwinger - dyson equations and universal structures in two - dimensional quantum gravity , '' 6,91,1385 . m. kontsevich , `` intersection theory on the moduli space of curves and the matrix airy function , '' 147,92,1 . s. kharchev , a. marschakov , a mironov , a. morozov and a. zabrodin , `` unification of all string models with @xmath7 , '' fian / td-9/91 . m. adler and p. van moerbeke , `` the @xmath264gravity version of the witten - kontsevich model , '' 147,92,25 . d. montano and g. rivlis , solving topological 2-d quantum gravity using ward identities `` ucb - pth-92 - 30 hep - th 9210106 . r. dijkgraaf , h. verlinde and e. verlinde , `` topological strings in @xmath7 , '' 352,91,59 . e. witten , `` on the structure of the topological phase of two dimensional gravity , '' 340,90,281 . e. witten , `` two dimensional gravity and intersection theory on moduli space , '' in cambridge 1990 , proceedings , surveys in differential geometry 243 - 310 . e. verlinde , and h. verlinde , ' ' a solution of two - dimensional topological quantum gravity , " 348,91,457 . k. aoki , d. montano , j. sonnenschein the role of the contact algebra in multimatrix models `` 7,92,1755 . y. lavi and j. sonnenschein , ' ' do we need the @xmath265 constraints to solve the @xmath8 models coupled to @xmath266 gravity ? , `` hep - th/9307073 , 414,94,329 . s. mukhi and c. vafa , `` two dimensional black hole as a topological coset model of @xmath5 string theory , '' hep - th/9301083 , 407,93,667 . o. aharony , o. ganor , j. sonnenschein and s. yankielowicz , `` c=1 string theory as a topological g / g model , '' hep - th/9302027 , 305,93,35 . r. dijkgraaf , g. moore and r. plesser , unpublished . r. dijkgraaf , g. moore and r. plesser , `` the partition function of 2d string theory , '' hep - th/9208031 , 394,93,356 . g. moore and r. plesser , `` classical scattering in @xmath267 dimensional string theory , '' hep - th/9203060 , 46,92,1730 . l. bonora and c.s . xiong , ' ' two - matrix model and @xmath5 string theory , `` sissa - isas 54/94/ep , hep - th/9405004 . k. takasaki , ' ' dispersionless toda hierarchy and two @xmath1 string theory , `` kucp-0067 , hep - th/9403190 . t. eguchi and h. kanno , ' ' toda lattice hierarchy and the topological description of the @xmath5 string theory , `` ut-674 , hep - th/9404056 . a. hanany , y. oz and r. plesser , ' ' topological landau - ginzburg formulation and integrable structure of @xmath1 string theory , " iassns - hep-94/1 , hep - th/9401030 . d. ghoshal and s. mukhi , `` topological landau - ginsburg model of two - dimensional string theory , '' mri - phy/13/93 , tifr / th/93 - 62 , hep - th/9312189 . d. montano and g. rivlis , solving ( 1,q ) kdv gravity `` 360,91,524 . r. dijkgraaf and e. witten , `` mean field theory , topological field theory , and multi - matrix models , '' 342,90,486 . p. di francesco , d. kutasov , `` world sheet and space time physics in two dimensional ( super ) string theory , '' hep - th/9109005 , 375,92,119 . for a review see , for example , i. klebanov , `` string theory in two dimensions , '' lectures delivered at ictp spring school on string theory and quantum gravity , trieste , april 1991 hep - th/9108019 . g. moore , r. plesser , and s. ramgoolam , `` exact s - matrix for 2d string theory , '' hep - th/9111035 , 377,92,143 . a. lossev , `` descendants constructed from matter fields in topological landau - ginzburg theories coupled to topological gravity , '' itep preprint print-92 - 0563 ( jan 1993 ) , hep - th/9211089 . i. klebanov and d. lowe , `` correlation functions in two - dimensional gravity coupled to a compact scalar field , '' 363,91,543 . m. bershadsky , s. cecotti , h. ooguri and c. vafa , ' ' kodaira - spencer theory of gravity and exact results for quantum string amplitudes , " hupt-93/a025 , hep - th/9309140 .

let be a continuous function . we show that if is `` homogeneously multifractal '' ( in a sense we precisely define ) , then is the composition of a monofractal function with a time subordinator ( i.e. is the integral of a positive borel measure supported by ) . when the initial function is given , the monofractality exponent of the associated function is uniquely determined . we study in details a classical example of multifractal functions , for which we exhibit the associated functions and . this provides new insights into the understanding of multifractal behaviors of functions . [ 1994/12/01 ]

0804.1887

local regularity and multifractal analysis have become unavoidable issues in the past years . indeed , physical phenomena exhibiting wild local regularity properties have been discovered in many contexts ( turbulence flows , intensity of seismic waves , traffic analysis , .. ) . from a mathematical viewpoint , the multifractal approach is also a fruitful source of interesting problems . consequently , there is a strong need for a better theoretical understanding of the so - called multifractal behaviors . in this article , we investigate the relations between multifractal properties and time subordination for continuous functions . the most common functions or processes used to model irregular phenomena are monofractal , in the sense that they exhibit the same local regularity at each point . let us recall how the local regularity of a function is measured . let @xmath5 . for @xmath6 and @xmath7 , @xmath1 is said to belong to @xmath8 if there are a polynomial @xmath9 of degree less than @xmath10 $ ] and a constant @xmath11 such that , locally around @xmath12 , latexmath:[\[\label{defpoint } exponent of @xmath1 at @xmath12 is @xmath14 the singularity spectrum of @xmath1 is then defined by @xmath15 ( @xmath16 stands for the hausdorff dimension , and @xmath17 by convention ) . hence , a function @xmath0 is said to be monofractal with exponent @xmath18 when @xmath19 for every @xmath20 . for monofractal functions @xmath1 , @xmath21 , while @xmath22 for @xmath23 . sample paths of brownian motions or fractional brownian motions are known to be almost surely monofractal with exponents less than 1 . for reasons that appear below , * we focus on monofractal functions associated with an exponent @xmath24 $ ] . * more complex models had to be used and/or developed , for at least three reasons : the occurrence of intermittence phenomena ( mainly in fluid mechanics ) , the presence of oscillating patterns ( for instance in image processing ) , or the presence of discontinuities ( in finance or telecommunications ) . such models may have multifractal properties , in the sense that the support of their singularity spectrum is not reduced to a single point . among these processes , whose local regularity varies badly from one point to another , let us mention mandelbrot multiplicative cascades and their extensions @xcite , ( generalized ) multifractional brownian motions @xcite and lvy processes @xcite ( for discontinuous phenomena ) . starting from a monofractal process as above in dimension 1 , a simple and efficient way to get a more elaborate process is to compose it with a time subordinator , i.e. an increasing function or process . mandelbrot , for instance , showed the pertinency of time subordination in the study of financial data @xcite . from a theoretical viewpoint , it is also challenging to understand how the multifractal properties of a function are modified after a time change @xcite . a natural question is to understand the differences between the multifractal processes above and compositions of monofractal functions with multifractal subordinators . a function @xmath0 is said to be the composition of a monofractal function with a time subordinator ( cmt ) when @xmath1 can be written as @xmath25 where @xmath26 is monofractal with exponent @xmath27 and @xmath28 is an increasing homeomorphism of @xmath4 . in this article , we prove that if a continuous function @xmath0 has a `` homogeneous multifractal '' behavior ( in a sense we define just below ) , then @xmath1 is cmt . hence , @xmath1 is the composition of a monofractal function with a time subordinator , and shall simply be viewed as a complication of a monofractal model . this yields a deeper insight into the understanding of multifractal behaviors of continuous functions , and gives a more important role to the multifractal analysis of positive borel measures ( which are derivatives of time subordinators ) . we explain in section [ self ] and [ multi ] how this decomposition can be used to compute the singularity spectrum of the function @xmath1 . let us begin with two cases where a function @xmath1 is obviously cmt : \1 . if @xmath1 is the integral of any positive borel measure @xmath29 , then @xmath30 , where the identity @xmath31 is monofractal and @xmath1 is increasing . remark that in this case , @xmath1 may even have exponents greater than 1 . any monofractal function @xmath32 can be written @xmath33 , where @xmath34 is monofractal and @xmath35 is undoubtably an homeomorphism of @xmath4 . these two simple cases will be met again below . to bring general answers to our problem and thus to exhibit another class of cmt functions , we develop an approach based on the oscillations of a function @xmath36 . for every subinterval @xmath37 , consider the oscillations of order 1 of @xmath1 on @xmath38 defined by @xmath39 * in the sequel , we assume that @xmath1 is continuous and for every non - trivial subinterval @xmath38 of @xmath4 , @xmath40 . * this entails that @xmath1 is nowhere locally constant , which is a natural assumption for the results we are looking for . it is very classical that the oscillations of order 1 characterize precisely the pointwise hlder exponents strictly less than 1 ( see section [ prel ] ) . let us introduce the quantity that will be the basis of our construction . for every @xmath41 , @xmath42 , we consider the dyadic intervals @xmath43 , so that @xmath44 , the union being disjoint . for every @xmath41 and @xmath42 , for simplicity we set @xmath45(@xmath46 since @xmath1 is @xmath47 ) . for every @xmath41 , let @xmath48 be the unique real number such that @xmath49 we then define the intrinsic monofractal exponent of @xmath1 @xmath50 as @xmath51 this quantity @xmath50 characterizes the asymptotic maximal values of the oscillations of @xmath1 on the whole interval @xmath4 . this exponent is the core of our theorem , because it gives an upper limit to the maximal time distortions we are allowed to apply . it is satisfactory that @xmath50 has a functional interpretation . indeed , if @xmath1 can be decomposed as ( [ decomp ] ) , then the exponent of the monofractal function @xmath2 shall not depend on the oscillation approach nor on the dyadic basis . in section [ secgen ] we explain that @xmath52 where @xmath53 and @xmath54 are respectively the besov space and _ oscillation space _ on the open interval @xmath55 ( see jaffard in @xcite for instance ) . for multifractal functions @xmath1 satisfying some multifractal formalism , the exponent @xmath50 can also be read on the singularity spectrum of @xmath1 . indeed ( see section [ secgen ] ) , @xmath50 corresponds to the inverse of the largest possible slope of a straight line going through 0 and tangent to the singularity spectrum @xmath56 of @xmath1 . these remarks are important to have an idea _ a priori _ of the monofractal exponent of @xmath2 in the decomposition @xmath57 . they also give an intrinsic formula for @xmath50 . let us come back to the two simple examples above : \1 . for the integral @xmath1 of any positive measure @xmath29 , @xmath58 , hence @xmath59 , which corresponds to the monofractal exponent of the identity @xmath60 from the oscillations viewpoint . the first difficulties arise for the monofractal functions @xmath32 . when @xmath32 is monofractal of exponent @xmath61 , then we do nt have necessarily @xmath62 . we always have @xmath63 ( see lemma [ lem0 ] in section [ prel ] ) , but it is always possible to construct wild counter - examples . nevertheless , we treat in details the examples of the weierstrass functions and the sample paths of ( fractional ) brownian motions in section [ mono ] , for which the exponent @xmath64 meets our requirements . unfortunately , the knowledge of @xmath50 is not sufficient to get relevant results . for instance , consider a function @xmath1 that has two different monofractal behaviors on @xmath65 and @xmath66 $ ] . such an @xmath1 can be obtained as the continuous juxtaposition of two weierstrass function with distinct exponents @xmath67 : we have @xmath68 , and @xmath1 can not be written as the composition of a monofractal function with a time subordinator . this is a consequence of lemma [ lemmonof ] , which asserts that two monofractal functions @xmath69 and @xmath70 of disctinct exponents @xmath71 and @xmath72 never verify @xmath73 for any continuous increasing function @xmath28 ( indeed , such an @xmath3 would `` dilate '' time everywhere , which is impossible ) . we need to introduce a homogeneity condition * c1 * to get rid of these annoying and artificial cases . this condition heuristically imposes that the oscillations of any restriction of @xmath1 to a subinterval of @xmath4 have the same asymptotic properties as the oscillations of @xmath1 on @xmath4 . * condition * c1 * : * + let @xmath74 , and @xmath75 . let @xmath76 be the function @xmath77 where @xmath78 is the canonical affine contraction which maps @xmath4 to @xmath79 . condition * c1 * is satisfied for @xmath1 when there is a real number @xmath18 such that for every @xmath74 and @xmath75 , @xmath80 . hence @xmath76 is a renormalized version of the restriction of @xmath1 to the interval @xmath79 . remark that @xmath81 does not depend on the normalization factor @xmath82 . although self - similar functions are good candidates to satisfy * c1 * , a function @xmath1 fulfilling this condition does not need at all to possess such a property . in order to guarantee that @xmath1 is cmt , we strengthen the convergence toward @xmath81 . * condition * c2 * : * + assume that condition * c1 * is fulfilled . there are two positive sequences @xmath83 and @xmath84 and two real numbers @xmath85 with the following property : 1 . @xmath83 and @xmath84 are positive non - increasing sequences that converge to zero , and @xmath86 for some @xmath87 . 2 . for every @xmath74 and @xmath88 , the sequence @xmath89 converges to @xmath90 ( it is not only a liminf , it is a limit ) with the following convergence rate : for every @xmath91 $ ] , @xmath92 assuming that @xmath93 is a limit is of course a constraint , but not limiting in practice , since this condition holds for most of the interesting functions or ( almost surely ) for most of the sample paths of processes . similarly , the decreasing behavior ( [ eq1 ] ) is not very restrictive : such a behavior is somehow expected for a @xmath94 function . the convergence speed ( [ eq1 ] ) is a more important constraint , but the convergence rate we impose on @xmath83 toward 0 is extremely slow , and is realized in the most common cases , as shown below . [ maintheo ] let @xmath95 be a continuous function . assume that @xmath1 satisfies * c1 * and * c2*. then @xmath1 is cmt and the function @xmath2 in ( [ decomp ] ) is monofractal of exponent @xmath50 . such a decomposition is of course not unique : if @xmath1 is cmt and @xmath96 is @xmath97 and strictly increasing , then @xmath98 , where @xmath99 is still a monofractal function of exponent @xmath100 and @xmath101 is an increasing function . nevertheless , if two decompositions ( [ decomp ] ) exist respectively with functions @xmath69 , @xmath70 , @xmath102 and @xmath103 , then @xmath69 and @xmath70 are necessarily monofractal with the same exponent @xmath50 . this is again a consequence of lemma [ lemmonof ] . an important consequence of theorem [ maintheo ] is that the ( possibly ) multifractal behavior of @xmath1 is contained in the multifractal behavior of @xmath3 . more precisely , since @xmath3 is an increasing continuous function from @xmath4 to @xmath4 , @xmath3 is the integral of a positive measure , say @xmath29 , on @xmath4 . the local regularity of @xmath29 is classically quantified through a local dimension exponent defined for every @xmath104 by @xmath105 where @xmath106 stands for the ball ( here an interval ) with center @xmath107 and radius @xmath108 , and @xmath109 is the diameter of the set @xmath110 ( @xmath111 ) . the singularity spectrum of @xmath29 is then @xmath112 it is very easy to see that if @xmath113 , then @xmath114 . hence for every @xmath115 , @xmath116 , i.e. there is a direct relationship between the singularity spectrum of @xmath1 and the one of @xmath29 . as a conclusion , theorem [ maintheo ] increases the role of the multifractal analysis of measures , since for the functions satisfying * c1 * and * c2 * , their multifractal behavior is ruled exclusively by the behavior of @xmath29 . as an application of theorem [ maintheo ] , we will prove the following theorem [ thself ] , which relates the so - called self - similar functions @xmath1 introduced in @xcite with the self - similar measures naturally associated with the similitudes defining @xmath1 . let us recall the definition of self - similar functions . let @xmath117 be a lipschitz function on @xmath118 $ ] ( we suppose that the lipschitz constant @xmath119 equals 1 , without loss of generality ) , and let @xmath120 be @xmath121 contractive similitudes satisfying : 1 . for every @xmath122 , @xmath123 ( open set condition ) , 2 . @xmath124 ( the intervals @xmath125 form a covering of @xmath4 ) . we denote by @xmath126 the ratios of the non trivial similitudes @xmath127 . by construction @xmath128 . let @xmath129 be @xmath121 non - zero real numbers , which satisfy @xmath130 [ defiself ] a function @xmath131 is called self - similar when @xmath1 satisfies the following functional equation @xmath132 relation ( [ cond1 ] ) ensures that @xmath1 exists and is unique @xcite . let us consider the unique exponent @xmath133 such that @xmath134 this @xmath135 is indeed greater than 1 , since @xmath136 and @xmath137 for all @xmath138 by ( [ cond1 ] ) . with the probability vector @xmath139 and the similitudes @xmath140 can be associated the unique self - similar probability measure @xmath29 satisfying @xmath141 [ thself ] let @xmath1 be defined by ( [ defself ] ) . then , either @xmath1 is a @xmath142-lipschitz function for some constant @xmath87 ( expliciteley found in section [ self ] ) , or @xmath1 is cmt and there is a monofractal function @xmath2 of exponent @xmath143 such that @xmath144),\ ] ] where @xmath29 is the self - similar measure ( [ defmu ] ) naturally associated with the parameters used to define @xmath1 . the multifractal analysis of @xmath1 follows from the multifractal analysis of @xmath29 , which is a very classical problem ( see @xcite ) . the paper is organized as follows . in section [ proof ] , theorem [ maintheo ] is proved , by explicitly constructing the monofractal function @xmath2 and the time subordinator @xmath3 . section [ secgen ] contains the possible extensions of theorem [ maintheo ] , the explanation of the heuristics ( [ form ] ) , and the discussion for exponents greater than 1 . in section [ mono ] , [ self ] and [ multi ] , we detail several classes of examples to which theorem [ maintheo ] applies . first we prove that the usual monofractal functions @xmath1 with exponents @xmath61 verify * c1 * and * c2*. we prove theorem [ thself ] in section [ self ] . finally we explicitly compute and plot the time subordinator and the monofractal function for a classical family of multifractal functions @xmath145 which include bourbaki s and perkin s functions . let us finish by the direct by - products and the possible extensions of this work : the reader can check that the proof below can be adapted to more general contexts : * the dyadic basis can be replaced by any @xmath146-adic basis . * if @xmath147 converges to zero ( without any given convergence rate ) , then ( under slight modifications of @xmath148 ) the same result holds true . we focused on a simpler case , but in practice , a convergence rate @xmath86 shall always be always obtained . * the fact the the quantities @xmath149 are limits is only used at the beginning of the proof . in fact , only the existence of the scale @xmath150 $ ] such that ( [ eq1 ] ) and ( [ eq1 ] ) hold true at scale @xmath150 $ ] is determinant . in particular , the conditions may be relaxed : we could treat the case where the @xmath149 are only liminf ( and not limits ) . again , in practice they are often limits , this is why we adopted this viewpoint . for every @xmath20 , let @xmath151 be the unique dyadic interval of generation @xmath152 that contains @xmath107 , and @xmath153 , @xmath154 . let us recall the characterization of the pointwise hlder exponents smaller than 1 in terms of oscillations of order 1 ( see for instance jaffard in @xcite ) . [ lem1 ] let @xmath155 a @xmath94 function , for some @xmath156 . assume that @xmath157 . then @xmath158 in lemma [ lem2 ] , we impose some uniform behavior of the oscillations of @xmath1 on a nested sequence of coverings of @xmath4 . this is used later to prove the monofractality property of the function @xmath2 in the decomposition @xmath57 ( section [ proof ] ) , and also to decompose self - similar functions ( in section [ self ] ) . [ lem2 ] let @xmath159 be a continuous function , @xmath160 and @xmath161 . suppose that there exists an infinite sequence @xmath162 of coverings of @xmath4 such that * each @xmath163 is a finite sequence of disjoint non - trivial intervals of @xmath4 , such that @xmath164 , * @xmath165 , * each interval @xmath166 in @xmath163 is contained in a unique interval @xmath167 of @xmath168 , * for every @xmath169 and @xmath170 , we have @xmath171 , for some positive sequence @xmath172 that converges to zero when @xmath173 . then : 1 . if there exists a positive sequence @xmath174 such that for every @xmath175 , @xmath176 , then for every @xmath20 , @xmath177 . 2 . if there exists a positive sequence @xmath174 such that for every @xmath175 , @xmath178 , then for every @xmath20 , @xmath179 . remark that in part ( 2 ) of this lemma , the property needs to be satisfied only for a subsequence @xmath180 of integers . let @xmath160 , and @xmath181 small enough . for every @xmath182 , @xmath107 belongs to one interval @xmath169 , that we denote @xmath183 . denote by @xmath184 the smallest integer @xmath185 so that @xmath186 . by construction , @xmath187 and @xmath188 ( since @xmath189 ) . by the fourth property of the sequence @xmath162 , we have @xmath190 . let us start by part ( 2 ) , which is very easy to get . we have @xmath191 . applying lemma [ lem1 ] , and using that @xmath192 and @xmath193 go to zero when @xmath108 goes to zero , we obtain @xmath179 . we now focus on part ( 1 ) , which is slightly more delicate . if @xmath194 , then we have @xmath195 . if @xmath196 , then there is an integer @xmath197 ( which depends on @xmath108 ) such that @xmath198 is covered by one interval @xmath199 and not covered by any interval of @xmath200 . using the same arguments as above , we get @xmath201 ( remark that @xmath202 ) . now we have @xmath203 since @xmath193 , @xmath192 , @xmath204 and @xmath205 converge to 0 as @xmath206 , lemma [ lem1 ] yields @xmath177 . let us begin with an easy upper - bound for @xmath50 . [ lem0 ] let @xmath159 be a non - constant continuous function . then @xmath207 . we can assume without loss of generality that @xmath208 . let @xmath41 . by construction , @xmath209 . in order to have ( [ defhj ] ) , we necessarily have @xmath210 . hence the result . [ lemmonof ] let @xmath69 and @xmath70 be two real monofractal functions on @xmath4 of disctinct exponents @xmath211 . there is no continuous strictly increasing function @xmath28 such that @xmath73 . suppose that such a function @xmath3 exists . let @xmath212 . this function @xmath3 is lebesgue - almost everywhere differentiable . there is a set @xmath213 of positive lebesgue measure such that for every @xmath214 , @xmath215 . around such a @xmath107 , we have @xmath216 . consequently , since @xmath217 , for every @xmath218 small enough we have @xmath219 this shows that @xmath220 . using again that @xmath217 , there is a sequence @xmath221 converging to zero such that for every @xmath182 , @xmath222 . choosing @xmath223 so that @xmath224 , we see that @xmath225 this holds for an infinite number of real numbers @xmath226 converging to zero . hence @xmath227 , which contradicts @xmath228 . note first that the previous results hold in the case where a @xmath146-adic basis , @xmath229 , is used instead of the dyadic basis . in fact , there is a functional interpretation of the exponent @xmath50 , independent of any basis , provided by the _ oscillation spaces _ of jaffard @xcite and the besov spaces . let us recall their definition , that we adapt to our context of nowhere differentiable functions . let @xmath1 be a @xmath94 function on @xmath55 , where @xmath94 is the global homogeneous hlder space and @xmath156 . since @xcite where the theoretical foundations of multifractal analysis of functions were given , a quantity classically considered when performing the multifractal analysis of @xmath1 is the scaling function @xmath230 . later , in @xcite , jaffard also proved the pertinency in multifractal analysis of his oscillation spaces @xmath231 , whose definitions are based on wavelet leaders ( we do not need much more details here ) . he also considered the associated scaling function @xmath232 . finally , still in @xcite , jaffard studied the spaces @xmath233 , which are closely related to our exponent @xmath50 , defined as follows : denote , for @xmath41 and @xmath234 , @xmath235}(z)$ ] , and consider the associated scaling function ( we assume hereafter that @xmath1 is nowhere differentiable , as in theorem [ maintheo ] ) @xmath236 for @xmath237 fixed , it is obvious that there is a constant @xmath238 such that @xmath239 since @xmath240 . as a consequence , @xmath241 comparing the definition of @xmath50 with this formula , we easily see that @xmath50 is the unique positive real number such that @xmath242 . the main point is that the three scaling functions @xmath243 , @xmath244 and @xmath245 coincide as soon as @xmath246 @xcite , and @xmath247 . using the property of the besov domains , we have @xmath248 consider the scaling function @xmath249 above . then for any function @xmath1 having some global hlder regularity @xcite , @xmath1 is said to obey the multifractal formalism for functions if its singularity spectrum is obtained as the legendre transform of its scaling function , i.e. @xmath250 in particular , since @xmath251 , we always have @xmath252 ( by using @xmath253 in the inequality above ) . moreover , assume that @xmath254 exists and that @xmath1 satisfies the multifractal formalism associated with @xmath244 at the exponent @xmath255 . this means that the inequality above holds true for @xmath256 , i.e. @xmath257 . from the two last properties we get that @xmath258 is the slope of the tangent to the ( concave hull of the ) singularity spectrum of @xmath1 , as claimed in the introduction . the functions @xmath2 and @xmath259 are constructed iteratively . first remark that since @xmath260 converges to zero , one can also assume , by first replacing @xmath261 by @xmath262 and then by imposing that @xmath260 is non - increasing , that the sequence @xmath260 satisfies : * for every @xmath41 , @xmath263 , * @xmath264 is now a non - decreasing sequence and @xmath265 when @xmath266 , * @xmath260 still satisfies ( [ eq1 ] ) and ( [ eq1 ] ) . assume that conditions * c1 * and * c2 * are fulfilled . the exponent @xmath267 is the limit of the sequence @xmath48 , so there exists a generation @xmath268 such that for every @xmath269 , @xmath270 we set @xmath271 , and by construction we have @xmath272 we then define the first step of the construction of the function @xmath3 : we set @xmath273 this function @xmath274 is strictly increasing , continuous and affine on each dyadic interval . moreover , @xmath275 . let us denote @xmath276 the image of the interval @xmath277 by @xmath274 , for every @xmath278 . the set of intervals @xmath279 clearly forms a partition of @xmath280 . one remarks that @xmath281 the first step of the construction of @xmath2 is then naturally achieved as follows : we set @xmath282 this function @xmath283 maps any interval @xmath276 to the interval @xmath284 , and thus satisfies : @xmath285 as a last remark , there are two real numbers @xmath286 such that for every @xmath138 @xmath287 . without loss of generality , we can assume that @xmath288 and @xmath289 ( @xmath290 and @xmath135 appear in condition * c2 * ) by changing @xmath290 into @xmath291 and @xmath292 , so that @xmath293 we perform the second step of the construction . let us focus on one interval @xmath294 , on which we refine the behavior of @xmath274 . by condition * c2 * and especially ( [ eq1 ] ) , we have @xmath295}-1 } ( \omega _ { [ j_0\eta_{j_0}],'k } ( z_{j_0,k}))^{1/h_1 } = 1,\ ] ] where @xmath296}(z_{j_0,k})$ ] satisfies @xmath297 . let @xmath298 $ ] , hence @xmath299 $ ] . remark that , by ( [ eq1 ] ) , we have for every @xmath300}-1\}$ ] @xmath301 \beta } \leq | \omega _ { [ j_0\eta_{j_0}],k ' } ( z_{j_0,k } ) | \leq 2^{- [ j_0\eta_{j_0 } ] \alpha}.\ ] ] now , remembering the definition of @xmath302 , we obtain that for every @xmath303 , @xmath304 consequently , ( [ eq5 ] ) is equivalent to @xmath305 and thus @xmath306 we now define the function @xmath102 as a refinement on @xmath274 on the dyadic interval @xmath294 . we set for every @xmath307 and for @xmath308 @xmath309 this can be achieved simultaneously on every dyadic interval @xmath294 , @xmath310 , by using the same generation @xmath311 for the subdivision ( indeed , condition * c2 * ensures that the convergence rate of @xmath312 does not depend on @xmath138 ) . the obtained function is again an increasing continuous function , affine on every dyadic interval of generation @xmath311 . let us denote @xmath313 the image of the interval @xmath314 by @xmath102 , for every @xmath315 . the set of intervals @xmath316 again forms a partition of @xmath280 . we get @xmath317 but the main point is that we did not change the size of the oscillations of @xmath274 on the dyadic intervals of generation @xmath318 , i.e. @xmath319 . the second step of the construction of @xmath2 is realized by refining the behavior of @xmath283 : set @xmath320 this function @xmath69 maps any interval @xmath313 to the interval @xmath321 , and thus satisfies : @xmath322 finally , we want to compare the size of the interval @xmath313 with the size of its father interval ( in the preceding generation ) @xmath323 . for this , let us choose @xmath324 and @xmath325 are such that @xmath326 ( hence @xmath138 can be written @xmath327 with @xmath328 ) . then , by ( [ eq21 ] ) , @xmath329 using ( [ eq20 ] ) we get @xmath330\beta / h_1}. \end{aligned}\ ] ] on the other side , we know by ( [ eq22 ] ) that @xmath331 , hence @xmath332 where the left inequality simply comes from the fact that @xmath326 . this procedure can be iterated . assume that the sequences @xmath333 , @xmath334 and @xmath335 are constructed for every @xmath336 , and that they satisfy : 1 . for every @xmath337 , @xmath338 $ ] and @xmath339 , 2 . for every @xmath337 , @xmath340 is a continuous strictly increasing function , affine on each dyadic interval @xmath341 and if we set @xmath342 , then @xmath343 is the unique integer such that @xmath344 , for @xmath345 , 3 . for every @xmath346 , the set of intervals @xmath347 forms a partition of @xmath280 . 4 . for every @xmath337 , if @xmath348 , then @xmath349 5 . for every @xmath337 , for @xmath350 , @xmath351 6 . for every @xmath337 , for every @xmath352 , we have @xmath353 for every @xmath354 . the last item ensures that once the value of @xmath355 at @xmath356 has been chosen , every @xmath357 , @xmath358 , will take the same value at @xmath356 . to build @xmath359 and @xmath360 , the procedure is the same as above . we use @xmath361 $ ] , and we focus on one interval @xmath362 . we have by ( [ eq1 ] ) @xmath363 where @xmath364}(z_{j_n , k})$ ] satisfies @xmath365 . we have for every @xmath366}-1\}$ ] @xmath367 \beta } \leq | \omega _ { [ j_n\eta_{j_n}],k ' } ( z_{j_n , k } ) | \leq 2^{- [ j_n\eta_{j_n } ] \alpha}.\ ] ] and @xmath368 the same manipulations as above yield @xmath369 then @xmath359 is a refinement on @xmath370 : for every @xmath371 and for @xmath372 @xmath373 remark that for every @xmath374 , for every @xmath375 , @xmath376 . this can be achieved simultaneously on every dyadic interval @xmath362 , @xmath377 , by using the same generation @xmath378 for the subdivision . the obtained function is again an increasing continuous function which is affine on every dyadic interval of generation @xmath378 . we then define @xmath360 by @xmath379 for @xmath350 . let @xmath380 be the image of the interval @xmath381 by @xmath359 , for every @xmath382 . this function @xmath360 maps any interval @xmath380 to the interval @xmath383 , and thus satisfies : @xmath384 with @xmath385 at this point , all the items of the iteration are ensured , except the item ( 4 ) . we prove it now . as above , let us choose @xmath386 and @xmath387 are such that @xmath388 , and let @xmath389 be such that @xmath390 . we have by ( [ eq31 ] ) @xmath391 then , by ( [ eq20 ] ) , @xmath392\beta / h_{n+1}}. \end{aligned}\ ] ] as above , since by ( [ eq22 ] ) that @xmath393 , we have @xmath394 the convergence of the sequence @xmath397 to a function @xmath3 is almost immediate . indeed , each @xmath370 is an increasing function from @xmath4 to @xmath4 , and by item ( 5 ) of the iteration procedure , for every @xmath41 , for every @xmath398 , @xmath399 is constant as soon as @xmath400 . recall that for every @xmath401 and @xmath138 , @xmath402 . by ( [ eq40 ] ) , and using ( [ eq20 ] ) , we obtain that @xmath403 and iteratively @xmath404 for some constant @xmath11 . hence the sequence @xmath405 converge exponentially fast to zero , with an upper bound independent of @xmath138 . as a consequence , if @xmath406 , then @xmath407 this cauchy criterion immediately gives the uniform convergence of the function series @xmath397 to a continuous function @xmath3 , whose value at each dyadic number is known as explained just above . the limit function @xmath3 is also strictly increasing , since it is strictly increasing on the dyadic numbers . the convergence of the functions sequence @xmath395 is then straightforward . indeed , each @xmath370 is an homeomorphism of @xmath4 , and admits a continuous inverse @xmath408 . we thus have , for every @xmath182 , @xmath409 . the series @xmath410 also converges uniformly on @xmath4 . since @xmath1 is uniformly continuous on @xmath4 , @xmath411 converges uniformly to a continuous function @xmath412 . remark that @xmath3 also admits an inverse function @xmath413 , and that @xmath414 . obviously , @xmath3 is a strictly increasing function from @xmath4 to @xmath4 , which is what we were looking for . all we have to prove the monofractality property of @xmath2 . this will follow from lemma [ lem2 ] . it has been noticed before that if we set , for every @xmath415 , @xmath416 , then every @xmath163 forms a covering of @xmath4 constituted by pairwise distinct intervals . we obviously have : * @xmath417 ( using the remarks of section [ secconv ] above ) , * @xmath162 is a nested sequence of intervals , * by item ( 4 ) of the iteration procedure , if @xmath418 , then we have @xmath419 , , with @xmath420 . this sequence @xmath172 converges to zero , since @xmath421 converges to zero and @xmath422 converges to @xmath61 . in order to apply lemma [ lem2 ] and to get the monofractality property of @xmath2 , it is thus enough to prove the last required properties , i.e. there is a positive sequence @xmath423 converging to zero such that for every @xmath169 , @xmath424 . for this , let @xmath182 and @xmath169 . this interval @xmath166 can be written @xmath425 for some @xmath426 . we have @xmath427 by construction , and @xmath428 . we just have to verify that @xmath429 , for some @xmath430 independent of @xmath138 . we have @xmath431 writing that @xmath432 and @xmath433 , we obtain @xmath434 let us denote @xmath435 , and @xmath436 for every @xmath401 and @xmath138 . comparing the last inequality with the desired result , all we have to show is that @xmath437 indepently of @xmath138 . this is obtained as follows : start from @xmath311 , that we suppose ( without loss of generality ) to be greater than 100 . recall that , by the remarks made at the beginning of section [ proof ] , we assumed that for every @xmath182 , @xmath438 . subsequently , every term @xmath439 is greater than @xmath440 , where @xmath441 is the sequence defined recursively by @xmath442 and @xmath443 . let us study the growth rate of such a sequence . it is obvious that @xmath444 . we set @xmath445 . we have @xmath446 for every @xmath185 , with @xmath447 that can be taken less than @xmath448 since @xmath449 is large enough . in particular , since @xmath450 , @xmath451 . recursively , if we assume that @xmath452 , then @xmath453 . hence the sequence @xmath441 converges to @xmath454 faster than @xmath455 , and thus faster than any polynomial @xmath456 . ( we could be more precise , and prove using the same arguments that the growth rate of @xmath441 is exactly @xmath455 . ) let us now find an upper bound for @xmath457 . using the item ( 1 ) of condition * c2 * , we have that @xmath458 . using the lower bound we found for @xmath459 with @xmath447 chosen small enough , we get that @xmath460 . the crucial point is that @xmath461 . now , since @xmath462 is a sequence increasing toward @xmath454 , rewrite the left term of ( [ eq9 ] ) as @xmath463 , where @xmath464 . by a classical caesaro method , we get that ( [ conv ] ) is true , independtly of @xmath138 . this directly implies , by ( [ eq9 ] ) , that independlty of @xmath138 , @xmath465 , for some sequence @xmath466 that converges to zero . we now apply lemma [ lem2 ] , which implies that @xmath2 is monofractal with exponent @xmath61 . let us finally say a few words about functions having regularity exponents greater than 1 . the presence of a polynomial in the definition ( [ defpoint ] ) of the pointwise hlder exponent is a source of problems when analyzing the local regularity after time subordination . indeed , suppose that a continuous function @xmath69 behaves like @xmath467 ( @xmath468 ) around a point @xmath12 , and that another continuous function @xmath70 behaves like @xmath469 ( @xmath470 ) around @xmath471 . then @xmath472 , @xmath473 , but @xmath474 , which is different than the expected regularity @xmath475 . applying the construction above and getting a decomposition of a function @xmath1 as @xmath57 , because of such problems , we did nt find any way to guarantee the monofractality of @xmath2 . this is related to the fact that , still for the just above toy example , @xmath476 when @xmath108 is small enough , while one would expect @xmath477 . the use of oscillations of order greater than 2 ( so that @xmath478 ) was not sufficient for us to prove theorem [ maintheo ] for exponents greater than 1 . an unsatisfactory result is the following : if @xmath1 has all its pointwise hlder exponents less than @xmath479 , then @xmath480 has all its exponents smaller than 1 ( @xmath481 is the weierstrass function ( [ defwei ] ) monofractal with exponent @xmath482 ) , and one shall try to apply theorem [ maintheo ] to this function . as a consequence , this problem is still open and of interest . it is satisfactory to check that classical monofractal functions verify the conditions of theorem [ maintheo ] , and that the exponent @xmath50 is actually equal to their monofractal exponent . the proofs below are also representative examples of the method used to get convergence rates for @xmath48 to @xmath50 . let @xmath468 , @xmath483 and @xmath484 be three real numbers . let @xmath485 be a bounded function that belongs to the global hlder class @xmath486 . consider the weierstrass - type function @xmath487 by @xcite , either the function @xmath1 is @xmath488 , or it is monofractal with exponent @xmath290 . for @xmath489 , we obtain the classical weierstrass functions monofractal with exponent @xmath290 . in fact , it is proved in @xcite that , if @xmath490 ( which is our assumption from now on ) , then there is a constant @xmath491 such that @xmath492 as a direct consequence , @xmath493 , and obviously @xmath494 . let us find the convergence rate of @xmath48 toward @xmath50 . we are looking for a value of @xmath212 and for a scale @xmath318 for which @xmath495 , for every @xmath269 . let @xmath41 . we have , by ( [ minor1 ] ) , @xmath496 for @xmath447 small , @xmath497 , and thus our constraint is reached as soon as @xmath498 . this leads to @xmath499 there is a generation @xmath318 such that the last inequality is realized by @xmath500 . subsequently , one necessarily has @xmath501 for every @xmath269 , since @xmath502 and the mapping @xmath503 is increasing with @xmath504 . hence @xmath505 . using the same method , we obtain @xmath506 for @xmath269 . finally , we have found @xmath318 large enough so that for every @xmath269 , @xmath507 , where we have set @xmath508 . for every @xmath509 and @xmath510 , we easily get the same convergence rates of @xmath511 toward @xmath290 from the self - affinity property of the weierstrass functions . more precisely , fix @xmath512 and @xmath513 , and let @xmath514 . remark that by construction of @xmath76 , we have @xmath515 . we are looking for a value of @xmath447 for which @xmath516 for every @xmath152 large enough . by ( [ minor1 ] ) ( used two times ) , and remarking that there are @xmath517 dyadic intervals of generation @xmath152 included in @xmath79 , we get @xmath518 the same computations as above yield that , if we impose @xmath519 and @xmath520 , then for every @xmath521 , @xmath522 . similarly , we obtain @xmath523 for @xmath521 . finally , for every @xmath521 , @xmath524 , and @xmath525 . consequently , the weierstrass functions satisfy * c1 * and * c2 * with @xmath526 , and they are also monofractal from our viewpoint . classical estimations on the oscillations of sample paths of brownian motions @xmath527 yield ( @xcite ) @xmath528 hence , by a classical borel - cantelli argument , with probability one , there is a generation @xmath529 such that for every @xmath530 , we have the bounds @xmath531 for the oscillations . the same computations as for the weierstrass functions show that there is a generation @xmath318 such that if @xmath532 , then @xmath533 where @xmath534 ( for some suitable constant @xmath11 ) . as a consequence , @xmath535 for every @xmath269 . the self - similarity property of brownian motions yields that for every @xmath509 and @xmath510 , for every @xmath536 , @xmath537 , where @xmath538 , for some constant @xmath539 independent of @xmath512 and @xmath513 . we omit the details here , that can be easily checked by the reader . consequently , a sample path of brownian motion satisfies with probability one * c1 * and * c2 * , with @xmath540 . similar estimations on the oscillations of fractional brownian motions @xmath541 of hurst exponent @xmath504 lead to the same almost sure result for the sample paths , which also satisfy almost surely * c1 * and * c2 * with @xmath542 . we consider the class of self - similar functions defined in definition [ defiself ] , with the parameters of @xmath1 and the contractions @xmath543 satisfying ( [ cond1 ] ) . the multifractal analysis of such a function @xmath1 is performed in @xcite . here we are going to prove that , under the conditions ( [ cond1 ] ) on the @xmath544 and the @xmath543 , @xmath1 is cmt , and that the multifractal behavior of @xmath1 can be directly deduced from this analysis . it is a case where our analysis provides a natural way to compute the singularity spectrum of @xmath1 . let us introduce some notations : for every @xmath182 , for every @xmath545 , we denote @xmath546 the interval @xmath547 . the integer @xmath185 being given , the open intervals @xmath548 are pairwise disjoint , and the union of the closed intervals @xmath549 equals @xmath4 . now fix an integer @xmath182 and a sequence @xmath545 . the interval @xmath549 has a length equal to @xmath550 . finally , by iterating @xmath185 times formula ( [ defself ] ) , we get that for every @xmath551 , @xmath552 recall that @xmath553 is defined in ( [ cond1 ] ) . [ lem3 ] let @xmath554 . then either @xmath1 is a @xmath142-lipschitz function , or there is a constant @xmath491 such that for every @xmath182 , for every @xmath545 , @xmath555 we first find an upper - bound for @xmath556 . we use the iterated formula ( [ decompf ] ) . let @xmath185 and @xmath545 . remark that when @xmath107 ranges in @xmath557 , @xmath558 ranges in @xmath4 . hence the oscillation of the first term of ( [ decompf ] ) is upper - bounded by @xmath559 . now , for every @xmath560 , when @xmath107 ranges in @xmath557 , @xmath561 ranges in @xmath562 . using that @xmath117 is a lipschitz function , we get that the oscillation of each term of the form @xmath563 is upper bounded by @xmath564 . finally , we obtain using ( [ cond1 ] ) @xmath565 \\ & \leq & | { \lambda}_{\ep_1 } { \lambda}_{\ep_2 } \ ! \cdot \ ! \cdot \ ! \cdot \ ! { \lambda}_{\ep_n } | \big [ 1 + \sum_{k=1}^{n-1 } \chi_{\max}^k \big ] \leq c_1 | { \lambda}_{\ep_1}{\lambda}_{\ep_2}\ \ ! \cdot \ ! \cdot \ ! { \lambda}_{\ep_n } |,\end{aligned}\ ] ] where @xmath566 . we now move to the lower bound . assume that @xmath1 is not @xmath142-lipschitz . there are two real numbers @xmath567 such that @xmath568 , for some @xmath569 . let @xmath185 and @xmath545 . let us call @xmath570 and @xmath571 . we obviously have @xmath572 , and thus @xmath573 . using again ( [ decompf ] ) , we get by the same lines of computations as above @xmath574 \\ \geq & & \hspace{-5 mm } | { \lambda}_{\ep_1 } { \lambda}_{\ep_2 } \ ! \cdot \ ! \cdot \ ! \cdot \ ! { \lambda}_{\ep_n } |\cdot |z(t'_0 ) -z(t_0)| \big [ 1- \frac { 1 } { \kappa + \eta}\sum_{k=1}^{+\infty } \chi_{\max}^k \big ] \geq c_2 \cdot | { \lambda}_{\ep_1}{\lambda}_{\ep_2}\ \ ! \cdot \ ! \cdot \ ! \cdot \ ! { \lambda}_{\ep_n } |,\end{aligned}\ ] ] where @xmath575 by assumption . finally , ( [ majmin1 ] ) is proved with @xmath576 . in order to prove that the function @xmath1 ( [ defself ] ) satisfies our conditions * c1 * and * c2 * , we introduce a self - similar measure @xmath29 , whose multifractal behavior will be compared with the one of @xmath1 , and the notion of multifractal formalism . let us consider the exponent @xmath133 such that ( [ defbeta ] ) holds and the associated self - similar measure @xmath29 defined by ( [ defmu ] ) @xmath577 in our case where the similitudes do not overlap , it is easily checked that by construction , for every @xmath185 and @xmath545 , we have @xmath578 . this class of measures has been extensively studied @xcite . for instance , the multifractal analysis of @xmath29 is very well known . for this , let us introduce the so - called @xmath579-spectrum of @xmath29 defined by @xmath580 we only recall the properties we need @xcite [ prop4 ] 1 . for every @xmath581 , @xmath582 is the unique real number satisfying the equation @xmath583 . the mapping @xmath584 is analytic on its support . moreover , the liminf used to define @xmath585 is in fact a limit for every @xmath586 such that @xmath585 is finite . 2 . there is an interval of exponents @xmath587 $ ] such that for every @xmath588 , @xmath589 , where @xmath590 is by definition the legendre transform of @xmath591 . 3 . if @xmath592 , then @xmath593 . 4 . there is @xmath594 such that for every @xmath595 large enough , @xmath596 . part ( 2 ) above is known as the multifractal formalism for measures , when it holds . let us come back to the function @xmath1 . the reader can check that such a function @xmath1 satisfies * c1 * and * c2 * , and is thus cmt . here we propose a quick proof of theorem [ thself ] , especially adapted to this case . the aim is to prove that @xmath1 can be written @xmath597 each dyadic interval @xmath598 is included in one dyadic interval @xmath599 , and contains a dyadic interval @xmath600 , such that @xmath599 and @xmath600 can be written respectively @xmath601 and @xmath602 with @xmath603 , for some constant @xmath11 independent of @xmath152 and @xmath138 . consequently , @xmath604 , and thus @xmath605 using now the self - similarity properties of the measure and the open set condition , we see that @xmath606 and @xmath607 . hence , combining this with the last double inequality , we obtain that for every @xmath152 and @xmath138 @xmath608 for another constant @xmath11 , i.e. proposition [ lem3 ] extends to all dyadic intervals . let us check that @xmath1 satisfies conditions * c1 * and * c2*. remark that , because of ( [ majmin3 ] ) , @xmath609 let @xmath610 . using part ( 4 ) of proposition [ prop4 ] to find upper- and lower - bounds for @xmath611 uniformly in @xmath138 , we obtain @xmath612 hence , the same computations as in the case of weierstrass functions lead to the following choice : for some @xmath318 large enough , we set @xmath613 , and thus for every @xmath614 , @xmath615 . let now @xmath616 be two integers , @xmath617 , and focus on @xmath81 . the same computations as above and as in the weierstrass case yield for @xmath618 @xmath619 first notice that @xmath620 . then we remark that @xmath621 combining these inequalities we get @xmath622 let us focus on @xmath623 . since @xmath133 , we have @xmath624 . this quantity is lower bounded by @xmath625 for some constant @xmath626 ( uniformly in @xmath138 and @xmath513 ) , since the ratio of the @xmath29-measures of a dyadic interval and its father ( in the dyadic tree ) is uniformly upper- and lower - bounded for our dyadic self - similar measure @xmath29 . finally , we obtain @xmath627 hence if we fix @xmath628 , then the sum above is greater than 1 as soon as @xmath629 $ ] and @xmath630 . thus for @xmath629 $ ] , @xmath631 with @xmath632 . similarly one shows that for @xmath629 $ ] , @xmath633 , and * c2 * holds true for @xmath1 . applying theorem [ maintheo ] yields that @xmath1 is cmt and can be written as @xmath634 , where @xmath2 is monofractal of exponent @xmath143 . applying directly the construction of section [ proof ] , we find a function @xmath2 monofractal with exponent @xmath143 and a strictly increasing function @xmath3 such that @xmath635 . one can even enhance this result as follows . following the proof of section [ proof ] , we see that for every @xmath246 , for every @xmath636 , @xmath637 where @xmath638 is a positive sequence decreasing to zero and @xmath29 is defined by ( [ defmu ] ) . let us denote by @xmath259 the integral of the self - similar measure @xmath29 , i.e. for @xmath104 @xmath639)$ ] . we claim that @xmath640 for some function @xmath2 which belongs to @xmath641 , for every @xmath569 . indeed , define for every @xmath104 @xmath642 . this is possible since @xmath259 is an homeomorphism of @xmath4 . by construction , for every @xmath246 , for every @xmath636 , @xmath643 , thus by the inequality above , @xmath644 where we used that @xmath645 . now the sets of intervals @xmath646 obviously forms a covering of @xmath4 to which lemma [ lem2 ] can be applied with @xmath647 . finally , we find that @xmath648 , where @xmath70 is clearly monofractal with exponent @xmath143 since @xmath2 and @xmath69 are monofractal respectively with exponents @xmath143 and @xmath649 ( the resulting function @xmath70 is monofractal since the oscillations of @xmath2 and @xmath69 are upper and most important lower bounded on every interval ) . we recall the contruction of multifractal functions of @xcite , which somehow generalizes the bourbaki s and perkin s functions . let us consider the function @xmath650 defined for @xmath651 as the limit of an iterated construction : start from @xmath652 on @xmath4 , and define @xmath653 recursively on @xmath4 by the following scheme : suppose that @xmath654 is continuous and piecewiese affine on each triadic interval @xmath655 $ ] , @xmath656 . then @xmath657 is constructed as follows : on each triadic interval @xmath655 $ ] , @xmath657 is still a continuous function which is affine on each triadic subinterval @xmath658 $ ] included in @xmath655 $ ] , and @xmath659 this simple construction is better explained by the figure [ fig1 ] . it is straightforward to see that the sequence @xmath660 converges uniformly to a continuous function @xmath650 as soon as @xmath661 . bourbaki s function is obtained when @xmath662 , while perkin s function corresponds to @xmath663 . for @xmath665 , the function is simply the integral of a trinomial measure of parameters @xmath666 , hence its singularity spectrum is completely known . we are going to explain why the functions @xmath650 , when @xmath667 , satisfy our assumptions , and thus can be written as the composition of a monofractal function @xmath2 ( with an exponent @xmath61 we are going to determine ) with an increasing function . we will also deduce from this study the singularity spectrum of @xmath650 . remark first that the slope of @xmath669 on @xmath670 $ ] is @xmath671 , it is @xmath672 on @xmath673 $ ] and @xmath671 on @xmath674 $ ] . iteratively , if @xmath675 and @xmath656 , we write @xmath676 , with @xmath677 . then the slope of @xmath654 on @xmath655 $ ] is simply @xmath678 where @xmath679 is the number of integers @xmath680 such that @xmath681 ( for @xmath682 ) in the triadic decomposition of @xmath683 . let us consider the trinomial measure @xmath684 of parameters @xmath685 . then it is obvious that the absolute value of the slope of @xmath654 on each triadic interval @xmath655 $ ] can be written as @xmath686)3^j ( 4a-1)^j.$ ] as a final remark , we also notice that the oscillations of @xmath650 on each triadic interval @xmath655 $ ] is the same as the oscillations of @xmath687 on each triadic interval @xmath655 $ ] , which is equal to @xmath688 times the slope , i.e. @xmath689 ) ( 4a-1)^j.\ ] ] let @xmath690 . let us compute the sum of the oscillations of @xmath650 at generation @xmath152 . we have @xmath691}(z_a))^{q } = \sum_{k=0}^{3^j-1 } ( \mu_a([k3^{-j } , ( k+1)3^{-j}]))^{q } 3^{qj\log_3 ( 4a-1)}.\ ] ] let us explain now how we easily compute the exponent @xmath692 such that ( [ defh ] ) holds true . for a multinomial measure @xmath684 ( in fact , for any positive borel measure ) , it is very classical in multifractal analysis to introduce the functions @xmath693 and the scaling function @xmath694 defined for @xmath690 as ( [ deftau ] ) but in the triadic basis : @xmath695))^{q } } { -j}.\end{aligned}\ ] ] in our simple case , it is easy to see that for every @xmath41 and @xmath690 @xmath696 what matters to us is the value of @xmath586 for which the sum in ( [ eq10 ] ) equals 1 . let us write this specific value @xmath586 as @xmath697 , for some @xmath18 . when this sum is 1 , then we have @xmath698))^{1/h } 3^{j(\log_3 ( 4a-1))/h}=1.\ ] ] let @xmath692 be the solution of the equation @xmath699 , which is equivalent to @xmath700 this solution is positive , unique , and strictly smaller than 1 . hence , in this case , the monofractal exponent @xmath692 is defined through an implicit formula . in order to get the whole condition * c2 * , it suffices to notice that any rescaled function @xmath701 ( as defined in ( [ eq00 ] ) , but here with triadic intervals ) is actually equal to @xmath650 ( if @xmath650 is increasing on @xmath702 $ ] ) or to @xmath703 ( if @xmath650 is decreasing on @xmath702 $ ] ) . hence @xmath704 is a limit for every @xmath616 , and is even constant equal to @xmath692 . thus * c2 * is satisfied . we can then apply theorem [ maintheo ] , and @xmath650 is the composition of a monofractal function @xmath705 of exponent @xmath692 with an increasing function @xmath650 . for @xmath662 , we see that @xmath706 is the solution to ( [ defha ] ) , since @xmath707 . we have plotted in figure [ fig2 ] the bourbaki s function @xmath708 , its corresponding time change @xmath709 and the corresponding monofractal function @xmath710 of exponent @xmath711 such that @xmath712 . in this case , we can even go further and compute the singularity spectrum of @xmath650 . the trinomial measure satisfy the multifractal formalism for measures , i.e. the singularity spectrum of @xmath29 is given by the legendre transform of @xmath713 : @xmath714 for every @xmath715 $ ] . [ fig2 ] on the left , the multifractal time change @xmath709 in the middle , and on the right the monofractal function @xmath710 of exponent @xmath711 such that @xmath716 . * bottom : * singularity spectra of @xmath717 on the left , of @xmath708 on the right.,title="fig:",width=529,height=188 ] + on the left , the multifractal time change @xmath709 in the middle , and on the right the monofractal function @xmath710 of exponent @xmath711 such that @xmath716 . * bottom : * singularity spectra of @xmath717 on the left , of @xmath708 on the right.,title="fig:",width=453,height=151 ] it is easy to see , using ( [ eq11 ] ) , that if @xmath684 has a local hlder exponent equal to @xmath290 at a point @xmath12 , then @xmath650 has at @xmath12 a pointwise hlder exponent equal to @xmath718 . hence the multifractal spectrum of @xmath650 is deduced from the one of @xmath684 by the formula @xmath719 for every @xmath720 $ ] . a more explicit formula is obtained as follows : for every @xmath690 , if @xmath721 , then @xmath722 . let us consider the value of @xmath726 such that @xmath727 . then @xmath728 , i.e. @xmath729 . when @xmath730 , the set of points @xmath107 for which @xmath731 is of lebesgue measure 1 , hence we recover the main result of @xcite : @xmath650 is differentiable on a set of lebesgue measure 1 . here we obtain in addition the whole multifractal spectrum of @xmath650 .

we present cn and ch indices and caii triplet metallicities for 34 giant stars and chemical abundances for 33 elements in 14 giants in the globular cluster m2 . assuming the program stars are cluster members , our analysis reveals ( ) an extreme variation in cn and ch line strengths , ( ) a metallicity dispersion with a dominant peak at [ fe / h ].7 and smaller peaks at.5 and.0 , ( ) star - to - star abundance variations and correlations for the light elements o , na , al and si and ( ) a large ( and possibly bimodal ) distribution in the abundances of all elements produced mainly via the-process in solar system material . following , we define two groups of stars , `` '' and ``-only '' , and subtract the average abundances of the latter from the former group to obtain a ``-process residual '' . this-process residual is remarkably similar to that found in m22 and in m4 despite the range in metallicity covered by these three systems . with recent studies identifying a double subgiant branch in m2 and a dispersion in sr and ba abundances , our spectroscopic analysis confirms that this globular cluster has experienced a complex formation history with similarities to m22 , ngc 1851 and centauri . [ firstpage ] stars : abundances galaxy : abundances globular clusters : individual : ngc 7089

1404.6873

photometric studies have revealed complex structure in the colour - magnitude diagrams ( cmd ) of galactic globular clusters ( e.g. , see review by @xcite ) . the subgiant branch region is of particular interest because differences in the luminosity of stars at this evolutionary stage require distinct ages and/or chemical compositions . any globular cluster that exhibits a broadened or split subgiant branch must therefore have experienced a complex , and likely prolonged , chemical enrichment history when compared to globular clusters with a single subgiant branch population . @xmath9 centauri and m22 ( ngc 6656 ) are two galactic globular clusters with multiple subgiant branches ( e.g. , @xcite ; @xcite ) . these two clusters are also notable for exhibiting a large star - to - star dispersion in the abundance of fe - peak and neutron - capture elements ( e.g. , @xcite ) . ngc 1851 is another globular cluster with multiple subgiant branches @xcite . although the difference in metallicity between the two populations , @xmath10[fe / h ] @xmath2 0.07 dex @xcite , is less pronounced in ngc 1851 compared to @xmath9 cen and m22 , a large star - to - star dispersion in the neutron - capture element abundances is also present ( e.g. , @xcite ) . while theoretical studies indicate that multiple population globular clusters could be formed through mergers or that some may be the remnants of dwarf galaxies ( e.g. , @xcite ) , understanding the sequence of events that produce multiple population globular clusters remains a major challenge ( e.g. , @xcite ) . an important step in advancing our knowledge of the formation of multiple population globular clusters is to understand the full range of phenomena and relative frequency present in the galactic globular cluster system . @xcite identified five new galactic globular clusters with broadened or split subgiant branches based on @xmath11 photometry . their sample included m2 ( ngc 7089 ) , a little - studied cluster . @xcite measured the strengths of the cn and ch molecular features in a sample of 19 m2 red giants . in addition to the usual bimodal distribution of cn band strengths @xcite , they noted that two objects are ch stars . ch stars are rare in globular clusters , and at the time of that paper , the only other clusters known to contain ch stars included the apparently normal cluster m55 as well as the peculiar systems m22 and @xmath9 cen . @xcite studied the cn and ch bands from sloan digital sky survey spectroscopy in a number of globular clusters including m2 . they did not identify any stars with unusually strong cn or ch in this cluster , and all of their program stars lie on the canonical red giant branch ( rgb ) . @xcite studied the cn and ch band strengths as well as the c and n abundances in a sample of 35 m2 red giants . they also noted the presence of an additional rgb in the @xmath12 versus @xmath13 cmd . both ch stars identified by @xcite are located on the anomalous rgb ( see figure 14 in @xcite ) . examination of the @xcite str " omgrenphotometry also confirms the peculiar nature of the rgb . while @xcite did not observe any stars on the anomalous rgb , in a subsequent study they obtained spectra for such stars @xcite . stars belonging to the two rgbs had distinct c , n , sr and ba abundances and @xcite argued that m2 has experienced a complex star formation history with similarities to @xmath9 cen , m22 and ngc 1851 . high - resolution spectroscopy and chemical abundance measurements for a larger suite of elements for stars on the canonical and anomalous rgbs of m2 are essential to reveal the true nature of this multiple population globular cluster . the purpose of this paper is to measure cn and ch indices and chemical abundances for a sample of stars in m2 belonging to the canonical and anomalous rgbs . the sample selection and observations are described in section 2 . section 3 contains the analysis . the results are presented in section 4 . section 5 includes a discussion on the nature of this cluster . the program stars were selected from the @xmath14 str " omgren photometry by @xcite . in figure [ fig : cmd100 ] , we present the @xmath15 , @xmath16 , and @xmath17 cmds . as noted by @xcite , we confirm the presence of an additional rgb sequence . such stars are highlighted in red and aqua in figure [ fig : cmd100 ] and were selected from the @xmath18 versus @xmath15 cmd ( upper panel in figure [ fig : cmd100 ] ) . we refer to these as anomalous rgb stars . ( the reason for using two sets of colours for the anomalous rgb stars will become clear when we present the chemical abundances for these objects : the red symbols are stars with [ fe / h ] @xmath2 @xmath31.5 and the aqua symbols are stars with [ fe / h ] @xmath2 @xmath31.0 . ) while @xcite showed that the two ch stars identified by @xcite were located on the anomalous giant branch , neither was included in the @xcite photometry . versus @xmath15 ( upper ) , @xmath19 versus @xmath16 ( middle ) and @xmath19 versus @xmath17 ( lower ) . the black symbols are stars that lie on the canonical rgb , agb or hb as well as one uv bright object . the red ( and aqua ) symbols are stars that lie on the anomalous rgb , and were selected from the @xmath18 versus @xmath15 cmd , upper panel . ( the aqua symbols are the unusually metal - rich objects as determined from high - resolution spectroscopy . ) crosses represent stars observed with the aat . square symbols ( subaru telescope ) and diamond symbols ( magellan telescope ) represent objects observed at high spectral resolution . [ fig : cmd100 ] ] we observed candidate m2 members using the aaomega multi - object spectrograph @xcite on the anglo - australian telescope as part of two separate observing runs . the two ch stars from @xcite were also observed . the first set of observations , obtained on 2010 september 30 , used the 1700b blue and 2000r red gratings . these provide spectral coverage of 3750 to 4440 and 5800 to 6300 at resolutions @xmath20 = @xmath21/@xmath22 = 3500 and 8000 for the red and blue arms , respectively . the second set of observations , from 2011 november 1 , employed the 580v blue and 1700d red gratings . these gratings provide a wavelength coverage from 3750 to 5500 at a resolution of 1300 in the blue arm , and 8350 to 8825 at @xmath20 = 10000 in the red arm . in each case the cluster observations were obtained together with flat - field and arc lamp calibration exposures . data reduction to wavelength - calibrated sky - subtracted individual stellar spectra was accomplished using the standard 2dfdr software . in generating the fibre configurations for the observations , high priority was given to the stars that lie on the anomalous giant branch . although the same input catalogue was used for both sets of observations , the combination of different field plates and different available fibre numbers meant that the two sets of stars observed are not identical . the first set contains 22 stars , the second 23 with 11 in common for a total sample of 34 candidate m2 members . the stars observed are listed in table [ tab : aao ] . we note that 14 stars belong to the anomalous rgb and the two ch stars from @xcite may also be regarded as anomalous rgb objects @xcite . based on their cmd location , we offer some comments on a handful of the stars observed with aaomega . star nr 82 , if a cluster member , would be classified as a asymptotic giant branch ( agb ) star rather than a rgb star . star nr 184 , if a cluster member , would be a uv - bright object lying more than a magnitude above the cluster blue horizontal branch . similarly , star nr 648 has the photometric colours and magnitude of a cluster blue horizontal branch star . however , the aaomega spectra ( from the second data set ) are that of a cluster - like red giant star . we have no straightforward explanation for this anomaly . star nr 707 , again if it is a cluster member , has a magnitude and colour that would suggest it is a red horizontal branch ( rhb ) star . m2 , however , is not normally considered to have a rhb population given its dominant blue hb morphology ( e.g. , @xcite ) . we have not considered these stars any further in the analysis . @llcccccrrrr@ name1 & name2 & ra . 2000 & dec . 2000 & p(% ) & flag versus @xmath15 cmd . all other stars lie on the canonical rgb . ] & @xmath12 & rv ( km s@xmath23 ) & @xmath24rv ( km s@xmath23 ) & @xmath25 & @xmath26 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) + + hi-240 & aiii-43 & 21 33 10.70 & @xmath300 51 09.67 & 99 & & 14.25 & 1.6 & 2.0 & 1.183 & 0.082 + hi-451 & & 21 33 39.11 & @xmath300 49 30.18 & & & 15.86 & 2.1 & 4.6 & 0.574 & 0.150 + nr 38 & & 21 33 28.91 & @xmath300 50 00.94 & & 1 & 13.60 & 1.1 & 2.2 & 0.520 & 0.045 + nr 76 & hi-104,aii-30 & 21 33 17.91 & @xmath300 48 19.82 & 99 & & 13.85 & @xmath31.5 & 4.9 & 0.291 & 0.049 + nr 81 & & 21 33 27.08 & @xmath300 48 19.41 & & 1 & 13.81 & @xmath316.8 & 2.2 & 0.350 & 0.157 + nr 124 & & 21 33 27.81 & @xmath300 47 30.43 & & & 14.21 & 4.3 & 2.3 & 0.421 & 0.009 + nr 132 & & 21 33 23.10 & @xmath300 48 11.53 & & 1 & 14.20 & 0.8 & 2.5 & 0.245 & 0.023 + nr 184nr 184 is a uv - bright star , nr 648 is a bhb star , nr 707 is a rhb ( or agb ) star and nr 82 is an agb star . ] & cr57 & 21 33 24.94 & @xmath300 50 41.42 & 99 & & 15.24 & 0.0 & 7.8 & 0.002 & @xmath30.113 + nr 216 & cr19 & 21 33 31.48 & @xmath300 49 06.33 & 99 & 1 & 14.82 & @xmath312.1 & 2.7 & 0.652 & 0.038 + nr 225 & hi-586,ai-58 & 21 33 29.27 & @xmath300 45 55.49 & 99 & & 14.82 & @xmath310.6 & 5.2 & 0.492 & 0.025 + nr 301 & aiv-37 & 21 33 32.80 & @xmath300 50 27.06 & 99 & 1 & 14.93 & @xmath36.3 & 3.2 & 0.693 & 0.098 + nr 358 & hi-521,ai-79 & 21 33 34.05 & @xmath300 47 32.10 & 99 & 1 & 15.06 & @xmath316.0 & 4.2 & 1.459 & 0.043 + nr 378 & ai-50 & 21 33 30.32 & @xmath300 47 24.54 & 99 & 1 & 15.22 & 3.2 & 5.2 & 0.363 & 0.086 + nr 386 & & 21 33 26.18 & @xmath300 49 21.35 & & 1 & 15.70 & @xmath33.4 & 3.7 & 0.057 & @xmath30.061 + nr 388 & & 21 33 27.16 & @xmath300 50 25.43 & & & 15.23 & @xmath36.0 & 6.0 & 0.293 & @xmath30.022 + nr 417 & cr76 & 21 33 23.48 & @xmath300 48 46.57 & 99 & & 15.39 & @xmath32.9 & 3.3 & 0.212 & 0.017 + nr 721 & & 21 33 24.32 & @xmath300 49 41.46 & & 1 & 15.79 & @xmath38.8 & 1.8 & 0.309 & 0.002 + nr 811 & & 21 33 22.85 & @xmath300 50 34.00 & & & 15.74 & @xmath39.3 & 5.2 & 0.388 & 0.013 + nr 847 & & 21 33 23.45 & @xmath300 46 24.34 & & & 15.79 & @xmath316.1 & 2.8 & 0.430 & @xmath30.008 + nr 915 & & 21 33 35.36 & @xmath300 49 57.45 & & & 15.83 & @xmath35.2 & 6.9 & 0.319 & 0.037 + nr 1178 & & 21 33 31.64 & @xmath300 49 59.80 & & 1 & 16.06 & 3.0 & 5.1 & 0.343 & 0.059 + nr 1204 & aiii-26 & 21 33 20.08 & @xmath300 50 13.76 & 99 & 1 & 15.82 & @xmath37.9 & 3.3 & 0.561 & 0.154 + + hi-240 & aiii-43 & 21 33 10.70 & @xmath300 51 09.67 & 99 & & 14.25 & 2.9 & 2.5 & 1.069 & 0.085 + hi-451 & & 21 33 39.11 & @xmath300 49 30.18 & & & 15.86 & 4.3 & 2.7 & 0.609 & 0.199 + nr 47 & cr12 & 21 33 28.52 & @xmath300 48 43.92 & 99 & 1 & 13.70 & 5.1 & 1.5 & 0.483 & 0.058 + nr 76 & hi-104,aii-30 & 21 33 17.91 & @xmath300 48 19.82 & 99 & & 13.85 & @xmath31.5 & 1.9 & 0.283 & 0.045 + nr 82@xmath27 & cr190 & 21 33 33.63 & @xmath300 50 29.50 & 99 & & 13.91 & @xmath33.1 & 1.7 & 0.424 & @xmath30.015 + nr 99 & aiii-86 & 21 33 23.59 & @xmath300 50 41.07 & 99 & & 13.69 & @xmath31.3 & 2.1 & 0.250 & 0.018 + nr 124 & & 21 33 27.81 & @xmath300 47 30.43 & & & 14.21 & 4.4 & 1.9 & 0.382 & 0.017 + nr 132 & & 21 33 23.10 & @xmath300 48 11.53 & & 1 & 14.20 & 1.2 & 2.2 & 0.337 & 0.053 + nr 207 & & 21 33 27.48 & @xmath300 49 51.35 & & 1 & 14.89 & @xmath31.1 & 2.0 & 0.219 & 0.004 + nr 225 & hi-586,ai-58 & 21 33 29.27 & @xmath300 45 55.49 & 99 & & 14.82 & @xmath310.8 & 2.0 & 0.430 & 0.049 + nr 254 & & 21 33 29.37 & @xmath300 49 42.84 & & 1 & 15.05 & 3.4 & 2.0 & 0.255 & 0.027 + nr 299 & ai-22 & 21 33 35.32 & @xmath300 49 22.13 & 99 & & 14.94 & 0.3 & 1.9 & 0.307 & 0.120 + nr 358 & hi-521,ai-79 & 21 33 34.05 & @xmath300 47 32.10 & 99 & 1 & 15.06 & @xmath316.7 & 1.2 & 1.328 & 0.114 + nr 375 & & 21 33 30.62 & @xmath300 50 08.33 & & & 15.20 & 2.1 & 1.7 & 0.322 & 0.022 + nr 378 & ai-50 & 21 33 30.32 & @xmath300 47 24.54 & 99 & 1 & 15.22 & @xmath32.3 & 2.0 & 0.432 & 0.081 + nr 403 & cr58 & 21 33 25.64 & @xmath300 50 43.12 & 99 & & 15.20 & @xmath315.3 & 1.8 & 0.331 & 0.015 + nr 648@xmath27 & & 21 33 26.32 & @xmath300 49 10.58 & & & 15.99 & @xmath31.6 & 1.9 & 0.194 & @xmath30.049 + nr 707@xmath27 & & 21 33 23.02 & @xmath300 48 56.69 & & & 15.81 & 0.3 & 1.9 & 0.095 & @xmath30.053 + nr 801 & & 21 33 27.46 & @xmath300 46 53.10 & & & 15.79 & @xmath32.3 & 2.2 & 0.305 & @xmath30.004 + nr 847 & & 21 33 23.45 & @xmath300 46 24.34 & & & 15.79 & @xmath33.9 & 2.0 & 0.377 & 0.034 + nr 915 & & 21 33 35.36 & @xmath300 49 57.45 & & & 15.83 & @xmath311.8 & 2.3 & 0.332 & 0.047 + nr 947 & aiii-10 & 21 33 18.75 & @xmath300 49 44.09 & 99 & & 15.79 & @xmath36.2 & 1.9 & 0.190 & 0.059 + nr 1204 & aiii-26 & 21 33 20.08 & @xmath300 50 13.76 & 99 & 1 & 15.82 & @xmath310.6 & 1.7 & 0.449 & 0.123 + three stars ( nr 76 , nr 81 and nr 132 ) were observed in service mode using the high dispersion spectrograph ( hds , @xcite ) on the subaru telescope on 2011 august 3 . six additional stars ( nr 47 , nr 99 , nr 124 , nr 207 , nr 254 and nr 378 ) were observed using hds in classical mode on 2013 july 17 . all nine stars were also observed with the aaomega instrument . for both sets of observations , we used the stdyb setting and the 08 slit which resulted in a wavelength coverage from @xmath284100 to @xmath286800 at a spectral resolution of @xmath29 . a telluric standard was also observed . the spectra were reduced using iraf adopting a similar approach as in @xcite . five stars ( nr 37 , nr 38 , nr 58 , nr 60 and nr 77 ) were also observed using the magellan inamori kyocera echelle ( mike ) spectrograph @xcite at the magellan telescope on 2012 august 26 . nr 60 is a likely agb star based on cmd location . full wavelength coverage was obtained ( @xmath283400 to @xmath289000 ) , and we used the 07 slit which provided a spectral resolution of @xmath30 in the blue arm and @xmath31 in the red arm , as measured from the thar lines . the spectra were reduced using the carpy pipeline and independently in iraf using the mtools package . one star , nr 77 , had a faint , nearby companion . this object was reduced in two different ways using iraf . in the first approach , we adopted a conservative aperture placement to try to avoid flux from the faint companion . in the second approach , the flux from both stars was extracted . in the subsequent section , we analyze both sets of spectra independently in order to quantify the contamination from the nearby companion . the program stars are listed in table [ tab : param ] . we note that of these 14 objects observed at high spectral resolution , six belong to the canonical rgb and eight are anomalous rgb stars . @llccccccrcccccc@ name1 & name2 & r.a . & dec . & p & flag versus @xmath15 cmd . all other stars lie on the canonical rgb . ] & run & @xmath12 & rv & @xmath24rv & @xmath32 & @xmath33 & @xmath34 & [ m / h ] refers to the metallicity used to generate the model atmosphere . ] & [ fe / h ] + & & & & ( % ) & & & & & ( k ) & ( cgs ) & ( km s@xmath23 ) & ( dex ) & ( dex ) + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) & ( 12 ) & ( 13 ) & ( 14 ) & ( 15 ) + + nr 37 & cr78 & 21 33 25.44 & @xmath300 48 53.73 & 99 & & m12 & 13.56 & @xmath315.3 & 1.0 & 4250 & 0.70 & 1.77 & @xmath31.6 & @xmath31.66 + nr 58 & cr30 & 21 33 32.17 & @xmath300 50 01.17 & 99 & & m12 & 13.55 & 11.8 & 1.0 & 4225 & 0.70 & 1.89 & @xmath31.6 & @xmath31.64 + nr 60 & cr28 & 21 33 32.57 & @xmath300 49 45.72 & 99 & & m12 & 13.55 & @xmath37.1 & 1.0 & 4325 & 0.30 & 2.19 & @xmath31.7 & @xmath31.75 + nr 76 & hi-104 & 21 33 17.91 & @xmath300 48 19.82 & 99 & & s11 & 13.85 & @xmath31.3 & 0.6 & 4375 & 0.90 & 1.73 & @xmath31.7 & @xmath31.69 + nr 99 & aiii-86 & 21 33 23.59 & @xmath300 50 41.07 & 99 & & s13 & 13.69 & @xmath31.5 & 0.6 & 4275 & 0.70 & 1.78 & @xmath31.6 & @xmath31.66 + nr 124 & & 21 33 27.81 & @xmath300 47 30.43 & & & s13 & 14.21 & 3.4 & 0.7 & 4425 & 0.85 & 1.81 & @xmath31.6 & @xmath31.64 + + nr 38 & & 21 33 28.91 & @xmath300 50 00.94 & & 1 & m12 & 13.60 & 3.7 & 1.3 & 4175 & 0.60 & 2.12 & @xmath31.6 & @xmath31.61 + nr 47 & cr12 & 21 33 28.52 & @xmath300 48 43.92 & 99 & 1 & s13 & 13.70 & 3.3 & 0.5 & 4050 & 0.65 & 1.77 & @xmath31.4 & @xmath31.42 + nr 77 & & 21 33 24.45 & @xmath300 48 36.29 & & 1 & m12 & 13.92 & 6.6 & 1.0 & 4350 & 1.00 & 2.25 & @xmath31.5 & @xmath31.46 + nr 81 & & 21 33 27.08 & @xmath300 48 19.41 & & 1 & s11 & 13.81 & @xmath322.0 & 0.5 & 4275 & 1.00 & 1.85 & @xmath31.6 & @xmath31.55 + + nr 132 & & 21 33 23.10 & @xmath300 48 11.53 & & 1 & s11 & 14.20 & 0.7 & 0.5 & 4325 & 1.30 & 1.88 & @xmath31.0 & @xmath30.97 + nr 207 & & 21 33 27.48 & @xmath300 49 51.35 & & 1 & s13 & 14.89 & @xmath32.1 & 0.4 & 4425 & 1.30 & 1.40 & @xmath31.1 & @xmath31.08 + nr 254 & & 21 33 29.37 & @xmath300 49 42.84 & & 1 & s13 & 15.05 & 3.2 & 0.5 & 4525 & 1.60 & 1.61 & @xmath31.0 & @xmath30.97 + nr 378 & ai-50 & 21 33 30.32 & @xmath300 47 24.54 & 99 & 1 & s13 & 15.22 & @xmath32.9 & 0.5 & 4750 & 1.50 & 1.68 & @xmath31.1 & @xmath31.08 + in figures [ fig : spec1s ] to [ fig : spec2 m ] , we plot regions of the high dispersion spectra for the program stars from both telescope+instrument combinations . these figures demonstrate that there are star - to - star variations in the strengths of zr and la lines which could be caused by differences in stellar parameters and/or chemical abundance ratios . in section 3.2 , we shall seek to quantify the stellar parameters and chemical abundances . . ) the positions of other atomic lines and the stellar parameters ( @xmath32/@xmath33/[fe / h ] ) are included . [ fig : spec1s ] ] , zr lines are highlighted and there are significant star - to - star differences in line strengths . the black spectra denote that those stars lie on the canonical rgb while red spectra represent stars that lie on the anomalous rgb . the positions of other atomic lines and the stellar parameters ( @xmath32/@xmath33/[fe / h ] ) are included . [ fig : spec1 m ] ] but for a region containing a la line used in the analysis . there are significant star - to - star differences in the line strength of la . the positions of cn lines are marked . [ fig : spec2s ] ] but for the magellan mike spectra . [ fig : spec2 m ] ] radial velocities were measured from the medium - resolution spectra by cross - correlating each program star against hi-240 . the radial velocity for hi-240 was determined by measuring the wavelengths of a small set of lines ( sodium doublet 5889.951 and 5895.924 and the calcium triplet 8498.03 , 8542.09 and 8662.14 ) . given the superior spectral resolution in the red arm , we adopted those values as the radial velocities and corrected for the heliocentric motion . we measured the @xmath25 and @xmath26 indices in the aaomega spectra using the definitions given in @xcite . calcium triplet line strengths were measured via gaussian line profile fits to the observed data for the two stronger caii triplet lines at 8542 and 8662 using the technique first described in @xcite . the heliocentric radial velocities and @xmath25 and @xmath26 indices are presented in table [ tab : aao ] . an assessment of the internal errors associated with these measurement can be obtained by consideration of the 11 objects observed on both runs . for the radial velocities , 10 of the 11 objects showed no evidence ( @xmath351-@xmath24 , i.e. , less than one standard deviation ) for radial velocity variation between the two observing runs . one star , nr 847 , exhibited evidence for radial velocity variability ; @xmath316.1 km s@xmath23 @xmath36 2.8 km s@xmath23 ( 2010 ) versus @xmath33.9 km s@xmath23@xmath36 2.0 km s@xmath23 ( 2011 ) . excluding nr 847 , the average difference in radial velocity for stars observed on both runs is 1.2 km s@xmath23 @xmath36 0.9 km s@xmath23 ( @xmath24 = 2.9 km s@xmath23 ) . for the @xmath25 and @xmath26 indices , we find mean differences for stars observed on both runs of 0.028 @xmath36 0.023 ( @xmath24 = 0.075 ) and 0.018 @xmath36 0.009 ( @xmath24 = 0.029 ) , respectively . since for the caii triplet spectra only single observations are available , we adopt the uncertainty in the pseudo - equivalent widths which results from the uncertainties in the gaussian fit parameters for the observed line profiles . an assessment of the systematic errors can be obtained by comparison of our measurements with literature values . for the radial velocities , five of our program stars were also observed by @xcite , noting that on average their measurement errors ( @xmath37rv@xmath38 = 16.5 km s@xmath23 ) are larger than ours ( @xmath37rv@xmath38 = 3.0 km s@xmath23 ) . for three of these five stars our radial velocity measurements are in agreement . the two stars with poor agreement are nr 132 , @xmath316.4 km s@xmath23 @xmath36 7 km s@xmath23 @xcite versus 0.9 km s@xmath23 @xmath36 3.4 km s@xmath23 ( this study ) , and nr 378 , @xmath360.3 km s@xmath23 @xmath36 5.8 km s@xmath23 @xcite versus 0.5 km s@xmath23 @xmath36 5.6 km s@xmath23 ( this study ) . these stars may be spectroscopic binaries . the @xmath25 and @xmath26 values are in good agreement with those of @xcite for the two stars in common . for hi-240 , our mean values are @xmath25 = 1.126 and @xmath26 = 0.084 and the @xcite values are 1.111 and 0.067 , respectively . for hi-451 , our mean values are @xmath25 = 0.592 and @xmath26 = 0.175 and the @xcite values are 0.571 and 0.165 , respectively . we note that the differences for @xmath25 and @xmath26 between this study and @xcite are comparable to mean differences for the 11 stars observed on both aaomega runs . equivalent widths ( ew ) were measured from the high - resolution spectra using routines in iraf and daospec @xcite . when using iraf to measure an ew , every line in every star was visually inspected . in a given star , lines regarded to be blended or poorly fit were excluded , and weak ( ew @xmath39 5 m ) and strong ( ew @xmath40 130 m ) lines were also removed from the analysis . when using daospec to measure ews , the continuum was the same as in the iraf analysis , i.e. , daospec did not re - adjust the continuum level . additionally , the set of lines measured using daospec was identical to those already measured , and visually inspected , using iraf . for the subaru and magellan spectra , there was good agreement ( @xmath24 = 1.5 m ) between the two sets of ew measurements , and we adopted the daospec values . for the magellan spectra , ew measurements could be compared between the carpy reduction and the iraf reduction . again , there was excellent agreement between the two sets of measurements ( @xmath24 = 1.4 m ) . the ew measurements and line list are presented in table [ tab : ew ] . [ cols="^,^,^,>,>,>,>,>,>,>,>,>,>,>,>,>,>,>,^ " , ] m2 shares similar , and peculiar , characteristics found in the unusual globular clusters m22 , ngc 1851 and @xmath9 cen , namely , a dispersion in metallicity and neutron - capture abundance ratios . if a subset of the metal - rich group are genuine cluster members , then m2 would host stars that span a range in metallicity from [ fe / h ] @xmath2 @xmath31.6 to [ fe / h ] @xmath2 @xmath31.0 , a factor of four . we note further that even if the most metal - rich stars are not members , there still remains a metallicity spread of order 0.25 dex among the stars for which we assert cluster membership . m2 appears to be different from m22 and ngc 1851 ; for the latter two clusters , the number of stars on the bright subgiant branch is similar to the number on the faint subgiant branch . in contrast , for m2 the canonical rgb stars represent the overwhelming majority of stars . as noted in section 4.4 , the relative numbers of canonical and anomalous rgb stars in m2 is probably comparable to the relative numbers of bright ( @xmath2896% ) and faint ( @xmath284% ) subgiant branch stars @xcite . regarding the light elements , even with our limited sample it is apparent that the @xmath8-only and @xmath7 groups both exhibit star - to - star abundance variations and correlations between [ na / fe ] and [ al / fe ] . the two populations in m22 and ngc 1851 both exhibit a o - na anticorrelation @xcite , and for @xmath9 cen , the o - na anticorrelation is present across a broad metallicity range @xcite . indeed , every well studied galactic globular cluster exhibits star - to - star abundance variations for the light elements c , n , o , f , na , mg and al ( e.g. , see reviews by @xcite ; @xcite ; @xcite ) . while these abundance variations are believed to be produced through hydrogen - burning , the specific site continues to be debated ( e.g. , @xcite ) . for the metal - rich group , the apparent absence of a star - to - star abundance variation for the light elements is intriguing , although the sample size is small . no such abundance spread would be expected if these were all field stars . on the other hand , a similar situation is present in the m54+sagittarius ( sgr ) system . while the o - na anticorrelation is evident in m54 , the more metal - rich sgr stars do not exhibit this pattern @xcite . if the four stars in the metal - rich group are indeed cluster members , then m2 would share this peculiar feature with m54+sgr . for the @xmath41 and fe - peak elements , there is no compelling evidence for a star - to - star abundance variation within a given group . additionally , the abundance ratios [ x / fe ] for a given star are compatible with field stars at the same metallicity . in other words , these elements appear to be well - behaved . the abundance of cu offers an important tool to distinguish between field stars and `` @xmath9 cen - like '' systems . for m2 , the cu abundance may help establish additional similarities with @xmath9 cen and potentially cluster membership , or otherwise , for the four metal - rich objects for the following reasons . in the metallicity regime @xmath32.0 @xmath42 [ fe / h ] @xmath42 @xmath30.5 , field stars exhibit a systematic increase in [ cu / fe ] with increasing metallicity ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? mono - metallic globular clusters in the same metallicity range appear to follow the field star trend @xcite . @xmath9 cen , however , displays a near constant cu abundance , [ cu / fe ] @xmath2 @xmath30.5 , over the range @xmath31.9 @xmath35 [ fe / h ] @xmath35 @xmath30.8 @xcite . at higher metallicities , @xmath31.2 @xmath35 [ fe / h ] @xmath35 @xmath30.4 , there is evidence for an increase in the [ cu / fe ] ratio in @xmath9 cen @xcite , although the rate of that increase is smaller than in field stars . chemical evolution models of @xmath9 cen and the milky way by @xcite attribute the nucleosynthesis of cu to massive stars and successfully reproduce the observed trends . in figure [ fig : cu ] , we plot [ cu / fe ] versus [ fe / h ] for m2 , field stars @xcite , mono - metallic globular clusters @xcite and @xmath9 cen @xcite . the cu abundances in m2 displayed in this figure have been adjusted in the following manner . following @xcite , the abundances from the 5105 and 5782 lines are referenced to solar values of @xmath43 = 4.21 and @xmath44 = 4.06 , respectively . such an approach reflects the different solar abundances obtained from these lines , and we note that the abundances we derive for program stars from the 5782 line are , on average , 0.21 dex @xmath36 0.03 dex ( @xmath24 = 0.11 dex ) higher than those from the 5105line . the gradient of the linear fit to m2 is not affected by these zero - point offsets . we also stress that although this figure includes data from numerous studies , the linear fit in each panel is performed upon data obtained from a single study ( for the lower left panel , m2 , m22 and ngc 1851 are from different studies but those data are not included in the linear fit ) . so long as each sample is analysed uniformly , the slopes should be robust and we can compare them in a quantitative manner . in figure [ fig : cu ] , the slopes for the field stars and mono - metallic globular clusters are in good agreement , and these slopes differ from that seen in @xmath9 cen . the behaviour of the slope of [ cu / fe ] versus [ fe / h ] in m2whether or not the @xmath7 stars are considered is different from the field stars , mono - metallic globular clusters and @xmath9 cen over the metallicity range @xmath31.7 @xmath45 [ fe / h ] @xmath45 @xmath46 . the metal - rich stars , relative to the @xmath8-only and @xmath7 groups , do not follow the field star trend , and this may be the strongest abundance - based evidence that they are cluster members . furthermore , if the metal - rich stars are indeed cluster members , then m2 does not share a similar chemical enrichment history to @xmath9 cen , at least for cu . figure [ fig : cu ] also demonstrates that the mean [ cu / fe ] ratios in ngc 1851 @xcite and the @xmath8-only groups in m2 and m22 @xcite match the trends established by the mono - metallic globular clusters . the cu in the @xmath7 group of stars may include small contributions from @xmath6-process nucleosynthesis , so we do not discuss cu in the @xmath7 group here . we now turn our attention to the neutron - capture elements . in the abundance analysis described in section 3.2 , we examined up to 122 lines of elements with atomic numbers @xmath47 38 in each of the program stars . all abundances were computed by matching synthetic spectra , generated using one dimensional plane - parallel model atmospheres , to the observed spectra under the assumption that lte holds in the line - forming layers . the abundances of sr and pb were derived from neutral lines . for the program stars , sr i and pb i are minority species , and lte calculations will tend to underestimate the populations of the lower levels of the sr i 4607 and pb i 4057 transitions . abundances of strontium and lead derived in lte from these lines are thus underestimated . calculations allowing for departures from lte in the line - forming layers by making reasonable assumptions for the photoionization cross sections have been made for stars with stellar parameters similar to those in our sample . these non - lte calculations suggest that our lte analysis may underestimate the strontium abundance by @xmath2 [email protected] dex @xcite and the lead abundance by @xmath2 [email protected] dex @xcite . the values presented in our tables and figures reflect the lte values . neglecting the non - lte corrections for these two elements should not significantly affect any of the abundance differences between the @xmath8-only and @xmath7 groups of stars that we shall discuss below . figure [ starplot ] illustrates the heavy element abundance patterns found in each star of our sample . the six stars shown in the left - hand panels are those on the canonical rgb , the four stars in the middle panels are the neutron - capture rich anomalous rgb stars and the four stars in the right - hand panels are the metal - rich anomalous rgb stars . for comparison , in each panel of this figure we overplot the heavy element abundance pattern found in the rich standard star ( normalised to the eu abundance ) , whose metallicity is only a factor of @xmath2 2.5 lower than the majority of stars in m2 . the stars on the canonical rgb have heavy element abundance patterns very similar to one another and to the pattern in , and the overall amounts of heavy elements are constant within their mutual uncertainties . we refer to these six stars ( nr 37 , nr 58 , nr 60 , nr 76 , nr 99 and nr 124 ) on the canonical rgb as the `` @xmath8-only group . '' the reasoning behind this name will be made clear shortly . as shown in the middle panels of figure [ starplot ] , the heavy elements in the neutron - capture rich anomalous rgb stars in m2 tell a different story . all heavy elements except europium in these four stars exhibit noticeable abundance enhancements relative to the stars on the canonical rgb and , therefore , enhancements relative to the standard . the pattern changes little from one star to the next , and the overall abundances in this group of stars are also constant within their mutual uncertainties . we refer to these four stars ( nr 38 , nr 47 , nr 77 and nr 81 ) as the `` @xmath7 group . '' the consistent patterns and levels of enhancement found within each of the @xmath8-only and @xmath7 groups suggests that we can average together their abundances to reduce the random uncertainties , which is especially helpful for elements whose abundances are derived from small numbers of lines . these mean abundance patterns are listed in table [ tab : meanabun ] and illustrated in figure [ meanplot ] . subtle differences between the stars in the @xmath8-only group and , ( e.g. , small overabundances in m2 for strontium , yttrium , zirconium , barium , cerium and neodymium , as well as small underabundances in m2 for ytterbium ) may simply reflect differing combinations of material produced by the so - called weak and main components of the enriching m2 and . this is plausible because the overall level of @xmath8-process enhancement relative to iron is different in and m2 , with [ eu / fe ] = + 0.9 and + 0.4 , respectively . regardless , figure [ starplot ] demonstrates that the heavy elements in the stars in the @xmath8-only group in m2 owe their origin to nucleosynthesis with little or no contributions . -only stars ( black circles ) and the four @xmath7 stars ( red triangles ) . ( only elements measured in more than one star are included , i.e. , we exclude mo , er and tm . ) the zirconium abundance derived from zr ii lines is shown . the gray line illustrates the abundances in the standard star normalized to the europium abundance . bottom panel : differences in these mean abundances . the dotted line indicates zero difference . [ meanplot ] ] in contrast , the middle panels of figure [ starplot ] demonstrate that the @xmath7 stars have abundance patterns that are inconsistent with nucleosynthesis alone . figure [ sfracplot ] demonstrates that the excess of heavy elements ( @xmath47 38 ) found in the @xmath7 group relative to the @xmath8-only group exhibits an unmistakable correlation with the fraction of each element attributed to an origin in solar system material . elements with a high fraction in the solar system are most overabundant in the @xmath7 group in m2 , and those with small fractions in the solar system show little excess . only a few percent of the europium in the solar system is attributed to nucleosynthesis , and this element shows a constant abundance in both the @xmath8-only and @xmath7 groups of stars ; the average @xmath48 ( eu ) and [ eu / fe ] ratios are very similar between these two groups . this suggests that there is a common abundance foundation in each star in the @xmath8-only and @xmath7 groups in m2 . ( we set aside , for now , the metal - rich group . ) following @xcite , we speculate that the stars in the @xmath7 group formed from additional material enriched by products of nucleosynthesis as well as iron . we emphasise that the abundance differences between these two groups of stars should be nearly insensitive to any non - lte effects given the modest range in stellar parameters ( @xmath32 , @xmath33 , [ fe / h ] ) spanned by the sample . we now consider the metal - rich group ( stars nr 132 , nr 207 , nr 254 and nr 378 ) shown in the right - hand panels of figure [ starplot ] noting that these stars may , or may not , be cluster members . the heavy element abundances in the metal - rich group closely resemble an pattern , despite the fact that the overall metallicity of these objects is a factor of @xmath2 4 higher than the other cluster members . furthermore , the [ x / fe ] ratios ( where x denotes any element with @xmath47 56 ) in the metal - rich group are on average 0.17 dex @xmath36 0.02 dex ( @xmath24 = 0.05 dex ) higher than in the @xmath8-only stars , i.e. , a factor of @xmath21.5 . if we assume that the overall metal content in an isolated stellar system increases monotonically with time , the metal - rich group should have formed later than either the @xmath8-only or @xmath7 groups . to the best of our knowledge , no isolated self - enriched stellar system shows a return to dominance after previous enrichment by a substantial amount of material . we conclude that the four metal - rich stars can not be easily understood as members of a single self - enriched stellar system . that said , the data do not preclude a scenario in which m2 is composed of independent fragments that experienced different chemical enrichment histories @xcite . if we assume that the enrichment is common to both the @xmath8-only and @xmath7 groups in m2 , we can subtract the average abundances found in the @xmath8-only group from those in the @xmath7 group to obtain the intrinsic abundance ratios of the material added to the @xmath7 group . these differences are shown in figure [ meanplot ] ( as a function of atomic number ) and in figure [ sfracplot ] ( as a function of the fraction of each element attributed to the @xmath6-process in solar system material ) . @xcite performed a similar calculation for the two stellar groups in m22 and the unrelated clusters m4 and m5 using data from @xcite . in figure [ sfracplot ] , we include the abundance differences for the m22 groups as well as the abundance differences when subtracting the mean values for m5 from those of m4 which we denote as `` m4 @xmath3 m5 '' @xmath49 @xmath31.2 , and m4 is known to exhibit a moderate enhancement in @xmath6-process element abundances compared to m5 @xcite . as in @xcite , subtracting the abundances for m5 from m4 attempts to quantify the @xmath6-process contribution to m4 . ] . in this figure we adopt the values from @xcite for the fraction of each element attributed to the @xmath6-process in solar system material . as noted , there is a clear trend between the abundance differences and the @xmath6-process fraction . as originally proposed by @xcite , we argue that the abundance residual represents @xmath6-process material . when considering all elements with @xmath50 @xmath51 38 ( i.e. , sr and heavier elements ) , the three `` systems , '' ( 1 ) @xmath52 @xmath3 @xmath53-only@xmath38 in m2 , ( 2 ) @xmath52 @xmath3 @xmath54-only@xmath38 in m22 and ( 3 ) @xmath55m4@xmath38 @xmath3 @xmath55m5@xmath38 , exhibit identical gradients within their mutual uncertainties . such a result is surprising given that the yields for the @xmath6-process elements in agb stars are mass and metallicity dependent ( e.g. , @xcite ; fishlock et al . , in preparation ) . if our interpretation that the abundance residuals in these systems represent @xmath6-process material is correct , then the implication is that these three systems , which span a range in metallicity from [ fe / h ] @xmath2 @xmath31.8 to [ fe / h ] @xmath2 @xmath31.2 , experienced enrichment by @xmath6-process material of indistinguishable composition . quantitative chemical evolution modelling is needed to test this intriguing hypothesis , and shingles et al . ( in preparation ) are investigating m22 and m4 and comparing the predicted and observed enrichment patterns taking into account yields from agb and massive stars . if we consider only elements with 38 @xmath35 @xmath50 @xmath35 45 ( i.e. , sr to rh ) , the gradients do not exhibit any consistent patterns . in sharp contrast , however , the elements from 56 @xmath35 @xmath50 @xmath35 72 ( i.e. , ba to hf ) exhibit identical gradients within their mutual uncertainties . for these elements , the measured abundance differences in each system are consistent with a single relation . this implies that the enrichment in m2 , m22 and m4 involved @xmath6-process material of remarkably similar composition despite the factor of @xmath28 4 difference in metallicity . the intrinsic ratios and indices ) and heavy ( @xmath56 ) abundances are [ @xmath57/fe ] @xmath58([y / fe ] @xmath59 [ zr / fe ] ) and [ @xmath56/fe ] @xmath60([la / fe ] @xmath59 [ nd / fe ] @xmath59 [ sm / fe ] ) . these include elements at the first ( sr , y , zr ) and second ( ba , la , ce , pr , nd ) peaks ; pb is the sole representative of the third peak . similarly , [ @xmath56/@xmath57 ] @xmath62 [ @xmath56/fe ] @xmath3 [ @xmath57/fe ] ] are [ pb / la]@xmath63 @xmath640.53 in m2 , @xmath590.18 in m22 and @xmath30.01 in m4 @xmath3 m5 ; [ @xmath56/@xmath57]@xmath63 @xmath65 0.02 in m2 , @xmath30.01 in m22 and @xmath30.50 in m4 @xmath3 m5 ; and [ pb/@xmath56]@xmath63 @xmath670.72 in m2 , @xmath590.29 in m22 and @xmath590.28 in m4 @xmath3 m5 . uncertainties on these ratios are typically [email protected] dex . these ratios and indices are largely insensitive to uncertainties in the atomic data and non - lte effects . for [ pb / la ] , m2 exhibits a higher ratio than m22 and m4 @xmath3 m5 . for [ @xmath56/@xmath57 ] , m4 @xmath3 m5 exhibits lower ratios than m22 and m2 , although this may reflect the higher metallicity of m4 and m5 relative to the other two clusters . for [ pb/@xmath56 ] , m2 exhibits a higher ratio than the other two systems . a number of studies have investigated nucleosynthesis in metal - poor stars on the agb ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * fishlock et al . , in preparation ) . while most of these models fail to offer an exact match for the metallicity of m2 , we can use them to get a sense of nucleosynthesis ratios predicted for metallicities higher and lower than m2 . we find encouraging agreement when comparing our results to the [ @xmath56/@xmath57 ] and [ pb/@xmath56 ] indices presented in figures c3 and c5 of @xcite for 3 and 5 m@xmath68 agb stars at the appropriate metallicities for m2 , m22 and m4 @xmath3 m5 . furthermore , we note that the yields of fishlock et al . ( in preparation ) for their 3 m@xmath68 and 3.5 m@xmath68 models for [ fe / h ] = @xmath31.2 bracket the [ pb / la]@xmath63 , [ @xmath56/@xmath57]@xmath63 and [ pb/@xmath56]@xmath63 ratios in m2 , m22 and m4 @xmath3 m5 . quantitative chemical evolution models based on their yields would be of great interest . overall , we reach the same conclusion drawn by @xcite : agb stars with masses less than 3 m@xmath68 can not reproduce the observed ratios unless the standard @xmath69c pocket efficiency in the models is reduced by factors of 30 or more . if we assume that the stars in the @xmath7 group in m2 formed later than the stars in the @xmath8-only group , and that the agb stars responsible for distributing this material in m2 formed simultaneously with the stars in the @xmath8-only group , this sets an upper limit on the amount of time that passed between the formation of the @xmath8-only group and the @xmath7 group . for a 3 m@xmath68 agb star , adopting the approximate stellar lifetimes computed by @xcite , this sets a limit of no more than 300 myr or so between the two groups . of course , this limit would be even smaller if higher - mass agb stars were the source of the material . finally , although we have focused on the neutron - capture elements , the difference in [ fe / h ] between the @xmath8-only and @xmath7 groups requires some source(s ) that produces the elements from si to zn ( and perhaps other elements ) to increase the abundances of these elements between these groups . in this paper we present a spectroscopic analysis of giant stars in the multiple population globular cluster m2 . our principal and novel results include the following . first , we identify a star - to - star dispersion in iron abundance with the anomalous rgb stars ( i.e. , stars lying redward of the dominant rgb ) being more metal - rich than the canonical rgb objects . the iron abundance distribution has a dominant peak at [ fe / h ] @xmath2 @xmath31.7 and smaller peaks at @xmath31.5 and @xmath31.0 , although membership for the latter group remains to be established . secondly , the neutron - capture element abundances exhibit a star - to - star dispersion with a possible bimodal distribution . in this regard , m2 is chemically similar to the globular clusters m22 , ngc 1851 and @xmath9 cen , whose subgiant branches exhibit multiple sequences . it is likely that m2 has therefore experienced a similarly complex formation history . thirdly , when subtracting the average abundances in the @xmath8-only group from those of the @xmath7 group , the abundance residual exhibits a striking correlation with the fraction of each element attributed to the @xmath6-process in solar system material . this residual is remarkably similar to that found in m22 and in m4 @xmath3 m5 . such a similarity would indicate that m2 , m22 , and m4 were enriched by @xmath6-process material of identical composition and potentially offers important observational constraints on the nature of the @xmath6-process in low metallicity environments . a comparison with theoretical predictions reveals that agb stars with masses less than 3 m@xmath68 are unlikely to have played a major role in the chemical enrichment of m2 . in addition to the agb star contribution , some source(s ) is needed to increase the abundances of the elements from si to zn in the @xmath7 group relative to the @xmath8-only group . additional studies are essential to understand the formation and evolution of this complex cluster . we thank k. cudworth for providing electronic data and the referee , kim venn , for helpful comments . some of the aat data were taken in time allocated to the aegis program ( pi keller ) . the authors are grateful to the aegis team for access to these data . d.y , g.d.c , a.i.k , j.e.n , a.f.m and a.p.m gratefully acknowledge support from the australian research council ( grants dp0984924 , ft110100475 , dp120100475 , dp120100991 and dp120101237 ) . funding for the stellar astrophysics centre is provided by the danish national research foundation . the research is supported by the asterisk project ( asteroseismic investigations with song and kepler ) funded by the european research council ( grant agreement no . : 267864 ) . , r. , shectman , s. a. , gunnels , s. m. , mochnacki , s. , & athey , a. e. 2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 4841 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . m. iye & a. f. m. moorwood , 16941704 , a. f. , sneden , c. , kraft , r. p. , wallerstein , g. , norris , j. e. , da costa , g. , milone , a. p. , ivans , i. i. , gonzalez , g. , fulbright , j. p. , hilker , m. , piotto , g. , zoccali , m. , & stetson , p. b. 2011 , , 532 , a8 , a. p. , bedin , l. r. , piotto , g. , anderson , j. , king , i. r. , sarajedini , a. , dotter , a. , chaboyer , b. , marn - franch , a. , majewski , s. , aparicio , a. , hempel , m. , paust , n. e. q. , reid , i. n. , rosenberg , a. , & siegel , m. 2008 , , 673 , 241 , w. , bridges , t. , gillingham , p. , haynes , r. , smith , g. a. , whittard , j. d. , churilov , v. , lankshear , a. , croom , s. , jones , d. , & boshuizen , c. 2004 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 5492 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . a. f. m. moorwood & m. iye , 389400 , 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

using a hydrodynamic simulation of a cold dark matter universe with a cosmological constant (cdm ) , we investigate the `` x - ray forest '' absorption imprinted on the spectra of background quasars by the intervening intergalactic medium ( igm ) , at redshift . in agreement with previous studies , we find that and produce the strongest absorption features . the strong oxygen absorbers that might be detectable with _ chandra _ or _ xmm - newton _ arise in gas with and overdensities that are characteristic of galaxy groups . future x - ray missions could detect weaker oxygen absorption produced by gas with a wider range of temperatures and the lower densities of unvirialized structures ; they could also detect x - ray forest absorption by carbon , nitrogen , neon , iron , and possibly silicon . if the igm metallicity is , as we assume in most of our calculations , then the predicted number of systems strong enough for a detection with _ chandra _ or _ xmm - newton _ is extremely low . however , scatter in metallicity increases the number of strong absorbers even if the mean metallicity remains the same , making the predictions somewhat more optimistic . our simulation reproduces the high observed incidence of ( 1032 , 1038 ) absorbers , and the most promising strategy for finding the x - ray forest is to search at the redshifts of known systems , thus reducing the signal - to - noise threshold required for a significant detection . however , while many absorbers have associated or absorption , the systems trace only the low temperature phases of the x - ray forest , and a full accounting of the strong and systems will require a mission with the anticipated capabilities of _ constellation - x_. the large effective area of the _ xeus _ satellite would make it an extremely powerful instrument for studying the igm , measuring x - ray forest absorption by a variety of elements and revealing the shock - heated filaments that may be an important reservoir of cosmic baryons .

astro-ph0203319

the primordial deuterium abundance , combined with the theory of big bang nucleosynthesis , implies a cosmic baryon density @xmath6 ( @xcite ; here @xmath7 ) , in agreement with recent estimates from the cosmic microwave background anisotropy measurements @xcite . this value is an order of magnitude higher than the estimated density of stellar mass ( e.g. , @xcite ) . at redshifts @xmath8 , it appears that most baryons reside in the diffuse , photoionized intergalactic medium ( igm ) that produces the forest : that is what hydrodynamic simulations predict @xcite , and it is what the measured opacity of the forest implies given reasonable estimates of the neutral fraction @xcite . by low redshift , according to the simulations , some of this gas makes its way into galaxies and some remains in a diffuse photoionized medium at @xmath9 , but a significant fraction is shock - heated to higher temperatures @xcite . @xcite coined the term `` warm - hot intergalactic medium '' ( whim ) to refer to intergalactic gas in the temperature range @xmath10 , and a variety of numerical simulations predict that 30 - 40% of baryons reside in the whim at @xmath11 @xcite . while the hot ( @xmath12 ) , dense ( @xmath13 ) gas in galaxy clusters can be readily detected in x - rays , most of the shocked igm lies at lower temperatures and densities , where its x - ray emission is relatively feeble . one of the few prospects for observing this component is via the x - ray _ absorption _ that its highly ionized metal lines would produce in the spectra of background quasars . the absence of continuous `` x - ray gunn - peterson '' absorption provided early limits on the density and temperature of the igm @xcite . anticipating the high spectroscopic resolution of chandra and xmm - newton , and of future , more powerful x - ray telescopes like constellation - x and xeus , ( * ? ? ? * hereafter hgm ) and @xcite proposed the concept of an `` x - ray forest , '' a pattern of resonant metal absorption lines analogous to the forest . using a hydrodynamic cosmological simulation , hgm predicted distribution functions of and absorption . @xcite and @xcite used analytic methods based on the press - schechter ( @xcite ) mass function to make similar predictions , considering additional ionic species and a range of cosmological models . x - ray afterglows of @xmath14ray bursts could also be used as background sources , in addition to bright quasars @xcite . this paper examines the x - ray forest phenomenon in a large , smoothed particle hydrodynamics ( sph ) simulation of the @xmath0cdm model ( inflationary cold dark matter with a cosmological constant ) . we extend the work of hgm by considering a wider range of ionic species , looking more closely at the physical state of the gas responsible for x - ray absorption , and examining the prospects for detection of the x - ray forest with current and next - generation x - ray telescopes . our study is similar in approach to the recent work of ( * ? ? ? * hereafter fbc ) , but there are significant differences in the atomic processes incorporated in the calculations ( see [ sec : ionfrac ] ) and differences in focus of the analysis . the recent detection of a high density of ( 1032 , 1038 ) absorbers @xcite offers a tantalizing hint of a baryon component that might correspond to the warm - hot igm . although we have not designed our study primarily for predictions , we do calculate the expected incidence of absorption and investigate the connections between absorbers and x - ray forest systems ( and ) . @xcite and @xcite provide detailed predictions using hydrodynamic simulations . the next section describes our sph simulation and our methods for calculating x - ray forest absorption . section 3 presents our predictions for the equivalent width distributions of a variety of metal lines and discusses the feasibility of detection . section 4 focuses on the physical properties of the strong oxygen absorbers , which appear to be the most promising targets for the near future . we summarize our results in 5 . the cosmological simulation analyzed here is identical to the simulation d1 described in @xcite . some illustrations of structure in this simulation appear in figures [ zoombox ] and [ boxmap ] , which are discussed in [ sec : physics ] below . we adopt a flat @xmath0cdm model with @xmath15 , @xmath16 , @xmath17 , @xmath18 , and the primordial power spectrum predicted by the inflationary cdm model with an inflationary spectral index @xmath19 and normalization @xmath20 . this cosmological model is consistent with observational constraints from the cosmic microwave background , large scale structure , galaxy clusters , the forest , and type ia supernovae , though the value of @xmath21 is somewhat higher than favored by the most recent measurements @xcite . the simulation box is a periodic cube of side - length @xmath22 comoving mpc . at the start of the simulation , it contains @xmath23 dark matter particles and an equal number of gas particles ; some of the gas particles are converted to collisionless star particles during the course of the calculation . the masses of individual dark matter and gas particles are @xmath24 and @xmath25 , respectively . the simulation is evolved from @xmath26 to @xmath11 using the parallel version @xcite of treesph @xcite , which utilizes a tree algorithm for gravitational forces and sph for hydrodynamic interactions . the gravitational force resolution is @xmath27 ( equivalent plummer softening ) , in comoving units . the spatial resolution of the hydrodynamic calculation varies with density because of the lagrangian nature of the sph algorithm ; all gas dynamical quantities are computed by smoothing with a symmetric spline kernel that encloses 32 sph particles , or at most @xmath28 of baryons , with a minimum sph smoothing length of @xmath29 ( comoving ) . the simulation incorporates shock heating and compton and radiative cooling , as well as adiabatic heating and cooling . radiative cooling rates are computed for a primordial composition gas , as described by @xcite . cold gas in regions with density comparable to the galactic interstellar medium is converted into stars ; since this has little impact on the igm at the densities and temperatures relevant to the x - ray forest , we refer the reader to @xcite for a discussion of the star formation algorithm . at the mass resolution of this simulation , inclusion of heating by a photoionizing background artificially suppresses the formation of galaxies @xcite , so the simulation was run without a photoionizing background . as a consequence , expansion of the universe cools unshocked gas to very low temperatures ( less than @xmath30 ) . in higher resolution simulations that incorporate a photoionizing background @xcite , unshocked gas at redshift @xmath11 follows a tight temperature - density relation , @xmath31 , which arises from the competition between photoionization heating and adiabatic cooling @xcite . ( the mean temperature for this cosmological model is higher than that in the @xcite [ @xcite ] simulations because of the higher baryon density . ) before analyzing the simulation , therefore , we raise the temperatures of all gas particles with baryon overdensity @xmath32 and @xmath33 so that they lie on this temperature - density relation . we do not adjust the temperatures of shock - heated gas particles that already lie above this relation . this after - the - fact adjustment is a good approximation to the gas temperatures we would obtain in a simulation with an ionizing background . most of the significant x - ray forest absorption will come from hotter , denser gas that is unaffected by this adjustment . in this paper , we concentrate entirely on the @xmath11 output of the simulation . owing to the difficulty of detecting x - ray forest lines at all ( as discussed in [ sec : predictions ] below ) , observational studies are likely to focus on the quasars with the highest apparent x - ray fluxes , and hence on low redshift . we obtain density , temperature , and peculiar velocity profiles along 1200 randomly chosen lines of sight through the simulation cube , 400 in each projection , using the algorithm described by @xcite , as implemented in katz & quinn s code tipsy . each line of sight is divided into 5000 bins , so that each bin has a width of 1 km / s in velocity units , which is much finer than the few hundred km / s spectral resolution typical of the current and next generation x - ray telescopes . the contribution of each gas particle to each bin of the profile is calculated by a line integral through its spherically symmetric spline kernel , with contributions going to zero beyond the smoothing scale that encloses 32 sph neighbors . we use the densities and temperatures to compute ionic species abundances as described below ( [ sec : ionfrac ] ) , then calculate absorption spectra in redshift space by accounting for peculiar velocities and thermal broadening . we treat the projection axis of the cube as the redshift direction and make use of the periodic boundary , so a feature that would be `` shifted out '' of the cube by its peculiar velocity reappears at the opposite side . except for the computation of ionic abundances , the calculation is entirely analogous to the calculation of forest absorption by @xcite . consider an absorption system of a given temperature , density , and path length . the strength of the absorption line produced by a given transition of a given ion is proportional to the path length , the transition oscillator strength , the abundance of the element in question , and the fraction of that element in the given ionization state , which depends in turn on the temperature , density , and the ionizing background radiation . the predicted number of lines per unit redshift also depends on the covering factor of absorbers with the temperature , density , and path length required to yield detectable absorption . to decide which lines to consider in our study , we first made a preliminary screening of all of the ions and transition lines listed in @xcite . we selected those with element abundance greater than @xmath34 and resonance lines with energy above 0.25 kev ( @xmath35 ) . for each ion we selected the resonance line with the greatest oscillator strength . we then calculated the total absorption produced along all 1200 lines of sight through the simulation box , i.e. , the summed equivalent widths of all of the absorption lines produced by these transitions , and ranked the ions accordingly . the ions that produced at least one of the ten strongest lines are listed in table [ ions ] . in addition to and , we found that , , , , , , , and produce relatively large amounts of absorption . detection of x - ray forest lines requires that the observed energy be larger than @xmath36 , both because of instrumental sensitivity and to avoid absorption by the galactic interstellar medium . lines in table [ ions ] with energy @xmath37 can therefore be observed only to a maximum redshift @xmath38 . @xcite have used press - schechter models to predict absorption by and by two ionic species with much higher transition energies , ( @xmath39 ) and ( @xmath40 ) , which can potentially be observed out to high redshifts ( see also @xcite ) . these last two ionization states become abundant only at the high temperatures ( @xmath41 and @xmath42 , respectively ) characteristic of rich galaxy clusters , which are poorly represented in our simulation box owing to its limited size . we therefore do not attempt to make predictions for these ions . the analytic method used by @xcite is well adapted to the study of these highly ionized species , which are likely to occur only in the virialized regions that are best described by the press - schechter approximation and which are too rare to be well represented in our hydrodynamic simulation . our numerical approach complements this analytic method by allowing more accurate investigation of lower ionization species , which may be found in lower density , unvirialized regions . the and absorption associated with forest lines implies that the metallicity of the low density igm is @xmath43 of solar at high redshift ( see , e.g. , @xcite ) . at low redshift ( @xmath44 ) , @xcite found [ c / h ] @xmath45 for hi absorbers with @xmath46 using hubble fos data . in the denser igm of x - ray clusters , on the other hand , the inferred metal abundance is as high as one third of solar @xcite . unfortunately , these observations give little guidance to the metallicity of the regions with @xmath47 that are responsible for the x - ray forest at @xmath11 . while metal enrichment can be tracked in a simulation either during dynamical evolution , as in hgm and @xcite , or in post - processing , as in @xcite , the numerical predictions are sensitive to the uncertain details of metal ejection from the immediate vicinity of galaxies . rather than adopt an explicit but model - dependent prescription for the metallicity as a function of environment , we will generally assume that the metallicity of the medium that produces x - ray forest absorption is @xmath48 , intermediate between the forest and cluster igm values . we also assume that the relative abundances of the different elements are similar to those in the sun ; some authors ( e.g. , fbc ) have adopted enhanced abundances of @xmath49-elements ( oxygen in particular ) , characteristic of type ii sn enrichment . since most absorbers are unsaturated , predicted equivalent widths are simply proportional to the assumed metallicity , but this proportionality breaks down for the strongest absorbers . we discuss the impact of different metallicity assumptions , in particular the effects of including a density - dependent metallicity and scatter about the mean metallicity - density relation @xcite , in [ sec : ewdist ] below . as discussed below in [ sec : ionfrac ] , ionic abundances are affected by both collisional ionization and photoionization . to compute the latter , we must adopt an intensity and spectrum of the x - ray background , which contains the ionizing photons that are relevant to these high ionization species . for most of our calculations , we adopt the background of ( * ? ? ? * hereafter mi98 ) , which was based on asca and rosat data : @xmath50 where @xmath51 . this has a higher intensity and slightly steeper slope than the ( * ? ? ? * hereafter bf92 ) background , which was adopted by hgm . the bf92 background spectrum is @xmath52 where @xmath53 and @xmath54 . figure [ xrb ] plots the bf92 and mi98 background spectra as dotted and short - dashed lines , respectively . extrapolation of these x - ray background fits to longer wavelengths substantially underestimates the flux expected from quasars and star - forming galaxies . while the uv photoionizing background makes little difference for high ionization x - ray lines , it does matter for determining the ionic abundances of hi , , and , at relatively low densities and temperatures , . we therefore include a uv background @xmath55 with @xmath56 , based on @xcite . this background is shown by the long - dashed line in figure [ xrb ] . we generally use the mi98 spectrum at energies @xmath57 and the uv background ( [ eqn : uvb ] ) at energies @xmath58 , where it exceeds the mi98 spectrum , though we have repeated some of our calculations for different background choices as discussed below . we calculate the ionization fraction for each ion at a number of temperature and density grid points with an assumed radiation background , using the publicly available code cloudy @xcite . the ionization fraction of each ion for any given temperature and density is then obtained by interpolation . figures [ cnofrac ] and [ nesifefrac ] illustrate the resulting ionization fractions that enter our calculations . figure [ cnofrac ] show results for the hydrogen - like , helium - like , and lithium - like states of oxygen ( top row ) , nitrogen ( middle row ) , and carbon ( bottom row ) . curves in the central panels are calculated using our standard radiation background for a gas density @xmath59 , which corresponds to a baryon overdensity @xmath60 at @xmath11 for our adopted cosmological parameters . right - hand panels show calculations for @xmath61 , an overdensity @xmath62 . left - hand panels show results for collisional ionization only , in which case ionization fractions are independent of density . ( we actually compute these curves using @xmath59 and a radiation background reduced in intensity by a factor of @xmath63 . ) since oxygen , nitrogen , and carbon all exhibit similar behavior , we restrict ourselves to oxygen in the following discussion . in the absence of photoionization , the fragile , lithium - like species exists only over a narrow temperature range , @xmath64 , and even in this range its fractional abundance never exceeds 0.25 . the helium - like state , , dominates a much wider temperature range , with a fraction exceeding 0.5 for @xmath65 . at higher temperatures , begins to give way to , though the fraction never exceeds @xmath66 because temperatures high enough to produce are also high enough to produce ( not shown on the plot ) . similarly , at low temperatures and give way to lower ionization species . at these densities , photoionization has a major impact on the predicted abundances . there is still a general trend of moving from to to as the temperature increases , but with photoionization included each species occupies a much wider temperature range , and there are more temperatures at which multiple species co - exist . for @xmath62 , the peak fraction occurs at @xmath67 , and the fraction peaks at @xmath9 . for @xmath60 , the recombination rates are higher relative to the photoionization rates , so the curves shift towards those of pure collisional ionization , but even in this case remains significant down to temperatures of @xmath68 and to @xmath69 . more generally , for the densities characteristic of intergalactic filaments or the outskirts of clusters and groups , photoionization allows a species to exist at temperatures far below those deduced from the collisional equilibrium ionization fractions . the behavior of nitrogen and carbon is entirely analogous to that of oxygen , except that all the curves are shifted towards lower temperature owing to the lower ionization energies . for the more highly ionized species shown in figure [ nesifefrac ] , photoionization is usually much less significant , since there are fewer high energy photons . however , it still makes a difference in some cases for @xmath62 . note the expanded temperature scale of figure [ nesifefrac ] relative to figure [ cnofrac ] . the highly ionized iron species , and , occur only at the high temperatures characteristic of rich clusters , though occurs at the lower temperatures characteristic of the whim . figure [ bfvsmi ] shows the dependence of the oxygen predictions on the adopted radiation background . in each panel , heavy and light lines show the ionization fractions for the mi98 and bf92 x - ray background spectra , respectively . upper panels show results including the uv background spectrum of equation ( [ eqn : uvb ] ) ( using the maximum of the x - ray background and uv background at each wavelength ) , and lower panels show results using the x - ray background only . there is little difference between the results that use the mi98 or bf92 spectra , indicating that uncertainties in the x - ray background should introduce little uncertainty in our predictions . however , the uv background makes a significant difference for and to some extent , particularly at higher densities and lower temperatures . uncertainty in the intensity and spectral shape of the uv background introduces some uncertainty in our predictions , but it should make little difference to our predictions of x - ray absorption . it is worth noting at this point the two main physical differences between our calculations and those of fbc . the first is that our simulation incorporates radiative cooling , which changes the temperature structure of collapsed regions and allows some of the gas to condense into galaxies , removing it from the igm . the second is that our analysis incorporates photoionization , while fbc only include collisional ionization . as we can see from figures [ cnofrac]-[bfvsmi ] , photoionization makes an important difference for oxygen , nitrogen , and carbon , though it is less significant for higher ionization species . fbc also assume a much higher metallicity , with an oxygen abundance of 0.5 solar , but this difference just shifts absorber column densities by a factor of five . our input physics is similar to that of hgm , who incorporate radiative cooling and photoionization using the bf92 x - ray background but do not include a uv background . the transmitted x - ray flux is @xmath70 where @xmath71 is the quasar s x - ray continuum multiplied by the instrument response function and @xmath72 is the optical depth at observed frequency @xmath73 . we will assume that the continuum can be determined and divided out , so we define @xmath74 in equation ( [ eqn : fnu ] ) and refer to the transmitted flux @xmath75 as the simulated spectrum . strong atomic features are rare in this wavelength regime , so continuum estimation is relatively straightforward if the signal - to - noise ratio of the spectrum is high enough . the absorption spectrum is determined by a convolution of the ion number density along the line of sight with the absorption line profile , @xmath76 where @xmath77 label the atomic number , the ionization stage , and the transition line of the ion , respectively , @xmath78 is the oscillator strength of the transition , and @xmath79 is the ion number density . if we assume that the relative abundances of heavy elements are similar to those of the sun , we can write @xmath80 where @xmath81 is the ionization fraction for the @xmath82 ion computed from the density and temperature at position @xmath83 , @xmath84 is the number density of hydrogen atoms , @xmath85 is the solar abundance of element @xmath86 by number relative to hydrogen , and @xmath87 is the metallicity relative to solar . ( we regret the use of @xmath86 for two different purposes in the same equation but see no better alternative ) . the line profile @xmath88 is @xmath89,\ ] ] where @xmath90 is the hubble redshift at @xmath83 , @xmath91 is the peculiar velocity , and @xmath92= \frac{e^{-(\nu-\nu_{l})^2/\delta\nu_d^2}}{\sqrt{\pi}\delta\nu_d}\ ] ] is the thermal broadening profile with doppler linewidth @xmath93 , where @xmath94 is the ion mass . we do not incorporate damping wings in the line profile , but they are negligible for these weak absorption lines . as described in 2.1 , the average density , temperature , and velocity of gas for each of the 5000 bins in a line of sight were extracted from the simulation using tipsy . for each ion of interest , we calculate the ionization fraction for these temperature and density values by interpolating among the cloudy outputs . the optical depth along each line of sight is then obtained by carrying out the convolution , equation ( [ eqn : spectrum ] ) , numerically . we consider each ionic species in isolation , so in practice we do not perform the sum in equation ( [ eqn : spectrum ] ) , only the integral . because x - ray forest lines are rare and the bright quasars amenable to study with present instruments are at low redshifts , there will usually not be much ambiguity in identification of features . figure [ simspec ] shows a sample absorption spectrum , one that contains a strong feature . as the lower panels demonstrate , this absorption feature arises in the hot gas halo of a moderate sized group , with a typical gas temperature of a few@xmath95 . for clarity , we have ignored peculiar velocities when calculating these profiles , but we include thermal broadening in the optical depth and flux profiles . the dashed line in the top panel shows the redshift - space spectrum , in which the absorption feature is shifted by peculiar velocity but not changed substantially in width or depth . in all of our statistical calculations below , we identify absorbers and measure their properties from the redshift - space spectrum , including peculiar velocities . we identify absorbers using a simple threshold algorithm : an absorber is defined as a region where all the pixels have absorption above a given threshold . because x - ray forest absorbers are likely to be unresolved or nearly unresolved , at least for the foreseeable future , more elaborate algorithms that deblend distinct peaks or fit superpositions of voigt profiles are unnecessary for our present purposes . we adopt an absorption threshold of 0.01 , hence absorbers consist of contiguous pixels with transmitted flux less than 0.99 . we have checked the effect of using different thresholds and found that while the number of weak absorbers is slightly suppressed by a higher threshold , the number of strong absorbers is insensitive to the threshold value . in figure [ simspec ] , the boundary of the absorber is marked by ticks on the spectrum . the equivalent width of an absorber is @xmath96 where @xmath97 , and @xmath98 is the optical depth . the equivalent width defined in this way has velocity units . conversions to other useful units are @xmath99 , where @xmath100 and @xmath37 are the observed line wavelength and energy , respectively . for optically thin absorbers , there is a linear relation between equivalent width and column density , @xmath101 where @xmath102 is the column density and @xmath103 is the transition oscillator strength . since x - ray forest lines are weak and fairly broad , equation ( [ col - width ] ) is usually a good approximation . however , the equivalent widths of the strongest absorbers are depressed by saturation . figure [ colo78 ] demonstrates this effect , plotting the ratio of actual equivalent width to the optically thin value of equation ( [ col - width ] ) for the and absorbers along our 1200 lines of sight . the upper panels show results using our standard metallicity @xmath4 , and the lower panels show results using @xmath104 , i.e. , a factor of ten increase in absorber column densities . for , saturation becomes significant above @xmath105 , @xmath106 , and by @xmath107 the equivalent width is typically depressed by a factor of two . there are very few absorbers in this regime for @xmath108 , but a significant number for @xmath104 . saturation of lines does not become significant until @xmath109 , and we find few absorbers of this column density even with @xmath104 . we will generally present our results in terms of equivalent width ( including saturation effects ) because this quantity is more directly determined from observations , but we also label our plots by the corresponding optically thin column density of equation ( [ col - width ] ) , to aid with physical interpretations and to compare with other numerical and analytic predictions . figure [ optdisto ] presents our principal quantitative prediction , the cumulative distribution of and absorbers as a function of equivalent width . we plot @xmath110 , which is equivalent to the number of lines per unit redshift at @xmath11 . adopting the bf92 background in place of the mi98 one , or omitting the additional uv background at longer wavelengths , makes a nearly imperceptible difference to these curves . as mentioned earlier , the number of weak absorbers would be lower if we adopted a higher threshold for the line identification algorithm , but the number of strong absorbers would be unaffected . we also find that the peculiar velocities have little effect on the equivalent width distribution , though they change the equivalent widths of individual absorbers to some extent . the solid lines in each panel show our standard case of uniform metallicity with @xmath4 . the shapes of the and distributions are similar , with somewhat fewer high equivalent width lines for . the number of lines drops to one per unit redshift at @xmath111 ( 25@xmath112 for ) and falls rapidly for larger @xmath113 . the paucity of stronger absorbers is not surprising : for @xmath4 , a 1 mpc path length through a medium with @xmath114 ( @xmath60 ) and an fraction of 0.5 yields a column density of @xmath115 , or @xmath111 , and longer path lengths at such high density and fraction are rare . dotted curves in figure [ optdisto ] show predictions for a solar metallicity igm . the increase in oxygen abundance boosts the column density of each absorber by the same factor , and at low @xmath113 the equivalent width distribution simply shifts to the right by a factor of ten . at high @xmath113 the rightward shift is smaller , since saturation begins to suppress the linear relation between equivalent width and column density ; the high-@xmath113 tails of the solar metallicity distribution functions are therefore steeper . because the equivalent width distribution falls steeply at high @xmath113 even for @xmath4 , a modest increase in metallicity can still translate into a large increase in the predicted number of systems above some threshold i.e. , a small horizontal shift creates a large vertical shift . labeled vertical error bars in figure [ optdisto ] show representative @xmath116 detection thresholds for several different x - ray satellites , which we discuss in [ sec : capabilities ] below . unfortunately , one can already see that a comprehensive study of the x - ray forest with current instruments will be challenging unless the igm has near solar metallicity , which seems unreasonably optimistic . the remaining curves in figure [ optdisto ] are discussed at the end of the section , where we consider the potential impact of non - uniform metallicity . the equivalent width distributions for the other ions listed in table [ ions ] are shown in figures [ optdistcn ] and [ optdistnesife ] , computed assuming 0.1 solar metallicity and the mi98+uv background . the number of high equivalent width absorbers is generally much smaller than for or . roughly speaking , one can think of the lower element abundances and , in some cases , weaker transition oscillator strengths as shifting the previous equivalent width distributions towards lower @xmath113 , though of course the distribution also depends on the occurrence of density and temperature regimes where these ions have a high fractional abundance . as a guide to relative observability , we list in table [ ions ] the equivalent width threshold above which we find one absorber per unit redshift in our simulated spectra for our standard assumptions . this threshold is roughly proportional to the adopted metallicity , though it would increase more slowly above @xmath117 because of saturation ( see figure [ colo78 ] ) . and have the strongest absorption after oxygen , but their strong resonance lines lie at 40.3 and 33.7 , respectively , which restricts their observability to relatively low redshift quasars and puts them in a range of poorer instrument sensitivity even at zero redshift . and produce the next strongest absorption ; their transition energies are higher , but the predicted equivalent widths are a factor @xmath118 lower than those of and . while the focus of this paper is x - ray absorption , the ( 1032 , 1038 ) uv doublet of is another potential tracer of intergalactic gas at these densities and temperatures , and is observable from hubble space telescope or the far ultraviolet spectroscopic explorer ( fuse ) . recent observations have detected a number of absorbers towards low redshift quasars , implying a high number density @xmath119 of lines with rest - frame equivalent width @xmath120 @xcite . figure [ o6 ] shows the cumulative equivalent width distribution of the 1032 line of absorbers predicted by our simulation , together with data points from @xcite and @xcite ( taken from figure 2 of @xcite ) . we have used the mi98+uv background , but the results are not very sensitive to this choice . as in figure [ optdisto ] , solid and dotted curves show predictions for uniform metallicity @xmath4 and @xmath104 , respectively ; long - dashed and dot - short dashed curves show predictions that incorporate non - uniform metallicity with typical @xmath121 , as discussed at the end of the section . for any of the metallicity models except @xmath104 , the simulation results are in good agreement with the observations , though we caution that our threshold algorithm may not be as good a match to the analyses of these higher resolution , uv data . the dot - long dashed curve of figure [ o6 ] shows the results of @xcite , who carried out similar calculations using a different numerical method , with a metallicity model similar to that shown by our dot - short dashed line . the agreement between the two calculations is remarkably good . these results also agree fairly well with those of @xcite , though cen et al . predict somewhat less absorption than observed . simultaneous detections of , , and could provide insight into the physical state of the absorbing medium ; we discuss the expected correlations between these ionization phases in [ sec : physics ] below . in principle , absorption can also be studied with x - ray telescopes , using the kll resonance of the photoionization cross - section , i.e. , excitation - autoionization via the 1s @xmath122 2p transition resulting in 1s2s2p ( kll ) resonance . the strongest transition has @xmath123 , @xmath124 @xcite . the dashed line at the lower left of figure [ o6 ] shows the equivalent width distribution for this absorption . it is weaker than the corresponding and absorption ( compare to figure [ optdisto ] ) , primarily because of the lower fractional abundance of . x - ray detection of intergalactic absorption is therefore unlikely , at least without a substantial increase in instrumental sensitivity . the closest point of comparison for our calculation is that of hgm , who used a different numerical method but similar input physics . they present their and predictions in the form of differential distributions , whereas we have shown cumulative distributions in figure [ optdisto ] , but we have also created differential plots ( not shown ) and compared results . at small equivalent width , we have excellent agreement with hgm . however , the number of strong absorbers in our simulation ( for our standard metallicity assumption @xmath4 ) falls short of the number in theirs by a factor of a few or , equivalently , for absorbers that have a frequency @xmath125 per unit redshift , we predict a lower equivalent width by a factor @xmath126 . the cosmological parameters adopted by hgm are slightly different from ours , in the direction of increasing the number of absorbers ( they have @xmath127 ) . our box size of @xmath128 is intermediate between those of the two simulations used by hgm , @xmath129 and @xmath130 , but close enough to their larger box that this difference seems unlikely to have a large systematic effect . the total number of absorbers in the simulations at the equivalent widths where we disagree is small , and it could be that the difference simply reflects statistical fluctuations in the structure present in the two simulation volumes . however , we think that the most likely source of difference is the treatment of igm metallicity : we assume a uniform metallicity of 0.1 solar , whereas hgm track the spread of metals in their simulation , so that the metallicity varies with spatial position . they comment that typical metallicities are @xmath131 in the density and temperature range that dominate x - ray forest absorption , but they do not plot the scatter in metallicity , and it could be that the strongest absorption occurs in regions where the metallicity is higher than average . @xcite find a substantial scatter in metallicity at fixed overdensity in their hydrodynamic simulations , with a mean of @xmath132 at @xmath133 and @xmath134 variations of a factor of 23 . this effect would rather naturally explain the difference between our results and hgm s , with agreement at low @xmath113 but disagreement in the steeply falling tail of the distribution . to address this point quantitatively , we show two additional curves in each panel of figure [ optdisto ] , calculated for a non - uniform metallicity with statistical properties based on figure 2 of @xcite . dashed curves show the impact of metallicity scatter alone . we compute them by drawing the metallicity of each line of sight independently from a log - normal distribution with @xmath135 and @xmath136 . because we assign a single metallicity to a given line of sight , the fluctuation above or below the mean is coherent across each absorber . at low equivalent widths , where the distribution functions are shallow , metallicity scatter has little impact . for rare systems , however , the value of @xmath113 at a given value of @xmath137 increases by a factor of @xmath138 , since there are more intrinsically low column density systems to scatter to high metallicity than vice versa . dot - dashed curves show the effect of incorporating a density - dependent mean metallicity , with @xmath139 , while retaining log - normal scatter about the mean metallicity as before . this change depresses the distribution function at lower @xmath113 , where systems typically arise in lower density gas . we conclude that the differences in metallicity treatment could plausibly account for most of the difference between our calculations and that of hgm , and that a metallicity distribution similar to that of @xcite would increase the predicted equivalent widths of rare systems ( at fixed line density ) by a factor @xmath118 relative to our uniform , @xmath4 case . we also show predictions for these non - uniform metallicity models in figure [ o6 ] ; the qualitative effect of non - uniform metallicity is similar , but the quantitative impact at the equivalent width thresholds of the observational data points is modest because these points do not lie on the extreme tail of the distribution . fbc predict a much higher occurrence of strong and absorption than either hgm or our standard model , but this difference is mainly attributable to their higher assumed oxygen abundance of @xmath140 , also adopted in the analytic calculations of fc and @xcite . ( these papers also use an older solar abundance for oxygen that is slightly higher than the one adopted here . in our units , their abundance would be @xmath141 . ) direct comparison of our figure [ optdisto ] to fbc s figure 13 is difficult because they show differential distributions of column density while we show cumulative distributions of equivalent width ( including saturation effects ) . however , t. fang has kindly provided the fbc data in numerical form , and we have made direct comparisons of the column density distributions . at high column densities ( @xmath142 ) , the fbc results for oxygen abundance @xmath143 lie very close to our results for @xmath104 . ( recall that in this regime , the equivalent widths in figure [ optdisto ] have typically been depressed by a factor of two or more relative to their optically thin values . ) the fact that the predictions match for metallicities that differ by a factor of 1.7 can plausibly be attributed to the presence in our simulation of radiative cooling , which converts some of the hot gas in collapsed halos into cold gas and stars . at lower column densities , our solar metallicity model predicts substantially more absorbers than fbc ( e.g. , by a factor @xmath144 at @xmath145 ) . the difference in the shapes of the distribution functions can be plausibly attributed to the impact of photoionization , which becomes important at the lower densities and temperatures characteristic of weaker systems . once we allow for differences in assumptions about metallicity , we find the general level of agreement between our results and those of hgm and fbc ( and @xcite and @xcite for ) to be encouraging . it suggests that , at the factor of two level , the x - ray forest predictions from hydrodynamic simulations are not sensitive to the numerical methodology , or even to the inclusion of radiative cooling and star formation , and that photoionization is significant for weaker absorption systems but not for the strongest absorbers . the largest uncertainty in the predictions is the poorly known metallicity distribution of the igm , and for , the strength of the metagalactic flux at @xmath146 9 ryd . fbc compare their numerical results to the analytic , press - schechter based approach of fc and @xcite in some detail . they find that the analytic approach works well for high - excitation ions ( e.g. , and ) and for the highest column density systems of lower excitation ions ( e.g. , @xmath147 for and ) , while it tends to underestimate the occurrence of weaker absorption , which often arises outside of the collapsed halos modeled by the press - schechter formalism . the best spectral resolution of the current and next generation x - ray telescopes is @xmath148 , or about @xmath149 in velocity units . most absorbers , therefore , are likely to be unresolved . in a quasar spectrum , the minimum detectable equivalent width for an unresolved absorption feature is @xmath150 where @xmath151 is the minimum signal - to - noise ratio required for detection , @xmath37 is the observed energy of the line , @xmath152 is the integration time , @xmath153 is the spectral resolution , @xmath154 is the effective area , and @xmath155 is the photon flux , e.g. , in units of photons @xmath156 . equation ( [ eqn : wmin ] ) yields an equivalent width in the same energy units used for @xmath37 and @xmath155 . an x - ray spectrum of a moderate redshift quasar might have @xmath157 resolution elements in which absorption could potentially fall , so the @xmath158 threshold for a `` blind '' detection must be fairly high to avoid the spurious identification of noise peaks as absorption systems . we therefore adopt @xmath159 as a threshold for secure detection in our discussion below . a @xmath160 feature would have a probability @xmath161 of arising by chance , assuming gaussian statistics , so it would be sufficient for a tentative detection . the characteristics of each x - ray telescope depend on the choice of instrument setting ( grating , detector , etc . ) . we will not discuss these issues in detail here , but we list in table [ instrument ] some representative value of effective area and resolving power for _ chandra _ , _ xmm - newton _ , _ constellation - x _ and _ xeus_. to assess detectability thresholds , we assume a quasar flux @xmath162 and we adopt as an example the x - ray bright quasar h1821 + 643 ( @xmath163 ) , with @xmath164 photons @xmath156 at @xmath165 , and @xmath166 ( s. mathur , private communication ; see @xcite for more detailed discussion of h1821 + 643 ) . we have not accounted for the effects of galactic hi absorption , which depresses the flux by @xmath167 at wavelengths relevant to or detection . values of @xmath168 for @xmath159 and an integration time of 500 ksec are listed in table [ equivwidth ] , for a variety of observed wavelengths . figures [ optdisto][optdistnesife ] show that for a given path length it is easiest to detect and absorbers . however , as shown in table [ equivwidth ] , even a 500 ksec integration with the brightness of h1821 + 643 yields a minimum detectable equivalent width of @xmath169 for _ chandra _ and @xmath170 for _ xmm - newton_. these limits are marked by vertical bars in figure [ optdisto ] . assuming a metallicity of 0.1 solar , the strongest absorbers in our simulated spectra have @xmath171 , which are at best marginally detectable by _ chandra _ and _ xmm - newton_. the curves with metallicity scatter are arguably more realistic than the constant metallicity case , for the reasons discussed by @xcite , and it is possible that enriched regions of the igm have enhanced ratios of @xmath49-elements like oxygen relative to iron . both of these changes would improve the prospects of detection . a search for and absorption associated with known absorption ( from uv observations ) holds considerably more promise than a blind search for x - ray forest absorption , for two reasons . first , when searching at a redshift that is known a priori , the required @xmath158 for a significant detection is lower , @xmath172 instead of @xmath5 , and at these equivalent widths the number of systems rises sharply with decreasing detection threshold . second , as we will show shortly , the strongest absorbers frequently have strong associated absorption , though this is less true for . _ constellation - x _ , which is likely to be the next major advance in capability for x - ray spectroscopy , is now under design . for the observing conditions assumed in table [ instrument ] , the minimum detectable width for _ constellation - x _ is about 25 km / s ; thus , for a metallicity of 0.1 solar , we predict that constellation - x could detect 2 - 3 absorbers and @xmath125 absorber per unit redshift along each random line of sight , at @xmath116 significance . the numbers rise slightly if one adopts the @xcite metallicity - density relation , and if the metallicity is as high as solar then tens of absorber should be detected per line of sight . the powerful _ x - ray telescope , with a typical equivalent width threshold of @xmath173 , should be able to find large numbers of absorbers . we now turn to the physical state of the gas producing oxygen absorption , which appears to be the element most amenable to observational study . figures [ zoombox ] and [ boxmap ] are a useful starting point . figure [ zoombox ] shows the gas particles in the simulation color - coded by temperature ( upper left ) , density ( upper right ) , ion fraction ( lower left ) , and ion fraction ( lower right ) . we have zoomed in on a @xmath174 sub - volume of the simulation in order to show finer details of the structure . gas in virialized halos has been shock - heated to typical temperatures of @xmath175 , while gas in the filaments connecting these halos is typically somewhat cooler , from @xmath176 to a few@xmath95 . the regions of high and ionization fractions trace the general structure of these filaments and halos . the fraction shows fairly sharp peaks in the dense inner regions of the halos , though there is by no means a one - to - one correspondence between density peaks and peaks of fraction . tends to dominate in the outer regions of halos and filaments , where the density is lower . the relatively massive , hot group at the center of the upper panels has low and fractions because its temperature is high enough to fully ionize the oxygen to . figure [ boxmap ] shows the projected column density of in the full @xmath128 simulation cube , with threshold column densities in the four panels of @xmath177 , @xmath178 , @xmath179 , and @xmath180 . the assumed igm metallicity is 0.1 solar . at the @xmath181 threshold , traces a rich network of shock - heated filaments . at @xmath182 , the stronger filaments remain , but the weaker ones disappear from view . by @xmath183 , the distribution consists mainly of distinct clumps , occasionally connected by a more diffuse medium . at @xmath180 , only the dense regions in collapsed halos are visible . the redshift path length through the simulation volume is only @xmath184 , so the covering factor for absorption to a quasar of redshift @xmath185 would be larger by @xmath186 , e.g. , by a factor of 18 for @xmath187 . the column density thresholds in the last three panels correspond roughly to the detectability thresholds for _ xeus _ , _ constellation - x _ , and _ chandra_/_xmm - newton _ , given the assumptions discussed in [ sec : capabilities ] . if the igm metallicity is @xmath188 , therefore , current generation x - ray telescopes have the potential to detect x - ray forest absorption in high - density gas , _ constellation - x _ can begin to reach into intergalactic filaments , and _ xeus _ can map the filamentary structure . if the igm metallicity is 0.3 solar , then the same panels represent column density thresholds that are @xmath189dex higher , so the _ chandra_/_xmm - newton _ threshold would correspond to the lower left panel , the _ constellation - x _ threshold to the upper right , and the _ xeus _ threshold to the complex network of filaments in the upper left . to quantify the physical properties of the oxygen absorption systems , we calculate the average temperature and overdensity of the gas associated with each absorber . we first define a _ pixel _ temperature and density , assigning to each pixel along the line of sight a temperature and density that is an average of the gas density and temperature contributing to that pixel weighted by optical depth , @xmath190^{-1 } \int dx \left(\frac{d\tau}{dx}\right ) n(x),\\ t_{\rm pix}(v ) & = & [ \tau(v)]^{-1 } \int dx \left(\frac{d\tau}{dx}\right ) t(x).\end{aligned}\ ] ] the absorber temperature and overdensity are then obtained by averaging ( with equal weight ) over the pixels belonging to that absorber . thermal broadening allows gas of different densities and temperatures to contribute to a given pixel , but the optical depth weighting ensures that the average is not dominated by gas that is not actually responsible for the corresponding absorption . figure [ ophysprpty ] plots the temperature and overdensity of each absorber against its equivalent width , for absorbers with @xmath191 . the assumed metallicity is @xmath108 ; for a higher assumed metallicity one would , in the optically thin approximation , simply shift each point horizontally by the same factor by which the metallicity is increased . most absorbers have temperature @xmath192 , absorbers have @xmath193 , and absorbers @xmath194 . for all three species , there is a correlation between equivalent width and gas overdensity , though there are some notable outliers from these correlations . in the case of absorbers , the median overdensity increases from @xmath195 at @xmath196 to @xmath197 at @xmath198 , and the three absorbers with @xmath199 all have @xmath200 . the system that produces the strongest and absorption has @xmath201 , but this system is weak in because of its high temperature ( @xmath202 ) . in figures [ ophysprpty][o1 - 678 ] , special symbols mark the five systems with the strongest absorption . the `` same '' absorber may have different density and temperature in different panels because of the optical depth weighting by the corresponding ionic species and because the velocity ranges defined in the different species may not be identical . figure [ ophysprpty ] confirms the impression from figures [ zoombox ] and [ boxmap ] that strong and absorption arises in high density gas , more characteristic of virialized halos than connecting filaments . at the virial boundary . for example , @xcite find @xmath203 for @xmath0cdm at the radius where the mean interior density is 200 times the _ critical _ density and @xmath204 at the radius where the mean interior density is 200 times the mean density . ( see their figure 3 . their @xmath0cdm parameters , @xmath205 , @xmath206 , and @xmath207 , are significantly different from ours , in a direction that produces denser halos . ) ] strong absorbers also tend to be fairly high overdensity , though there are some strong systems with @xmath208 , and the highest overdensity absorbers tend to be weak in because they are too hot . for each species , the strongest absorbers usually have temperatures fairly close to the value where the corresponding ionic abundance peaks in collisional equilibrium ( see figure [ cnofrac ] ) , but there is a spread in temperature at every value of @xmath113 , and for all three species there is a tail of weaker , lower temperature , lower density absorbers for which photoionization plays an important role . our conclusions about agree with those of @xcite and @xcite , in that collisional ionization dominates for the strongest absorbers but photoionization becomes important at slightly lower equivalent widths . our conclusions about and appear consistent with those of hgm , to the extent that we can compare them . the collisional equilibrium assumption used by fbc in their simulation analysis should be fairly accurate for the strongest and absorbers , but it breaks down for the lower density absorbers associated with filamentary structures . with sufficiently good data , it should be possible to measure absorption by multiple ionization stages of oxygen , or to obtain limits on the relative absorption by different species . figure [ ocorr ] shows the predicted correlations between and equivalent widths ( top ) , and ( middle ) , and and ( bottom ) . there is some ambiguity in the definition of these correlations , since the threshold algorithm will , in general , choose somewhat different velocity ranges to mark the same absorber in different species , and in some cases there may be an absorber above threshold in one species but no corresponding absorber in the other . in figure [ ocorr ] , crosses represent absorption in the velocity range defined by the ion on the abscissa , and squares represent absorption in the velocity range defined by the ion on the ordinate . in many cases , the absorbers defined in these two ways nearly coincide , so the square and cross are close to each other , but in some cases , especially in the vs. panel , there are absorbers in one species with no obvious counterpart in the other . the top panel shows a fairly good correlation between and absorption , as one might expect on the basis of figures [ cnofrac ] and [ bfvsmi ] : there are substantial ranges of temperature where both species have high fractional abundance , and in this regime the density of both species increases in proportion to the gas density . however , the spread in igm temperatures does produce significant spread in the / ratio . in particular , there are some strong systems with relatively weak ; these correspond to high density gas where the temperature is too low to produce substantial by collisional ionization . the vs. diagram shows a tight core of points with a well defined correlation , tracing the underlying correlation of both species with density ( see figure [ ophysprpty ] ) . the strongest absorbers tend to have strong associated , which suggests that following up known absorbers is a promising way to find x - ray forest absorption . however , there are some strong absorbers with weak , corresponding to dense , relatively cool gas in which lower ionization species dominate . moreover , the strong absorbers ( @xmath209 ) have a wide range of strength , including quite low values , since the fraction drops to nearly zero once the temperature exceeds @xmath210 . thus , while is a useful signpost for absorption , it will not mark the positions of all or even most strong absorbers , nor should the existence of strong absorption with weak come as a surprise . despite the correlations between and and and , there is virtually no correlation between and ; the variations in temperature at fixed density that produce much of the scatter in the top two panels drive the / and /ratios in opposite directions . the lack of correlation in the bottom panel is not surprising in light of figure [ cnofrac ] and [ bfvsmi ] , which show only narrow ranges of temperature for which and both have substantial fractional abundance . figure [ myhist ] displays these predictions in a form that is useful for interpreting x - ray forest searches at the redshifts of known absorbers . upper panels show the distributions of and equivalent widths for systems with equivalent width above @xmath211 ( 103 m ) . both distributions are bimodal , with a main peak of cooler systems that have weak or and a secondary peak of hotter systems that have stronger x - ray absorption . lower panels show the corresponding distributions for an equivalent width threshold of @xmath212 ( 34 m ) . here the distributions are more dominated by weak and systems , though there is still a tail of stronger absorbers . once again it is clear that following up strong absorbers is a useful way of finding x - ray forest absorption , but that many absorbers , especially the weaker ones , have little associated or . figure [ o1 - 678 ] shows the predicted correlations of , , and equivalent width with ( ly@xmath49 ) equivalent width , where in all cases we take the velocity ranges defined by the absorbers . for there is a tight correlation at low equivalent widths , reflecting the increase of both and with density in photoionized gas . there is a spread of points towards higher associated with shock - heated gas raising the temperature always decreases the hi fraction at a given density , but it can increase the fraction . at high equivalent widths the correlation flattens and spreads , but it remains the case that most of the strong systems have strong associated , and vice versa . there are hints of the same correlation present in the diagram , but in this case the scatter is considerably larger , and at the high equivalent width end ( defined either by high or by high ) there is an enormous spread in / , with essentially no discernible correlation . the - diagram shows no significant correlation or anti - correlation , except that the strongest systems all have absorption with @xmath213 . given the radically different temperature ranges required for and to exist in substantial abundances , it seems likely that the absorption is arising in the environs of the hotter gas producing the absorption , rather than in the same gas . unfortunately , the large spread in / and / suggests that it will be difficult to infer the metallicity of x - ray forest systems directly from line ratios . the prospects are better for inferring the metallicity in absorption systems , but these trace only the cooler portions of the forest . the advent of x - ray telescopes with large effective areas and instruments for high resolution spectroscopy opens a new window on the intergalactic medium , one that could reveal a large reservoir of previously hidden baryons . we have used a hydrodynamic simulation of the @xmath0cdm cosmological model to predict the properties of the x - ray forest at @xmath11 . in agreement with previous studies based on numerical simulations ( hgm ; fbc ) and analytic methods @xcite , we find that the oxygen species and produce the strongest x - ray forest lines , thanks to the high cosmic abundance of oxygen , the oscillator strengths of the transitions at 21.60 ( ) and 18.97 ( ) , and the relatively frequent occurrence of temperature and density regimes where these species have large fractional abundance . if the igm metallicity at these temperatures and densities is @xmath4 , as we assume in most of our calculations , then and are the only x - ray forest lines likely to be detectable with existing instruments , and even these will be difficult to study in any comprehensive fashion . other promising lines for study with future instruments include the helium - like and hydrogen - like species of carbon , nitrogen , and neon ( , , , , , ) , , which has lower abundance but a large oscillator strength , and possibly . we have not examined highly ionized species like and , which occur only at high temperatures that are poorly represented in our @xmath128 box ( see @xcite for analytic predictions ) . our conclusions about the detectability of these x - ray forest lines are summarized by the rightmost column of table [ ions ] , which lists the equivalent width threshold at which we predict one line per unit redshift , and table [ equivwidth ] , which gives approximate @xmath116 detection thresholds for observations with _ chandra _ , _ xmm - newton _ , _ constellation - x _ , and _ xeus _ , assuming 500 ksec exposures and the source spectrum of the x - ray bright quasar h1821 + 643 ( @xmath163 ) . target - of - opportunity searches that observe blazars in very high states could potentially reach lower equivalent width thresholds ( nicastro , private communication ) . the full distribution functions for absorber equivalent widths are shown in figures [ optdisto][optdistnesife ] . the largest uncertainty in the predictions is the unknown metallicity distribution of the igm at the relevant temperatures and densities . this distribution can itself be predicted by tracking the dispersal of metals in cosmological simulations ( as done by , e.g. , hgm and @xcite ) , but the uncertainties in the escape of metals from the galaxies themselves make these results quite uncertain . we predict a lower frequency of detectable x - ray forest lines than fbc , mainly because we assume @xmath4 instead of the higher metallicity found in the central regions of hot x - ray clusters . for the same metallicity , the two simulations give similar predictions for the number of strong absorbers , though the inclusion of radiative cooling and photoionization in our calculations tends to reduce the column densities of the strongest systems and substantially increase the number of weak absorbers . our predicted equivalent width distribution agrees with that of hgm at low @xmath113 but falls below it at high @xmath113 , perhaps because of differences in the assumed cosmological parameters and in the simulation methods , but perhaps because their strongest absorbers arise in regions of higher than average metallicity . our assumption of a uniform metallicity of @xmath132 with solar abundance ratios ( instead of the @xmath49-enhanced ratios characteristic of type ii sn enrichment ) is probably conservative , and perhaps unduly pessimistic . the predicted equivalent width thresholds in table [ ions ] are approximately proportional to the assumed metallicity , except that they grow more slowly for @xmath214 as saturation becomes significant ( see figure [ colo78 ] ) . as shown in figure [ optdisto ] , the predicted number of systems at high equivalent width is sensitive to the scatter in igm metallicity as well as its mean value . for example , if we include 0.4 dex of scatter about 0.1 solar metallicity , then the and equivalent width thresholds corresponding to one line per unit redshift are about a factor of two higher than the values quoted in table [ ions ] . because the intrinsic equivalent width distribution is steep , the first x - ray forest systems to be detected are likely to lie on the high tail of the igm metallicity distribution we have devoted much of our attention to the physical state of the gas producing and absorption , as illustrated in figures [ zoombox][o1 - 678 ] , especially figure [ ophysprpty ] . strong and absorbers , with @xmath215 , arise mainly in gas with @xmath3 and temperature @xmath2 , where collisional ionization dominates over photoionization . weaker absorbers arise in lower density gas ( @xmath216 for @xmath217 ) over a wider range of temperatures , with photoionization often playing a significant role . for @xmath213 , all absorbers and most absorbers arise in the temperature regime @xmath10 identified by @xcite as the `` warm - hot igm , '' which contains a substantial fraction of the low redshift baryons in cosmological simulations @xcite . however , one should keep in mind that the absorbers potentially detectable with _ chandra _ or _ xmm - newton _ trace only the high density regions of the whim , with overdensities characteristic of virialized groups rather than intergalactic filaments . in agreement with @xcite and @xcite , we find that the @xmath0cdm model can plausibly account for the high incidence of ( 1032 , 1038 ) absorbers at low redshift @xcite , as shown in figure [ o6 ] . the absorbers exhibit a well - defined trend between equivalent width and overdensity , though there are some extreme outliers ( figure [ ophysprpty ] ) . at every equivalent width there is a substantial range of gas temperatures ( about a factor of ten ) , with photoionization dominating in the cooler systems . at @xmath217 ( @xmath218 ) , much of the absorption arises in gas with overdensity @xmath195 . in searches with _ chandra _ or _ xmm - newton , absorbers will play a crucial role as signposts for or absorption . if the igm metallicity is @xmath132 , or even @xmath219 , then the predicted number of x - ray forest systems strong enough to be detected at the @xmath5 level needed in a `` blind '' survey is very small . however , if one is searching at the redshift of a known system , then a @xmath220 detection is already significant , and the steepness of the equivalent width distribution translates a small change in detection threshold to a large change in predicted line density . ly@xmath49 absorption can also serve as a signpost , but the higher density of ly@xmath49 lines and the large scatter between and or ( figure [ o1 - 678 ] ) makes it less useful for this purpose than . recent observations provide several tentative detections of intergalactic and absorption . the most convincing detections are those of @xcite , who find absorption features of , , and in their _ chandra _ spectrum of pks 2155 - 304 , at wavelengths consistent with zero redshift . the implied and column densities are @xmath221 . _ xmm - newton _ and _ chandra _ observations also show @xmath11 absorption along lines of sight towards mrk 421 and 3c 273 @xcite . these observations suggest that highly ionized gas is fairly ubiquitous along lines of sight through the local group , as one might expect when looking out from a local maximum in the igm density ( see , e.g. , @xcite ) , though at least some of the absorption could be galactic in origin @xcite . our simulations predict that oxygen lines of this strength should be rare along random lines of sight beyond the local group ( fig . [ optdisto ] ) , and here the observational situation is more ambiguous . @xcite find a statistically significant absorption feature in the _ chandra _ spectrum of pks 2155 - 304 that could correspond to absorption coincident with a known galaxy group at @xmath222 , with an implied column density @xmath223 . however , the _ xmm - newton _ spectrum appears to rule out the existence of an absorption feature at this level ( rasmussen , private communication ; see @xcite ) . @xcite use a _ chandra _ spectrum of h1821 + 643 to search for absorption at the redshifts of previously known systems @xcite , obtaining apparent detections of one system in and another in both and . the statistical significance of the features is only @xmath224 in each case , but because the absorption redshifts are known _ a priori _ from the measurements , the probability that they are simply noise fluctuations is small ( @xmath225 ) . the rest - frame equivalent widths are @xmath226 and @xmath227 for the lines and @xmath228 for the line , with corresponding column densities @xmath229 ( ) and @xmath230 ( ) , assuming no saturation . the existence of two such strong systems in a path length @xmath231 is surprising relative to our predictions in figure [ optdisto ] , unless the metallicity in these systems is substantially higher than 0.1 solar , though it is worth noting that this line of sight also has an unusually high incidence of absorption @xcite . further measurements of or limits on and absorption associated with will take us much further towards understanding this newly discovered population of low redshift absorbers , providing constraints on their densities , temperatures , and ionization mechanisms and on their contribution to the cosmic baryon budget . however , absorbers trace only the low temperature end of the x - ray forest ; many strong and systems have very weak absorption , and , conversely , many strong systems will be undetectable in or ( figure [ ocorr ] ) . a full accounting of the strong and absorbers will probably require a mission with the capabilities of _ constellation - x _ , which may also begin to detect x - ray absorption from other elements in the same systems . even at the @xmath232 threshold of _ constellation - x _ , the detectable x - ray forest absorbers at @xmath4 correspond mainly to gas in the outer regions of groups and poor clusters , with some contribution from more diffuse gas in filaments . the powerful _ satellite would allow a comprehensive investigation of x - ray forest absorption by a number of ionization species , providing great insight into the distribution of shock - heated baryons and the enrichment of the intergalactic medium . we are grateful to smita mathur for helpful advice at many stages in this work , especially in matters related to observational capabilities . we also thank jordi miralda - escud for several valuable discussions and joop schaye for comments on the manuscript . we thank todd tripp for providing the observational data points used in figure [ o6 ] and taotao fang for providing the numerical results from @xcite and fbc ( discussed in [ sec : ewdist ] and shown in figure [ o6 ] ) . this work was supported by nasa ltsa grant nag5 - 3525 , nsf grant ast-9802568 , and chandra observatory grant g01 - 2118x from smithsonian astrophysical observatory . x.c . was supported at osu by the department of energy under grant de - fg02 - 91er40690 , and at itp / ucsb by the nsf under grant phy99 - 07949 . d.w . acknowledges the hospitality of the institute for advanced study and the financial support of the ambrose monell foundation during the completion of this work . r.d . was supported by nasa through hubble fellowship grant number hst - hf-01128.01-a awarded by the space telescope science institute , which is operated by aura , inc . , under nasa contract nas5 - 26555 . the simulation was performed at the san diego supercomputer center and ncsa . barlow , t. a. & tytler , d. , 1998 , , 115 , 1725 burles , s. , & tytler , d. 1997 , , 114 , 1330 burles , s. , & tytler , d. 1998 , , 507 , 732 cen , r. , miralda - escud , j. , ostriker , j. p. , & rauch , m. , 1994 , , 427 , l9 hellsten , u. , dav , r. , hernquist , l. , weinberg , d. h. , & katz , n. 1997 , , 487 , 482 hernquist , l. , & katz , n. 1989 , , 70 , 419 hernquist , l. , katz , n. , weinberg , d. h. , & miralda - escud , j. , 1996 , , 457 , l51 press , w. h. , & schechter , p. 1974 , , 187 , 425 rasmussen , a. , kahn , s. m. , & paerels , f. 2003 , to appear in the igm / galaxy connection : the distribution of baryons at @xmath11 , ( kluwer : dordrecht ) , astro - ph/0301183 equivalent widths corresponding to @xmath159 , computed from eq . ( [ eqn : wmin ] ) using the resolution and effective area values of table [ instrument ] , a 500 ksec integration time , and a quasar spectrum @xmath233 similar to that of the @xmath163 quasar h1821 + 643 .

high precision kepler photometry is used to explore the details of agb light curves . since agb variability has a typical time scale on order of a year we discuss at length the removal of long term trends and quarterly changes in kepler data . photometry for a small sample of nine sr agb stars are examined using a 30 minute cadence over a period of 45 months . while undergoing long period variations of many magnitudes , the light curves are shown to be smooth at the millimagnitude level over much shorter time intervals . no flares or other rapid events were detected on the sub - day time scale . the shortest agb period detected is on the order of 100 days . all the sr variables in our sample are shown to have multiple modes . this is always the first overtone typically combined with the fundamental . a second common characteristic of sr variables is shown to be the simultaneous excitation of multiple closely separated periods for the same overtone mode . approximately half the sample had a much longer variation in the light curve , likely a long secondary period . the light curves were all well represented by a combination of sinusoids . however , the properties of the sinusoids are time variable with irregular variations present at low level . no non - radial pulsations were detected . it is argued that the long secondary period variation seen in many sr variables is intrinsic to the star and linked to multiple mode pulsation .

1408.4323

the general catalog of variable stars lists multiple groups of pulsating variable stars of late spectral type with periods of several tens of days or longer . while historical divisions by period and spectral type were made in the absence of knowledge about stellar evolution , we now recognize that these long period variable ( lpv ) stars are evolved low- and intermediate - mass stars . during the red giant branch ( rgb ) , asymptotic giant branch ( agb ) , and post - agb phases stars are characterized by highly extended stellar envelopes providing the basis for long - period pulsation . the variability time scale of lpvs requires monitoring over years to decades to properly describe the light changes . in the 1990s long term , rapid cadence monitoring of variable stars in selected fields occurred as a byproduct of ground - based photometric surveys searching for microlensing events . this data was transformative , providing large data sets of high quality light curves for stars in the magellanic clouds @xcite . the magellanic cloud variables are at a known distance , and hence have known luminosities , permitting the construction of period - luminosity ( p - l ) diagrams for a large sample of late - type stars . while previously a p - l relation had been known for the large amplitude lpvs , the miras @xcite , period - luminosity relations for all late - type variables are apparent in the lmc data @xcite . in this paper we focus on those lpvs that are luminous with periods of @xmath0100 days or longer but are not miras . in the parlance of the @xcite p - l relation these are sequence c and c @xcite . in the classical variable star nomenclature these are semi - regular ( sr ) variables . the gcvs classification is unfortunately blurred across a large range of temperature and luminosity with sr variables found across a large part of the cool star p - l diagram . here we will discuss those that have agb luminosity . other recent papers have discussed sr variables that are less luminous and are on the rgb @xcite . the improved p - l diagram results in new astrophysical challenges . the c and c variables are pulsating radially in the fundamental mode , a low overtone mode , or a combination of modes . the origin of agb semi - regular behavior , both in cycle length and amplitude , remains unclear , but their light curves can in some cases be reproduced by the combination of several periods ( e.g. * ? ? ? * ; * ? ? ? the major challenge , however , is an explanation of the long secondary periods ( lsp ) found in many of the overtone pulsators . with a typical time scale of several hundreds to thousands of days the secondary periods are significantly longer than the fundamental mode periods and therefore can not be explained by radial pulsation . various solutions have been discussed , but a satisfactory explanation has yet to be found ( e.g. * ? ? ? one route to further understanding of these stars is precision , continuous photometric monitoring . recent studies faced the typical problems of ground - based observations , time series of very limited length , of limited photometric accuracy , and observational gaps resulting from both diurnal and seasonal cycles . a space mission obtaining photometry continuously over a time span of several years is , therefore , clearly relevant . one of the many side benefits of the kepler mission was the ortunity to obtain long and precise time series of variables , including lpvs with uninterrupted sampling extending over months . @xcite presented the first extensive study on the variability of both rgb and agb m giants using the long and continuous light curves provided by kepler . in that paper the authors focused on the global characteristics of the light variations of these cool giants . their study confirmed the presence of several pulsation modes based on the patterns seen in the petersen diagram and it revealed a clear distinction between solar like oscillations and large amplitude , mira - like pulsations . this transition seems to appear at log p(days ) @xmath0 1 . the kepler mission was optimized to search for planet transits , not slow variations spanning weeks to years . as a consequence , the aim of our paper is twofold . in the first part , we investigate the applicability of the kepler data products to the construction of long base line light curves by combining individual data - sets of three months length . we will focus on a small sample of 12 stars , three of which have small amplitudes that are included especially to identify instrument specific effects . the light curves are then used in the second part of the paper to discuss the light change of sr variables on the basis of these uninterrupted and photometrically precise light curves . the definition of the sr variable class limits the visual amplitude to less than 2.5 magnitudes . the mira variables are defined as having visual amplitude in excess of 2.5 magnitudes . thus the sr class excludes the large amplitude , tip - agb variables . the intent of our survey was to study agb sr variables over as full a range of variability as possible . unfortunely , most sr variables have not been extensively studied in the past . the better known objects have been assigned a subclass a or b reflecting a more or less regular light curve . however , many late - type stars have time series of ground based photometry that are too short or otherwise inadequate to classify the variability . the srb class has been associated with lsp phenomenon . a sample spanning these variability classes was of special interest with the goal of seeing if actual differences existed between the subclasses . the general catalogue of variable stars ( gcvs * ? ? ? * ) list of variables in lyrae and cygnus was searched for late - type sr variables in the kepler field . other surveys , for instance the asas catalog @xcite , sample a much shorter time interval and were less useful in selecting a variety of variables . the two brightest srs in the kepler field are af cyg , an srb with spectroscopically and photometrically studied lsp @xcite , and aw cyg , a carbon - rich srb variable . observation of these two bright stars required special target pixel masks and hence were not included in our original target list . however , @xcite selected af cyg as a target and developed a special pixel mask . we included that data . to represent the sra class we identified v588 lyr , v1670 cyg , and v1766 cyg . for the srb class v1953 cyg , eg lyr , and bu lyr were selected . we included two sr stars without sub - classifications , v1253 cyg and v2412 cyg . three low amplitude variable stars located in the field of the old open cluster ngc 6791 were analyzed to test the reduction process . the stars have periods in the range @xmath010 - 20 d @xcite with amplitudes of less than 0.1 magnitude . the brevity of the ground based ngc 6791 survey results in fairly uncertain periods . the general catalogue of variable stars ( gcvs * ? ? ? * ) classifies the test stars as red giants with semi - regular periods , i.e. stars that are likely members of sequence a or b in the lpv l - p diagram . the spectral types are correspondingly earlier than for the program stars with the test stars probably on the rgb . spectroscopy was carried out to determine the spectral types for the brighter of the program stars . the observations were made at the kitt peak national observatory using the 4-meter mayall telescope in may 2011 and the coude feed telescope in october 2010 , march 2011 and may 2011 . the resulting spectral type classifications can be found in tables [ t : star_list ] . when possible these are compared to the classifications found in the general catalog of variable stars ( gcvs ) . v616 lyr was observed with the 4-meter only , v588 with both the 4-meter and the coude feed , and the other observations are from the coude feed . spectral types and luminosity classes were assigned based on @xcite . the spectral types for v616 lyr , v588 lyr , v1766 cyg and v1953 cyg are presented here for the first time . our work compares favorably with the types in the gcvs for eg lyr . however , we classified bu lyr as m4-m5 iii while the gcvs lists it as an m7 star . this could result from changes in the spectral type with the pulsation phase . this has been reported for semiregular variables by @xcite . phase dependent changes can be mitigated by repeated observations . eg lyr , bu lyr , and v1766 cyg were observed on two or more dates separated by several weeks or months . six stars in table [ t : star_list ] were too faint on the dates of our observations for spectroscopic work . spectral types were derived from v - k colors . the reddening is not known for these stars so the corresponding spectral temperatures are likely too cool . these data show that all the program sr stars are , as expected , m stars ( table [ t : star_list ] ) . the low amplitude test objects ( table [ t : star_list ] ) are earlier spectral types . rclcc 2437359 & v616 lyr & srs & g9iii + 2569737 & v607 lyr & & k0 + 2570059 & v621 lyr & srs & g8 + 3431126 & eg lyr & srb & m4 - 5iii & m5iii + 5176879 & bu lyr & srb & m4 - 5iii & m7 + 5614021 & v588 lyr & sra & m2iiie & + 6127083 & v1670 cyg & sra & m4iii & + 7842386 & v1766 cyg & sra & m3 - 5iii & + 8748160 & v1253 cyg & sr : & m6iii & + 9528112 & af cyg & srb & m5iii & m4 iii + 10034169 & v2412 cyg & sr & m7iii & + 12215566 & v1953 cyg & srb & m3iii & + the kepler space telescope has an effective aperture of 0.95-meter , a primary mirror with 1.4 meter diameter , and a field - of - view ( fov ) of 105 deg@xmath1 . it was positioned in an earth - trailing heliocentric orbit with a 372.5 days period . the data discussed here is from the original kepler mission to observe a section of the cygnus region along the orion arm centered on galactic coordinates ( 76.32@xmath2,+13.5@xmath2 ) or ra=19:22:40 , dec=+44:30:00 ( 2000 ) . the photometer is an array of 42 charge coupled devices ( ccds ) , with each 50x25 mm ccd having 2200x1024 pixels . to prevent saturation the ccds were read out every 6 seconds . the kepler observatory has two observing modes , either the short cadence mode ( sc ) recording one image every 0.98 minutes or the long cadence mode ( lc ) recording one observation every 29.4 minutes . the data were downloaded in quarterly blocks divided in monthly sections . after each of the quarterly blocks the satellite was rotated so that the next quarter was observed with a different part of the ccd array . for our project we requested only the lc mode . for the lc mode various data products can be downloaded . details can be found in the kepler archive manual . here is a brief summary of the data typically available for a star observed with kepler : * simple aperture photometry ( sap ) : in the kepler asteroseismic science consortium ( kasc ) archive these data are called raw . the sap data come with no correction of differences in the flux zero point between the quarterly observation blocks . such differences stem from changes in the ccd pixels used , the use of different target masks , changes of the seasonal thermal conditions causing focus changes and changes of the zodiacal light . * pre - search data conditioning ( pdc or pdcsap ) : these are preliminary light curves provided by the kepler science operations center science processing pipeline . for searching for transiting planets various errors including systematic errors are corrected , flux outliers removed , and data gaps partially filled . the pipeline producing the pdc data removes all long time trends , so that these data are of limited use for the analysis of lpvs . * target pixel data ( tpd ) : the tpd deliver flux information for each pixel in the window on the ccd used to obtain the star s brightness . the size of the window is neither the same for all stars nor necessarily constant for a given object . formats between 4@xmath33 to 13@xmath359 pixels have been found for our sample stars . for the bright target af cyg a maximum window size of 15@xmath3587 pixels was used . * superstamps : for two areas in the kepler field , around the open clusters ngc 6791 and ngc 6819 , a complete image is provided in lc mode . these `` superstamps '' are provided with the same sampling frequency as the tpd data . * full field images ( ffi ) : these were taken monthly at the end of an observation cycle . corrected flux light curves by the kasc working groups using their own specific pipelines for the data reduction are sporadically available . to examine the shorter period astrophysical signals the instrumental effects that dominate the light curves first must be removed . in addition to zero point differences ( above ) there are also trends in the flux values within a quarter . one technique for removing these effects is ` cotrending ' , described in detail by @xcite , and based on the identification of common trends in a selected set of light curves on a given chip . cotrending is most effective at removing instrumental signal while preserving astrophysical signal for periods of a few tens of days . for the lpvs in our set , with typical periods generally greater than 100 days , limitations of the cotrending algorithm preclude reliable results . the alternative way is to leave the trends in the light curve and try to identify and remove them during period search . tests with low amplitude variables show that the first three cotrending basis vectors for each individual quarter result in a good match to the instrumentally dominated overall shape of the light curve and the thermal features . this gives us confidence that cotrending is removing instrumental effects but not introducing significant errors into the light curve . introduction of the fourth and fifth basis vector start to show overfitting of the light curves and loss of astronomical signal . however , once the individual quarters have been cotrended a significant difference between the median flux levels for each quarter remains . a significant limitation of cotrending for the lpv kepler light curves lies in preserving and identifying very long period variations where the method typically results in ambiguous conclusions . in contrast to other variable classes , the study of long period variables with kepler requires a very precise linking of the various quarters . errors introduced in linking the quarters can have a major impact on the derived long periods and their amplitudes . unfortunately , this linking is hampered by the gaps of 0.9 to 3 days in time coverage that occur between the quarters . in some cases , where the star has not been observed continuously in all quarters , very large gaps can occur . therefore , we focused in our study on the various approaches to close the gap between the quarters correctly and , if possible , in an objective , unambiguous , and easily reproducible way . first , we consider sap data . a first simple approach is to simply stick together the various quarters starting with the first one , always using the flux of the last data point of the previous quarter and the first data point of the following quarter . if cotrending was applied first , scaling of each quarter to the same median level can be used to reduce the discontinuities between the quarters . this approach is particularly helpful for faint targets where the increased scatter in the light curve complicates the scaling . one source of the differences in the flux levels stems from the variable mask sizes used to extract the photometric data for the sap light curves @xcite and the quarterly change of the ccds and their different characteristics . fortunately , the kepler archive also includes target pixel data ( tpd ) that allow one to check the automatically chosen mask for an individual star . as mentioned above , the size of the mask is not constant for the four quarters and is not always covering the target correctly . for illustration , we present some example tpd around the star v621 lyr ( kic 2570059 ) in figure[f : v621lyrtpd ] . in the four quarters q6 to q9 we see a shift of the expected center of the star indicated by the black cross and the actual central pixel . consequently the target mask , indicated by red squares , does not include the star s center . therefore , neither the sap nor the pdc light curve show the correct light variation of this star . using tpds we could correct this mismatch , and by extracting the data from missing pixels in q6 and q9 from the superstamps segment 930 ( see below ) we could even complete the tpd data set . furthermore we also extracted and added the missing quarters q2 to q5 and q10 to q13 to get a much longer time coverage for this star . as a consequence of these mismatches we decided to use a flexible aperture introducing a threshold value to select the pixels included for each individual observation . the threshold value was based on the flux of pixels not belonging to the star , which typically have significantly lower values and show almost no variation . depending on the star s position in a crowded field or not threshold values between 45 to 220e@xmath4/s give the best results . in the case of the example shown in figure[f : v621lyrtpd ] , we see that there are four to eight more pixels significantly contributing to the total light , compared with the kepler default target aperture . a careful definition of the aperture is therefore of critical importance to avoid systematic errors in the photometric data . our approach permitted us to reduce strongly the influence of the telescope jitter on the extracted light curve . figure[f : v621lyrsap ] compares the result of our approach using tpd and the sap data directly from the archive with and without amplitude rescaling . linking of the various quarters unambiguously can also remain a challenge in the case of the tpd data . a simple connecting , as described above , can lead to results that are likely incorrect . this problem occurs primarily for the gap between q7 and q8 which is much larger than the typical ones lasting for more than 15 days . here we use the light curve shape from the nearby cycles shifted to lie over the gap to estimate the needed adjustment immediately after the gap . the various sources producing the offset between quarters are a mixture of multiplicative and additive contributions that can not be disentangled . to estimate the resulting range of uncertainty we tested both approaches , i.e. a purely multiplicative and a purely additive correction . for the period determination , both methods give about the same results , but the amplitude variations are naturally different . for a low flux star like v621 lyr the amplitude difference between rescaling using a multiplicative factor or using an additive constant is about @xmath52% , whereas for a highly variable and high flux object like v1766 cyg we notice changes up to 3.76% . in the following we assume that the major effects are of multiplicative nature and use this for scaling . while the tpd achieves a better result in some cases problems can still occur . for instance , for q6 to q9 the row of pixels adjacent to the center of v621 lyr is missing in the tpd data . in such cases the superstamps discussed in the next paragraph can provide a solution . data for missing tpd observations also can be recovered from the superstamps . the superstamps for ngc 6791 consist of 20 images of 100 x 20 pixels in the format of the target pixel data at the 29.4 minute long cadence ( lc ) intervals . several of our targets in the constellation lyra have superstamp data available . we used these data to close gaps in the time coverage of the sap and tpd light curves . to make sure we can combine the superstamp data with the other data sources we extracted quarter 10 data of v607 lyr , where we have also tpd data for the same time . the agreement is excellent . one disadvantage of this method lies in the tpd like structure of the superstamp data . for one complete set of 100 x 20 pixels only one quality flag is available , and in many cases it shows values far above the good quality limit of 1000 . for the tpd analysis we ignored all observations with non - zero quality flags . for the superstamps the same restriction results in just a handful of observations in a quarter . an acceptable solution was found by accepting observations with a summed up quality flag up to 9000 and adding some additional formal checks . for instance we remove data with zeros in the date or flux and extreme outliers identified by a two - point difference function . a few obviously wrong data still in the light curve after these checks had to be removed manually . our analysis revealed that the best way to construct complete and reliable light curves from the kepler data for long period variables is to apply a mix of methods combining various types of data provided in the kepler archive . in difference to the work of @xcite we use tpd data instead of sap data in our analysis . as a primary period search method we used classical fourier analysis combined with a least square method optimization provided by the period04 software package @xcite . the search was limited to a maximum frequency of 0.2 cycles per day , i.e. a lower period limit of 5 days . up to 4 periods were used to fit the light change . while this upper limit for the number of periods included in the model fit was set somewhat arbitrarily , it was well motivated by our experience that including further periods usually does not lead to a significant improvement of the fit . furthermore , in many of our sample srvs we observed a clear drop of the s / n ratio for additional periods in the fourier power spectrum ( see also table [ t : pstars ] ) . however , the existence of further significant periods can not be totally excluded . the fourier analysis was done on the light curves without co - trending . in addition to fourier analysis , other methods have proven useful in light curve analysis , revealing additional information beyond the periods present in the light curves @xcite . for stars with part of the fundamental group description including irregularities in the light curve autocorrelation and related self - correlation methods have been found to be useful in quantifying the cycle - to - cycle variation in addition to the finding the dominate period @xcite . a regularly periodic light curve will have strong correlations for time lags that are multiples of the period with the strength of the peak correlation staying high over many cycles . semi - regular stars will generally have a strong correlation for the first few multiples of the period but the coefficient drops off over time . the persistence of the autocorrelation can then be used as an indicator of the regularity present in a semi - regular star . light curves which have multiple periods will tend to have weaker autocorrelation peaks although there can be complex patterns emerging from the beat phenomena between the dominate periods . we performed an autocorrelation analysis based on the algorithm of @xcite . the traditional autocorrelation routine requires a perfectly evenly spaced time series , therefore we used an autocorrelation routine adapted for astronomical time series developed by matthew templeton at the aavso ( private communication ) . for the longer period sr stars , we used a lag timing step of 0.5 days and maximum lag times up to 1000 days . while lag times larger than half of the total data range have more limited accuracy , we include these longer lag times for some stars as an indication of the possible longer term behavior but always recognizing that the accuracy is diminished . the autocorrelation coefficient can range from + 1.0 for a perfectly correlated signal to -1.0 for a perfectly anti - correlated signal @xcite . in this analysis we consider a strong correlation to be one with a peak coefficient above 0.75 while levels below 0.25 are classified as weak correlation . in our analysis of these stars , the periods found using autocorrelation are generally in agreement with the periods found through fourier analysis but we additionally comment of the regularity in the light curves as measured through the strength and persistence of the autocorrelation coefficient . the resulting periods ( numbers of brackets give the s / n ratio derived from the fourier analysis ) and supplemental data for each star are listed in table [ t : tstars ] . in the text we will refer to the periods as p1 to p4 , listed left to right in the tables , with p1 having the largest amplitude . transformation of kepler fluxes into kepler magnitudes was done using the relation given on the kepler webpage . previous analyses of kepler data ( e.g. * ? ? ? * ) identified a signal of approximately one year length in most light curves . this so called kepler - year " of 372.5 days length has an origin that has not been fully explained . it is obviously related to the orbital motion of the satellite . focus changes due to a varying satellite temperature are a possible cause . this could be augmented by seasonal variations of the background zodiacal light . the kepler - year is prominent in our test sample variables which have low amplitudes . however , for the program stars , all of which have large amplitudes , the kepler - year has either a minor impact or could not be detected at all . we performed additional test reductions on stars in fig.5 of @xcite showing the kepler year . using tpd data instead of sap data , the kepler - year variations disappeared in five of eight stars and thus could be partly an artifact introduced by the sap reduction pipeline . a detailed investigation of this difference could shed more light on the origin of this feature in the kepler light curves but is beyond the scope of the present paper . llrrrrr 2437359 & v616 lyr & 20.9 & 370.11 ( 8) & 189.10 ( 1 ) & long & 16.5 ( 1 ) + 2569737 & v607 lyr & 9.6 & 371.31 ( 14 ) & 185.36 ( 6 ) & 487.38 ( 4 ) & 13.54 ( 3 ) + 2570059 & v621 lyr & 10 & 365.31 ( 7 ) & 841 : ( 3 ) & long & + a combination of a long trend ( @xmath65000d ) and [email protected] , resulting from fourier analysis , gives an excellent fit to the light curve ( fig.[v616lyrplot ] ) . adding [email protected] , which is approximately half the value of p1 , gives a visible improvement . however , p1 is very close to the kepler year and p1 and p2 are almost certainly not intrinsic variations of the star . cotrending was very effective in removing this long period . also obvious in the data is a very short period of @xmath016 days . @xcite and @xcite found similar periods . autocorrelation and fourier analysis both show that the 16 - 16.5d period is not a single period . the typical full amplitude of the 16 d period variation is around 0.01 magnitudes with the amplitude changing over time . the strongest signal resulting from fourier analysis is a 372 days ( p1 ) period . the fit can be improved by adding essentially the half value of p1 , [email protected] , and [email protected] . since p1 is very close to the kepler year , p1 and p2 are almost certainly not intrinsic to the star . the origin of p3 is not clear . as with v616 the data contain an obvious very low amplitude variation ( p4 ) . in the case of v607 lyr the period is @xmath013.5 days . p4 agrees with the 13.6 d period from the ground - based data of @xcite but differs from the 9.6 day value of @xcite . the typical full amplitude of the oscillation is roughly 0.01 mag with the amplitude changing with time . the light curve shown in figure [ v607lyrplot ] contains a gap over quarters 11 and 12 when it was not observed as a target aperture with kepler . we filled that gap using superstamp data as described in sect.[quarters ] . it is interesting to note that the short period variation is much less obvious in these data . a reason for this difference is not understood yet . the amplitude of the kepler light curve ( fig.[v621lyrplot ] ) is rather small . a one year period ( 365.31d , p1 ) is very clear but could be identical with the kepler year . a combination of p1 , [email protected] , and a long time trend , clearly exceeding the total length of our time series , gives a very good fit to the kepler light curve . p2 and and the additional long time trend are not well defined . to further examine the data we applied cotrending to the sap light curves for the individual quarters . a satisfactory cotrend fit required the first five cotrending basis vectors in this case since the third and fourth basis vector still seemed to show some of the original unwanted thermal features . the cotrend fits matched the shape of the variations of v621 lyr very well . autocorrelation and fourier analysis of the cotrended light curves reveal no significant short period signals . the very small amplitude @xmath010 d variations reported by @xcite and @xcite were not present in agreement with the results published by @xcite . the kepler data for v621 lyr contain no intrinsic variations , i.e. at the time of the kepler mission this object was not a variable star . in the three very low amplitude test stars the kepler - year artifact provides the dominant signal . figure [ vxxx_lyr_kpyri ] shows the tpd light curves for the three low amplitude test stars and the zoomed and shifted kepler background extracted from v607 lyr . the kepler - year periods in these stars differ by at most 2% from the nominal kepler - year period of 372.5 days with amplitudes of less than 0.3 kepmag . the kepler - year signals are not in phase and all have specific kepler - year distortions even when recorded on the same ccd . this agrees with the findings on the kepler - year of @xcite . in addition to the 372.5 d kepler year our data are frequently better fit by including an instrumental period of a half kepler - year and periods longer than the kepler - year . in addition , a long time trend at the 0.1 kepler magnitude scale with a period not defined in the 4 years of kepler data is seen in all three stars . in the test stars we were able to detected intrinsic periods in the range 13 to 20 days with amplitudes as small as 0.005 kepler magnitude . here the analysis of the individual program stars is discussed . we include brief summaries of what we know about these stars from the literature . the resulting periods and supplemental data for each star are listed in table [ t : pstars ] . rcccrrrr 3431126 & eg lyr & 236 & & 228.9 ( 16 ) & 258.2 ( 13 ) & 137.7 ( 5 ) & + 5176879 & bu lyr & 259 & & 156.5 ( 32 ) & 158.9 ( 31 ) & 347.5 ( 6 ) & + 5614021 & v588 lyr & & & 116.4 ( 9 ) & 217.9 ( 9 ) & 198.3 ( 6 ) & + 6127083 & v1670 cyg & 188 & 193 & 187.2 ( 17 ) & 94.0 ( 4 ) & & + 7842386 & v1766 cyg & & 113.7 & 122.2 ( 6 ) & 754.0 ( 6 ) & 104.2 ( 4 ) & + 8748160 & v1253 cyg & & 187 & 197.2 ( 13 ) & 857.4 ( 3 ) & 99.4 ( 2 ) & + 9528112 & af cyg & 92.5 , 175.8 , 941.2 & & 177.4 ( 6 ) & 1867 . ( 8) & 93.6 ( 5 ) & 441.3 ( 3 ) + 10034169 & v2412 cyg & 250 & & 255.9 ( 16 ) & 241.2 ( 8) & 140.4 ( 2 ) & + 12215566 & v1953 cyg & & 95.2 & 163.1 ( 6 ) & 847.5 ( 6 ) & 93.4 ( 4 ) & + @xcite found that eg lyr had an m5 iii spectrum and rejected the earlier identification as an rv tau variable . @xcite classified eg lyr ( using the alias vv 221 ) as a lpv with an uncertain period of 267 days . the 4th edition of the gcvs @xcite reclassified eg lyr as an srb star with a period of @xmath0236 days . @xcite found the colors to be in agreement with an m6 iii . the asas observations of eg lyr cover about two cycles with a period roughly consistent with the gcvs value . however , asas classified the star as aperiodic . the kepler data show a strong light variation ( see fig.[eglyrplot ] ) but the periodicity is not obvious . the full amplitude of the oscillations , from minimum to maximum , range from about 0.2 to 0.8 magnitudes . in general the peaks in the light curves are fairly narrow across the whole light curve . the minima are fairly narrow in the first half of the light curve but become much wider with evidence of greater structure in the second half of the light curve . three periods , p1=228.9 d , p2=258.24 d , and p3=137.74 d , give a reasonably good fit and predict the times of the observed maxima and minima as well as the overall shape of the light curve . after subtraction of the periods the residuals suggest the existence of a long time trend , considerably longer than the observed time series , and a short period with a length around 78 days but without a strict periodicity . autocorrelation analysis similarly shows a strong correlation at 238 days with a less well resolved period at 140 days . the gcvs period of 236d is a good match to the autocorrelation period and falls between our fourier periods p1 and p2 providing overall good agreement on the period length . bu lyr was initially identified as a variable by @xcite from a one magnitude difference in brightness measure 20 years apart . it subsequently was identified by @xcite as a bright near - ir star , @xmath82.95 , with an infrared index @xmath94.39 . @xcite confirm that this is a very late - type star with a spectral type of m7 and a stable mean period of 259 days . the light curve appeared to have a double maximum . the brightness of the main maximum is fairly stable but the brightness and phase of the second maximum and the minima is variable . however , asas observations of bu lyr show variability without a clear periodic pattern and this survey classified bu lyr an aperiodic variable . the gcvs classifies bu lyr as an srb variable with a visual amplitude of 1.4 magnitude . the kepler light curve looks irregular ( see fig.[bulyrplot ] ) but the amplitude is quite large . typical variations are in the general range of 0.3 magnitudes with variations just over 0.5 magnitudes for the full light curve . autocorrelation analysis suggests a weakly correlated period of about 150 days . fourier analysis of the data set gives two periods very close together : 156.52d ( p1 ) and 158.90d ( p2 ) . adding a variation on a time scale of about one year ( p3 ) reproduces some aspects of the light curve including the correct prediction of most of the maximum and minimum times . however , much of the shape is not properly reproduced . no clear periodicity is visible in the residuals after subtracting p1 to p3 . the double maximum structure reported by @xcite is not seen in the kepler data . instead , the brightness of the maxima and minima seems to change in an irregular way . similarly we do not see a 259d period . the residuals o - c over the time span of our kepler observations suggest a tendency towards an increasing period . @xcite identified v588 lyr as a possible mira of period 221 days and a range from 12.5 magnitudes to 14.0 magnitudes . the gcvs lists v588 lyr as an sra variable . asas observations found it to be aperiodic with no clear light curve shape at the time of observation . the kepler light curve of v588 lyr ( fig.[v588lyrplot ] ) shows strong oscillations covering slightly over 1.1 kepler magnitude of variation with individual maxima and minima levels varying considerably from cycle to cycle . during the time of observation the maxima and minima had alternate bright and faint cycles . the combination of two periods resulting from a fourier analysis , p1(116.39d ) and p2 ( 217.93d ) , permits a fit to the alternating depths , and the autocorrelation analysis agrees with these periods with even peaks more strongly correlated than the odd peaks . the addition of p3 ( 198.33d ) slightly improves the result but a few maxima / minima are not well reproduced . the residuals do not suggest the presence of an additional long period trend or further short periods , but rather non - sinusoidal variations and additional cycle - to - cycle irregularities . p2 is almost identical to the value given by @xcite . hoffleit and sands @xcite classified v1670 cyg as an sr star with a period of 187.5 days . @xcite rediscovered v1670 cyg classifying it as a 191 day mira with a visual magnitude ranging from 12.4 15.0 . asas found v1670 cyg to be quasi - periodic with a period of 193 days and visual amplitude of 1.99 magnitudes . the light change observed by kepler ( fig.[v1670cygplot ] ) is very regular except for a very bright maximum at the beginning of the time series and a second bright maximum shortly before the end of our time series . there are variations in the depth of the minima from cycle to cycle . autocorrelation indicates a strong correlation at a period of 187.5 days . fourier analysis gives a single period of 187.19d ( p1 ) , consistent with the autocorrelation analysis . subtracting p1 from the light curve reveals a further , rather regular change corresponding to p2 ( 94.01d ) which is very close to half of p1 . no further periodicity was found in the residuals . p1 is in nice agreement with the literature values for the period . @xcite found v1766 cyg to be a fairly regular sra variable with a period of 119 days and a visual light range between 10.3 and 11.5 magnitudes . v1766 is classified as an ms star @xcite . based on infrared colors @xcite suggests that v1766 cyg is an extrinsic s - star , i.e. its enrichment in s - process elements results from mass transfer from a more massive companion . asas lists the star as quasi - periodic with a period of 113.7 days and a visual amplitude of 0.80 mags . in this case , periodicity is not very strongly expressed in the kepler light curve ( fig.[v1766cygplot ] ) . typical variations are generally in the range of 0.2 magnitudes with variations just over 0.4 magnitudes for the full light curve . autocorrelation analysis suggests periods of 111d and 120d with only moderate to weak correlation . the fourier diagram shows several peaks around 100 days period , a combination of 122.25d ( p1 ) and 104.21d ( p3 ) reproduces at least some features of the light curve . they are close to the asas value of 113.7d . double maxima are present , so we suspect that variations on a somewhat shorter time scale are present but they do not show up clearly in the fourier diagram . a long time trend , which we fit with a period of 754d ( p2 ) , seems to be present as well . @xcite found an uncertain period for v1253 cyg of 188 to 226 days with a @xmath10 magnitude range of 11.8 to less than 15th and provided a a classification of sr . @xcite list the star as quasi - periodic with a period of 187 days . the kepler light curve ( fig.[v1253cygplot ] ) shows the brightness of the maxima and minima vary smoothly with time and a bump on the ascending branch . the data set is somewhat shorter than the others covering quarters q6 through q15 . autocorrelation suggests a strong correlation with a period of 196.5d . using fourier analysis the times of maxima and minima can be reproduced very well with a single period of 197.15d ( p1 ) . the changing maximum and minimum brightness and the bump can be qualitatively reproduced by adding a long period of 857.39d ( p2 ) and 99.35d ( p3 ) . p2 is , of course , very poorly defined due to the limited length of the kepler time series . p1 is not very far off the gcvs value . p3 is approximately half the period length of p1 . its s / n in the fourier spectrum is only 2 . @xcite noted that the bright ( 7.4 9.4 mag ) m5 star now referred to as af cyg was variable . the spectral type at maximum light was confirmed m 5 iii by @xcite . the variability has been extensively studied over the last 100 years . the principal period is @xmath093d . @xcite first commented on the long secondary period ( lsp ) of 960 days . using aavso data @xcite found periods of 92.9 and 165.9d . an analysis of four sets of visual observations by @xcite confirmed the periods of 93 and 163d and also confirmed a lsp of 921d . the gcvs lists periods of 92.5 , 175.8 , and 941.2 days . @xcite discuss the instabilities observed in the historic light curve . these include alternating dominance of the 93 and 176d periods , intervals of regularly strong weak maxima , and variations in times of the 93 day maximum from 79.4 to 97.4d . for the data analyzed they found a dominant 94.2d period with clusters of other periods present . @xcite observed af cyg spectroscopically and derived an ` orbital ' period of 926 @xmath536d matching the lsp . however , they note that the orbits for all lsp systems look the same so the motion is likely not orbital in origin . @xcite provide a parallax of [email protected] mas which allows rough placement on a period luminosity diagram . the kepler light curve of af cyg ( fig.[afcygplot ] ) shows a non - strict alternation of strong and weak maxima that could be interpreted as secondary maxima . the lack of a strict periodicity of this phenomenon suggests that instead this is an interplay of two excited periods of comparable amplitude with a period ratio close to an integer value . several minor bumps are also present in particular on the descending part of the light curve . the light curve appears to have a long time variation exceeding the length of our monitoring . the typical amplitude of the light change is 0.5mag in the kepmag system . a fourier analysis reveals 4 main periods , and their combination gives a reasonable fit of the kepler light curve although some deviations remain . the short term variability is accounted for by p1 ( 177.4d ) and p3 ( 93.6 ) . these two periods show a ratio close to 2 , and thus agree with the above mentioned suspicion on the origin for the alternating bright and weak maxima . the long time change is represented by a period of 1867d ( p2 ) . while p2 is a strong contributor to the light curve it is not adequately sampled . a further significant improvement of the fit can be achieved by including a fourth period of 441.3d . similar length periods were seen in the test data and may be related to the kepler year . autocorrelation analysis contains significant structure indicating evidence of multiple periods that are beating against each other . the highest correlation occurs at roughly 186 days but there are clearly indications of other periods present . increasing anti - correlation between 800 - 1000 days is consistent with the longer period of 1867 days shown in the fourier analysis . for af cyg , which has been very well studied from the ground , we note the agreement of the two well sampled kepler short periods with the literature periods ( 93 and 176d ) and the disagreement of the poorly sampled kepler long periods ( 441 and 1867d ) with the literature period ( 941d ) . the 941d period is not apparent in the kepler data . a fourier analysis on the residual light curve after the removal of p1 through p4 gives an unclear pattern of additional frequencies between 0 and 0.03 cycles per day ( c / d ) and no significant peaks for frequencies above 0.03 c / d . the variability of v2412 cyg was discovered in 2001 as a part of the misao project @xcite but the only additional information was a possible coincidence with nsvs 5651487 @xcite . nsvs 5651487 has a light curve with an @xmath0250 day period . asas @xcite lists v2412 cyg as aperiodic . the gcvs lists a 250 day period for the star and an sr : classification . the kepler time series for v2412 cyg is not continuous ( see fig.[v2412cygplot ] ) . this star was observed from quarter q2 through quarter q6 , then again during quarters q8 and q9 , and finally in quarter q14 . in part the gaps are due to the placement of v2412 cyg on the failed kepler module 3 . the large gaps make it extremely difficult to determine an accurate flux level after each of these gaps . the behavior during the five continuous quarters gives us some clues to the proper treatment for this star . during these quarters we have strong oscillations covering over 1.5 kepler magnitudes . the shifts to adjust for discontinuities between these continuous quarters were relatively small . we assume that the corrections in flux for quarters q8 and q14 would also be relatively small and we leave them uncorrected . times of maxima and minima can be reproduced reasonably well using one period of 255.91d ( p1 ) only . variations in the brightness of the maxima and minima can be fitted nicely using a beat period of 241.19d ( p2 ) . a third period of 140.42d ( p3 ) comes with a somewhat low s / n ratio , but including it helps to reduce the residuals significantly . the light curve shows an asymmetric oscillation with a steeper descent into minima than the rise to maxima and a slight bump in the ascending side which seems to shift its position from cycle to cycle . including period p3 partly compensates for these bumps in the rising branches although some difference remains . unfortunately an enticing wiggle in the ascending side during quarter q14 @xmath11 is cut off before the star reaches maximum sometime during the missing quarter q15 . p1 and p2 are both close to the gcvs value of 250 days . @xcite found that v1953 cyg had slow semiregular variations with a photographic amplitude of 14.8 - 15.6 mag and a period of @xmath0150 days . the gcvs classifies v1953 cyg as an srb . the asas observations of v1953 cyg list the star as quasi - periodic with a period of 95.2 days . although the kepler measured light curve for v1953 cyg ( fig.[v1953cygplot ] ) looks quite irregular with a typical full amplitude of 0.4 mag , a combination of only three periods of 163.07d ( p1 ) , 847.51d ( p2 ) , and 93.41d ( p3 ) permits a fit to its main shape . p2 is poorly defined . many details of the light curve , however , can not be reproduced properly . a fourier analysis on the residuals gives a number of peaks of similar amplitude between 500 and 50 days . autocorrelation analysis finds a moderately correlated period of 179.5 days , but the correlation decays rapidly indicating that no single period dominates the light curve signal . p3 is similar to the asas period . p1 may be identical to the 150d variation seen by @xcite . the kepler data provide a new way to explore agb pulsation through continuous , high - precision , photometric time - series spanning multiple years without significant interruption . diurnal and seasonal effects in ground - based data result in time series that are too short or too fragmented for proper handling of the complex variability patterns observed in stars with very long periods ( e.g. * ? ? ? we were interested in exploring the various radial modes involved in sr pulsation , looking for long secondary periods , and searching for the existence of any low amplitude , short period events or pulsations . for the latter goal we were especially interested in looking for any periods that might be attributable to non - radial pulsations . the kepler data , of course , have limitations . interruptions of the kepler time series observations occurred due to telescope rotation every three months . sr variables have periods of order months to years @xcite and extracting periods longer than a kepler quarter posed special challenges in data reduction . linking the quarters is surprisingly complicated due to a number of effects that can introduce photometric errors and we have explored optimum and non - subjective techniques for the construction of light curves spanning the whole kepler monitoring time span . another unfortunate reality of kepler data , impacting this paper as well as the planet search goal of the mission , is that most of the stars observed are moderately faint . as a consequence most of the variable stars are not among the best studied objects of their classes . in spite of these problems , we find good agreement with the main periods detected in the kepler data and ground - based results including the asas data set ( see table [ t : pstars ] ) . the differences appear to be astrophysical and we will discuss them below . physical interpretation of the observed pulsational behavior of red giants has been a topic of research and debate for several decades . recent results for low luminosity red giants , based also on kepler data @xcite , suggest that solar - like oscillations are observed in these stars . for the red giants less luminous than the rgb - tip , three parallel p - l - sequences are detected @xcite that can be interpreted as radial low overtone modes plus non - radial ( l=1 ) and ( l=2 ) p - modes @xcite . asteroseismology using short cadence kepler data provided the possibility of identifying core helium burning red giants on the basis of frequency separation of gravity modes @xcite . above the rgb tip where the classical long period variables are found the pulsation behavior is different . in the more luminous stars with their larger light and velocity amplitudes , the role of non - radial modes for the observed ( semi-)periodic behavior is expected to be comparably low ( e.g. * ? ? ? * ) , although some recent results by @xcite may initiate a new discussion on this question . various observational studies , more recently by @xcite and @xcite , have demonstrated the difference in the variability behavior between rgb stars and the more luminous agb stars . as pointed out by @xcite , stability analysis of pulsation modes in red giants shows a change from higher overtone radial modes in low luminosity red variables to lower overtone radial modes in the more luminous ones . this is in agreement with observations from globular cluster giants @xcite . model calculations , in particular of wood and collaborators ( * ? ? ? * and references therein ) , find the variability of the more luminous stars dominated by radial modes that are self - excited rather than stochastically excited . coupling between convection and oscillations has been identified as the most likely excitation mechanism @xcite . @xcite find that irregular pulsations are driven by energy exchange between two nonadiabatic modes in a 2:1 resonance . pulsation models nicely reproduce the period - luminosity sequences of luminous red giants ( e.g. * ? ? ? * ) in terms of a sequence of the fundamental radial mode and several low order overtone modes . the only exception is sequence d , the long secondary periods , which we will discuss in more detail below . the observed amplitudes in our program stars place them in the group of luminous red variables . the periods and parallax of af cyg demonstrate this unambiguously @xcite . hence we will focus on radial pulsation modes in this paper . @xcite find that strange modes may develop in red giants beside the normal radial modes , but that they are always damped and thus should not be visible beside the normal radial modes on the upper giant branch . however , they might affect the period shown by a radial mode . the kepler light curves typically appear at first examination very complex . however , in all our stars the combination of two or three periods was sufficient to predict the times of maxima and minima correctly . the exception is bu lyr where this is correct for only the second half of the kepler data set . however , in most cases a single period was not sufficient . this implies , as others have concluded , that even behavior classified irregular is multiperiodic @xcite . for the detailed fitting of the complex light change shapes we used a combination of several periods . the continuity and the high photometric precision lowered the noise level in the fourier power spectra significantly compared with ground - based surveys . in some cases , the observed changes in amplitudes and period lengths could be traced reasonably well , while in other stars the large residuals remained after subtraction of three to four frequencies , just like in the case of stars well observed from the ground . this is in agreement with other studies that note an irregular component in the light curves @xcite . the two independent period search methods we applied lead to very similar results for period fits to the light curve . the period detection is robust for up to four periods at the precision and time span of the kepler data . there are differences between the periods detected in our study and values in the literature based on different data sets for the same star . to explore this further , we downloaded the asas data set for one of our stars , v1953 cyg , from the asas website . we selected v1953 cyg because its light curve is one of the most irregular ones in our sample . the fit produced using our kepler derived periods can be used to model almost all maxima and minima in the asas data set correctly and the combination of three periods reproduces a remarkable fraction of the changes in amplitude and period length . naturally , the photometric uncertainties of the asas measurements are much larger than those of the kepler measurements . therefore , it is not possible to test the quality of the fit of the artificial light curve with similar precision . in the asas data base v1953 cyg is listed with a period of 95.2d although the phased light curve using this period shows considerable scatter . our fourier analysis of the same asas data gives a period of 95.1d as the first peak . an analysis of the residuals results in a second period of 66.1d . the third period is a long period with almost 600d length . the kepler data show a third peak at 93.4d and a second peak at 847.5d . there is no fourier peak visible in the asas data at the prime 163.1d kepler peak . attempting to fit the kepler light curve using the three periods derived from the asas light curve does not result in a good fit . we find that the periods derived for semiregular variables are always connected to the star s behavior at a specific time interval but are not necessarily transferable to a different data set observed at a different time . to identify the dominant period(s ) of an sr variable an analysis of several pieces of the light curve obtained in different years seems to be necessary . multiple periods in sr variables are known from ground - based data . the kepler light curves of the sr variables provide extremely clear examples . the multiple periods are manifested in three ways . most of the program stars have two pulsation modes present that are separated by roughly a factor of two ( table [ t : pstars ] ) . historical light curves show that the dominant mode can switch between the two excited modes @xcite . in 30 50% of stars a third long period that is @xmath09@xmath54 times longer than the dominant period can also be present @xcite . in addition , each of the two pulsation modes can be composed of multiple periods spanning a small range of values @xcite . the result is that for any given epoch the resulting light curve can appear irregular or quite periodic . an excellent example is the sra variable v1766 cyg which is described in the literature as fairly regular with a period of 119 days @xcite . this is the one star star in our program that does not have a second mode present . nonetheless during the time of the kepler observations the light curve appears to be fairly irregular this is due to the presence of a number of periods spanning the range from 104 to 122d . on the other hand , two stars in our sample had sr : or srb classifications , v1253 cyg and v2412 cyg , which imply that historically the light curves have been irregular ( or badly sampled ) . these stars do indeed have multiple modes / periods but during the interval of the kepler observations light curves were quite regular . time variations in the periods of late - type stars pulsating in multiple modes are well known from comparisons of observed and average periods . changes in the strengths of radial modes also are a feature of type ii cepheids and rv tauri stars @xcite . in these stars as well as in lpvs variations of this type have been attributed to chaotic processes @xcite . using combined sinusoid fits to the lpv light curves from space missions like kepler or corot with a limited number of periods will not result in vastly different insight over ground - based analysis of these objects but does produce more reliable secondary periods . a substantial collection of high precision data could form the basis for testing statistical models of interior processes in lpvs ( see for example * ? ? ? the asas catalog @xcite of variable stars in the kepler field lists a total of 947 variables . of these approximately 43% are in the asas category qper . the qper category consists mainly of sr variables , many with short periods , although we find that some of our program stars are classified as aper . miras make up another 6 percent of the asas catalog variables . figure[f : histogram ] shows the period distribution of the qper and mira classes . a period distribution is the projection of the period - luminosity diagram onto the period axis . we are unable to produce a period - luminosity diagram since the luminosities of the kepler field stars are unknown . for most of the program stars the dominant period is around 200 days with a shorter period of around 100 days . the longer periods match the long period sr distribution which is shared by the miras . the miras are fundamental mode pulsators @xcite . the figure suggests that for most , if not all , of the program stars the long period is a fundamental mode . 114 day period is shown as an open triangle . long secondary periods ( over 500 days ) are not shown.,scaledwidth=80.0% ] to explore the periodicity patterns detected in our semiregular variables the period ratios are shown on a petersen diagram in figure [ petersen ] . the petersen diagram sets the ratio of the longer to the shorter period in relation to the longer period @xcite . this approach has become widely used for the study of large sets of lpv light curve data ( e.g. * ? ? ? * ; * ? ? ? * ) . in our figure [ petersen ] we see a clustering of data points at period ratios of 1.1 and around 1.8 . the ratios with the very long periods have been excluded . we show the theoretical ratios from fundamental mode ( p0 ) and low overtone mode pulsation models ( pn ) given in figure 4 of @xcite . the period ratios around 1.8 fall between the theoretical sequences for fundamental to first overtone mode and first to third overtone mode , respectively . all our sample srvs with the exception of v1766 cyg are found in that region with a pair of periods . our result is consistent with the petersen diagram presented in @xcite for long period variables in the lmc . in analogy to their result we conclude that most of the srvs in our sample are fundamental mode pulsators which also have the first overtone mode excited . the period ratios clustering around 1.1 , i.e. pairs of periods very close together , are not the result of two different low order radial pulsation modes ( cf . * ? ? ? @xcite detected that about 35% of the lmc semiregular variables in the ogle database show such a behavior . in analogy to similar observations found in rr lyr stars and cepheids , these authors attribute such periods to non - radial oscillations . however , non - radial oscillations of the size required for m giant sr variability would result in unrealistic distortions @xcite . non - radial modes also are not expected to provide significant velocity variations @xcite . in the case of af cyg the velocity amplitude is @xmath0 5 kms@xmath12 . similar results are found for the related stellar sample presented by @xcite . @xcite , while excluding non - radial oscillations as the origin of multiperiodic behavior , pointed out that the clustering of peaks in the fourier power spectra of some lpvs has a striking similarity to the power spectra seen for stochastically excited pulsators like our sun . these are peaks from the same pulsation mode but with slightly different periods . however , @xcite point out that the low mass and high luminosity of agb stars rules out a stochastic origin and favored chaotic pulsation dynamics arising from nonlinear interaction between two resonant pulsation modes . the light curves show that these peaks can co - exist or come and go without changing the overall brightness . @xcite noted a possible connection to convection . this has been discussed by @xcite , @xcite and most recently by @xcite . @xcite note that convection carries most of the energy through most of the envelope but is poorly modeled resulting in large computational uncertainties . ) . period ratios above three ( long secondary periods ) have been excluded . lines indicate theoretical period ratios from @xcite . , scaledwidth=80.0% ] as noted in the introduction the sr variables exhibit long secondary period ( lsp ) variations which have periods longer than the fundamental and typically @xmath08 10 times longer than the dominant pulsation period @xcite . while these long periods have been known for a long time ( * ? ? ? * and references therein ) , lsp became an astrophysical problem when @xcite identified period - luminosity sequences covering fundamental and overtone radial modes , as well as an lsp sequence , in macho observations of lmc agb stars . stars with lsp are multimode pulsators with a low overtone pulsation . in the agb stars these are sr variables . recent surveys of luminous red giants by @xcite and @xcite show that lsp is common . @xcite suggests very low amplitude lsps exist in up to 50% of variable agb stars . the kepler mission proved too brief to reliably measure lsp for any star . however , in agreement with @xcite about half of our light curve fits require a long period of a few hundred days length . in the case of v1953 cyg an lsp of @xmath0600d is detectable from ground - based data . we tentatively identify lsps of 800 + days for this star and 700 + days and 800 + days for v1766 cyg and v1253 cyg , respectively . these stars appear indistinguishable from the other srvs in terms of global parameters like color . among the srvs with a long period in our sample , only one ( v1766 cyg ) shows in addition a pair of periods of very similar length ( 104 and 122 days ) . the other four stars with such pairs of periods do not show an lsp , however , three of them show a third period about half the length of the two similar periods . the third period of bu lyr is approximately two times larger than its pair of periods . af cyg is well known from ground based observations to have a lsp @xcite . in our data it shows an interesting characteristic of lsp not previously reported . the ground based lsp is @xmath0960d which is also seen spectroscopically @xcite . however we can not fit , even by attempting to force a fit by hand , this period to the kepler data . on the other hand a period of roughly half the lsp is present . in the few stars where the lsp has been monitored spectroscopically the lsp is never the dominant photometric signal but is the dominant spectroscopic signal @xcite . archival photometry of these stars suggests that the lsp can at times be difficult to observe . the physical cause of lsps remains unknown @xcite . no clear non - radial modes were detected in our data in agreement with the previous discussion . other investigators have excluded a number of other causes with remaining explanations including star spots @xcite and long - period convective cycles @xcite . long - period convective cycles were mentioned as a possible cause discussed above for multiple periods in overtone modes . the requirement of dual mode pulsation for a long period suggests that the mode switching mechanism is involved with the long period . coupling between convection and oscillation in sr pulsation has been invoked by both @xcite and @xcite the large amplitude lpv variables are miras and these have been proven to be fundamental single - mode variables @xcite . @xcite also find that most srvs are double mode variables with the fundamental mode and first overtone simultaneously excited . our data similarly show most srvs pulsating in two modes . as discussed in @xcite agb stars evolve through instability stages from overtone mode pulsation to fundamental mode pulsation as the luminosity increases . the miras in the lmc sample investigated by @xcite , @xcite and others are single - mode variables and do not have an lsp . since the sr class contains fundamental mode pulsators , the implication is that either lsp behavior limits the amplitude of pulsation or that a large pulsation amplitude somehow excludes lsp . both explanations require that lsps are an intrinsic feature of an agb star related to overtone pulsation . this agrees with the conclusion reached by @xcite using other lines of evidence . the 30 minute sample interval of the kepler observations permits a search for rapid outbursts or very short time variations that have been discussed for lpvs on the basis of hipparcos data @xcite . a detailed search by eye of the program star kepler light curves for such outbursts lasting between hours to a few days did not result in any detections . the time around the transition between two quarters was excluded . this result adds to other recent studies of this phenomenon that also did not detect such variations @xcite . we support the conclusion of these works that outbursts of duration less than a day in long period variables are very rare events . kepler data requires considerable care in reduction when used for multi - year high precision photometry . the technical problems have been discussed at length and solutions are presented . the resulting light curves of the sr variables are remarkably featureless at high time resolution and high precision . no variations were detected on ` rapid ' time scales of one day or shorter . we have undertaken various types of period analysis on the data . while many of the light curves appear irregular the analysis shows this to be the result of combining two or three regular periods , suggesting that even the most irregular late - type variables are multiperiodic sr variables . multiperiodicity is a ubiquitous property of the sr variables that we analyzed . using a petersen diagram the sr variables in the current sample can be shown to be fundamental plus first overtone multimode radial pulsators . the multiple periods of the program stars are not solely the result of pulsation in more than one mode . these stars are in fact multiperiodic in three ways : they are generally pulsating in both the fundamental and an overtone mode . they have multiple periods in the dominant pulsation mode . approximately half of the stars also have a long secondary period 8 to 10 times the length of the dominant pulsation period . the strong connection between lsp and multimode overtone pulsation suggests that lsp is related to the convective processes that drive mode switching and multiperiod modes in sr variables . one of our goals was to use kepler data to compare the different subclasses of sr variables . all the sample stars are multiperiodic with no apparent differences between the sra , srb , or sr : subclasses . we interpret the apparent regularity of some light curves as the result of constructive interference between beating periods . the implication , which is not surprising , is that the sr classification scheme is very insensitive to the location of a variable in the p - l diagram . when gaia data becomes available it will be possible to place the milky way lpvs on the p - l diagram . mira variables have been separated from the sr variable by pulsation amplitude but are also single mode pulsators . previously sr variables have been seen as simply low amplitude versions of the fundamental mode mira pulsators @xcite but this is clearly an incomplete description . it will be of interest to see how the extensively studied late - type variables fit on a p - l diagram where multimode versus single mode and large amplitude versus small amplitude variables can be compared . the work of tl has been supported by the austrian science fund under project number p23737-n16 . eh thanks mark taylor for stilts support . we wish to thank steve howell for his helpful comments and for providing the mayall 4 m spectra . we also thank the kitt peak directors for their allocations of time on the coude feed telescope . finally , we acknowledge the undergraduate students at scsu who participated in the initial analysis of the kepler data . support to scsu was provided through the nsf paare award ast-0750814 and nasa kepler awards nnx11ab82 g and nnx13ac24 g . the national optical astronomy observatory is operated by the association of universities for research in astronomy ( aura ) under cooperative agreement with the national science foundation .

we examine the scattering of acoustic-mode waves from a thin magnetic fibril embedded in a gravitationally stratified atmosphere . the scattering is mediated through the excitation of slow sausage waves on the magnetic tube , and only the scattering of the monopole component of the wavefield is considered . since such tube waves are not confined by the acoustic cavity and may freely propagate along the field lines removing energy from the acoustic wavefield , the excitation of fibril oscillations is a source of acoustic wave absorption as well as scattering . we compute the mode mixing that is achieved and the absorption coefficients and phase shifts . we find that for thin tubes the mode mixing is weak and the absorption coefficient is small and is a smooth function of frequency over the physically relevant band of observed frequencies . the prominent absorption resonances seen in previous studies of unstratified tubes are absent . despite the relatively small absorption , the phase shift induced can be surprisingly large , reaching values as high as 15 for modes . further , the phase shift can be positive or negative depending on the incident mode order and the frequency .

1106.5078

it is well established that sunspots and magnetic plage are strong absorbers and scatterers of the sun s acoustic @xmath0-mode waves ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the absorption is unexpectedly large , reaching values as high as 70% for sunspot groups and a lesser , but still substantial , value of 20% for magnetic plage . equally strong phase shifts are also observed for sunspots , with shifts as large at 100@xmath1 being possible . interestingly , plages lack measurable phase shifts , despite possessing significant absorption @xcite . initial attempts to model the interaction of acoustic waves with magnetic features concentrated on the absorption caused by sunspots , as the observed signal was strong and easy to measure . a host of absorption mechanisms were investigated , including resonant absorption ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , mode mixing @xcite and mode conversion ( e.g. , * ? ? ? * ; * ? ? ? the present belief is that mode conversion from acoustic waves to slow magnetosonic waves within an appreciably inclined magnetic field is responsible @xcite . under this paradigm , acoustic energy is transformed into slow waves which propagate along the field lines removing energy from the @xmath0-mode cavity . the conversion process is most efficient when the phase velocity of the incident acoustic wave matches the local phase velocity of the slow mode . such strong coupling of the phase velocities occurs when two conditions are met : the wavenumber vector of the incident acoustic wave is nearly parallel to the magnetic field and the alfvn and sound speeds are comparable . in a sunspot , the alfvn and sound speeds are equal to each other in a layer just below the solar surface and the ray paths of the @xmath0 modes in this layer are such that the strongest conversion occurs in regions with field lines that are inclined 2030 degrees from vertical ( such as within penumbrae ) . in addition to successfully explaining the absorption observed for sunspots , mode conversion models have also succeeded in reproducing the large phase shifts @xcite . so , while we presently have a fairly good understanding of how a flaring , monolithic , magnetic structure such as a sunspot can absorb and scatter acoustic waves , we have made little progress in modelling the observed absorption by plage . the absorption within a plage is smaller than that of a sunspot by only a factor of two , yet the mode conversion process is unlikely to work in the same manner , as the field within a plage has a fractured and fibril nature , and probably has less inclined field of a structured geometry in the photospheric layers . the limited theoretical work that has been done modelling the effects of fibril magnetic fields indicates that multiple scattering may play a very important role @xcite . @xcite examined multiple scattering between a small number of flux tubes ( 23 ) in an unstratified atmosphere and found that resonant absorption by the collection of tubes can be greatly enhanced if the tubes are tightly spaced . @xcite performed a similar calculation but considered incident @xmath0-mode waves trapped within a solar - like wave guide and their interaction with thin flux tubes . tirry permitted tube waves excited on the magnetic fibrils to freely cross the bottom boundary of the acoustic waveguide and discovered that such wave leakage destroyed the absorption resonances observed by @xcite . both studies lacked gravitational stratification . neither of these studies included a sufficiently large number of tubes to reliably estimate the expected phase shifts measured in plage . the lack of measurable phase shift is likely to be the result of incoherent scattering from a large number of tubes . @xcite examined whether the wave leakage mechanism explored by tirry is sufficiently large on its own to explain the absorption observed within plage . they modeled a plage as a large collection of thin magnetic tubes embedded in a stratified atmosphere and considered each tube to absorb acoustic waves in isolation from the other members of the collective . the mhd tube waves excited on each tube by the incident @xmath0 mode freely propagate along the fibril out of the acoustic cavity despite the gravitational stratification . they found that high levels of absorption ( potentially greater than 50% ) were easily achieved . however , they did not take into account scattering between tubes or the screening that is likely to occur for tubes located far from the edge of the plage . to date few studies have examined both the _ absorption _ and _ scattering _ of acoustic waves by magnetic tubes residing in a _ gravitationally stratified _ atmsophere . notable are those of @xcite and @xcite which focused on the scattering of the dipole component of incident @xmath2 modes from thin magnetic tubes through the excitation of kink waves , and the study by @xcite that , through numerical simulation , examined axisymmetric scattering from tubes of general thickness . here we examine scattering and absorption of the wavefield s monopole component of both @xmath2 and @xmath0 modes from a thin magnetic tube in a gravitationally stratified atmosphere . the scattering is mediated through the excitation of axisymmetric sausage waves that travel up and down the fibril . the paper is organized as follows : in section [ sec : formalism ] we present our scattering formalism . in section [ sec : matrices ] , we calculate the scattering matrices . section [ sec : absorption ] describes the absorption achieved and the properties of the mode mixing across mode order that occurs , while section [ sec : phases ] presents the resulting phase shifts . finally , in section [ sec : discussion ] we discuss our findings and conclusions , including what our results for a single tube adumbrate for plage . we consider a compact magnetic fibril that is embedded within a field - free atmosphere . the fibril is axisymmetric , straight and aligned with gravity . under such conditions the acoustic wavefield in the surrounding , nonmagnetized atmosphere can be expressed as the sum of three components , @xmath3 here @xmath4 can represent any physical variable we might choose to consider ; however , we will restrict our attention to a plane - parallel atmospheric model that is neutrally stable to convection a good approximation within the sun s convection zone . for such an atmosphere , acoustic waves can be described using a displacement potential and we will let @xmath4 be this potential . the first term , @xmath5 , is the unperturbed @xmath0-mode wavefield that would exist in the absence of the magnetic flux concentration ( i.e. , the _ incident _ wavefield ) . the second term , @xmath6 represents a discrete set of far - field _ scattered _ waves in the form of propagating @xmath0 modes . finally , @xmath7 is the contribution due to the continuous spectrum of laterally evanescent , near - field , _ jacket _ modes ( see * ? ? ? we can express an arbitrary wavefield of incident @xmath0 modes as a fourier - bessel expansion in cylindrical polar coordinates , @xmath8 , where the origin is centered on the flux concentration , @xmath9 in this expression and in the subsequent two we have chosen to apply temporal fourier transforms @xmath10 and to work in frequency space . our fourier convention is such that our solutions have a temporal dependence of @xmath11 . the coefficients @xmath12 are arbitrary complex amplitudes that characterize the incident wavefield , @xmath13 is the bessel function of the first kind , @xmath14 is the vertical eigenfunction for the @xmath15th - order @xmath0 mode , and @xmath16 is the wavenumber eigenvalue for the @xmath15th - order @xmath0 mode with frequency @xmath17 . since we have assumed that the magnetic fibril is axisymmetric , incident waves of azimuthal order @xmath18 only scatter into waves with the same order . therefore , for such an axisymmetric scatterer , the scattering only occurs across mode order @xmath15 , and the explicit form of the scattered waves can be represented as follows : @xmath19 in these expressions @xmath20 is the hankel function of the first kind corresponding to an outward propagating cylindrical wave , and @xmath21 is the modified bessel function of the second kind , with @xmath22 being the lateral decay length for the jacket modes . there are two separate scattering matrices : @xmath23 is the far - field scattering matrix that represents scattering into outgoing , propagating @xmath0 modes , whereas @xmath24 is the near - field matrix and represents the excitation of the acoustic jacket . in this formalism , mode mixing appears as nonzero values for the off - diagonal elements of the far - field scattering matrix ( i.e. , @xmath25 for @xmath26 ) . the functions @xmath27 describe the vertical behavior of each jacket mode and are oscillatory with depth . we note that the @xmath0 mode eigenfunctions @xmath28 and the jacket modes @xmath29 form a complete , orthogonal set . we shall make use of this orthogonality property in the following section to compute the scattering matrices . appendix [ app : eigenfunctions ] describes in detail the @xmath0 mode eigenfunctions and jacket waves appropriate for the neutrally - stable , polytropic atmosphere . in general , the two scattering matrices , @xmath30 and @xmath31 , tell us everything we need to know about the wave - tube interaction , including the absorption coefficient @xmath32 and phase shift @xmath33 . for a particular mode ( i.e. , a particular @xmath18 and @xmath15 ) the absorption coefficient is defined as one minus the ratio of the outgoing power to the ingoing power contained by that mode , @xmath34 the phase shift is defined in the usual way , as the phase difference between the ingoing and outgoing components , @xmath35 if we consider only a single incident wave , we can relate the absorption coefficient and phase shift to the scattering matrices in a straight forward manner . noting that an incident wave component is comprised of outgoing and ingoing parts of equal amplitude , @xmath36 , the outgoing and ingoing amplitudes , @xmath37 and @xmath38 , are then given by , @xmath39 .\end{aligned}\ ] ] if we now insert these expressions for the ingoing and outgoing power into equations ( [ eqn : abscoef_def ] ) and ( [ eqn : phaseshift_def ] ) , we find @xmath40 for realistic wave fields ( e.g. , as sampled by helioseismic observations ) these simple formulae are invalid . firstly , multiple incident waves of all mode orders @xmath15 exist simultaneously . secondly , mode mixing ( the scattering from one mode order @xmath15 to a different mode order @xmath41 ) will result in the outgoing component of any mode order being comprised of the scattered waves from all incident mode orders . one goal of this paper is to assess the importance of mode mixing in the energy budget of an incident wave . fortunately this can be accomplished by examining each incident mode in isolation , and computing the off - diagonal elements of the scattering matrix . we will leave for a subsequent study the effects that a complex , multi - component incident wave field has on measured absorption coefficients and phase shifts . further , we will only consider mode mixing across mode orders . we presume that the scatterer is axisymmetric ; therefore , scattering from one azimuthal order @xmath18 to another @xmath42 does not occur . finally , for simplicity we have chosen to examine only the axisymmetric component ( @xmath43 ) of the incident wavefield , leaving higher - order waves to subsequent studies . the index @xmath18 will be dropped from most expressions from here forward . scattering matrices have been computed previously for axisymmetric magnetized tubes within unstratified atmospheres ( e.g. , * ? ? ? * ; * ? ? ? however , calculating the scattering matrices for a magnetic tube in a gravitationally stratified atmosphere is formidable and can only be accomplished in general through numerical wave - field simulations . however , if we assume that the magnetic flux tube is thin and untwisted , substantial progress can be made without resorting to computer simulation . by _ thin _ , we mean that all characteristic scale lengths such as the wavelength of the external acoustic oscillations and the density scale height are much larger than the radius of the tube @xmath44 . such a thin flux tube , although in hydrostatic balance , is unable to support internal lateral forces ; hence , the total pressure will be uniform across the tube and continuous with the external value . this mandates that the total pressure has the same scale height inside and outside the tube . if we further assume that the temperature is the same within the tube as outside , the plasma parameter @xmath45 , defined as the ratio of the gas pressure to the magnetic pressure , is constant with height inside the tube . for photospheric and deeper layers , this assumption that the tube is in thermal equilibrium with its surroundings is consistent with observations @xcite . thin tubes with constant @xmath45 are theoretically attractive , as they permit analytic treatment of the waves that propagate along them and prove to be weak scatterers for which asymptotic analysis may be used . the scattering matrices are determined by the application of two physical matching conditions . at the interface between the magnetic tube and the surrounding field - free atmosphere , both the total pressure perturbation and the normal component of the fluid displacement must be continuous . in general , thin flux tubes permit two types of magnetosonic waves , sausage and kink waves . the sausage waves are axisymmetric ( @xmath43 ) , longitudinal , slow surface waves with pressure as the primary restoring force . the kink waves are sinuous , transverse , fast waves with @xmath46 and a restoring force composed of magnetic tension and buoyancy . since we will only be considering axisymmetric incident waves , the solution within the magnetic tube will be composed of only sausage waves . we mention for completeness that magnetic tubes also permit torsional alfvn waves which are also axisymmetric oscillations . however , acoustic - gravity waves within a neutrally - stable atmosphere are irrotational and are unable to couple to torsional waves . the propagation of sausage waves on magnetic fibrils is reasonably well understood ( e.g. , * ? ? ? * ; * ? ? ? such waves have the property that the total pressure is nearly constant across the cross section of the tube and matches the overpressure in the field - free surrounding media . furthermore , the normal component of the fluid displacement at the boundary of the tube is a small quantity of the same size as the radius of the tube ( which is presumed small by the thin tube approximation ) . the smallness of the normal displacement enables the solution inside the tube to be obtained to leading order with only the enforcement of pressure continuity ; the continuity of the displacement appears only in higher - order corrections @xcite . here we explicitly consider these higher order terms in order to compute the scattered wave field . outside the flux tube , a small argument expansion of the bessel functions appearing in equation ( [ eqn : incident ] ) reveals that , to lowest order in the radius of the flux tube @xmath47 , the pressure perturbation of the incident wave is a zero - order quantity @xmath48 , whereas the normal displacement is first order @xmath49 . thus , the ordering of terms for the incident waves is the same as the sausage waves . however , similar expansions of equations ( [ eqn : scattered ] ) and ( [ eqn : jacket ] ) reveal that the far - field and near - field scattered waves have a pressure perturbation whose leading - order terms are proportional to @xmath50 or @xmath51 , respectively , where @xmath52 and @xmath53 indicate the leading - order behavior of the scattering matrices . the normal displacements for these waves are proportional to @xmath54 and @xmath55 . the size of each of these terms for each wave component is summarized in table 1 . careful scrutiny reveals that the manner in which a balance can be successfully achieved , where both continuity conditions are satisfied , is if the scattering matrices are to leading order proportional to @xmath56 . with this behavior , pressure continuity is maintained by a balance between the sausage wave and the incident wave ; the scattered waves play no role to leading order . continuity of the normal displacement involves a balance between the sausage wave and all three external wave components ( incident , far - field scattered and near - field scattered ) . table 1 : leading order of each wave component [ cols="^,^,^",options="header " , ] by these arguments , the sausage waves are excited by the incident wave through pressure continuity and the scattered waves ( and the scattering matrices ) are generated by the sausage waves through continuity of the normal component of the displacement . the normal displacement outside the tube @xmath57 is a linear combination of wave components , @xmath58 \ ; , \ ] ] where @xmath59 is the normal displacement for the incident wave of order @xmath15 and @xmath60 is a scale depth ( chosen to be the depth of the photosphere in our field - free atmospheric model see appendix [ app : eigenfunctions ] ) . the displacements @xmath61 and @xmath62 are the far - field and near - field scattered waves generated by incident mode order @xmath15 . all displacements are evaluated at the flux - tube interface @xmath63 and any given displacement can be computed from the associated displacement potential through the relation , @xmath64^{-1 } \ ; , \ ] ] where @xmath65 is the unit vector normal to the surface . the leading order behavior of each of these terms is found by small argument expansion of the bessel functions in equations ( [ eqn : incident])([eqn : jacket ] ) , @xmath66 \ ; , \label{eqn : ninc } \\ \nonumber \\ n_{{\rm sca},n}(\omega;z ) & = & \frac{z_0 ^ 2}{r(z ) } \ ; \frac{2i}{\pi } \sum_{n'=0}^\infty \ ; s_0^{n\to n'}(\omega ) \ ; q_{n'}(\omega;z ) \ ; , \label{eqn : nsca } \\ \nonumber \\ n_{{\rm jac},n}(\omega;z ) & = & - \frac{z_0 ^ 2}{r(z ) } \ ; \int_{0}^\infty d\lambda \ ; t_0^{n\to \lambda}(\omega ) \ ; q(\omega,\lambda;z ) \ ; . \label{eqn : njac}\end{aligned}\ ] ] each order of incident wave excites a separate sausage wave on the magnetic fibril , @xmath67 appendix [ app : sausagewaves ] provides details on how to compute each @xmath68 . continuity of the normal displacement requires that @xmath69 . if we insert equations ( [ eqn : ninc])([eqn : njac ] ) into this relationship and multiply by @xmath70 , we obtain an equation for the mismatch " between the incident and tube waves , @xmath71 \ ; , \label{eqn : dn } \\ & = & -\frac{2i}{\pi } \sum_{n'=0}^\infty s_0^{n \to n'}(\omega ) \ ; q_{n'}(\omega;z ) + \int_0^\infty d\lambda \ ; t_0^{n \to \lambda}(\omega ) \ ; q(\omega , \lambda ; z ) \ ; . \end{aligned}\ ] ] figure [ fig : wave_functions ] illustrates the two displacement components for an incident @xmath72 mode , the components for the resulting sausage wave , and the mismatch between them @xmath73 . we can now solve for the two scattering matrices separately by invoking the orthogonality of the eigenfunctions and the jacket waves . in appendix [ app : eigenfunctions ] we discuss that the weighting function for the orthogonality integrals is proportional to the mass density . if we make use of a dimensionless depth @xmath74 , the density within a neutrally stable atmosphere is a power law @xmath75 , where @xmath76 is the photospheric density and the polytropic index @xmath77 is related to the ratio of specific heats @xmath78 through @xmath79 . multiply equation ( [ eqn : dn ] ) by @xmath80 and the complex conjugate of either a @xmath0-mode eigenfunction @xmath81 or a jacket wave @xmath82 and integrate over all depths to obtain , @xmath83 the limits of integration correspond to the photosphere ( @xmath84 ) and infinitely deep into the solar interior . each element of the scattering matrices can therefore be thought of as the projection of the mismatch between the incident and scattered waves onto each @xmath0-mode and jacket - mode eigenfunction . the matrix associated with the far - field scattering , @xmath85 can be easily obtained by direct numerical integration of equation ( [ eqn : smatrix ] ) . we have used a bulirsch - stoer numerical integrator to perform these integrations and the results are discussed in the following sections . the integral that must be computed to obtain @xmath86 is fundamentally more difficult to numerically compute since jacket wavefunctions are not confined to an acoustic cavity and remain propagating to any depth . therefore , while the magnitude of the integrand does vanish as @xmath87 , it does so slowly as a power law . this is unlike the integrand of equation ( [ eqn : smatrix ] ) which behaves like the product of a power law times a rapidly decaying exponential as @xmath87 . therefore , we delay the calculation of the near - field scattering matrix to a subsequent time . fortunately , the far - field and near - field scattering matrices are independent of each other due to the orthogonality of the @xmath0 modes and the jacket modes ; thus , using this technique to calculate the far - field scattering matrix avoids the necessity of computing the acoustic jacket simultaneously . under many reasonable boundary conditions , one can demonstrate that the far - field scattering matrix @xmath88 is symmetric under exchange of @xmath15 and @xmath89 ( see appendix [ app : symmetrys ] for a derivation ) . here we apply a stress - free condition at the photosphere ( for both the @xmath0 modes and the sausage waves ) and finite energy ( for the @xmath0 modes ) and radiation ( for the sausage waves ) boundary conditions deep in the atmosphere ( as @xmath87 ) . these boundary conditions are among those for which the scattering matrix is symmetric . this means that we need not calculate all of the matrix elements , only a little more than half need to be computed directly with the remaining obtained by symmetry . the absorption coefficient is often considered to be a measurement of energy loss or dissipation . when mode mixing occurs , this simple scenario needs to be modified as in addition to true absorption ( or removal of energy from the acoustic wavefield ) , energy can be shuffled between modes by mode mixing . in this section we calculate the energy budget for a single incident wave and account for all the ways that the incident wave energy can be reprocessed by the scatterer . when the scatterer is an axisymmetric thin flux tube , there are only three ways in which energy can be extracted . ( 1 ) the single incident wave can excite tube waves which subsequently propagate out of the acoustic cavity . this mechanism is a true absorption process whereby acoustic energy is removed from the system . ( 2 ) the incident wave can generate an acoustic jacket that surrounds the flux tube . this mechanism is only significant for initial value problems and not for a spectral treatment as performed here . ( 3 ) the incident wave is scattered into outgoing waves of a different mode order . this last effect is mode mixing which is not true absorption as the acoustic energy remains in the cavity , albeit at a different wavenumber . we will assess the relative importance of the first and third mechanisms and leave the second to subsequent work . consider a single incident wave of order @xmath15 , with a unit amplitude ( @xmath90 ) . this incident wave generates a set of outgoing waves with mode orders @xmath89 . these outgoing waves can be written as follows , where we have included both the far - field scattered waves and the outgoing component of the incident wave , @xmath91 h_m^{(1)}(k_{n'}r ) \ ; q_{n'}(\omega;z ) \ ; .\ ] ] the term in square brackets is the amplitude of the outgoing wave of mode order @xmath89 , and is composed of a contribution from the incident wave ( @xmath92 ) and from a scattered wave ( @xmath93 ) . if the vertical eigenfunctions @xmath94 are normalized according to equation ( [ eqn : orthog_discrete ] ) , each outgoing wave carries a lateral energy flux @xmath95 with the following form , @xmath96 note , the energy flux only depends on the mode order @xmath89 ( and the azimuthal order @xmath18 ) through the amplitude . therefore , the fraction @xmath97 of the ingoing wave s energy that is re - emitted in mode order @xmath89 has similar dependence on the scattering matrix as exhibited by the absorption coefficient , @xmath98 for this example with only a single incident mode order , the absorption coefficient is by definition 1 minus the diagonal elements of this fractional energy flux , @xmath99 of course this equation is identical to equation ( [ eqn : abscoef_singlemode ] ) . by energy conservation , we can decompose this absorption coefficient into two parts , a true absorption from the excitation and escape of sausage waves @xmath100 and a redistribution term arising from mode mixing @xmath101 , @xmath102 the contribution from mode mixing @xmath101 can be directly written down using the off - diagonal elements of the fractional energy loss @xmath103 , @xmath104 therefore , by combining equations ( [ eqn : alpha ] ) and ( [ eqn : emodemix ] ) , the loss of energy due to the excitation of tube waves must have the following expression , @xmath105 before we can use any of these expressions , we need to consider the order of each of their terms . since we have computed the scattering matrix to only leading order in the radius of the tube , @xmath106 , for self - consistency we must drop the last term in equations ( [ eqn : alpha ] ) and ( [ eqn : etubewaves ] ) , @xmath107 to lowest order , the absorption coefficient and the excitation rate of tube waves are identical . therefore , our calculation of the absorption coefficient is only compatible with tube - wave generation . the terms due to mode mixing appear at higher orders which are not represented self - consistently . fortunately , we can evaluate the mode mixing terms separately using the far - field scattering matrix through equations ( [ eqn : efrac ] ) and ( [ eqn : emodemix ] ) . figure [ fig : absorb ] illustrates the absorption coefficient ( solid curves ) evaluated through the use of equation ( [ eqn : efrac_orderr2 ] ) . the fraction of the incident energy , @xmath101 , mixed into all other modes is represented with the dotted curves . in figure [ fig : modemix ] , the energy @xmath103 that is scattered into each individual mode is shown . from this figure , one can see that the amount of energy lost to mode mixing decreases as the order of either the scattered mode @xmath89 or the incident mode @xmath15 increases . the symmetry of the scattering matrix leads to the symmetry of @xmath97 , which is apparent in figure [ fig : modemix]@xmath77 as the equal height of pairs of symbols of different color and in figure [ fig : modemix]@xmath108 as the overlying of curves . for low - frequencies , the mode - mixing losses are rather inconsequential compared to the sausage - wave losses . however , for higher frequencies the amount of energy lost to mode mixing can be a relative high fraction of the total energy loss , perhaps reaching as high as a third for some parameters . furthermore , for the @xmath2 mode at high frequencies the excitation of sausage waves becomes anomolously inefficient and the mode - mixing terms can be the dominant source of energy loss . a more in depth discussion of tube wave excitation by the @xmath2 mode appears later in [ subsec : absorption ] . the computation of the scattering matrix also allows us to estimate phase shifts @xmath109 through equation ( [ eqn : phaseshift_singlemode ] ) . these phase shifts are illustrated in figure [ fig : phase_shift ] . for @xmath0 modes , both positive and negative phase shifts are possible , although all shifts become positive for sufficiently high frequency which , depending on the value of @xmath45 may be beyond the physically relevant band of frequencies . the @xmath2 mode is fundamentally different . not only is the phase shift for the @xmath2 mode negative for all frequencies , but the phase shift is large , reaching values as high as 15@xmath1 for high frequencies and for tubes with low values of the plasma parameter @xmath45 . the phases of all the scattered waves are correlated in a precise manner with the phase of the incident wave . the difference in phase @xmath110 between each individual scattered wave and the incident wave can also be derived from the scattering matrix , @xmath111 these relative wave phases are uniformly small ( see figure [ fig : scatphase ] ) . the difference is positive for @xmath112 even and negative for @xmath112 odd . the symmetry of the scattering matrix to exchange of @xmath15 and @xmath89 appears in figure [ fig : scatphase ] as the overlying of curves and the equal height of symbols . in general , scattering to and from the @xmath2 mode results in larger phase differences , exceeding scattering between @xmath0 modes by an order of magnitude . we have calculated by semi - analytic means the scattering matrix for axisymmetric acoustic waves encountering a vertically - aligned , magnetic tube . we have assumed that the tube is slender and invoked the thin flux - tube approximation to compute the scattering matrix to leading order in the tube s radius @xmath47 for mode orders up to @xmath113 . in the following two sections we discuss the implications of the results of our calculation , both in terms of the energy budget of the incident and scattered waves and in terms of the phases of the outgoing waves . when encountering a thin flux tube of the type discussed here , a @xmath0 mode of a particular radial order @xmath15 may lose energy in either of two ways . the incident acoustic wave can excite mhd tube waves on the magnetic fibril that freely propagate along the tube carrying energy out of the acoustic cavity . this represents a true absorption as acoustic energy is removed from the wavefield . the tube may also scatter the incident @xmath0 mode into outgoing waves of different radial order @xmath89 . mode mixing of this sort does not represent a true absorption as the energy is merely redistributed between radial orders . despite this fact , mode mixing can modify the observed absorption coefficient , as the definition of this coefficient does not take power redistribution into account . these two mechanisms are both inherently weak for thin tubes . the thin tube approximation mandates that the excitation of tube waves generates an absorption coefficient that scales linearly with the magnitude of the scattering matrix , which in turn scales like the square of the tube s radius @xmath56 , a small quantity by definition . the energy losses due to mode mixing are even smaller , scaling as the square of the magnitude of the scattering matrix , or as @xmath114 . therefore , mode mixing can only be important energetically when other sources of absorption ( i.e. , tube wave excitation ) are peculiarly small . over the physically relevant range of frequencies ( @xmath115 mhz ) , our calculations reveal that mode - mixing dominates the absorption coefficient only for high - frequency @xmath2 modes interacting with low-@xmath45 tubes . any model with a reflective upper boundary can have special frequencies for which the excitation of tube waves is unusually tiny . if the upper boundary reflects all wave energy , the excitation can formally be zero at these frequencies . such excitation nulls " arise because of total destructive interference . a localized driver , such as an incident @xmath0 mode , excites waves that propagate both up and down the tube with equal energy . at these special frequencies , the upward propagating wave reflects off the upper surface with just the phase needed to cancel the downward propagating wave @xcite . in such a circumstance the downward energy flux vanishes and no energy is carried away from the acoustic cavity by tube waves . near such an excitation null , the energy losses arising from mode mixing could be the dominant source of energy loss not because mode mixing is large , but because the other source of absorption is small . for the stress - free upper boundary condition applied here and for reasonable atmospheric and tube parameters , excitation nulls do not appear within the physically relevant frequency band ( @xmath116 mhz ) . however , formally for the @xmath2 mode such a null exists in the limit of infinite frequency and it has extremely broad influence , causing the absorption coefficient to asymptote to zero as the frequency becomes large . the frequency regime for which this asymptotic behavior is valid depends on @xmath45 , with lower @xmath45 pushing the behavior to lower frequencies . physically , such behavior exists because the @xmath2 mode is a surface wave that becomes increasingly confined to the photospheric layers as the frequency increases . the @xmath2 mode s evanescence length @xmath117 can be derived directly from its dispersion relation @xmath118 , revealing that the vertical extent of the @xmath2 mode scales as @xmath119 . the wavelength @xmath120 of the sausage wave at the photospheric surface can be derived from equation ( [ eqn : homogeneoustubewave ] ) , @xmath121^{1/2 } \frac{\pi}{\omega } \ ; .\ ] ] as the frequency increases , @xmath120 shrinks , scaling as @xmath122 ; however , it decreases more slowly than the scale length of the @xmath2 mode decreases ( which scales as @xmath119 ) . therefore , as the frequency increases , compared to the tube wave the driver looks more and more like a delta function located at the surface . a delta - function driver generates waves that propagate in each direction with identical amplitude and phase . if one of those components subsequently reflects off a surface and passes back through the driving layer , there is the potential for destructive interference ( as discussed previously ) . destructive interference occurs if the phase of the reflected wave advances by an odd integer factor of @xmath123 . this phase advance can come from a phase change on reflection or from propagation of the wave to the reflecting layer and back . for the @xmath2 mode , one can show that the phase change on reflection asymptotes to @xmath124 as the frequency becomes large ( see appendix [ app : highfrequency ] ) . therefore , the @xmath2 mode , in the limit of infinite frequency , satisfies the necessary resonance condition because the distance between the driving layer and the photospheric reflection layer vanishes . the @xmath0 modes do not evince such high - frequency behavior for the simple reason that at high frequency , the depth of the cavity becomes roughly constant with frequency . the wavenumber eigenvalue @xmath125 scales linearly with the frequency , thereby causing the horizontal phase speed @xmath126 and the lower turning point of the @xmath0 mode to approach constants . since , the depth of the cavity and hence the vertical extent of the driving region does not change as the frequency changes , the wavelength of the sausage waves becomes much shorter than that of the @xmath0 modes . thus , for incident @xmath0 modes , the driving region remains vertically distributed and the arguments that we made previously for the @xmath2 mode are not valid . figure [ fig : highfreq ] illustrates this behavior for @xmath127 . while the figure extends to frequencies as large as 30 mhz , one must bear in mind that frequencies greater than 6 mhz are greater than the sun s photospheric acoustic cutoff ; thus , such waves are no longer trapped and the construction of our solar model is suspect . further , one must be careful to not over - interpret the results at such extreme frequencies as the perturbation scheme has become invalid ( see figure [ fig : highfreq ] ) . despite this , for the @xmath2 mode , trapped waves of low frequency ( where the perturbation analysis remains valid ) can still sense the presence of the excitation null at infinity , causing a high - frequency fall - off of the absorption . the frequency of waves affected depends on the plasma parameter @xmath45 of the tube because the wavelength of the tube wave depends on @xmath45 . we can estimate the lowest frequencies that should be affected by requiring that the wavelength of the sausage waves @xmath120 be two or three times larger than @xmath2 mode s evanescence length @xmath117 . if we require that the ratio of these length scales satisies @xmath128 , we discern that for @xmath127 we would expect the high frequency regime to exist for frequencies above 3 mhz , whereas for @xmath129 the regime begins around 10 mhz . this discussion only holds for absorption and mode mixing resulting from the excitation of the slow sausage wave . similar calculations performed with fast kink oscillations reveal that excitation nulls can appear for all incident mode orders at frequencies less than 6 mhz @xcite ; also see @xcite for a similar phenomenon in an unstratified model . further , the calculations of the scattering matrix elements arising from coupling to kink oscillations ( @xmath130 ) @xcite show significant oscillatory modulation in the magnitudes as a function of frequency , whereas our scattering calculation for the axisymmetric sausage wave ( @xmath88 ) show smooth frequency variation . in fact one could fit the magnitude of the scattering matrix calculated here with a single power law in frequency that would be roughly valid over all but the highest frequencies . in a real plage , we do not expect such excitation nulls to have an easily identifiable effect other than a diminution of absorption at high frequencies for the @xmath2 mode . not only does mode mixing tend to fill in the absorption profile , but the plage is likely to be comprised of a large number of tubes with a distribution of properties , including photospheric radius @xmath131 and plasma parameter @xmath45 . therefore , each tube in the collection will have suppressed absorption over a different range of frequencies . the added affect of all of the tubes will be to mitigate the effect of nulls @xcite . in general , we expect most flux tubes in a plage to have a near equipartition of magnetic and gas pressure , thus tubes with @xmath132 should predominate . a quick examination of figure [ fig : absorb ] reveals that only the most extreme frequencies are likely to be affected . to summarize , we find that for thin tubes , mode mixing is a small effect compared to tube wave excitation . the scattering matrix itself has elements that are proportional to the cross - sectional area of the tube . tube wave excitation extracts energy at a rate proportional to the scattering matrix ( or proportional to the cross - sectional area of the tube ) while mode mixing redistributes energy at a rate proportional to the square of the scattering matrix . phase shifts are generally defined as the difference in phase between the outgoing wave in the presence of the scatterer and the phase that would have existed in the absence of the scatterer . a magnetic tube can generate a phase shift by three distinct mechanisms : ( 1 ) the ingoing acoustic wave reflects off the lateral surface of the tube , shortening the total path length traveled by the wave . ( in the absence of the tube , the wave would travel all the way to the coordinate axis and back . ) since the wave travels a shorter distance , the presence of the tube causes a phase lag resulting in a negative phase shift or travel - time difference . ( 2 ) the wave is at least partially transmitted through the lateral surface and propagates at a different speed within the magnetic tube . if the phase speed increases , a negative phase shift is achieved . ( 3 ) upon reflection off the lateral surface of the tube , the wave can undergo a phase change . off course if a flux - tube boundary could be treated as a rigid , impenetrable surface , the phase change would be 180 degrees . but , in general a flux tube is elastic and the phase change may differ from 180 degrees and may be either positive or negative . for our thin flux - tube model , the only wave permitted within the tube is a sausage wave corresponding to a slow surface wave ( see * ? ? ? * ; * ? ? ? for such a wave , the thin flux - tube approximation precludes lateral variation by implicitly assuming that the wave crossing time is much shorter than the wave period . thus , for all tubes the crossing time for the tube should be ignored , resulting in a common negative phase shift , independent of @xmath45 . obviously the phase shift whould be equal to the phase lag that would have occured in the unmagnetized model for a wave propagating a distance equal to the tube diameter . this shows that for thin flux tubes , mechanisms 1 and 2 discussed in the previous paragraph are essentially the same . this is not true for thicker tubes for which the travel time across the tube is significant . a rough estimate of the expected phase shift would be @xmath133 , where @xmath131 is the photospheric radius of the tube and we have _ assumed _ that the phase lag should be evaluated at the photospheric surface . this is clearly appropriate for the @xmath2 mode as it is a surface wave confined to this layer . for @xmath0 modes one should probably consider a depth weighted average of the tube radius . however , for our purposes , the previous expression will suffice . estimates made this way for the @xmath2 mode give a phase shift of @xmath134 , which for waves at 3 mhz give a phase shift of -15@xmath1 . this surprisingly large shift is about a factor of four larger than the phase shifts we obtain with our model ( see [ fig : phase_shift ] ) , indicating that a phase change upon reflection is likely to be at work as well . we point out that despite the large size of the estimated shift , it is quite similar in magnitude and of the same sign as actually measured phase shifts for small magnetic features . @xcite measured the average scattering from roughly 2500 small magnetic elements , and generated a typical travel time kernel from their measurements . one can obtain a typical phase shift of -14@xmath1 by the following process . multiply the largest value of the measured kernel in figure 6 from @xcite ( roughly -20 s kg@xmath135 mm@xmath135 ) by the average magnetic flux of the studied features ( 0.6 kg mm@xmath136 ) to obtain a travel time lag of -12 s. convert this time lag into a phase shift by dividing by a wave period ( 300 s ) and multiplying by 360 degrees . the phase change that occurs at reflection from the lateral boundary of the flux tube is difficult to estimate , as the reflection depends on both the incident wave and the driven sausage wave over a range of heights . we would like to receive guidance from calculations performed for unstratified , thin tubes , for which analytic solutions are possible ( e.g. , * ? ? ? . however , the regime in frequency wavenumber space for which slow surface waves exist and for which propagating external acoustic waves exist are disjoint . thus , for unstratified tubes a propagating acoustic wave can not excite slow surface waves . the stratification in our model is what makes the coupling possible . we can safely say , however , that the phase shift caused by mechanisms 1 and 2 always result in negative shifts for thin tubes . further , we would expect the phase change upon reflection ( mechanism 3 ) to increase for tubes of higher @xmath45 since such tubes are more elastic . for the @xmath2 mode this is indeed what we achieve ; the phase shift is large and negative at all frequencies ( probably as a result of a reduced path length ) and becomes less negative as @xmath45 increases ( probably due to an increasing positive phase change at reflection ) . however , for @xmath0 modes we obtain shifts of both signs ; negative phase shifts at low frequency and positive shifts at high frequency . clearly for the @xmath0 modes a positive phase change upon reflection at the flux - tube boundary must occur for at least high frequencies . we thank tom duvall for his helpful suggestions in how to convert measured travel - time kernels into phase - shift estimates . we thank tom bogdan for a useful conversation during which the proper ordering of the different wave components became apparent . bwh acknowledges support from nasa through grants nng05gm83 g , nnx08aj08 g , nnx08aq28 g and nnx09ab04 g . rj acknowledges stfc ( uk ) for partial funding . the field - free atmosphere that surrounds the magnetic flux tube is treated as a plane - parallel , neutrally stable polytrope . this atmosphere has been utilized extensively in the past as its a good representation of a stellar convection zone and simplifies the wavefield by imposing a condition of irrotationality . gravity is uniform , acting in the downwards direction @xmath137 , with the height coordinate @xmath138 increasing upward . the atmosphere is polytropic below the height @xmath139 , which corresponds to the model s photosphere . the gas pressure , mass density and sound speed vary with height as power laws . if we employ a nondimensional depth @xmath74 , the background thermodynamic variables become , @xmath140 the quantities @xmath76 and @xmath141 are the photospheric values of the mass density and gas pressure . the value of the polytropic index @xmath77 is set such that the stratification is neutrally stable to convection ; this requires @xmath142 , where @xmath78 is the ratio of specific heats . above the photosphere @xmath143 , we assume the existence of a hot vacuum ( @xmath144 with @xmath145 ) , with the property that the gas pressure ( @xmath146 ) is finite and continuous across the @xmath143 layer . following @xcite and @xcite , we specify the depth of the photosphere @xmath60 and the photospheric density @xmath76 ( and therefore the reference pressure @xmath141 ) by matching our model photosphere to the @xmath147 level of a solar model by @xcite . at this layer in the solar model @xmath148 cm s@xmath149 , @xmath150 g @xmath151 , and @xmath152 g cm@xmath135 s@xmath149 . we adopt a polytropic index of @xmath154 , which is consistent with an adiabatic index of @xmath155 . since the buoyancy frequency in such an atmosphere is zero by definition , acoustic waves propagating within the atmosphere are irrotational . this permits such waves to be expressed using a displacement potential @xmath4 . if we assume oscillatory time dependence @xmath156 and harmonic functions with wavenumber @xmath157 in the horizontal direction , the vertical portion of the acoustic eigenfunction satisfies the following sturm - liouville equation , @xmath158q(\nu;s ) & = & 0 \ ; , \label{eqn : sturm_liouville } \\ \lambda & \equiv & 2 k z_0 \ ; , \\ \nu^2 & \equiv & \frac{a \omega^2 z_0}{g } \ ; , \end{aligned}\ ] ] where @xmath159 is a dimensionless horizontal wavenumber , @xmath160 is a dimensionless frequency and @xmath81 represents either a discrete eigenfunction @xmath161 for real @xmath162 or a spectrum of jacket modes @xmath163 for continuous , imaginary values of @xmath164 . the general solutions to this equation are proportional to whittaker @xmath165 and @xmath166 functions @xcite . the discrete @xmath2 and @xmath0 modes reside within an acoustic cavity and therefore become evanescent deep in the atmosphere . thus , the @xmath166 functions are relevant as they decay exponentially as @xmath167 , @xmath168 here @xmath169 is a normalization constant and @xmath170 the quantization of the horizontal wavenumber @xmath171 ( or equivalently @xmath172 ) arises from the requirement that the lagrangian pressure perturbation vanish at the model photosphere . the laterally evanescent jacket waves are surface waves ( with a lateral decay length of @xmath22 ) that propagate up and down the outside of the flux tube . they are not trapped within the acoustic cavity and can freely propagate to any depth . if we select only those jacket waves that satisfy the condition that the lagrangian pressure perturbation vanishes at the model photosphere , the jacket waves are standing waves that can be represented as a sum of whittaker @xmath165 functions with imaginary argument ( see * ? ? ? * ) , @xmath173 \ ; , \ ] ] where @xmath174 is a normalization constant and @xmath175 unlike the @xmath2 and @xmath0 modes , for which the dimensionless horizontal wavenumber @xmath159 is purely real and only takes on discrete values @xmath172 , for the jacket modes @xmath159 is purely imaginary and continuous . therefore , the argument and @xmath176-index of the whittaker functions in equation ( [ eqn : jacketmodes ] ) are purely imaginary as well . since both the discrete modes and the jacket modes satisfy hermitian boundary conditions ( vanishing lagrangian pressure perturbation at both the photosphere @xmath84 and at infinity @xmath167 ) , we automatically know that the discrete , acoustic modes and the jacket modes are all mutually orthogonal with a weight function @xmath80 ( which is proportional to the mass density ) , @xmath177 in writing these last equations , we have chosen @xmath169 and @xmath174 such that the discrete modes and jacket modes are orthonormal . we have assumed the magnetic fibril is untwisted , straight , axisymmetric , vertically aligned and thin . thin flux tubes are unable to support internal lateral structure and hence the total pressure is uniform across the tube and continuous with its external value . if we further assume that the tube is the same temperature as its surroundings , the total pressure must have the same scale height inside and outside the tube and the plasma parameter @xmath45 , defined as the ratio of the gas pressure to the magnetic pressure , is constant with height inside the tube . since we only consider tubes below the photosphere ( i.e. , @xmath178 ) , we need not worry about the rapid flaring of the tubes into a magnetic canopy that occurs within the chromosphere . we ignore lateral variation of the magnetic field strength , and describe the tube s internal gas pressure @xmath179 , mass density @xmath180 , sound speed @xmath181 , and field strength @xmath182 by their axial values . these four quantities as well as the tube s radius @xmath44 can be described uniquely by the total magnetic flux @xmath183 contained by the tube and by the plasma @xmath45 , @xmath184 observations indicate that typical field strengths for small photospheric flux tubes are between 1 and 2 kg within both quiet sun and plage ( e.g. , * ? ? ? * ; * ? ? ? such field strengths indicate rough equipartition between magnetic and gas pressure . for a flux tube with @xmath185 embedded in the polytropic atmosphere described in appendix [ app : eigenfunctions ] , the magnetic field strength at the photosphere is @xmath186 kg . by inserting the external pressure @xmath187 into the expression for the cross - sectional area of the tube ( [ eqn : cross - area ] ) , one obtains the explicit dependence of the tube radius on depth , @xmath188 where @xmath189 is the photospheric radius of the tube . for all illustrations we will assume @xmath190 km , a fairly typical value for the size of magnetic elements within solar plage @xcite . thin flux tubes support a longitudinal magnetoacoustic wave known as the sausage wave . here , these oscillations are driven by the pressure perturbation imposed by external @xmath2- and @xmath0-mode oscillations ( e.g. , * ? ? ? using the formulation of @xcite , to lowest order in the radius of the flux tube , the vertical displacement @xmath191 within the tube can be described by the following equation , @xmath192 where @xmath193 is the cusp or tube speed which depends on both the sound speed @xmath194 and alfvn speed @xmath195 within the tube , @xmath196 the right - hand side of equation ( [ eqn : parallel ] ) is due to the forcing by the incident wave and is evaluated at the flux tube s axis , @xmath197 . this forcing arises from the enforcement of continuity of total pressure across the boundary of the flux tube . we are allowed to consider only the incident waves in this pressure balance because the far - field and near - field scattering amplitudes are smaller than the incident wave by a factor of @xmath198 . the forcing provided by each individual incident @xmath0 mode can be obtained by evaluating the expression for the incident wavefield , equation ( [ eqn : incident ] ) , at @xmath199 and inserting the result into the sausage wave equation . after fourier transforming in time @xmath200 , one obtains @xmath201 note that only the @xmath0 modes with @xmath202 contribute to the forcing of the sausage waves . all other azimuthal components vanish in the limit @xmath203 . we adopt a non - dimensional form for this equation by the following substitutions , @xmath204 producing @xmath205 the solution of this equation is expressed using a green s function , @xmath206 + \psi_\parallel^*(s)\left[{\cal i}_n-{\cal j}_n(s)\right ] \right\ } \ ; . \label{eqn : zetan}\end{aligned}\ ] ] in this equation , @xmath207 and @xmath208 are the two solutions to the homogeneous equation , @xmath209 and @xmath210 are interaction integrals over the driver , and @xmath211 is a parameter that specifies the complex amplitude of the wave that reflects from the upper surface and thus determines the boundary condition at the model photosphere ( see * ? ? ? * for details ) , @xmath212 if we apply a stress - free boundary condition on the sausage wave at the photosphere , the boundary condition parameter has the following form , @xmath213 figure [ fig : wave_functions]@xmath214 illustrates the resulting sausage wave as well as the driving @xmath0-mode eigenfunction when the incident wave is a @xmath72 wave . we can derive an expression for the normal displacement inside the tube using the definition for the divergence of the displacement vector @xmath215 in cylindrical coordinates , @xmath216 where @xmath217 is the radial displacement . in this equation we have utilized the axisymmetry of the sausage mode . this can be combined with the taylor series expansion of the radial displacement , @xmath218 which we evaluate at @xmath219 to obtain @xmath220 if we now use equation a9 from @xcite , which is a statement about pressure equilibration of the tube with its external environment , @xmath221 and combine equations ( [ eqn : ncomp ] ) and ( [ eqn : hpar ] ) , we obtain after some manipulation @xmath222 .\ ] ] this displacement can be decomposed into seperate components driven by each incident mode , by utilizing equations ( [ eqn : incident ] ) and ( [ eqn : zetacomponents ] ) , @xmath223 . \label{eqn : nparn}\end{aligned}\ ] ] the eigenfunction for the @xmath2 mode is simple enough that we can perform high - frequency asymptotics to derive approximate analytic solutions for the interaction integral @xmath224 and all quantities derived from it . the @xmath2 mode eigenfrequency is always @xmath118 independent of the details of the stratification . the associated eigenfunction is just an exponential that grows with height @xmath225 . in our dimensionless variables , these relations take the form @xmath226 and @xmath227 . the @xmath2 mode s evanescence length is given by @xmath228 and as the frequency becomes large , the evanescence length becomes tiny , scaling as @xmath229 . the @xmath2 mode s eigenfunction is , therefore , progressively more confined to the photospheric surface as the frequency becomes large . the tube wave on the other hand has a wavelength that diminishes much more slowly . from equation ( [ eqn : homogeneoustubewave ] ) the wavelength of the sausage wave can be seen to scale as @xmath230 . thus , in the interaction integral @xmath224 see equation ( [ eqn : interactionintegral ] ) , the integrand is only significant near the lower limit of integration ( near the photosphere ) and the sausage wave can be considered long wavelength . after inserting the @xmath2-mode eigenfunction and the homogeneous sausage wave solution equation ( [ eqn : homogeneoustubewave])into equation ( [ eqn : interactionintegral ] ) , we obtain @xmath231 since the exponential term in the integrand has significant magnitude only very close to the lower limit of integration and the sausage wave is long wavelength compared to the exponential scale length , we may expand the function @xmath232 in a taylor series about @xmath233 , @xmath234 if we define @xmath235 and perform the necessary derivatives , the following expression is obtained , @xmath236 ( s-1)^2 + \cdots\end{aligned}\ ] ] after inserting this expansion into the integral appearing in equation ( [ eqn : i0 ] ) , each term in the expansion can be integrated separately by analytic means , producing the following expansion for the interaction integral , @xmath237 \ ; .\ ] ] the order of the missing terms was deduced by noting that for large frequency the large - argument expansion of the hankel function in equation ( [ eqn : little_h ] ) is appropriate . therefore , @xmath238 and the leading order frequency dependence for @xmath224 is @xmath239 . to lowest order , the interaction integral should decay rapidly as the frequency increases . the complex amplitude of the wave that reflects from the photosphere is given by the boundary condition parameter @xmath240 . for the stress - free upper boundary with an @xmath2-mode incident wave , the boundary condition parameter @xmath241 takes on the following form , @xmath242 if we now combine equations ( [ eqn : i0_asymp ] ) and ( [ eqn : omega0 ] ) we find asymptotically , @xmath243 \label{eqn : omega0_asymp}\end{aligned}\ ] ] from equation ( [ eqn : zetan ] ) we can easily see that well below the driving layer , as @xmath87 , the amplitude of the downward propagating sausage wave becomes proportional to @xmath244 . oddly , if we add expression ( [ eqn : omega0_asymp ] ) to the complex conjugate of equation ( [ eqn : i0_asymp ] ) and collect terms of like order , all of the terms cancel to the order calculated . therefore , for an incident @xmath2 mode , in the limit of infinite frequency , the reflected wave and the directly generated wave undergo complete destructive interference and infinite frequency should be considered an excitation null " . in this limit , the driver is essentially a delta function located at the surface . therefore , the phase difference between the reflected wave and the directly generated downward wave can only accrue from a phase change upon reflection off of the photosphere . for destructive interference , this phase change must be @xmath245 . the actual phase change turns out to be @xmath246 , and this can be verified asymptotically . the phase change can be expressed using the ratio of the complex amplitudes of the upward propagating wave and the reflected wave . these can be obtained by evaluating equation ( [ eqn : zetan ] ) at @xmath84 , @xmath247 since the frequency is large , the hankel functions represented by @xmath248 can be expressed using their large argument expansions , @xmath249 if we now compute the ratio @xmath250 using these same large argument expansions we find , @xmath251 combine these last two expressions to find that @xmath246 to leading order . in this appendix we will demonstrate that under several commonly used boundary conditions the far - field scattering matrix , @xmath88 , is symmetric to interchange of @xmath15 and @xmath89 @xmath252 . we start with equation ( [ eqn : smatrix ] ) , using equations ( [ eqn : ninc ] ) , ( [ eqn : tuberadius ] ) and ( [ eqn : nparn ] ) this expression can be reduced to a sum of two terms : a term that depends solely on the incident @xmath0-mode eigenfunction and a term that depends on the excited sausage waves , since we have applied hermitian boundary conditions to the @xmath0 modes , one can easily verify that the operator @xmath259 is also hermitian with respect to the same set of functions ; thus , @xmath261 . so , we now only need consider the second term in the braces in equation ( [ eqn : smatrix_2parts ] ) : is this term symmetric upon exchange of @xmath15 and @xmath89 ? this formulation is different from the one used in appendix b because we will exploit symmetries possessed by the green s function and the present formulation makes those symmetries explicit . in particular , note that the function @xmath266 is symmetric upon the interchange of @xmath267 and @xmath268 . in expression ( [ eqn : greenssolution ] ) , @xmath211 is a complex constant that determines the boundary condition applied to the sausage wave at the surface of the truncated polytrope . for a stress - free surface the appropriate value of @xmath211 is given by , if one inserts equation ( [ eqn : greenssolution ] ) into expression ( [ eqn : smatrix_2parts ] ) , integrates by parts , and uses the fact that the @xmath0-mode eigenfunctions and the tube wave solutions vanish at infinity ( @xmath87 ) , one can obtain the following expression after some manipulation , it is trivial to demonstrate that @xmath273 is symmetric . since the variables @xmath267 and @xmath268 in the integrals are dummy variables , we can exchange them ( @xmath274 and @xmath275 ) and then exchange the order of integration . further , the function @xmath266 is symmetric to interchange of its two arguments . therefore , after performing these exchanges we find the forestated symmetry , @xmath276 . thus , the symmetry of the far - field scattering matrix hinges on the properties of the matrix @xmath277 , which contains boundary terms arising from the integration by parts as well as a contribution that comes from the homogeneous solution appearing in the first term in equation ( [ eqn : greenssolution ] ) . under some common boundary conditions for the sausage waves , this matrix is symmetric , but in general it is not . one can readily verify by directly inserting expressions ( [ eqn : bcstress - free ] ) or ( [ eqn : bcmax - flux ] ) into equation ( [ eqn : bmatrix ] ) , and using the identity for the wronskian of two hankel functions , that both of the boundary conditions mentioned previously ( stress free and minimum reflectivity ) produce a symmetric matrix , @xmath279 . therefore , at least for these two boundary conditions , the far - field scattering matrix is also symmetric . the crucial properties that were required for this symmetry are ( 1 ) the @xmath0 modes satisfy hermitian boundary conditions , ( 2 ) the sausage wave vanishes deep in the atmosphere ( @xmath87 ) , and ( 3 ) the sausage wave satisfies a boundary condition at the surface of the truncated polytrope ( @xmath233 ) that causes the boundary term matrix @xmath277 to be symmetric ( note , not necessarily vanishing ) . this proof is very reminiscent of the calculation that is performed to demonstrate that a differential operator is hermitian ; however , in this case the operator in question is actually an integro - differential operator ( because of the integration over the green s function ) . lagg , a. , solanki , s. k. , riethmller , t. l. , martnez pillet , v. , schssler , m. , hirzberger , j. , feller , a. , borrero , j. m. , schmidt , w. , del toro iniesta , j. c. , bonet , j. a. , barthol , p. , berkfield , t. , domingo , v. , gandorfer , a. , knlker , m. , & title , a. m. 2010 , , 723 , l164

we prove that every geometric lorenz attractor satisfying a strong dissipativity condition has superpolynomial decay of correlations with respect to the unique srb measure . moreover , we prove the central limit theorem and almost sure invariance principle for the time-1 map of the flow of such attractors . in particular , our results apply to the classical lorenz attractor .

1311.5017

the statistical point of view on dynamical systems is one of the most useful tools available for the study of the asymptotic behavior of transformations or flows . statistical properties are often easier to study than pointwise behavior , since the future behavior of an initial data point can be unpredictable , but statistical properties are often regular and with simpler description . one of the main concepts introduced is the notion of _ physical _ ( or _ sinai - ruelle - bowen _ ( srb ) ) measure for a flow ( or transformation ) . an invariant probability measure @xmath0 for a flow @xmath1 is a physical probability measure if the subset of points @xmath2 satisfying for all continuous functions @xmath3 @xmath4 has positive volume in the ambient space . these time averages are in principle physically observable if the flow models a real world phenomenon admitting some measurable features . in 1963 , the meteorologist edward lorenz published in the journal of atmospheric sciences @xcite an example of a polynomial system of differential equations @xmath5 as a very simplified model for thermal fluid convection , motivated by an attempt to understand the foundations of weather forecast . numerical simulations performed by lorenz for an open neighborhood of the chosen parameters suggested that almost all points in phase space tend to a _ chaotic attractor _ , whose well known picture can be easily found in the literature . the mathematical study of these equations began with the geometric lorenz flows , introduced independently by aframovi _ et al . _ @xcite and guckenheimer & williams @xcite as an abstraction of the numerically observed features of solutions to ( [ e - lorenz - system ] ) . the geometric flows were shown to possess a `` strange '' attractor with sensitive dependence on initial conditions . it is well known , see e.g. @xcite , that geometric lorenz attractors have a unique srb ( or physical ) measure . tucker @xcite showed that the attractor of the classical lorenz equations is in fact a geometric lorenz attractor ( see remark [ rmk : tucker ] below ) . for more on the rich history of the study of this system of equations , the reader can consult @xcite . an invariant probability measure @xmath0 for a flow is mixing if @xmath6 as @xmath7 for all measurable sets @xmath8 . mixing for the srb measure of geometric lorenz attractors was proved in @xcite and , by @xcite , this includes the classical lorenz attractor @xcite . results on the speed of convergence in the limit above , that is , of rates of mixing for the lorenz attractor were obtained only recently : a first result on robust exponential decay of correlations was proved in @xcite for a nonempty open subset of geometric lorenz attractors . however , this open set does not contain the classical lorenz attractor . also , it follows straightforwardly from @xcite that a @xmath9-open and @xmath10-dense set of geometric lorenz flows have superpolynomial decay of correlations ( in the sense of @xcite ) . it is likely , but unproven , that this open and dense set includes the classical lorenz attractor . in this paper , we introduce an additional open assumption , _ strong dissipativity _ , that is satisfied by the classical lorenz attractor , under which we can prove superpolynomial decay of correlations . we consider @xmath11 vector fields @xmath12 on @xmath13 possessing an equilibrium @xmath14 which is _ lorenz - like _ : the eigenvalues of @xmath15 are real and satisfy @xmath16 we say that @xmath12 is _ strongly dissipative _ if the divergence of the vector field @xmath12 is strictly negative : there exists a constant @xmath17 such that @xmath18 for all @xmath19 , and moreover the eigenvalues of the singularity at @xmath14 satisfy the additional constraint @xmath20 . for the classical lorenz equations , we have @xmath21 so the conditions and strong dissipativity are satisfied . let @xmath22 denote the open set of @xmath10 vector fields having a strongly dissipative geometric lorenz attractor @xmath23 ; see section [ sec : geometr - aspects - lore ] for precise definitions . given @xmath24 , let @xmath1 denote the flow generated by @xmath12 and let @xmath0 denote the unique srb measure supported on @xmath23 . [ thm : rapid ] let @xmath24 . then for all @xmath25 , there exists @xmath26 and @xmath27 such that for all @xmath28 observables @xmath29 and all @xmath30 , @xmath31 by @xcite , geometric lorenz flows satisfy the central limit theorem ( clt ) for hlder observables . a stronger property is the clt for the time-@xmath32 map @xmath33 which is only partially hyperbolic . by theorem [ thm : rapid ] , @xmath34 has superpolynomial decay of correlations . following @xcite , we use this information to prove the clt for time-@xmath32 maps of geometric lorenz flows thereby verifying conjecture 4 in @xcite . [ thm : clt ] let @xmath35 . then there exists @xmath27 such that for all @xmath28 observables @xmath36 there exists @xmath37 such that @xmath38 \xrightarrow{\mathcal d } \mathcal n(0,\sigma^2)\ ] ] where the convergence is in distribution . moreover , if @xmath39 , then for every periodic point @xmath40 , there exists @xmath41 ( independent of @xmath42 ) such that @xmath43 . [ rmk - sigma ] since there are infinitely many distinct periodic solutions in @xmath23 , it follows from the final statement of theorem [ thm : clt ] that the family of @xmath28 observables @xmath44 for which @xmath39 forms an infinite codimension family in the space of all @xmath28 observables . by @xcite , geometric lorenz flows satisfy also an almost sure invariance principle ( asip ) for vector - valued observables @xmath45 . such a result is currently unavailable for the time-@xmath32 map @xmath34 , but we are able to prove a scalar asip . [ thm : asip ] let @xmath35 . there exists @xmath27 such that for all @xmath28 observables @xmath46 the asip holds for the time-@xmath32 map : passing to an enriched probability space , there exists a sequence @xmath47 of iid normal random variables with mean zero and variance @xmath48 ( as in theorem [ thm : clt ] ) , such that @xmath49 the asip implies the clt and also the functional clt ( weak invariance principle ) , and the law of the iterated logarithm together with its functional version , as well as numerous other results . see @xcite for a comprehensive list . in section [ sec : geometr - aspects - lore ] , we recall basic properties of geometric lorenz attractors . in section [ sec : tempor - distort - funct ] , we define the temporal distortion function and prove a result about the dimension of its range . this is the main new ingredient in the proof of theorem [ thm : rapid ] in section [ sec : fast - mixing - decay ] . in section [ sec - asipsemiflow ] , we prove a general result on the asip for time-@xmath32 maps of nonuniformly expanding semiflows . in section [ sec - sigma ] , we prove that the asip is typically nondegenerate . in section [ sec - asipflow ] , we prove theorems [ thm : clt ] and [ thm : asip ] . it is natural to extend all these results to more general singular - hyperbolic attractors ( formerly referred to as lorenz - like flows ) , that is , transitive attracting sets of three - dimensional flows having finitely many lorenz - like singularities and a volume hyperbolic structure ; see e.g. @xcite for the precise definitions . indeed , analogously to the geometric lorenz case , it is possible to reduce the dynamics of these attractors to a piecewise expanding @xmath50 one - dimensional map ; see e.g. ( * ? ? ? * chapter 6 ) or ( * ? ? ? * section 4 ) for a detailed presentation . [ conj : singhyp ] let @xmath51 denote the open set of @xmath10 vector fields having a singular - hyperbolic attractor on a given compact three - dimensional manifold . then the results stated in theorems [ thm : rapid ] , [ thm : clt ] and [ thm : asip ] are true for all @xmath52 . there exists a natural generalization of singular - hyperbolicity for higher - dimensional attractors , known as sectional - hyperbolicity ; see e.g. ( * ? ? ? * sections 5.2 & 8.2 ) and also @xcite . in this setting both the stable and the unstable manifolds of points in the attractor need not be codimension one embedded submanifolds , which makes analysis of these singular flows challenging . [ conj : highdim ] let @xmath51 denote the open set of @xmath10 vector fields having a sectional - hyperbolic attractor in a given compact finite dimensional manifold . then the results stated in theorems [ thm : rapid ] , [ thm : clt ] and [ thm : asip ] are true for all @xmath52 . throughout , @xmath53 is used to denote a constant whose value may change from line to line . we define here the open set @xmath51 of @xmath11 vector fields exhibiting strongly dissipative geometric lorenz attractors and we describe the basic structure of such attractors ; see e.g. @xcite . let @xmath12 be a strongly dissipative @xmath11 vector field on @xmath13 possessing a lorenz - like equilibrium , which we suppose without loss to be at @xmath54 . we assume that the flow @xmath1 is @xmath50 linearizable in a neighborhood of @xmath54 which , by a suitable choice of coordinates , can be assumed to contain the cube @xmath55 ^ 3 $ ] . ( it follows from ( * ? ? ? * theorem 12.1 ) that smooth linearizability holds for an open and dense set of such vector fields . ) choose coordinates @xmath56 corresponding to the eigenspaces of @xmath57 , @xmath58 , @xmath59 respectively . we define the cross - section @xmath60 we write @xmath61 where @xmath62 is the poincar first return time to @xmath63 , also referred to as the roof function . we assume that there exists a global exponentially contracting @xmath64-invariant _ stable foliation_. that is , there is a compact neighborhood @xmath65 of @xmath66 satisfying @xmath67 and a partition @xmath68 of @xmath69 consisting of @xmath10 one - dimensional disks called _ stable leaves _ ( including the `` singular leaf '' @xmath70 ) . let @xmath71 denote the stable leaf containing @xmath72 . then it is required that @xmath73 for all @xmath74 and that there exist constants @xmath26 , @xmath75 such that @xmath76 for all @xmath77 in the same leaf and all @xmath78 . moreover , we assume that @xmath79 is a @xmath50 foliation , meaning that @xmath69 can be chosen so that there is a @xmath50 change of coordinates from the interior of @xmath69 onto @xmath80 transforming stable leaves into vertical lines . shrinking @xmath69 if necessary , we can arrange that each stable leaf intersects @xmath81 $ ] in a single point . define the @xmath50 projection @xmath82 given by holonomy along the stable leaves ( so @xmath83 ) . quotienting along stable leaves , we obtain a @xmath50 one - dimensional map @xmath84 with a singularity at @xmath54 : @xmath85 . [ le : prop2.6 ] let @xmath86 . 1 . @xmath87 is hlder on @xmath88 : @xmath89 with @xmath90 ; 2 . the roof function has a logarithmic singularity at @xmath54 : @xmath91 with @xmath92 and @xmath93 . in addition , we assume that @xmath94 is _ uniformly expanding _ : there are constants @xmath95 and @xmath96 such that @xmath97 for all @xmath98 and @xmath99 . as in @xcite , we assume further that @xmath100 is _ locally eventually onto ( l.e.o . ) _ ; namely that for any open set @xmath101 , there exists @xmath102 such that @xmath103 contains @xmath104 . ( more generally , it suffices that almost every point in @xmath105 has dense preimages in @xmath105 . however , the l.e.o.property is standard in the literature and holds for the classical lorenz attractor @xcite . ) considering @xmath106}(x)$ ] we obtain a closed neighborhood of @xmath55 ^ 3 $ ] and , in what follows , we denote by @xmath107 the _ geometric lorenz attractor _ of the vector field @xmath12 . it can be shown that @xmath23 is compact , volume hyperbolic , has a dense regular orbit and has zero volume ( lebesgue measure in @xmath13 ) ; see e.g. @xcite . we recall that , in our three - dimensional setting , volume hyperbolicity means that there exists a @xmath108-invariant singular - hyperbolic splitting of the tangent bundle over @xmath23 . that is , there is a vector bundle splitting @xmath109 with @xmath110 , @xmath111 , and there are constants @xmath112 , @xmath113 , such that for all @xmath114 , @xmath30 , * _ the splitting is dominated _ : @xmath115 * _ @xmath116 is uniformly contracting _ : @xmath117 * _ the area along @xmath118 is uniformly expanded _ : @xmath119 . the existence of the stable foliation @xmath68 of any small cross - section to the flow of @xmath12 ( such as @xmath63 ) is a consequence of volume hyperbolicity for three - dimensional smooth flows ; see e.g. ( * ? ? ? * chapter 3 , section 3 ) . an important consequence of domination , uniform contraction along the stable direction @xmath116 and strong dissipativity for the attractor @xmath23 is the existence of a @xmath50 global exponentially contracting @xmath1-invariant foliation @xmath120 , defined in a neighborhood ( which we may take to be @xmath121 ) of @xmath23 . [ lem - foliate ] the strong stable foliation @xmath120 is @xmath50 for some @xmath122 . we apply ( * ? ? ? * theorem 6.2 ) adapted to our setting , since only domination and uniform contraction is used in its proof . indeed , a sufficient condition to obtain @xmath50 regularity for the strong stable foliation is that for some @xmath30 , @xmath123 for all @xmath114 . ( we note that the statement in @xcite covers only the case @xmath124 , but it is standard that their result extends to the case @xmath122 . ) for each @xmath125 we define @xmath126 , @xmath127 note that @xmath128 is a continuous family of continuous functions each of which is subadditive , that is , @xmath129 . let @xmath130 denote the set of flow - invariant ergodic probability measures on @xmath23 . we claim that for @xmath122 sufficiently small , and for each @xmath131 , the limit @xmath132 exists and is negative for @xmath133-almost every @xmath114 . it then follows from ( * ? ? ? * proposition 3.4 ) that there exists constants @xmath134 such that @xmath135 for all @xmath30 , @xmath114 . in particular , for @xmath136 sufficiently large , @xmath137 for all @xmath114 . hence condition is satisfied for such @xmath138 and @xmath136 and the result follows . it remains to verify the claim . let @xmath139 denote the dirac delta concentrated at @xmath54 and let @xmath140 . we deal with the cases @xmath141 and @xmath142 separately . each @xmath141 has a zero lyapunov exponent in the flow direction and two further lyapunov exponents @xmath143 and @xmath144 associated with the vector bundles @xmath116 and @xmath118 respectively . fix @xmath141 . for @xmath133-a.e . @xmath114 we have @xmath145 on the other hand , it follows from dissipativity that @xmath146 for all @xmath72 . hence we deduce from that @xmath147 . moreover , @xmath148 , so for @xmath122 sufficiently small ( uniformly in @xmath133 ) @xmath149 . using , and together with the definition of @xmath150 , it follows that @xmath151 for @xmath133-almost every @xmath114 . it remains to consider the dirac measure @xmath139 . by strong dissipativity , for @xmath138 sufficiently small , @xmath152 for all @xmath136 as required . by lemma [ lem - foliate ] , we may consider the cross - section @xmath153 in the place of the original cross - section @xmath63 . since the strong stable foliation is @xmath50 and the cross - section @xmath63 is foliated by stable leaves over the smooth disk @xmath154 , it follows that @xmath63 is a @xmath50 embedded surface in @xmath13 . all the properties described so far are retained , with the useful advantage that @xmath155 for all @xmath156 and ( c ) : : the first return time @xmath62 of any given point in @xmath157 ( where @xmath70 now represents the leaf of @xmath68 through the point @xmath158 ) to @xmath63 is constant on the leaves of @xmath68 , that is , @xmath159 for all @xmath160 . since the cross - section @xmath63 is a @xmath50 embedded surface in @xmath13 , the roof function @xmath62 retains the properties mentioned in lemma [ le : prop2.6](2 ) ; in particular @xmath161 is a @xmath50 function with a logarithmic singularity at @xmath70 . we keep the notation @xmath162 for the holonomy along the leaves of @xmath68 to @xmath105 and also @xmath100 for the one - dimensional @xmath50 quotient map of @xmath163 over @xmath68 . another consequence of volume hyperbolicity is that there exists a field of cones @xmath164 having width @xmath165 containing the @xmath118 subbundle over @xmath23 which admit a continuous @xmath108-invariant extension @xmath166 to a neighborhood of @xmath23 . for the geometric lorenz flow we can assume without loss that this neighborhood coincides with @xmath121 . the invariance means that @xmath167 for @xmath72 in an open neighborhood @xmath121 of @xmath23 and @xmath30 , where @xmath165 is small enough . then the cones @xmath168 on @xmath169 are also @xmath170-invariant and defined on the whole of @xmath171 . we say that a @xmath172 curve @xmath173 in @xmath63 is a _ @xmath174-curve _ if @xmath175 for all parameter values @xmath176 . the @xmath170-invariance of the field of cones @xmath177 ensures that the image by @xmath64 of every @xmath174-curve is sent into another @xmath174-curve . moreover , the tangent direction to the stable leaves @xmath178 is not contained in the @xmath179 cone and makes an angle bounded away from zero with any vector inside @xmath179 , for all @xmath156 , by the volume hyperbolicity assumption ; see @xcite . [ rmk : tucker ] tucker @xcite showed that the classical lorenz equations have a robust nontrivial attractor @xmath23 containing the equilibrium at the origin . it follows from morales _ et al . _ @xcite that @xmath23 is a singular hyperbolic attractor that ( in their words ) `` resembles a geometric lorenz attractor '' . in particular , it is immediate that all of the properties listed above are satisfied except possibly for ( i ) strong dissipativity , ( ii ) the l.e.o property , and ( iii ) smoothness ( class @xmath50 ) of the contracting foliations @xmath120 and @xmath79 for the flow and poincar map respectively . we note that property ( i ) is immediate for the classical lorenz equations and condition ( ii ) was verified in @xcite . regarding ( iii ) , it is claimed in ( * ? ? ? * section 2.4 ) that @xmath79 is a smooth foliation but no details are provided . smoothness of the contracting foliations @xmath120 and @xmath79 is not part of the definition of singular hyperbolic attractor , and hence is not discussed in @xcite . however , proofs of existence of an srb measure with good statistical properties rely heavily on the smoothness of @xmath79 . although this foliation is of codimension one , the fact that the poincar map @xmath64 is singular means that an extra argument is required ; see for example robinson @xcite and rychlik ( * ? ? * section 4 ) . in particular , our results apply to the open set of flows considered by @xcite , but these do not include the classical lorenz equations . however , as shown above in lemma [ lem - foliate ] , the properties established by @xcite combined with strong dissipativity guarantee smoothness of @xmath120 , and hence of @xmath79 , for the classical parameters and nearby parameters . the geometric lorenz attractor can be written as a suspension flow @xmath180 given by @xmath181 on the space @xmath182 where @xmath183 . indeed , we can take the conjugacy as @xmath184 ( which is smooth ) and the roof function @xmath62 has a logarithmic singularity at all points of @xmath185 , is smooth elsewhere and @xmath64 is a non - uniformly hyperbolic map with invariant stable foliation @xmath68 . we denote also by @xmath54 the point @xmath186 in what follows and since @xmath187 we also write @xmath161 for the restriction @xmath188 . in particular , since @xmath189 is a @xmath50-piecewise nonuniformly expanding map , there exists a subset @xmath190 and an inducing time @xmath191 such that @xmath192 is a @xmath50-piecewise expanding markov map with partition @xmath193 ; see e.g. ( * ? ? ? * theorem 4.3 ) . since @xmath100 is l.e.o . , for any specified point @xmath194 we can choose @xmath195 to be an arbitrarily small open interval containing @xmath72 ; see e.g. @xcite , where an inductive construction procedure is described showing that we can build a full branch markov map @xmath196 as long as @xmath195 is a neighborhood of a point with dense preimages . during the paper , we will consider various inducing schemes . all of these are full branch on an interval except for the one constructed in section [ sec - double ] which is the combination of two such inducing schemes . let @xmath197 denote the @xmath198th refinement of @xmath193 , and set @xmath199 , so that @xmath200 . the most important features of @xmath100 are the following backward contraction and bounded distortion properties . there exist constants @xmath201 such that for each @xmath78 backward contraction : : @xmath202 is a @xmath50 diffeomorphism and if @xmath203 , then @xmath204 moreover there is _ slow recurrence to the singular point _ @xmath205 bounded distortion : : if @xmath203 , then @xmath206 we note that the induced map can be obtained by the methods presented in @xcite and conditions and follow from the definition of hyperbolic times ( c.f . definition 10 in @xcite with @xmath207 ) . next we construct a piecewise uniformly hyperbolic map @xmath208 with infinitely many branches , which covers @xmath209 , as follows : define @xmath210 to be the union of the stable leaves through @xmath195 and define the poincar return map @xmath211 for @xmath212 . we let @xmath213 denote the measurable partition of @xmath214 whose elements are @xmath215 with @xmath216 . also , we extend @xmath217 to a function on @xmath214 by setting @xmath218 . let @xmath219 be the unique @xmath220-invariant absolutely continuous probability measure on @xmath195 . it is well - known that @xmath221 . it is then standard that there exist unique invariant measures @xmath222 for @xmath208 , @xmath223 for @xmath224 and @xmath225 for @xmath226 satisfying @xmath227 , @xmath228 and also @xmath229 and @xmath230 . we have @xmath231 and @xmath232 , hence @xmath233 ; see e.g. ( * ? ? ? * section 3 ) for more details . here we obtain an almost everywhere defined local product structure for the induced hyperbolic map @xmath208 . we have already seen that the poincar map @xmath224 has a stable foliation @xmath68 with leaves that cross @xmath63 , and hence the induced map @xmath208 has stable manifolds @xmath234 that cross @xmath214 . in the next proposition , we construct local unstable manifolds for @xmath118 of uniform size , defined almost everywhere . [ pr : wu_crosses ] for @xmath222-almost every @xmath212 , there exists a local unstable manifold @xmath235 that crosses @xmath214 . in particular , @xmath236 for @xmath222-almost every @xmath212 . we begin with the local unstable manifolds @xmath237 for the flow @xmath1 . it follows from pesin theory ( see e.g. @xcite ) that almost every point @xmath14 of @xmath23 with respect to the srb measure @xmath0 admits a local unstable manifold @xmath238 which is a @xmath50-curve containing @xmath14 in its interior . by definition , @xmath239 if and only if @xmath240 for all @xmath241 ( recall that @xmath113 and also that the constant @xmath242 depends on the leaf ) ; see e.g. @xcite . since @xmath23 is an attractor , unstable leaves are contained in @xmath23 . the smooth conjugacy @xmath243 sends these leaves into unstable leaves for the suspension flow @xmath244 , which can be written locally as @xmath245\}$ ] , where @xmath246\to x$ ] and @xmath247\to{{\mathbb r}}$ ] are @xmath50 diffeomorphisms into their images . by its definition , the curve @xmath173 is the local unstable manifold @xmath248 through @xmath249 with respect to @xmath64 , that is , @xmath250 if and only if @xmath251 , for all @xmath252 . we observe that the inverse images of @xmath72 and @xmath253 are all well defined since these points belong to the attractor @xmath254 which is @xmath244-invariant . moreover , the curve @xmath173 is a graph @xmath255 and @xmath256 is contained in a cone @xmath257 for some @xmath258 by the domination condition on @xmath64 , consequence of the existence of dominating splitting for the flow on the attractor . we remark that @xmath248 is a @xmath174-curve which coincides with @xmath259 if @xmath72 also belongs to @xmath260 , and so the statements above hold for @xmath222 almost every point @xmath212 . indeed , on the one hand , @xmath248 is formed by points whose preorbit is asymptotic to the preorbit of @xmath72 , hence these preorbits contain the preorbits with respect to @xmath118 , and so @xmath261 . on the other hand , this inclusion shows that @xmath248 and @xmath259 coincide in a neighborhood of @xmath72 inside @xmath259 . since these unstable manifolds are contained in the attractor then , repeating the argument around each point @xmath262 , we see that the two manifolds coincide . in addition , the stable leaves through points @xmath263 are transverse to @xmath264 and the angle between @xmath265 and @xmath266 is bounded away from zero . hence @xmath267 is a neighborhood of @xmath268 in @xmath195 for @xmath222-a.e . @xmath269 . let @xmath270 denote the element of the @xmath198th refinement @xmath271 that contains @xmath272 . this is well - defined for all @xmath78 for @xmath219-a.e . moreover , @xmath274 is a neighborhood of @xmath272 for all @xmath78 for @xmath222-a.e . @xmath269 . since @xmath220 is full - branch , @xmath275 for all @xmath198 . by the poincar recurrence theorem , we may assume without loss that @xmath269 is recurrent : there exists @xmath276 such that @xmath277 . therefore , for all large enough @xmath278 we have @xmath279 and hence the iterate of a connected piece of the unstable manifold of @xmath269 defined by @xmath280 is a @xmath174-curve that crosses @xmath214 . the sequence @xmath281 has a convergent subsequence to @xmath282 by the arzel - ascoli theorem and by the recurrence assumption on @xmath269 we have @xmath283 . we claim that @xmath284 , which completes the proof that @xmath222-almost every point has an unstable manifold crossing @xmath214 . the last statement of the proposition is a simple restatement of this conclusion . to prove the claim , we consider @xmath285 and sequences @xmath286 such that @xmath287 . fix @xmath288 and choose @xmath289 so that @xmath290 for all @xmath291 . by the definition of @xmath281 and since @xmath292 , we have uniform backwards contraction . thus @xmath293 where @xmath294 is such that @xmath295 . hence @xmath296 . to obtain @xmath297 we have used that all the iterates @xmath298 of @xmath299 are @xmath174-curves and so their length is comparable to the length of their projection @xmath300 on @xmath105 ; and then take advantage of the backward contraction property ( [ eq : backcontraction ] ) associated to the partition @xmath193 with the same contraction rate @xmath301 . finally , since these constants are independent of @xmath278 , letting @xmath302 gives @xmath303 for each given fixed @xmath288 . this completes the proof of the claim and finishes the proof of the proposition . using this geometric structure we can also prove the following : [ prop - localprod ] the induced map @xmath208 has a _ local product structure _ : for any partition element @xmath304 there exists a measurable map @xmath305:y\times a\to a$ ] defined for all @xmath306 and @xmath222 almost every @xmath212 such that @xmath307\in w_f^u(y)\pitchfork w_f^s(y')\ ] ] consists of a unique point . in addition , the map @xmath305 $ ] is constant along unstable manifolds in the first coordinate , and constant along stable manifolds in the second coordinate . furthermore , @xmath305 $ ] is @xmath50 in the second coordinate . from proposition [ pr : wu_crosses ] we have that for @xmath222 almost every point @xmath269 the local unstable manifold @xmath264 crosses @xmath214 . from the definition of geometric lorenz attractor , @xmath308 crosses @xmath309 transversely to @xmath264 , for every @xmath310 . hence @xmath311 $ ] is well defined for @xmath222-almost every @xmath212 and every @xmath310 . since @xmath309 is a union of local stable manifolds , it is immediate that if @xmath312 then @xmath311\in a$ ] . we note that if @xmath311 $ ] is defined , then @xmath313=[y , y ' ] { \quad\text{and}\quad}w\in w^s_f(y')\mapsto [ w , s]=[y , y ' ] \end{aligned}\ ] ] which shows that @xmath311 $ ] is constant along unstable manifolds on the first coordinate and stable manifolds on the second coordinate . in addition , the stable manifolds @xmath308 depend continuously in the @xmath50 topology on the base point @xmath314 ( by the partial hyperbolicity of the attractor ) and the unstable manifolds @xmath264 depend measurably on @xmath269 ( by nonuniform hyperbolicity ) . hence @xmath305 $ ] is a measurable map and is @xmath50 along the second coordinate . [ prop - periodic ] suppose that @xmath315 are such that @xmath311 $ ] is well - defined . then , there is a sequence of periodic points @xmath316 for @xmath118 such that ( i ) @xmath317 , ( ii ) @xmath318 $ ] is well - defined for all @xmath278 , and ( iii ) @xmath318\to [ y , y']$ ] . we use proposition [ pr : wu_crosses ] : we can assume without loss that @xmath269 is recurrent . let us fix a neighborhood @xmath121 of @xmath269 given by @xmath319 , where @xmath320 is an open subinterval of @xmath55\setminus\{0\}$ ] and @xmath321 is an open subset of @xmath55 $ ] . we fix a similar smaller neighborhood @xmath322 such that closure of @xmath323 is contained in @xmath324 . we can regard @xmath325 as a neighborhood of @xmath326 . in our setting , this ensures the existence of a sequence @xmath276 such that @xmath277 , @xmath327 is a neighborhood of @xmath326 and @xmath328 ; see the proof of proposition [ pr : wu_crosses ] . the density of periodic points for @xmath118.,width=283 ] hence , there exists a stable leaf @xmath329 for some @xmath330 which is sent inside itself by @xmath331 , by the uniform contraction of the stable leaves of @xmath332 . since we can assume without loss that @xmath333 , then taking @xmath334 big enough , we claim that @xmath335 . indeed , due to the domination assumption on @xmath64 , the unstable manifold @xmath336 crosses @xmath214 and its angle with respect to the horizontal direction is uniformly bounded from above , so @xmath337 , and the claim follows by uniform contraction of the stable leaves ; see figure [ fig : perdense ] . thus we have a fixed point @xmath338 of @xmath331 in @xmath339 . since @xmath121 belongs to a fundamental system of neighborhoods of @xmath269 , this proves item ( i ) in the statement of the proposition . all periodic points of @xmath64 are hyperbolic of saddle type , thus @xmath340 is well - defined and @xmath341 is a neighborhood of @xmath342 . if @xmath343 , then clearly @xmath344 crosses @xmath309 and so @xmath318 $ ] is well - defined . we claim that this is always the case . for otherwise , if @xmath345 , then since @xmath346 is an expanding diffeomorphism and @xmath340 is @xmath331-invariant , the length of @xmath347 grows while this set is contained in @xmath348 . thus @xmath341 covers @xmath327 , as claimed . finally , for the continuity statement ( iii ) , since @xmath317 as @xmath302 with @xmath338 periodic points of @xmath118 , it is enough to show that @xmath349 as smooth curves that cross @xmath309 . since each curve @xmath350 is a @xmath174-curve , then there exists a accumulation point @xmath173 which is also a @xmath174-curve ( by the arzel - ascoli theorem ) . we show that @xmath173 contains @xmath351 . indeed , @xmath173 contains @xmath269 . by uniform backward contraction , if @xmath352 converge to @xmath353 , then we can argue similarly to ( [ eq : cont - u - curves ] ) since there are @xmath354 so that @xmath355 and @xmath356 . hence , for a given fixed @xmath357 we get @xmath358 and for limit points @xmath359 of @xmath360 and @xmath361 letting @xmath302 , we obtain ( since @xmath118 is smooth in @xmath309 ) @xmath362 hence , because @xmath27 was arbitrary , we see that @xmath363 , as needed . the proof is complete . we define the _ induced roof function _ @xmath365 , @xmath366 . since @xmath161 , and hence @xmath364 , is constant along stable leaves , we also denote by @xmath364 the quotient induced roof function @xmath367 . it follows in a completely analogous way to ( * ? ? ? * section 4.2.2 ) that @xmath367 satisfies @xmath368 where @xmath369 is the set of all inverse branches of @xmath370 . indeed , let @xmath371 , @xmath372 be an inverse branch of @xmath220 with inducing time @xmath373 and let us fix @xmath374 . then @xmath375 recall from the construction of the inducing partition using hyperbolic times that conditions and are satisfied ( here we require these conditions only with @xmath376 ) . by , @xmath377 moreover , by , @xmath378 and so , by lemma [ le : prop2.6](2 ) , @xmath379 for some @xmath380 depending only on @xmath64 . altogether this implies that @xmath381 establishing . [ prop - rdecay ] there exists @xmath112 such that @xmath382 . by * section 4.2.1 ) , @xmath383 . the result follows since @xmath384 is bounded , see for example ( * ? ? ? * proposition 4.5 ) . ( we caution that our @xmath385 is denoted by @xmath386 in @xcite ) . [ rmk - rdecay ] proposition [ prop - rdecay ] is not necessary for the results in this paper but simplifies the exposition , see remark [ rmk - simplify ] . for @xmath387 define the _ separation time _ @xmath388 to be the least integer @xmath389 such that @xmath390 are in distinct partition elements of @xmath193 . for any given @xmath391 we define the _ symbolic metric _ @xmath392 on @xmath195 . let @xmath393 denote the lipschitz constant of the quotient induced roof function @xmath394 with respect to @xmath395 . we write @xmath396 for the sum @xmath397 in what follows . [ le : lip - symbolic ] there exists @xmath26 such that for all @xmath387 with @xmath398 and @xmath399 we have @xmath400 as a consequence , there exists @xmath391 such that @xmath401 and also @xmath402 . let us consider @xmath387 such that @xmath403 . then @xmath203 and so @xmath404 . thus , from the choice of the cross - section , ensuring that @xmath161 is constant on stable leafs , together with lemma [ le : prop2.6](2 ) , we can write @xmath405 \\ & = \sum_{\ell=0}^{\tau(y)-1 } \big [ -\lambda_u^{-1}(\log|\bar f^\ell(y)|-\log|\bar f^\ell(y')| ) + h(\bar f^\ell(y))-h(\bar f^\ell(y ' ) ) \big ] . \end{aligned}\ ] ] combining ( [ eq : backcontraction ] ) together with ( [ eq : slow - recurrence ] ) we obtain @xmath406 \\ & \le c \sum_{\ell=0}^{\tau(y)-1 } \left [ c_0\frac{\lambda^{\tau(y)-\ell}}{\sqrt{\lambda}^{\tau(y)-\ell } } |\bar f(y)-\bar f(y')| + \|h\|_\epsilon \lambda^{\epsilon(\tau(y)-\ell ) } |\bar f(y)-\bar f(y')|^\epsilon \right ] \\ & \le \kappa\lambda^{\epsilon/2}\cdot \frac{1-\lambda^{\frac{\epsilon}2\cdot\tau(y)}}{1-\lambda^{\epsilon/2 } } |\bar f(y)-\bar f(y')|^\epsilon,\end{aligned}\ ] ] for some constant @xmath407 depending on the flow only . this implies in particular the first statement of the lemma . finally , because @xmath408 is at least @xmath32 , we also have @xmath409 which , combined with the previous inequality , gives another constant @xmath410 depending only on the flow satisfying @xmath411 we can find @xmath412 and a constant @xmath413 so that @xmath414 for all @xmath78 and , since @xmath415 is bounded above , the proof is complete . we are now ready to prove a useful consequence of backward contraction for the quotient map and expansion of the flow in the linearizable region . we keep the choice of @xmath416 from lemma [ le : lip - symbolic ] . also , we define @xmath417 for points @xmath315 by setting @xmath418 . [ le : expansion - flow ] there exist constants @xmath419 such that for all @xmath315 satisfying @xmath420 and all @xmath421 , we have @xmath422 taking @xmath423 , we have @xmath424 . there exist @xmath425 such that @xmath426 \quad\text{and}\quad u\in[r_{k'-1}(y'),r_{k'}(y ' ) ] . \end{aligned}\ ] ] from lemma [ le : lip - symbolic ] we know that @xmath427 . hence for @xmath417 small enough , @xmath428 for all @xmath429 . it follows that @xmath430 , and we may suppose that @xmath431 . hence , there are two cases to consider . estimating the distance between @xmath432 and @xmath433.,width=321 ] [ [ the - case - kk1 ] ] * the case @xmath434 : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we have @xmath435 and so the orbit of @xmath269 has already had @xmath436 returns to the cross - section @xmath63 , while the orbit of @xmath314 has only returned @xmath437 times . we estimate the distance between the points with the distance between the @xmath438 returns of both orbits to @xmath63 as follows . setting @xmath439 and @xmath440 , we get @xmath441 where we have used the uniform contraction of the stable foliation of the attractor , together with lemma [ le : lip - symbolic ] and the uniform backward contraction of iterates of @xmath100 ; see figure [ fig : case_kprime ] . again from lemma [ le : lip - symbolic ] and the choice @xmath412 we obtain @xmath442 where @xmath443 is the length of the largest stable leaf in the cross - section @xmath63 . [ [ the - case - kk ] ] * the case @xmath444 : * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + now both points are past their @xmath445th return and we again estimate the distance comparing with the distance of their @xmath445th returns . setting @xmath446 and @xmath447 and assuming without loss that @xmath448 we get @xmath449 and @xmath450 as before , while @xmath451 the uniform contraction along stable leaves and the relation @xmath452 allows us to write @xmath453 and since @xmath454 , we are left to prove that @xmath455 for this we use the construction of the geometric lorenz attractor with the linearizable lorenz - like singularity to explicitly calculate trajectories . in this way , we easily see that the distance between the pair of stable leaves @xmath456 and @xmath457 ( on @xmath63 ) is expanded by @xmath458 , that is , @xmath459 as long as @xmath460 and @xmath461 remain in the linearizable region around the singularity , for @xmath462 . the flight time from @xmath63 to @xmath463 ( the boundary of the linearizable region ) is given by @xmath464 and @xmath465 , and their difference is uniformly bounded since , by the backward contraction ( [ eq : backcontraction ] ) and slow recurrence ( [ eq : slow - recurrence ] ) properties , @xmath466 hence , we have expansion of the distance between @xmath467 in the linear region , and the flow from @xmath463 back to @xmath63 is performed in a uniformly bounded time for all points of the attractor . thus , this last non - linear action of the flow distorts the distance by at most some constant factor ( a bound on the norm of the derivative of the flow on a bounded interval of time ) . therefore , we have shown that the left hand side of ( [ eq : expansionflow ] ) is bounded by a constant factor of @xmath468 . this last difference is bounded by @xmath469 which is bounded by the expression on the right hand side of ( [ eq : expansionflow ] ) . the proof is complete . in subsection [ sec : lorenz - attractors - as ] , we saw that the geometric lorenz flow can be modelled as a suspension flow @xmath180 where @xmath63 is the poincar section and @xmath161 is the first hit time . shrinking the cross - section to @xmath214 and using the induced roof function @xmath364 ( which need not be the first hit time ) , we have the alternative model of the geometric lorenz flow as the suspension flow @xmath470 over a uniformly hyperbolic map @xmath208 with integrable but unbounded return time function @xmath385 . again the suspension flow is given by @xmath471 computed modulo identifications , and the probability measure @xmath472 is @xmath244 invariant . similarly , we define the quotient suspension semiflow @xmath473 with invariant probability measure @xmath474 . the next result enables us to pass from the ambient manifold @xmath13 to @xmath475 by means of the projection @xmath476 , @xmath477 . [ prop - p ] let @xmath478 . then @xmath479 without loss , we can suppose that @xmath480 . by the mean value theorem , there is a @xmath481 between @xmath174 and @xmath482 such that latexmath:[\ ] ] using and , @xmath935 , and @xmath936 also @xmath937 and @xmath938 . hence @xmath939 as required . [ cor - cm3 ] there exists a constant @xmath53 ( depending on @xmath357 , @xmath416 and @xmath523 ) such that @xmath940 for all mean zero @xmath776 . on one hand , by proposition [ prop - dual ] and lemma [ lem - power ] , @xmath941 on the other hand , @xmath942 using the first estimate , @xmath943 using the second estimate , @xmath944 combining these gives the required estimate . now we can complete the proof of theorem [ thm - asipsemiflow ] . by proposition [ prop - dual ] , @xmath945 . hence conditions and are satisfied for @xmath946 . by corollary [ cor - cm3 ] , condition is satisfied for @xmath877 . finally , as promised in remark [ rmk - iii ] , we show how condition ( iii ) can be relaxed . indeed it suffices that * @xmath947 , where @xmath948 for @xmath949 , @xmath950 . we begin by recalling a standard estimate for gibbs - markov maps which we did not explicitly make use of earlier : there is a constant @xmath26 such that @xmath951 for all @xmath952 , @xmath953 . note that condition ( iii ) was only used in the estimate of term @xmath954 in the proof of lemma [ lem - power ] . but alternatively , we compute that @xmath955 where @xmath956 now @xmath957 hence @xmath958 and @xmath959 hence , assuming condition ( iii@xmath960 ) instead of ( iii ) , we obtain an estimate analogous to the one in lemma [ lem - power ] with the conclusion replaced by the estimate @xmath961 for all @xmath914 . in this section , we prove that the degenerate case @xmath39 in theorem [ thm - asipsemiflow ] is of infinite codimension . suppose as in theorem [ thm - asipsemiflow ] that @xmath882 for some @xmath883 . following ( * ? ? ? * proposition 2 ) , we define @xmath962 and @xmath963 . then @xmath964 is a continuous cocycle for the semiflow @xmath783 ; that is @xmath965 for all @xmath966 . moreover , @xmath967 and @xmath968 , so @xmath964 is an @xmath969 coboundary . in the axiom a setting of @xcite it now follows from a livic regularity theorem of @xcite that @xmath970 has a hlder version . hence if @xmath601 is a periodic point of period @xmath971 for @xmath783 , then @xmath972 . since @xmath783 has infinitely many periodic orbits , this places infinitely many restrictions on @xmath42 . in the nonuniformly expanding case , the situation is similar once we have a livic regularity theorem for nonuniformly expanding semiflows . as we now show , this is a straightforward combination of results of @xcite for gibbs - markov maps and @xcite for uniformly expanding semiflows . in the remainder of this section , we suppose as in section [ sec - asipsemiflow ] that @xmath196 is a full - branch gibbs - markov map and that @xmath367 is a roof function satisfying conditions ( i ) , ( ii ) and ( iii@xmath960 ) . ( the full - branch condition is relaxed in remark [ rmk : livsic ] . ) given a continuous cocycle @xmath964 on @xmath735 , we define @xmath973 @xmath974 [ lem : livsic ] let @xmath975 be a cocycle for @xmath783 such that @xmath976 satisfies @xmath977 suppose that there exists @xmath978 measurable such that for all @xmath914 : @xmath968 a.e . then @xmath970 has a version that is continuous . we begin by following the proof of ( * ? ? ? * theorem 3.3 ) . note that the set of zero measure where @xmath968 fails for each @xmath914 can be made independent of @xmath136 ; see e.g. @xcite . hence for almost every @xmath979 , @xmath980 then @xmath981 , @xmath982-a.e . and so by fubini s theorem there exists @xmath983 such that @xmath984 for @xmath219-a.e @xmath985 . since @xmath975 is a cocycle , @xmath986 substituting in ( [ eq : cobd ] ) with @xmath987 and using @xmath988 we obtain @xmath989 where @xmath990 is measurable . we have shown that @xmath991 satisfies the hypotheses of ( * ? ? ? * theorem 1.1 ) . it follows that @xmath694 has a version that is continuous ( even lipschitz ) on @xmath195 . let us now define @xmath992 as in the proof of ( * ? ? ? * theorem 3.3 ) , it follows from the definitions that @xmath993 is a well - defined function on @xmath735 ( ie @xmath994 ) and that @xmath993 is a version of @xmath970 . since @xmath995 and @xmath694 are continuous , it follows that @xmath993 is continuous as required . [ cor : livsic ] suppose that @xmath996 satisfies @xmath882 for some @xmath883 . then @xmath997 for all periodic points @xmath998 of period @xmath971 . define @xmath962 and @xmath963 , so @xmath968 . we claim that @xmath999 satisfies the assumption in lemma [ lem : livsic ] . then @xmath970 has a continuous version . hence it follows as in the axiom a setting that @xmath997 for all periodic points @xmath998 of period @xmath971 . it remains to verify the claim . for @xmath1000 , we compute that @xmath1001 . for @xmath304 , @xmath1002 , and taking @xmath1003 , @xmath1004 hence @xmath1005 and so @xmath1006 as required . [ rmk : livsic ] the condition that @xmath871 is full - branch can be relaxed as in @xcite . the function @xmath949 constructed in the proof of lemma [ lem : livsic ] will no longer be continuous in general , but it is continuous on each partition element of the partition @xmath1007 generated by the images @xmath1008 of the elements of @xmath213 . we conclude that @xmath970 has a version that is continuous on @xmath1009 for each @xmath1010 . hence in corollary [ cor : livsic ] we obtain that @xmath997 for periodic points @xmath998 of period @xmath971 such that the orbit of @xmath601 intersects one of the partition elements @xmath1011 . in this section we prove theorem [ thm : asip ] ( and as a consequence theorem [ thm : clt ] ) , by reducing from the geometric lorenz flow to the quotient flow , enabling the application of theorem [ thm - asipsemiflow ] . to achieve this reduction , we modify the argument in ( * ? ? ? * appendix a ) which deals with the axiom a case and bounded roof function . we note that the argument in @xcite is unnecessarily complicated , since having reduced without loss to the situation where @xmath161 depends only on future coordinates , the quantity @xmath571 in ( * ? ? ? * proposition 5 ) is identically zero . on the other hand , the situation for geometric lorenz attractors is made complicated since ( a ) there is no convenient metric on the symbolic flow @xmath475 , and ( b ) the roof function is unbounded . to deal with ( a ) , we reduce directly to @xmath735 . for ( b ) , we make crucial use of lemma [ le : lip - symbolic ] and proposition [ prop - p ] . we recall that @xmath1019 denotes the measure - preserving semiconjugacy between the flows @xmath470 and @xmath1020 . recall also that the projection @xmath162 along stable manifolds restricts to a projection @xmath1021 . also , we defined @xmath1022 provided that @xmath1023 . this induces a measure - preserving semiconjugacy @xmath1024 between the flow @xmath470 and the semiflow @xmath875 . let @xmath44 be a @xmath1025 mean zero observable . by theorem [ thm - sinai ] , @xmath1026 uniformly on @xmath475 , where @xmath1027 . hence the asip for @xmath42 is equivalent to the asip for @xmath834 and follows from theorem [ thm - asipsemiflow ] . it remains to verify the statement about the degenerate case @xmath39 in theorem [ thm : clt ] . by theorem [ thm - asipsemiflow ] , @xmath1028 for some @xmath1029 . working still on @xmath735 , define @xmath1030 , @xmath1031 , so @xmath968 . by lemma [ lem : livsic ] , @xmath970 has a continuous version . now suppose that @xmath601 is a periodic point of period @xmath1036 for the geometric lorenz flow @xmath1 . then @xmath1037 for some @xmath1038 , @xmath1039 $ ] . since @xmath364 is not necessarily the first return time to @xmath214 it need not be the case that @xmath1040 has period @xmath1036 under @xmath244 . however , certainly there exists @xmath41 , an integer multiple of @xmath1036 , such that @xmath1041 . by , @xmath1042 as required . recall that the projection @xmath476 is given by @xmath1043 . note that if @xmath860 then @xmath1044 . however , we caution that for general @xmath30 , @xmath1045 it is _ not _ the case that @xmath1046 and @xmath1047 coincide . when proving ( iii ) , we note that the formula for @xmath1056 is identical to that for @xmath1015 with @xmath42 replaced by @xmath1057 throughout . hence it suffices to consider the case @xmath1058 and to prove that @xmath1059 for @xmath1060 . we again use exponential contraction along stable directions as in to show that @xmath1071 be the lap number for @xmath212 under @xmath1 , so @xmath1072 and @xmath1073 . then the @xmath198th term of @xmath1074 has the form @xmath1075 where @xmath1076 note that @xmath1077 . let @xmath1078 + 2 $ ] . suppose that @xmath1079 . choose @xmath69 so large that @xmath1080 . then @xmath1081 for all @xmath1082 $ ] . hence for this range of @xmath703 , the intervals @xmath1083 $ ] and @xmath1084 $ ] almost coincide ( the initial points are within distance @xmath1085 , as are the final points ) . it follows that the lap numbers @xmath703 and @xmath1086 in satisfy @xmath1087 for all @xmath1088 . the estimation of the terms in @xmath1089 now splits into three cases . when @xmath1090 , we obtain the term @xmath1091 hence by proposition [ prop - p ] , @xmath1092 if @xmath1093 , then @xmath1094 so that @xmath1095 yielding the same estimate as in . similarly for the case @xmath1096 . hence in all three cases , we obtain the estimate . summing over @xmath198 , we obtain that latexmath:[\[\begin{aligned } \label{eq - a1 } to deal with the @xmath198th term @xmath1098 we need to introduce four lap numbers . first let @xmath1099 be the lap number corresponding to @xmath1100 , so @xmath1101 then let @xmath1102 where @xmath1103 is the lap number corresponding to @xmath1104 under @xmath1105 . altogether , we obtain @xmath1106 similarly , we write @xmath1107 again , we consider the three cases @xmath1090 , @xmath1108 , @xmath1109 separately . for example , if @xmath1093 , then @xmath1110 and the calculation proceeds as for . hence we obtain @xmath1111 . hence holds with @xmath1112 completing the proof . v. arajo and m. j. pacifico . , volume 53 of _ ergebnisse der mathematik und ihrer grenzgebiete . 3 . folge . a series of modern surveys in mathematics [ results in mathematics and related areas . 3rd series . a series of modern surveys in mathematics]_. springer , heidelberg , 2010 . v. arajo , e. r. pujals , m. j. pacifico , and m. viana . singular - hyperbolic attractors are chaotic . , 361:24312485 , 2009 . p. hartman . , volume 38 of _ classics in applied mathematics_. society for industrial and applied mathematics ( siam ) , philadelphia , pa , 2002 . corrected reprint of the second ( 1982 ) edition [ birkhuser , boston , ma .

barchan dunes crescent - shaped dunes that form in areas of unidirectional winds and low sand availability commonly display an asymmetric shape , with one limb extended downwind . several factors have been identified as potential causes for barchan dune asymmetry on earth and mars : asymmetric bimodal wind regime , topography , influx asymmetry and dune collision . however , the dynamics and potential range of barchan morphologies emerging under each specific scenario that leads to dune asymmetry are far from being understood . in the present work , we use dune modeling in order to investigate the formation and evolution of asymmetric barchans . we find that a bimodal wind regime causes limb extension when the divergence angle between primary and secondary winds is larger than , whereas the extended limb evolves into a seif dune if the ratio between secondary and primary transport rates is larger than . calculations of dune formation on an inclined surface under constant wind direction also lead to barchan asymmetry , however no seif dune is obtained from surface tilting alone . asymmetric barchans migrating along a tilted surface move laterally , with transverse migration velocity proportional to the slope of the terrain . limb elongation induced by topography can occur when a barchan crosses a topographic rise . furthermore , transient asymmetric barchan shapes with extended limb also emerge during collisions between dunes or due to an asymmetric influx . our findings can be useful for making quantitative inference on local wind regimes or spatial heterogeneities in transport conditions of planetary dune fields hosting asymmetric barchans . barchan dunes , dune asymmetry , wind erosion , sand transport , dune model

1304.6573

the classical symmetric shape of barchan dunes is far from being prevalent in nature . a wide variety of barchan morphologies on earth and mars are asymmetric , with one extended limb ( cf . [ fig : sketch_barchan_asymmetry ] ) @xcite . since the pioneering works by @xcite , various conceptual models have been proposed in order to explain the existing asymmetric dune morphologies . dune asymmetry has been most predominantly attributed to asymmetric bimodal winds @xcite , topography @xcite , dune collisions @xcite or asymmetric sand influx @xcite . indeed , understanding dune asymmetry constitutes an important issue in planetary science , as asymmetric dunes could potentially serve as a proxy for local wind regimes or variations in sand supply or topography . however , the significance of the different causes of barchan asymmetry is still poorly understood , whereas the potential range of dune morphologies resulting in each case remains unknown . although some insights into the dynamics of asymmetric dunes could be gained from field monitoring within a time span of a few years @xcite , assessment of dune shape evolution is a difficult task owing to the long timescales involved in dune processes . moreoever , diverse asymmetric dune morphologies @xcite might be the outcome of different factors in concurrent action , which poses a further challenge in the investigation of the developmental stages of asymmetric dunes through field observations . numerical modeling which has become indispensable in the investigation of aeolian and dune processes on earth and other planetary bodies @xcite could provide a helpful tool in the study of barchan dune asymmetry . in order to shed light on the factors competing for the appearance of various asymmetric dune shapes , a minimal model that accounts for a mathematical description of the wind over dunes , as well as for the equations for grain transport and landscape evolution is required . such a model has been developed in the course of the last decade @xcite . this model combines an analytical model for the average turbulent flow over the sand landscape with a continuum model for sand transport . the model has proven to reproduce the shape of different types of dunes with good quantitative agreement with measurements @xcite . the model has been further extended in order to model longitudinal seif dunes and diverse unusual martian dune forms under bimodal wind regimes @xcite . in the present work , we use the dune model to investigate the role of different asymmetry causes for the diversity of asymmetric barchan shapes reported previously from observations of dune fields on earth and mars @xcite . we calculate the evolution of a symmetric barchan dune under the separate action of the potential causes for dune asymmetry . this paper is organized as follows . in section [ sec : model ] the dune model is described . in sections [ sec : bimodal_winds]-[sec : dune_collision ] we present the results obtained from calculations of barchan dune asymmetry due to bimodal wind regimes , topography , asymmetric sand supply and dune collision , respectively . moreover , in section [ sec : discussion ] we compare the results of our simulations with the shapes of real asymmetric barchans and discuss on the application of the model outcomes to the research of extraterrestrial dunes . finally , conclusions are presented in section [ sec : conclusions ] . the model used in the calculations of the present work consists of a set of mathematical equations that compute : the average turbulent wind field ( the surface shear stress ( @xmath2 ) over the topography ; the mass flux ( @xmath3 ) of saltating particles due to the shear stress , and the time evolution of the surface resulting from particle transport @xcite . in this section , we present a brief description of the model equations and of the calculation procedure . the first step of the model calculations consists of describing quantitatively the average three - dimensional turbulent wind field over the dune . the average shear stress field ( @xmath4 ) is calculated through solving a set of analytical equations developed by @xcite . in the absence of dunes , the wind velocity @xmath5 within the atmospheric boundary layer increases logarithmically with the height ( @xmath6 ) above the flat ground . that is , @xmath7\,{\mbox{ln}}(z / z_0),\ ] ] where @xmath8 is the von krmn constant and @xmath9 is the wind shear velocity , which is used to define the ( undisturbed ) shear stress @xmath10 , with @xmath11 standing for the air density @xcite . furthermore , @xmath12 is the surface roughness , which scales with the average grain size ( @xmath13 ) composing the sand bed . we take @xmath14 based on a recent theoretical work on modeling saturated sand flux @xcite . a smooth hill or dune introduces a perturbation in the wind field . the fourier - transformed longitudinal and transverse components of the shear stress perturbation @xmath15 ) due to the local topography are computed using following equations , @xmath16 @xmath17 where @xmath18 and @xmath19 are parallel , respectively , perpendicular to the wind direction , @xmath20 , @xmath21 and @xmath22 are the components of the wave vector @xmath23 , i.e. the coordinates in fourier space , @xmath24 and @xmath25 are modified bessel functions and @xmath26 is the fourier transform of the height profile ; @xmath27 is the vertical velocity profile which is suitably non - dimensionalized , @xmath28 is the depth of the inner layer of the flow and @xmath29 is a typical length scale of the hill or dune and is given by @xmath30 the mean wavelength of the fourier representation of the height profile . in the presence of the saltating grains , the roughness of the dune s surface increases to an apparent value @xmath31 , the aerodynamic roughness @xcite . the wind velocity over the dune is calculated using @xmath32 mm , a value based on experimental observations @xcite . the surface shear stress field is obtained , then , with the equation , @xmath33 where @xmath34 is the undisturbed shear stress over the flat ground . flow separation at the dune brink due to the strong local curvature of the surface gives rise to a zone of recirculating flow @xcite , which can not be described by the analytical model @xcite . this problem is overcome introducing the so - called `` separation bubble '' : for each longitudinal slice of the dune , a separation streamline connecting the brink to the reattachment point is introduced at the lee . each separation streamline is fitted to a third - order polynomial , the parameters of which are determined as described in detail by @xcite . the wind model @xcite is then solved for the smooth `` envelope '' that comprises the separation bubble and the dune surface @xcite . thereafter , the shear stress in the recirculating zone within the separation bubble is set as zero , since the net transport within the bubble essentially vanishes . next , the mass flux of particles in saltation which is the dominant transport mode and consists of grains hoping in ballistic trajectories and ejecting new particles upon collision with the ground is computed . the saltation cloud is regarded as a thin , fluid - like layer that can exchange sand with the immobile sand bed @xcite . when the wind shear stress exceeds a minimal threshold and saltation begins , the sand flux first grows exponentially due to the multiplicative process inherent to the saltation process . however , since saltating grains accelerate at cost of aeolian momentum , the flux can not increase beyond a maximal value , the so - called saturated sand flux @xcite which is reached after a saturation transient where the air shear stress within the saltation cloud equals the minimal value for sustained saltation , i.e. the impact threshold , @xmath35 @xcite , defined in terms of the impact threshold shear velocity @xmath36 . in the continuum model , the derivation of the three - dimensional equations for the sand flux is performed by explicitly accounting for the decrease in aeolian shear stress due to the growth of the number of particles in saltation ( the `` feedback effect '' @xcite ) , which leads to the saturation transient of the flux as described above . the following equation is obtained for the sand flux ( @xmath3 ) over the terrain , @xmath37 where @xmath38({{\tau } - { \tau}_{\mathrm{t}}})$ ] is the saturated flux ; @xmath39{\tau}_{\mathrm{t}}{({{\tau } - { \tau}_{\mathrm{t}}})}^{-1}$ ] is the characteristic length of flux saturation ; @xmath40 is gravity ; @xmath41 and @xmath42 are empirically determined model parameters @xcite , and the steady - state velocity of the particles in saltation ( @xmath43 ) is calculated numerically from the balance between aeolian drag , gravitational and bed friction forces on the particles @xcite . the outcome of the calculation is a two - dimensional field ( @xmath44 ) ) that gives the height - integrated , average mass flux of saltating particles over the terrain . finally , mass conservation is used in order to compute the evolution of the local height ( @xmath45 ) through the equation , @xmath46 where @xmath47 is the bulk sand density . ( [ eq : mass_conservation ] ) implies that the deposition ( erosion ) occurs at those places where the flux locally decreases ( increases ) downwind . _ avalanches _ wherever the local slope exceeds the angle of repose of the sand ( @xmath48 ) , the surface is relaxed through avalanches in the direction of the steepest descent . avalanches are considered to be instantaneous as their time - scale can be regarded as negligible compared to the time - scale of the surface evolution due to aeolian transport . the flux of avalanches along the slip - face is given by , @xmath49}{\frac{{\nabla}h}{|{\nabla}h| } } , \label{eq : avalanche_flux}\ ] ] where @xmath50 is a parameter and @xmath51 is the so - called `` dynamic '' angle of repose , which characterizes the surface after relaxation @xcite . the update in the local height is obtained by solving eq . ( [ eq : mass_conservation ] ) using the flux due to avalanches given by eq . ( [ eq : avalanche_flux ] ) . the calculation is repeated until the local slope is below @xmath52 . summarizing , the model consists of iteratively performing the following calculations : 1 . the average shear stress ( @xmath2 ) over the surface is computed using eqs . ( [ eq : wind_a ] ) , ( [ eq : wind_b ] ) and ( [ eq : shear_stress ] ) ; 2 . next , the height - integrated average mass flux @xmath3 over the terrain is calculated by solving eq . ( [ eq : sand_flux ] ) ; 3 . the change in the local surface is computed with eq . ( [ eq : mass_conservation ] ) ; wherever the local inclination is larger than @xmath53 , the flux due to avalanches is calculated using eq . ( [ eq : avalanche_flux ] ) and the topography updated again using eq . ( [ eq : mass_conservation ] ) . the initial surface is a smooth hill of gaussian shape , which is subjected to a constant upwind shear stress of value @xmath54 in the following , we call the upwind shear stress simply @xmath55 . calculations are performed with open boundaries and an influx @xmath56 at the inlet , which is a fraction of the saturated flux , @xmath57 . indeed , field observations show that transport in interdune areas is typically in the undersaturated regime and is dependent upon several factors , predominantly terrain type and the size and shape of dunes upwind @xcite . we consider that the interdune flux in a field of barchans migrating on top of bedrock is dictated by the average output flux of the dunes in the field ( that is , the flux of sediment being released from the barchans limb ) . since the output flux of barchan dunes produced with the model is about @xmath58 of the saturated flux @xcite , we choose @xmath59 . we note that this choice for @xmath60 is fairly consistent with measurements of drift rates by @xcite in the jafurah sand sea in saudi arabia . the authors reported monthly - averaged drift rates downwind of a mature barchan varying within a broad range between @xmath61 and about @xmath62 of the value on top of a sand sheet ( where the flux is saturated ) , with annual mean about @xmath58 ( cf . data for trap @xmath63 in table 2 by @xcite ) . however , we further note that the same authors also reported significantly larger drift rates ( about @xmath64 ) downwind of domes and small dunes with incipient slip - face , while nearly vanishing drift rate values were obtained from measurements in interdune areas consisting of salt - encrusted sandy plains . therefore , it should be remarked that our choice for @xmath56 should be reasonable for mature barchans separated by bedrock interdune areas . for other situations , like dunes surrounded by sand patches that can serve as sediment source or evolving on a terrain containing vegetation or moisture ( thus hindering interdune sediment transport ) , the value of @xmath56 should be adjusted in order to adequately model the particular physical conditions . a list of the main relevant parameters of the model with their respective values used in our calculations is displayed in table [ tab : parameters ] . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] most attempts to model barchan asymmetry have concentrated on the role of asymmetric bimodal wind regimes for the elongation of one barchan limb @xcite ( see also @xcite ) . the bimodal wind is said to be asymmetric when the transport rates of both wind components are not equal . the first conceptual model was by @xcite . according to this model , a symmetric barchan , originally formed by a gentle wind , becomes asymmetric if a storm wind blows from a secondary direction , making an accute divergence angle with the primary one . the limb exposed to the storm wind elongates as it enters the sand stream of the limb at the opposite side , thus evolving into a longitudinal seif dune @xcite . according to bagnold s model , the seif dune formed in this manner aligns approximately parallel to the direction of the storm this conceptual model was referred to by many authors in the past @xcite , though reports on supporting field examples were scarce @xcite . a different model was proposed later by @xcite : a gentle wind ( not a storm wind ) blows from the secondary direction , and the limb that elongates is the one opposite to the secondary gentle wind @xcite . some field observations exist that support this model @xcite . in our calculations , we simulate a bimodal wind by periodically alternating the orientation of the field between two directions forming a divergence angle , @xmath65 @xcite such as in the turntable experiments of ripples and subaqueous dune formation on a sediment bed under bimodal flow regimes @xcite . the wind model is solved considering a constant wind blowing over the rotated landscape , while the separation bubble adapts to the wind direction following the rotation of the field @xcite . primary _ wind direction , i.e. the one that forms the barchan , has duration @xmath66 and upwind shear stress @xmath67 . the _ secondary _ wind direction makes an angle @xmath65 with the primary wind , and lasts for a time @xmath68 with upwind shear stress @xmath69 . we begin our study with an obtuse divergence angle ( @xmath70 ) , and consider two models : _ model @xmath71 _ : both wind directions have the same upwind shear stress ( @xmath72 ) , however the primary wind has a longer duration ( @xmath73 ) a scenario reminiscent of the experiments by @xcite with the difference , however , that in the simulations the ground is not covered with sand . fig . [ fig : duran ] shows snapshots of the dune evolution obtained from the calculations . as can be seen from fig . [ fig : duran ] , the barchan develops an asymmetric shape , which depends on the ratio @xmath74 . if @xmath75 , then the limb at the side opposite to the secondary wind elongates into the _ resultant _ wind direction ( fig . [ fig : duran ] ) to form a longitudinal dune . it is interesting that the longitudinal alignment also appears to be lost in turntable experiments of ripples on a sand bed when the transport rates differ by a factor larger than 4 @xcite . a theoretical model following the concept of maximum gross transport @xcite could shed light on the origin of the threshold value of @xmath76 for seif dune elongation . here we could not find a significant dependence of this threshold on @xmath65 or @xmath56 . when @xmath77 , the barchan dune gives place to a symmetric longitudinal dune @xcite . , blows from the left ) . the barchan shape depends on @xmath78 , i.e. the duration @xmath68 of the secondary wind relative to @xmath66 . in these simulations , @xmath66 is about @xmath79 of the migration time of the barchan dune , @xmath80 . the shapes shown for @xmath81 are _ stationary _ shapes , whereas for @xmath75 , the elongated dune horn increases with time . simulation snapshots correspond to time @xmath82 of the order of @xmath83 . the barchan obtained in the simulation with @xmath84 has length and width of approximately 120 m. ] _ model @xmath85 _ : the primary wind has a larger shear stress ( @xmath86 ) as in the conceptual model by @xcite while @xmath87 . in this case , the dune shapes do not differ much from the ones in fig . [ fig : duran ] . the limb opposite to the secondary wind elongates when the relative values of bulk sand flux in the secondary and primary wind directions , respectively @xmath88 and @xmath89 , are such that @xmath90 . indeed , we also considered the situation where the storm wind is the secondary one ( @xmath91 ) , such as in the model by @xcite . in this case , primary and secondary winds simply exchange their roles , whereas the same dune shapes such as in fig . [ fig : duran ] are obtained . in summary , bagnold s hypothesis that the elongating limb is the one exposed to a secondary storm wind is not supported by the simulation results . in a general manner , the condition for elongation of the asymmetric limb due to a bimodal wind of obtuse divergence angle , as found from our calculations , reads : @xmath92}/{\left[{q_1 \cdot t_{{\mathrm{w}}1}}\right ] } > 1/4 , \label{eq : r}\ ] ] where the indices 1 and 2 refer to the primary and to the secondary wind , respectively . equation ( [ eq : r ] ) allows us to estimate the onset for emergence of asymmetry in barchan dunes , where both the relative duration and shear stress values of the wind components are accounted for quantitatively . next , we extend the calculations to different values of the divergence angle @xmath65 . in fig . [ fig : asymmetric_wind2 ] , the dune shape emerging from an asymmetric bimodal wind with @xmath93 is shown as a function of @xmath65 and @xmath94 , which is the duration of the primary wind , @xmath95 , rescaled by the migration or reconstitution time of the barchan , @xmath96 @xcite . the dune turnover time is roughly the time needed for the dune to cover a distance of its width ( @xmath97 ) @xcite . as can be seen from fig . [ fig : asymmetric_wind2 ] , dunes orient transversely or longitudinally to the resultant transport trend , depending on whether @xmath65 is acute or obtuse , respectively . bimodal wind regimes with @xmath98 lead to rounded ( or oblate ) barchans , whereas larger values of @xmath65 yield asymmetric dunes that elongate in the resultant transport direction @xcite . ) of the bimodal wind , and of the rescaled duration of the primary wind ( @xmath99 ) , for @xmath93 , where @xmath74 is the ratio between the duration @xmath68 of the secondary wind and the duration of the primary wind ( @xmath66 ) . the directions of primary and secondary winds are indicated by the arrows ( primary wind blows to the right ) . simulation snapshots correspond to time @xmath100 , except those obtained with @xmath101 , which correspond to time @xmath102 . ] however , the elongating limb is not stable if @xmath65 is within the range @xmath103 . in this case , if the influx is sufficiently small , the extended limb separates from the barchan , giving rise to a `` mixed state '' @xcite , as depicted in fig . [ fig : asymmetric_wind3]a . in fact , mixed states analogous to the ones found for asymmetric barchans have been also found in turntable experiments of aeolian ripples and subaqueous bedforms on a sand bed for the same range of @xmath65 @xcite . at very large divergence angles , the resulting morphology resembles an asymmetric dune of `` slim '' shape @xcite , as shown in the diagram of fig . [ fig : asymmetric_wind2 ] and also illustrated with the calculation using @xmath104 in fig . [ fig : asymmetric_wind3]c . the dunes in fig . [ fig : asymmetric_wind3]c are `` reversing '' dunes , which are formed by winds coming from opposing directions . this dune form has been discussed by @xcite , who also gave examples of reversing dunes occuring in south africa @xcite . [ fig : asymmetric_wind2 ] ) . time is in units of the turnover time of the barchan ( @xmath80 ) . ( a ) the divergence angle is @xmath105 , the duration of the primary wind ( to the right ) relative to @xmath80 is @xmath106 , and the influx ( @xmath56 ) is @xmath107 ( top ) and @xmath58 ( bottom ) ; ( b ) @xmath70 , @xmath108 , @xmath109 ( top ) and @xmath110 ( bottom ) ; ( c ) @xmath104 , @xmath108 , @xmath106 . ] asymmetric barchans such as the ones in fig . [ fig : duran ] form when the relative duration of the primary wind ( @xmath111 ) is within the range @xmath112 . when @xmath113 is smaller than about @xmath114 , no limb elongation occurs ; both limbs , as well as the slip face , disappear and a dome - like shape is obtained ( cf . [ fig : asymmetric_wind2 ] ) . as @xmath113 approaches unity , any change due to the secondary wind is fully compensated by the primary wind during every cycle of the bimodal wind regime the resulting morphology is essentially a barchan that is periodically reformed by a unimodal wind of strength @xmath67 , as can be seen from fig . [ fig : asymmetric_wind2 ] . the dune shapes shown in fig . [ fig : asymmetric_wind2 ] for @xmath109 are thus `` transitional '' shapes @xcite . the oblique incidence of the secondary wind leads to destabilization of the dune surface and the emergence of smaller barchans , which detach from the larger dune and migrate on the bedrock thereby alternating between both directions of the bimodal wind . figure [ fig : asymmetric_wind3]b shows the calculation of a barchan under asymmetric bimodal wind regime with @xmath70 for @xmath109 ( top ) and @xmath106 ( bottom ) . on the basis of fig . [ fig : asymmetric_wind3]b , it is possible to understand the coexistence of barchans and other complex bedforms shaped by multimodal wind regimes . if the dune is small enough , it has a small turnover time ( @xmath115 large @xmath113 ) and can readapt to the prevailing transport trend notwithstanding the complexity of the wind system @xcite . field evidence of barchan limb elongation caused by topography are unclear @xcite . we investigate two cases where topography is observed to trigger dune asymmetry under both constant wind direction and sediment supply . _ sloping terrain _ we consider a barchan dune migrating on a sloping terrain that makes an angle @xmath116 with the horizontal ( fig . [ fig : topography1]a ) . the simulation starts with a symmetric barchan of height @xmath117 and width @xmath97 , which was shaped on a flat surface under constant upwind shear stress @xmath55 and constant influx ( @xmath59 ) . then , the barchan is let to evolve on the inclined surface . due to gravity , sand transport on the tilted surface has now a component perpendicular to wind trend . figure [ fig : topography1]b shows that the emerging barchan shape is asymmetric , yet none of its limbs elongate to form a longitudinal dune @xcite . the asymmetry arises from the combined effect of both downwind sand transport and gravity - driven mass flow in the direction orthogonal to the wind . sand transport along the lower limb has a small net component downhill due to the tilting , and so this limb stretches downwhill . in contrast , the upper limb is not strechted because the sand transported laterally due to the tilting is trapped at the slip face . a consequence of the surface inclination is the downhill extension of the barchan . the migration velocity of the barchan dune orthogonal to the wind trend increases with @xmath116 , as shown in the main plot of fig . [ fig : topography1]c . for moderate values of @xmath116 within the range @xmath118 , the ratio between transverse and longitudinal migration velocities , @xmath119 and @xmath120 , respectively , can be described by the equation : @xmath121 where the coefficient @xmath122 has a slight dependence on the upwind sand flux , @xmath60 ( cf . inset of fig . [ fig : topography1]c ) . the ratio @xmath123 for barchans migrating on a sloping terrain is thus a function of the slope @xmath116 , being independent of the wind velocity . the topography - induced barchan asymmetry can be quantified in terms of the relative width @xmath124 , where @xmath97 is the total dune width , and the width @xmath125 is the distance measured in the direction orthogonal to wind trend between the border of the dune at the lowest elevation and the dune s windward foot ( cf . [ fig : topography1]d ) . if there is no surface tilting ( @xmath126 ) , then @xmath127 since the dune is symmetric along its central axis . as @xmath116 increases within the range @xmath118 , @xmath125 increases roughly linearly with @xmath128 , as shown in fig . [ fig : topography1]d . within this moderate range of @xmath116 , the ratio @xmath124 can be described by the equation , @xmath129 where the value @xmath130 is obtained from the best fit to the simulation data ( cf . [ fig : topography1]d ) . the aspect ratio of the dune ( @xmath131 ) , which is a function of the relative shear stress ( @xmath132 ) @xcite , is essentially independent of @xmath116 . in other words , @xmath132 and @xmath116 can be approximately estimated from @xmath131 and @xmath124 of the asymmetric barchan shape , respectively , provided other asymmetry causes ( i.e. bimodal wind , dune collisions or influx asymmetry ) are not relevant at the dune field considered . _ topographic rise _ asymmetric barchans occasionally occur approaching and crossing a topographic break in slope @xcite . we investigate , using the dune model , the evolution of a barchan crossing a ridge that is placed obliquely to the direction of motion ( fig . [ fig : topographic_raise ] ) . the dune is subjected to an unimodal wind regime and an influx that is @xmath58 of the saturated flux . we consider that the longitudinal axis of the ridge forms a moderate angle with the direction orthogonal to the wind ( about @xmath133 ) . the ridge has gaussian cross section , width @xmath134 and height @xmath135 , while the barchan has a height @xmath136 m. s , @xmath137w . illumination is from the top and the arrow indicates the north direction . the box indicates a barchan that extends its eastern limb across the topographic rise . the sand rose at the top - right corner is for san juan de marcona , located near @xmath138s , @xmath139w , about @xmath140 km from the dunes in the image . wind regime in the area is bimodal resulting in the extension of western limbs , except where influenced by topography . image credit : google earth . ( b ) asymmetric barchan dune of height @xmath141 m produced with the model using a topographic rise of gaussian cross section which has along - wind width 200 m and height 6 m. ( c ) simulation snapshots of the evolution of the barchan dune . time increases from left to right and wind blows from the bottom . the simulation was performed with @xmath142 and @xmath59.,title="fig : " ] s , @xmath137w . illumination is from the top and the arrow indicates the north direction . the box indicates a barchan that extends its eastern limb across the topographic rise . the sand rose at the top - right corner is for san juan de marcona , located near @xmath138s , @xmath139w , about @xmath140 km from the dunes in the image . wind regime in the area is bimodal resulting in the extension of western limbs , except where influenced by topography . image credit : google earth . ( b ) asymmetric barchan dune of height @xmath141 m produced with the model using a topographic rise of gaussian cross section which has along - wind width 200 m and height 6 m. ( c ) simulation snapshots of the evolution of the barchan dune . time increases from left to right and wind blows from the bottom . the simulation was performed with @xmath142 and @xmath59.,title="fig : " ] indeed , the height - to - width ratio @xmath143 ( or aspect ratio ) of the ridge can not be too large , as the ridge would act as a barrier for sand transport . on the contrary , a ridge of too small aspect ratio has negligible effect on the dune shape . an asymmetric dune shape with limb elongation occurs when the aspect ratio of the ridge is about @xmath144 . in figs . [ fig : topographic_raise]b and [ fig : topographic_raise]c , the ridge has height @xmath145 m and width @xmath146 m. we have examined satellite images of the dune field of fig . [ fig : topographic_raise]a and noted that the cross section of the low ridges can indeed vary within the broad range between 50 and 150 m. thus , both in the field and in the simulations , the ridge s cross section is of the same order of the cross - wind width of the barchan dune . as the barchan approaches the ridge , deposition occurs at the lee of the ridge , on the side corresponding to the closest limb . this limb then elongates once it arrives at the lee of the obstacle . the taller the ridge , the more accentuated the extension . as the other limb approaches sufficiently close to the ridge , a similar process occurs and thus that limb also elongates . the deposition at the ridge s lee occurs due to a decrease in wind strength just downwind of the crest . when the sand released through the limbs crosses the ridge s crest , it can not escape the lee since the transport rate there is reduced . as the dune advances and more limb sand is deposited , the wind strength at the lee suffices to extend the limb , yet accumulation ensues . thus , from the sand released through this limb , a sand patch forms downwind of the ridge s crest ( second frame from left to right ) , which develops into an elongated limb ( cf . third and fourth frames ) . then , in the fifth frame , the left limb is crossing the ridge s crest which also leads to a sand patch in the ridge s wake . this process causes , then , elongation of the left limb as we can see in the subsequent frame . it is interesting to note that in fig . [ fig : topographic_raise]a not all barchans elongate the same limb . for instance , some barchans which have the right ( left ) limb elongated are upwind ( downwind ) of the ridge s crest , which is in agreement with the prediction of fig . [ fig : topographic_raise]c as discussed above . however , this behavior can be explained only if we assume that the dunes are well isolated from each other ( i.e. they are not participating at a collision ) . for instance , the dune shape in the most lower right of fig . [ fig : topographic_raise]a , which seems to result from a collision between dunes , has limbs of nearly equal size . furthermore , the dune on the most lower left has an elongated arm opposite to that of the example barchan with the black box drawn around it . however , the elongated arm of the lower left dune too has a slip face possibly indicating that this dune too is being affected by a collision . therefore , application of our results to the field should be made with care , since we did not account for the effect of dune collisions or influx asymmetry , which can influence the dune shape and will be addressed separately in later sections . future studies should focus on the asymmetric dune resulting from simulations using different angles between ridge orientation and wind direction , as well as different shapes of the ridge . indeed , the results of our calculations should be valid for ridges which have a smooth cross section , where the slope of the surface does not exceed the critical angle for causing flow separation ( about @xmath147 ) @xcite . if the ridge s cross section has a sharp edge , then flow separation occurs and the flow is diverted parallel to the ridge along the lee side @xcite . in this case , sand can be transported in the direction parallel to the ridge wherever the wind speed within the separation zone exceeds the transport threshold . moreover , the size of the zone of recirculating flow increases with the crest - to - brink distance of transverse rigdes @xcite , and thus the effect of secondary flow on the lateral transport of sand at the lee of the ridge should also depend on the shape of the ridge at its windward side . however , due to the assumption of our model that net flow ceases within the separation zone , the sand deposited at the lee of a ridge with sharp edge can not be transported at all . thus , three - dimensional flow patterns at the lee of dunes or obstacles should be accounted for in the modeling of dune asymmetry when the topographic rise has a sharp slope @xcite . the role of asymmetric sediment supply for barchan asymmetry has remained uncertain in the few investigations undertaken @xcite . it is admittedly a difficult task to conduct systematic field studies on the barchan shape as a function of the degree of spatial asymmetry in the upwind sediment budget . the sand flux onto the windward side of a barchan is largely dependent upon the spatial distribution of the upwind dunes or other sand sources , e.g. at a sand sea margin @xcite . spatial inhomogeneities in the local physical properties of the interdune terrain may also play an important role for the incoming flux @xcite . furthermore , the shape of a barchan migrating in a field is inevitably affected by wind trend variations in time @xcite , as well as by the interaction with neighbouring dunes through merging , breeding and lateral linking . with the help of simulations , the barchan shape resulting from asymmetric upwind flux alone can be studied without regard to other physical factors . we consider a symmetric barchan initially migrating within an area where the influx @xmath148 is approximately a constant fraction ( @xmath149 ) of the saturated flux ( @xmath57 ) the value of @xmath150 is , in fact , within average values measured at interdune areas in typical barchan dune fields @xcite . the barchan has height @xmath151 m and width @xmath152 m , and has been shaped by a constant shear stress @xmath153 . at @xmath154 , the upwind influx becomes asymmetric . the influx on the side of the barchan s left limb has a larger value , @xmath155 ( dark areas in fig . [ fig : asymmetric_influx]a ) . this model for asymmetric influx constitutes a simple model for a scenario typical for real dune fields , where in some areas the presence of a dome , a sand sheet or a larger dune upwind may significantly increase the local interdune flux . figure [ fig : asymmetric_influx]a shows different snapshots of the evolution of the barchan dune subject to the asymmetric influx . because more sand deposits in the areas of larger influx ( @xmath156 ) , the dune adapts to the asymmetric influx by increasing its relative volume within the areas under influx @xmath156 . different ( transient ) asymmetric dune shapes appear as the dune reforms its shape in order to accomodate the larger influx @xmath156 , as can be seen in fig . [ fig : asymmetric_influx]a . the shorter limb is the one exposed to the larger influx value , because a large upwind flux prevents erosion at the windward foot and downwind motion of the dune . thus , the barchan limb on the side of lower influx appears advanced in relation to its `` fat '' @xcite counterpart . the fraction of the dune width ( @xmath124 , cf . [ fig : asymmetric_influx]b ) that is within the area of larger influx ( @xmath156 ) increases in time , as can be seen in fig . [ fig : asymmetric_influx]c . in this figure , @xmath119 denotes the rate with which @xmath124 grows in time . firstly , @xmath124 increases roughly as a linear function of time which means that in this initial stage @xmath119 is approximately constant for given values of @xmath148 and @xmath156 relative to @xmath57 . the inset of fig . [ fig : asymmetric_influx]c shows that @xmath119 is indeed much smaller than the longitudinal migration velocity of the barchan ( @xmath157 ) even for large @xmath158 . indeed , the main plot of fig . [ fig : asymmetric_influx]c suggests that , after a sufficiently long time , @xmath124 asymptotically approaches a maximal value ( and thus @xmath159 ) . since the increase of @xmath124 with time becomes extremely slow after long time , and since the dune is increasing rapidly in volume as @xmath156 approaches @xmath57 , it is difficult to reliably estimate the steady - state value of @xmath124 through numerical simulations . however , we can approximately estimate the steady - state value of @xmath124 ( which we call @xmath160 ) by means of a simple calculation , as we will explain in the next paragraph . a barchan under constant influx @xmath56 is an unstable object . while the dune is gaining sand due to @xmath56 , it is also losing mass through the limbs . the mass balance determines whether the dune grows or shrinks with time . as shown previously @xcite , there is a critical value of @xmath56 , denoted by @xmath161 , above which the dune volume increases in time and below which the dune shrinks . this critical influx is approximately equal to @xmath162 of the saturated flux , @xmath57 @xcite . the rate @xmath163 at which the cross - wind width @xmath97 of the barchan changes in time approximately scales with @xmath164}/w$ ] @xcite . using this relation , we can approximately estimate the growth rate of @xmath125 and of @xmath165 , which denote the fractions of the dune width under influx values @xmath156 and @xmath148 , respectively ( cf . fig . [ fig : asymmetric_influx]b ) . the growth rates @xmath166 and @xmath167 should be proportional to @xmath168}/w$ ] and @xmath169}/(w - w)$ ] , respectively . the steady - state value of @xmath125 , i.e. @xmath170 is achieved when both rates equal , which gives , @xmath171 with @xmath172 as mentioned above . figure [ fig : asymmetric_influx2 ] shows how @xmath160 depends on @xmath158 for different values of @xmath173 , with @xmath174 . for values of @xmath148 close to @xmath161 , @xmath160 approaches unity as @xmath156 increases , which means that the dune enters almost entirely into the region of larger influx and is nearly symmetric again . this situation is indeed observed in the simulation ( see e.g. fig . [ fig : asymmetric_influx]a for @xmath175 ) . however , as the influx becomes very large the barchan shape gives place to a flat dome - like dune without slip - face , which may serve as source of sand for smaller dunes @xcite . we note that an increase of @xmath148 implies a decrease in the influx asymmetry , since @xmath156 is within the range @xmath176 . thus , as @xmath148 increases , the steady - state shape of the dune becomes less symmetric ( though the barchan shape gives place to a dome as discussed above ) and @xmath177 . evidently , for @xmath178 there is no asymmetry in the influx and the dune shape is thus symmetric ( @xmath170 is equal to @xmath179 ) . of the fraction @xmath124 of the dune cross - wind width within the region of larger influx ( @xmath156 ) , as a function of @xmath158 for different values of @xmath173 . ] dune collisions are well - known factor for the emergence of complex , asymmetric dune morphologies @xcite . the dynamics of collisions between two barchans have been systematically investigated in previous studies using the present model @xcite . it was shown that , if the symmetry axes of both dunes are aligned , then the collision dynamics is determined by the volume ratio of the dunes , whereas the resulting morphology is symmetric @xcite . however , if the collision occurs with a lateral offset ( cf . [ fig : dune_collision1 ] ) , asymmetric dune shapes can arise @xcite . figure [ fig : dune_collision2 ] shows snapshots of calculations using different values of volume ratio and lateral offset defined as @xmath180 , where @xmath181 ( @xmath19 ) is the crest s position of the large ( small ) dune in the direction orthogonal to the wind , and @xmath97 is the width of the large dune ( fig . [ fig : dune_collision1 ] ) . the asymmetric dune shapes produced due to collisions are different from the ones in section [ sec : asymmetric_sand_supply ] , since the collision not only implies an asymmetric influx but also leads to a ( dynamic ) modification in the shear stress field during the interaction between the dunes @xcite . the hybrid , asymmetric dunes depicted in fig . [ fig : dune_collision2 ] are transient shapes . a sufficiently long time after the collision , dunes are well separated and adapt to constant influx and wind direction again . this time - scale which the dune needs to adapt its shape after the collision is governed by the dune reconstitution time , @xmath80 , which scales with the barchan s cross - wind width @xmath97 divided by its migration velocity @xmath157 @xcite . within this time - scale , the dune migrates a downwind distance of the order of its own width @xmath97 . ) approaches a larger one ( width @xmath97 ) from behind with lateral offset . the small and the large barchans are centered at @xmath19 and @xmath181 , respectively . ] and volume ratio @xmath182 . calculations with @xmath183 are shown for ( a ) @xmath184 , ( c ) @xmath185 and ( e ) @xmath186 ; calculations with @xmath187 are shown for ( b ) @xmath184 , ( d ) @xmath185 and ( f ) @xmath186 . ] the simulations corroborate previous observations that dune asymmetry could result from limb merging as two barchans link laterally , thereby causing downwind extension of the coalesced limbs @xcite . such a situation is found for sufficiently large offset ( @xmath188 ) , as can be seen in figs . [ fig : dune_collision2]e and [ fig : dune_collision2]f . it is interesting to notice that the result of the lateral coalescence is the release of a small dune from the merged limbs , whereas the colliding barchans merge then into a single asymmetric dune . the release of the small dune from the merged limbs is signature of the so - called `` solitary wave - like behaviour '' of sand dunes discussed in previous studies of binary collisions between barchans @xcite . as the smaller , faster dune approaches the larger one from the upwind , sand from the downwind dune is trapped in the wake of its companion . thus , the smaller dune upwind gains sand from the downwind dune thereby increasing in volume and becoming slower . due to erosion downwind of the smaller dune s wake region ( i.e. after the `` separation bubble '' of the small dune ) , a small dune is released from the larger barchan . in effect what happens is that the colliding dunes merge to form a single ( asymmetric ) dune while a small barchan is ejected from the surface of the downwind dune . moreover , we see in figs . [ fig : dune_collision2]e , f that the volume ratio of dunes colliding with large offset can yield distinct asymmetric dune shapes . if the upwind dune is small , it merges laterally to the lowest portions of the downwind barchan s limb which is involved in the collision ( cf . [ fig : dune_collision2]e ) . this limb appears then deformed ( with a somewhat `` sinuous '' shape ) and its cross - wind width also increases . however , no significant limb extension in the wind direction occurs , as we can see in the last frame of fig . [ fig : dune_collision2]e . a different situation occurs if the size of the upwind dune participating at the collision is large , as in the simulation of fig . [ fig : dune_collision2]f . as we can see in the last frame of this figure , the upper limb of the larger barchan left behind ( i.e. the limb _ opposite _ to the side where the binary collision occurred ) appears _ advanced _ downwind . the reason for this behaviour is that the lower limb is much larger and slower than the upper one , thus appearing `` retarded '' in relation to its companion . asymmetric dune patterns very different from the ones in fig . [ fig : dune_collision2 ] can result from collisions between multiple dunes . figure [ fig : barchan_field]a shows one snapshot of the simulation of a barchan dune field emerging from a flat sand surface subjected to a saturated influx . the simulations of the genesis of barchan fields were discussed in detail in previous works @xcite . as shown in these works , the small transverse bedforms emerging from the instabilities which develop on the flat hill upwind give place to small barchans after reaching the bedrock due to the cross - wind instability inherent to transverse dunes @xcite . the barchan dunes , which are subjected to a strong influx , grow in size during their migration , thereby experiencing multiple collisions with their counterparts and producing complex , asymmetric dune shapes . different patterns produced in the simulation include limb extension of the upwind barchan , as we can see in fig . [ fig : barchan_field]a . while the simulation of fig . [ fig : barchan_field]a denotes a model for an immature dune field ( which is at the early stages of its development ) , we have also performed simulations using periodic boundary conditions in the wind direction such as to model the steady - state dynamics of a dune corridor @xcite . such simulations also lead to asymmetric barchan shapes which can not be obtained with binary collisions ( cf . [ fig : barchan_field]b ) . the complex asymmetric shapes shown in figs . [ fig : barchan_field]a and [ fig : barchan_field]b result from collisions between multiple barchans which have different sizes and offset values , and are further subjected to an asymmetric influx depending on the spatial distribution of dunes upwind . the systematic study of collisions between multiple dunes in a field is out of the scope of the present study . however , on the basis of our calculations , we can conclude that binary collisions can produce a rather limited range of asymmetric dune patterns . the only previously reported asymmetric pattern produced by a binary collision is the limb elongation due to dunes merging laterally ( fig . [ fig : dune_collision2]e , f ) . other asymmetric patterns induced by collisions can result if the collisions involve multiple dunes during the evolution of a barchan dune field . as can be seen in figs . [ fig : barchan_field]a and [ fig : barchan_field]b , collisions between dunes lead to the formation of groups of barchans , in which upwind barchans releasing sand through their extended limbs act as source of sediment to the dunes in the front , as previously noted by @xcite . indeed , the complex patterns emerging from collisions between barchans are transient dune shapes , as shown in fig . [ fig : dune_collision2 ] and in previous works @xcite . further modeling work is required in order to deepen our understanding of the collision dynamics of multiple dunes and of the resulting dune patterns . the results of our simulations can help researchers in the future to discern among the most relevant factors competing for diverse types of barchan asymmetry . for instance , our model predicts that only asymmetric bimodal wind regimes can lead to seif dune formation due to an elongating barchan limb ( cf . section [ sec : bimodal_winds ] and the condition for seif dune elongation in eq . ( [ eq : r ] ) ) . barchans subjected to an asymmetric influx , crossing a topographic rise or undergoing a collision with other dunes can experience different types of limb extension , however a seif dune can not appear from an asymmetric barchan due to these factors alone . furthermore , our simulations show that topographic breaks in slope and collisions can trigger extension of one or the other limb depending on the location of the dune relative to the topographic obstacle or to the surrounding dunes ( cf . figs . [ fig : topographic_raise ] , [ fig : dune_collision2 ] and [ fig : barchan_field ] ) . in contrast , dunes subjected to the same asymmetric bimodal wind regime should elongate the same limb , provided other asymmetry causes are not influencing the dune morphology . application of the results of this paper to the field should be made with care because typically asymmetric barchans result from the concurrent action of more than one asymmetry causes . in the present paper , we list three examples of barchan dunes displaying different asymmetric shapes , the origin of which could be attributed to one specific asymmetry cause , based on our simulations . these examples are dicussed separately in sections [ sec : example_1]@xmath189[sec : example_3 ] , while in section [ sec : planetary_discussion ] we discuss on how our findings could help the research of extraterrestrial dunes . the dune shown in fig . [ fig : comparison_bimodal_wind]b is in bir lahfan and provides an example of seif dune formation from a barchan dune due to obtuse bimodal wind regime @xcite . the sand rose in the inset shows two main wind directions with a divergence angle about @xmath190 . we see that the dune displays only one limb , which aligns nearly longitudinally to the resultant transport trend . the growth and migration of the longitudinal seif dunes in bir lahfan was discussed in detail in previous publications ( see e.g. @xcite for a review ) . and @xmath106 ( see fig . [ fig : asymmetric_wind3]a ) . the influx is @xmath108 of the saturated flux . we compare this dune with the asymmetric barchan in * ( b ) * , which is at bir lahfan in sinai , near @xmath191n , @xmath192e . the sand rose shown in the inset is for an area between bir lahfan and jabel libni , sinai . * ( c ) * image of the dune field at bir lahfan showing that the dune in ( b ) , enclosed by the box , is surrounded by other asymmetric barchans which are also giving rise to seif dunes . both in ( b ) and ( c ) , north is at top as indicated by the arrows ( images credit : google earth).,title="fig : " ] and @xmath106 ( see fig . [ fig : asymmetric_wind3]a ) . the influx is @xmath108 of the saturated flux . we compare this dune with the asymmetric barchan in * ( b ) * , which is at bir lahfan in sinai , near @xmath191n , @xmath192e . the sand rose shown in the inset is for an area between bir lahfan and jabel libni , sinai . * ( c ) * image of the dune field at bir lahfan showing that the dune in ( b ) , enclosed by the box , is surrounded by other asymmetric barchans which are also giving rise to seif dunes . both in ( b ) and ( c ) , north is at top as indicated by the arrows ( images credit : google earth).,title="fig : " ] the dune in fig . [ fig : comparison_bimodal_wind]a was obtained from a simulation using an asymmetric bimodal wind regime with @xmath93 . it is the third dune from left to right in fig . [ fig : asymmetric_wind3]a bottom ( @xmath108 of the saturated flux ) . this dune indeed displays some characteristics that are found in the real bir lahfan dune , namely it has only one limb , which is elongating into a seif dune . the wind regime that generated the dune of fig . [ fig : comparison_bimodal_wind]a differs from the one in the sand rose of fig . [ fig : comparison_bimodal_wind]b as in this sand rose both main wind directions have nearly equal transport rates . indeed , it was shown in experiments @xcite and numerical simulations @xcite , that symmetric bimodal flow regimes ( with @xmath193 ) lead to _ symmetric _ longitudinal dunes rather than asymmetric barchans . however , we note that the sand rose in fig . [ fig : comparison_bimodal_wind]b also displays secondary transport directions , which are not taken into account in the simulations . the role of these smaller transport directions for the dune shape is still uncertain . simulations using multidirectional wind regimes , which are out of the scope of the present work , should be performed in the future in order to clarify the relevance of secondary transport directions for the shape and elongation of asymmetric barchans . possibly , the presence of more than two wind directions as well as influx asymmetry due to the presence of other dunes in the field ( see fig . [ fig : comparison_bimodal_wind]c ) could explain the meandering of the barchan limb visible in fig . [ fig : comparison_bimodal_wind]b . this meandering is namely absent from the asymmetric barchan of fig . [ fig : comparison_bimodal_wind]a and also from symmetric longitudinal dunes on top of bedrock generated in experiments @xcite and numerical simulations @xcite . the second example is provided in fig . [ fig : topographic_raise]a , which shows asymmetric barchans in peru , while the sand rose of the dune field is shown in the inset . the barchans are crossing topographic rises , which suggests topography as cause for their asymmetric shape @xcite . indeed , as discussed in section [ sec : topography ] , we found that when the barchans are _ upwind _ of the ridge s crest , the limb closest to the ridge appears extended , while the other limb elongates once the dune has crossed the ridge , i.e. when the dune is _ downwind _ of the ridge s crest ( see fig . [ fig : topographic_raise]c ) . although this behavior could explain the asymmetric shape of some of the dunes in fig . [ fig : topographic_raise]a ( in particular see the simulation of fig . [ fig : topographic_raise]b and the more detailed discussion in section [ sec : topography ] ) , for many of the dunes in peru other asymmetry causes can not be excluded , in particular collisions @xcite . we see in fig . [ fig : barchan_field]c the third example of barchan asymmetry , which is caused by collisions . many of the dunes in fig . [ fig : barchan_field]c that are well separated from other dunes , and are thus not participating of a collision , have approximately the same shape , which is nearly symmetric and consistent with the wind regime indicated by the sand rose in the inset . other dunes in the barchan corridor , especially those migrating close to other dunes , display different types of asymmetry . we see that several barchans have the upper limb extended , while others have the lower limb extended . moreover , in some cases either limb is giving rise to a small barchan , thus indicating that dune _ calving _ @xcite is a relevant mechanism for the generation of small dunes in the field . the simulations of figs . [ fig : barchan_field]a and [ fig : barchan_field]b , performed with strictly unidirectional wind regimes ( wind blows from left to right ) , produce barchan asymmetry behavior with the characteristics found in the examples of fig . [ fig : barchan_field]c , thus indicating that complex ( asymmetric ) wind regimes are not a pre - requisite for the asymmetric dune shapes in the referred field . asymmetric barchan dunes occur in many dune fields on mars @xcite . it was shown by @xcite that the origin of a variety of martian asymmetric barchans could be understood on the basis of the four main asymmetry causes tested in the present paper . indeed , the insights of our simulations should be applicable for understanding the formation of seif dunes on mars and also the influence of topography and collisions on complex dune patterns occurring within martian barchan corridors . in martian dune fields hosting asymmetric barchans with one limb elongating into a seif dune , figs . [ fig : asymmetric_wind2 ] and [ fig : asymmetric_wind3 ] could be used for estimating the divergence angle @xmath65 and the ratio @xmath76 between the transport rates associated with the dominant sand - moving wind directions . we note that a diversity of exotic martian dune shapes occurring on top of bedrock could be indeed simulated with the present model using obtuse bimodal wind regimes with @xmath77 @xcite . on the other hand , due to the lower gravity of mars the influence of topography on the shape of barchan dunes on the red planet should be correspondingly smaller than on earth . for instance , we expect dunes crossing a sloping terrain on mars to migrate laterally as they do on earth , however the value of @xmath194 ( cf . ( [ eq : vt_vl ] ) ) , which gives the ratio of transverse to longitudinal migration rates , should be smaller on mars for a given terrain s slope . furthermore , simulations of dune collisions such as the ones in figs . [ fig : dune_collision2 ] and [ fig : barchan_field ] but using attributes of sediment and fluid valid for martian sand systems could shed light on the origin of complex asymmetric patterns occurring on different locations ( and thus environmental conditions ) on mars . however , in order to perform quantitative simulations , the model should be further improved by accounting for recently noted differences between saltation on mars and earth as well as for the recently gained insights into the physics of the threshold wind velocity required to sustain sediment transport on mars @xcite . moreover , future simulations should incorporate a model for the saturation length @xmath195 ( cf . ( [ eq : sand_flux ] ) ) that encodes the relevant mechanisms of flux saturation of sediment transport in extraterrestrial environments @xcite . finally , asymmetric barchans leading to seif dunes due to limb extension could be identified in cassini images of the surface of titan @xcite , thus indicating asymmetric bimodal wind regimes on this saturn s moon . the formation of the equatorial linear dune fields of titan has been indeed attributed to seasonally varying winds @xcite . thus , the modeling presented here could be used to investigate the ratios of transport rates and divergence angles of titan s bimodal wind regimes at those locations hosting asymmetric barchans . in conclusion , we have investigated , by means of morphodynamic modeling of aeolian dunes , the respective roles of bimodal wind regimes , topography , influx asymmetry and dune collisions on the formation and evolution of asymmetric barchan dune shapes . the conclusions of our numerical simulations can be summarized as follows : * _ bimodal wind regimes _ the limb opposite to the secondary wind elongates into the resultant transport direction if the divergence angle of the bimodal wind is obtuse and the ratio @xmath196}/{\left[{q_1t_{{\mathrm{w}}1}}\right]}$ ] where @xmath89 ( @xmath88 ) and @xmath66 ( @xmath68 ) stand for the bulk sand flux and the duration of the primary ( secondary ) wind , respectively exceeds @xmath1 . these conditions are the same as the ones for oblique alignment of bedforms in dense sand beds @xcite , the asymmetric barchan being the corresponding morphology under low amount of available sand . * _ topography _ a barchan crossing a topographic rise can become asymmetric ; the limb closest to the topographic rise elongates downwind . the migration velocity of a barchan that is on a tilted surface has a downhill component proportional to the tilting slope ; the preferential limb extension is downhill rather than downwind . * _ influx asymmetry _ the side of the barchan subjected to the larger influx increases in volume , whereas the opposite limb elongates downwind . the typical asymmetric morphology is a barchan with an elongated arm ( the one subjected to lower influx ) . the asymmetry is of transient nature , since the dune migrates laterally towards the region of higher influx . * _ dune collisions _ asymmetry can be triggered due to barchan collisions with lateral offset . binary collisions with large offset can lead to extension of the limb resulting from the merging of the two limbs participating at the collision . collisions between multiple barchans in a field can trigger more complex asymmetric patterns . some patterns produced in these simulations involve the elongation of one limb of the upwind dune , as reported from previous observations @xcite . in order to improve the quantitative assessment of asymmetric barchans , the model should be extended in order to account for secondary flow effects at the dune lee @xcite , which may be relevant for the dynamics of limb elongation due to bimodal wind regimes or due to topographic ridges with sharp slopes @xcite . indeed , the separation bubble model of section [ sec : wind_field ] does not account for the occurrence of three - dimensional flow structures within the zone of recirculating flow , which affect the dune shape at the lee and can contribute to the elongation of the limb . realistic simulations of bimodal wind regimes should further account for multiple secondary wind trends in consistence with the complex sand roses of real dune fields @xcite . moreover , real winds can be very gusty , with changes in direction of 45 degrees or more taking place in most cases in less than 15 minutes , while in some instances the changes may be gradual or abrupt @xcite . this effect of changes in wind directions should be considered in the future . finally , we also note that there are examples of seif dunes formation from vegetated linear dunes that lost their vegetation due to human impact @xcite . the formation of seif dunes in the presence of a dynamic vegetation cover remains to be investigated in the future . the results of our calculations are potentially useful for inferring local wind regimes or spatial variations either in topography or in availability of mobile sediments in planetary dune fields where asymmetric barchans occur . this work was supported in part by nasa mdap grant nnx10aq35 g , by funcap , capes and cnpq ( brazilian agencies ) , by funcap grant 0011 - 00204.01.00/09 , by swiss national foundation grant nf 20021 - 116050/1 and eth grant eth-10 09 - 2 . we also thank the german research foundation ( dfg ) for funding through the collaborative research initiative `` additive manufacturing '' ( sfb814 ) and the cluster of excellence `` engineering of advanced materials '' . arajo , a.d . , parteli , e.j.r . , pschel , t. , andrade jr . , j.s . , herrmann , h.j . numerical modeling of the wind flow over a transverse dune . scientific reports 3 , 2858 . doi:10.1038@xmath197srep02858 . bourke , m.c . , goudie , a.s . , 2009 . varieties of barchan dunes in the namib desert and on mars . aeolian research 1 , 45 - 54 . bourke , m.c . , lancaster , n. , fenton , l.k . , parteli , e.j.r . , zimbelman , j.r . , radebaugh , j. , 2010 . extraterrestrial dunes : an introduction to the special issue on planetary dune systems . geomorphology 121 , 1 - 14 . bristow , c.s . , bailey , s.d . , lancaster , n. , 2000 . the sedimentary structure of linear sand dunes . nature 406 , 56 - 59 . durn , o. , parteli , e.j.r . , herrmann , h.j . , 2010 . a continuous model for sand dunes : review , new developments and application to barchan dunes and barchan dune fields . earth surface process and landforms 35 , 1591 - 1600 ( 2010 ) . doi:10.1002/esp.2070 . sullivan , r. , greeley , r. , kraft , m. , wilson , g. , golombek , m. , herkenhoff , k. , murphy , j. , smith , p. , 2000 . results of the imager for mars pathfinder windsock experiment . journal of geophysical research 105 , 24547 - 24562 .

we study scalar and vector modulation instabilities induced by the vacuum fluctuations in birefringent optical fibers . to this end , stochastic coupled nonlinear schrdinger equations are derived . the stochastic model is equivalent to the quantum field operators equations and allow for dispersion , nonlinearity , and arbitrary level of birefringence . numerical integration of the stochastic equations is compared to analytical formulas in the case of scalar modulation instability and non depleted pump approximation . the effect of classical noise and its competition with vacuum fluctuations for inducing modulation instability is also addressed .

quant-ph0412096

in the early 1980s single - mode silica optical fibers were recognized as a privileged medium for experiments in quantum optics because they exhibit a well defined transverse mode and very low losses . the kerr nonlinearity of silica has been used to produce non classical states of light . squeezed state production in fibers was pioneered by levenson _ @xcite in 1985 . since then a large number of squeezing experiments have been performed with cw - light or short optical pulses ( for a review see @xcite ) . more recently , fiorentino _ et al . _ demonstrated that optical fibers may also be used to produce _ twin - photons _ pairs @xcite . this new kind of twin - photon source is well suited for fiber - optic quantum communication and quantum cryptography networks . in contrast with the traditional twin - photon sources , based on the @xmath0 down - conversion process , it avoids large coupling losses occurring when the pairs are launched into long distance communication fibers . the physical process used in @xcite to generate twin - photon pairs is a four - wave mixing ( fwm ) phase - matched by the interplay of the optical kerr effect and the chromatic dispersion , called ( scalar ) modulation instability ( mi ) . this process can be basically understood as the destruction of two pump photons at frequency @xmath1 and the creation of stokes ( red - shifted ) and anti - stokes ( blue - shifted ) photons at frequencies @xmath2 and @xmath3 satisfying the energy conservation relation @xmath4 . in the time domain , the beating of the pump , signal and idler waves produces a fast modulation of the pump envelope . mi is a spontaneous phenomena which can be described in the framework of classical nonlinear optics @xcite . in this classical framework , one usually considers that mi is induced by some incoherent noise initially present on the pump wave . twin - photon pair production requires however a very low input noise power to be efficient ( otherwise the twin - photons are buried in the background photon noise ) . such a regime is dominated by vacuum fluctuations and requires a quantized field theory to be described properly . a small - perturbation approach of this problem @xcite ( non depleted pump approximation ) shows that vacuum fluctuations can induce the mi even in the absence of any classical input noise . this is an ideal situation for twin - photon pairs production , but it never occurs in a real life experiment . in practice , classical input noise and vacuum fluctuations compete for inducing mi . understanding the dynamical development of mi from a quantum point of view is an important issue for the design of fiber - optics twin - photon pairs sources . we present an unified approach to this problem based on the _ stochastic nonlinear schrdinger equations _ ( snlse ) @xcite . this formalism is equivalent to quantum - field heisenberg equations but has two main advantages . first , it is very suitable for numerical simulations of complex situations where classical noise and vacuum fluctuations act together . second the _ correspondence principle _ of quantum physics , i.e. the transition from quantum to classical world , looks very natural in the snlse formalism . the mi observed in standard non - birefringent single - mode silica fibers is often referred to as _ scalar _ modulation instability ( s - mi ) because polarization of light plays no role in this process . in birefringent fibers different kinds of mi may appear because of the interplay of nonlinearity , chromatic dispersion and birefringence . this case will be referred as _ vector _ modulation instability ( v - mi ) . twin - photon pairs sources based on v - mi have not been demonstrated yet , in contrast with s - mi based sources @xcite . in this article , we will however address both issues because we recognize that v - mi twin - photon pairs sources could be more practicable than s - mi based ones . the theoretical and computational tools that we developed apply to all kind of mi , including the low - birefringence , high - birefringence and scalar limits . this article is divided into four sections , beginning with the present introduction . the second section consists of a review of the approach based on a perturbation analysis around the steady state solution . this is the simplest method of approaching the problem of mi . it allows us to set the stage for the more sophisticated approach we then develop and to make contact with earlier work in the field . then in sec . [ sec:3 ] we exhibit the snlse for scalar and vector mi . contrary to the approach of sec . [ sec:2 ] these equations are not based on a perturbation analysis and are valid in the strongly non linear regime . they describe both mi originating from classical noise and from vacuum fluctuations ( quantum noise ) . the snlse we obtain generalize the earlier results of @xcite . in particular the snlse we obtain does not require the birefringence to be either small or large as in earlier work @xcite , but are valid for all values of the birefringence . for the interested reader we present a self contained derivation of these equations in the appendix . in sec . [ sec:4 ] we use the _ split - step fourier _ method to integrate the snlse derived in sec . [ sec:3 ] , and illustrate our algorithm on the cases of scalar mi and of vector mi both for weak , intermediate , and strong birefringence . in order to interpret the results of the numerical integration it is essential to introduce the notion of _ mode_. this allows us to compare the numerical results and the analytical solutions derived from perturbation theory . we also compare in detail the characteristics of mi induced by classical noise and by quantum noise . in a companion article @xcite we shall show that our numerical results are in very good quantitative agreement with experimental results . as pointed out in the introduction several kind of mi can occur in optical fibers . from a general point of view , mi occurs when the continuous wave ( steady - state ) propagation is unstable . the most straightforward way to get some insight on how the mi develops is to perform a small - perturbation analysis of the steady - state . however this method has limitations : it can not address the strongly non linear regime and simple analytical formulas can not be always obtained by this method . in this section we illustrate the perturbation method in the case of s - mi , which is the simplest . s - mi occurs in isotropic fibers in the anomalous dispersion regime . in what follows we will review the most important aspects of this approach , focussing on the comparison between the classical and quantum descriptions , and on the limitations of this approach . in sec . [ sec:4 ] , we will compare the analytical formulas for s - mi derived from the perturbation analysis to numerical results from the snlse derived in sec . [ sec:3 ] . the discussion of mi in birefringent fibers is also postponed to sec [ sec:3 ] and sec . [ sec:4 ] . in an isotropic single - mode fiber , the electric field may be written @xmath5\mathbf{\hat{x}}+\mathrm{c.c.}\ ] ] where @xmath6 is a unit vector orthogonal to the fiber axis ( @xmath7-axis ) , @xmath1 the carrier angular frequency and @xmath8 the associated propagation wave number ( modal propagation constant ) . @xmath9 stands for the mode profile function and @xmath10 for the field envelope . the field is supposed to be polarized linearly . this is not restrictive because of the isotropy assumption . the complex envelope @xmath11 evolves according to the _ nonlinear shrdinger equation _ ( nlse ) that can be established from maxwell s electromagnetic theory @xcite . if the envelope is normalized in such way that @xmath12 is equal to the instantaneous power flowing through the plane @xmath13 at time @xmath14 , the nlse is obtained in the following form : @xmath15 the group - velocity dispersion ( gvd ) parameter and @xmath16 is fiber nonlinearity parameter defined as @xmath17 where @xmath18 is the fiber mean linear index of refraction , @xmath19 the mode effective area and @xmath20 is the diagonal element of the fiber @xmath21 nonlinearity tensor . s - mi is observed when a continuous wave ( cw ) or a quasi - cw optical pulse is launched into the fiber . in the cw - case , when an optical power @xmath22 at frequency @xmath1 in injected , a first order perturbation analysis of eq . ( [ nls ] ) shows that the cw steady - state solution @xmath23 becomes instable in the anomalous dispersion regime ( @xmath24 ) . the instability manifests itself by a parametric gain at frequencies @xmath25 with @xmath26 . the maximal gain @xmath27 occurs for @xmath28 @xcite . noise at these frequencies is strongly amplified . as a result , the optical spectrum at the fiber output exhibits two sidebands at frequencies @xmath29 . this analysis supposes that the pump power remains constant ( undepleted pump approximation ) and a cw regime . however , numerical simulations of eq . ( [ nls ] ) show that the above formula also hold for quasi - cw ( for instance nanosecond ) optical pulses ( see sec . [ sec:4 ] ) . sideband generation can be interpreted as a fwm process phase - matched by the interplay of dispersion and kerr nonlinearity . from this point of view , the mi process can be seen as the destruction of two pump photons at frequency @xmath1 followed by the creation of stokes and anti - stokes photons at frequencies @xmath30 and @xmath31 respectively . in the cw - case and non depleted pump approximation , the output power spectral density at frequencies @xmath2 and @xmath3 can be computed using analytical formulas @xcite for given initial conditions . these formulas also hold for quasi - cw pulses provided one replace stokes and anti - stokes power spectral densities by the number @xmath32 and @xmath33 of stokes and anti - stokes photons located in the same temporal mode as the pump wave ( see sec . [ sec:4b ] for a detailed analysis of this issue ) . for example , assuming an incoherent initial noise , one finds [ s - micl ] @xmath34 where @xmath35 is the propagation distance . these equations show that classical incoherent noise is amplified exponentially with gain @xmath27 . noise induced s - mi was first demonstrated experimentally by tai _ et al . _ in 1986 @xcite . this experiment confirmed two main predictions of the classical theory : the amplification of stokes and anti - stokes photons and the power - dependence of their frequency shift . however the classical theory of s - mi only holds when the initial number of noise photons is high or in a stimulated regime when a coherent probe pulse at stokes or anti - stokes frequency is injected together with the pump pulse @xcite . ( [ s - micl ] ) are unable to explain generation of stokes and anti - stokes twin - photons reported in @xcite . neither are they valid when the mean photon numbers @xmath36 and @xmath37 are of the order of one or lower . in this regime vacuum fluctuations play a central role and field quantization is required . to take into account vacuum fluctuations , fields must be quantized . the quantum counterpart of the nlse ( [ nls ] ) is known as the quantum nonlinear shrdinger equation ( qnlse ) @xcite : @xmath38 where @xmath39 is the quantum operator corresponding to the field envelope and @xmath40 its hermitic conjugated . @xmath39 and @xmath40 satisfy the bosonic _ equal - space _ commutation rule : @xmath41=\hbar\omega_0 \delta(t - t').\ ] ] the normalization constant @xmath42 stems from the normalization chosen for the field envelope . in the quantum propagation theory , the expectation value of the optical power flowing through the plane @xmath13 at time @xmath14 is given by @xmath43 . in a one dimensional system , space and time play a symmetrical role . the dynamics can be described either in terms of spatial wave - packet _ evolution _ ( evolution picture ) or in terms of temporal wave - packet _ propagation _ ( propagation picture ) . the first picture is the must common in quantum field theory and leads to equal - time commutation rules between the envelope fields @xmath39 and @xmath40 . in this section however , we chose to work in the propagation picture ( which is usual in nonlinear optics ) in order to get a closer correspondence between quantum ( [ qnls ] ) and classical ( [ nls ] ) equations . in this case equal - space commutation rules ( [ commut1 ] ) must be imposed to @xmath39 and @xmath40 @xcite . ( in contrast , the evolution picture is used in the appendix [ app : a ] to derive the stochastic equations of sec . [ sec:3 ] . ) in the cw regime , the basics of the quantum theory of s - mi can be understood by performing a first - order perturbation analysis of the steady state solution of eq . ( [ qnls ] ) . because the pump field contains a large number of photons one can treat it as a classical coherent wave . using this approximation the steady - state solution is just the classical one : @xmath23 . the disturbed field can be written as : @xmath44 the disturbance operator @xmath45 is defined by eq . ( [ perturb ] ) . it satisfies the same commutation rule as @xmath46:@xmath47= \hbar\omega_0 \delta(t - t')$ ] . injecting the ansatz ( [ perturb ] ) into eq . ( [ qnls ] ) one obtains a propagation equation for the disturbance @xmath45 . supposing that the disturbance is small ( non depleted pump approximation ) this equation can be linearized and solved analytically in the fourier domain @xcite . using this method one finds that the quantum and classical theory predict the same frequency dependence of the parametric gain . in contrast the quantum and classical theory differ in that they do not predict the same growth law for stokes and anti - stokes photon numbers . one easily finds that the quantum counterparts of eqs . ( [ s - micl ] ) are [ s - miqu ] @xmath48 \sinh^2(\gamma p_0l),\\ n_a(l)&=&n_a(0 ) \cosh^2(\gamma p_0 l)\nonumber\\ & \phantom{=}&+[n_s(0)+1]\sinh^2(\gamma p_0l),\end{aligned}\ ] ] where @xmath32 and @xmath33 stand now for the _ expectation values _ of the photon number operators : @xmath49 , @xmath50 . these equations show that s - mi can be observed even in absence of any classical input noise . setting @xmath51 in eqs . ( [ s - miqu ] ) gives the number of stokes and anti - stokes photons produced by the sole action of vacuum fluctuations : @xmath52 eqs . ( [ s - miqu ] ) hold only for perfectly phase - matched photons at frequencies @xmath2 and @xmath3 and uncorrelated initial noise but can be easily generalized to get round these restrictions . although eqs . ( [ s - miqu ] ) and ( [ s - mivf ] ) give a good insight into the physics of vacuum - fluctuations induced s - mi and photon pairs generation , they are not suitable for quantitative predictions in the pulsed regime . this is because the effective value of the pump power @xmath22 depends on pump pulse shape , duration , and spectral width . furthermore the energy spectral density of stokes and anti - stokes waves deduced from ( [ s - mivf ] ) is highly dependent on the precise definition of modes . this will be discussed in sec . [ sec:4b ] . in the next sections we will show how to get around these difficulties by introducing the snlse and solving it numerically . one can go beyond the perturbation analysis of s - mi by solving the qnlse ( [ qnls ] ) numerically . such a plan could seem cumbersome because eq . ( [ qnls ] ) is a field - operator equation . the problem can be bypassed by converting the operator equation ( [ qnls ] ) into c - number equations . this can be performed by choosing a representation for the electromagnetic field . in this article , we will use the positive @xmath53-representation ( @xmath54 ) introduced by drummond and gardiner @xcite . the c - number equations obtained in this way are not standard deterministic partial derivative equations but stochastic ( langevin - type ) ones . using the @xmath54 representation , it can be shown @xcite that the qnlse ( [ qnls ] ) is equivalent to the following set of stochastic equations : [ snls ] @xmath55a(z , t)+\sqrt{i\gamma \hbar \omega_0}~\zeta_1(z , t)~a(z , t),\\ \frac{\partial a^\dag}{\partial z}+\frac{1}{v_g}\frac{\partial a^\dag}{\partial t}&= & + i\frac{\beta_2}{2}\frac{\partial^2a^\dag}{\partial t^2}-i\gamma [ a^\dag(z , t ) a(z , t)]a^\dag(z , t ) + \sqrt{-i\gamma \hbar \omega_0}~\zeta_2(z , t)~a^\dag(z , t).\phantom{mm}\end{aligned}\ ] ] eqs . ( [ snls]a ) and ( [ snls]b ) look like the classical nlse ( [ nls ] ) and its complex conjugated , except for the last terms which accounts for vacuum fluctuations . these last terms contains two independent zero - mean gaussian white noise random fields @xmath56 and @xmath57 characterized by the following second order moments : @xmath58 with @xmath59 . because the random fields @xmath60 and @xmath61 are not complex conjugated of each other , the envelope fields @xmath11 and @xmath62 are only complex conjugated `` in mean '' and have to be treated as different mathematical objects . ( [ snls ] ) can be solved on a computer . the numerical methods used for this task will be briefly explained in sec . [ sec:4a ] . note that solving eqs . ( [ snls ] ) gives a single realization of the stochastic process . in order to calculate the expectation value of a quantum observable , a statistical average on many realizations is required . thus one generates a large number of realizations @xmath63}(z , t),a^\dag_{[n]}(z , t))$ ] , @xmath64 . in order to calculate the expectation value of a quantum observable one then carries out a statistical average over the many realizations . for example , the expectation value of the energy spectral density of a pulse is given by : @xmath65}^\dag(z,\omega)\tilde{a}_{[n]}(z,\omega),\ ] ] where @xmath66 designates the fourier transform of the field @xmath67 and @xmath68 is the detuning from the pump angular frequency @xmath1 . in practice , a few realizations are enough when the number of photons per mode is high . however in a regime dominated by vacuum fluctuations hundreds of realizations are typically required to get precise values . a comparison between stokes and anti - stokes photon production predicted by the snlse ( [ snls ] ) and the analytical formulas ( [ s - miqu ] ) will be presented in sections [ sec:4b ] and [ sec:4c ] . the v - mi occurs in birefringent fibers . these fibers are characterized by different propagation constants @xmath69 and @xmath70 and different group velocities @xmath71 and @xmath72 for the @xmath73- and @xmath74-axis polarized modes . the numerical study of vacuum fluctuations induced v - mi requires an extension of eqs . ( [ snls ] ) that takes into account the phase mismatch parameter @xmath75 as well as the group - velocity mismatch @xmath76 . such extensions have been established in earlier works for the low - birefringence @xcite and the hight - birefringence @xcite limits . we generalized these results for an _ arbitrary _ level of birefringence and obtained the following stochastic coupled nonlinear schrdinger equations : [ scnls ] @xmath77 a_x + i\gamma b ( a_y)^2a_x^\dag e^{-2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{i\gamma \hbar \omega_0}\left[\zeta_1 a_x+\sqrt{b } \zeta_3a_ye^{-i\delta\beta_0 z}\right],\\ \frac{\partial a_x^\dag}{\partial z}+\frac{1}{v_{gx}}\frac{\partial a_x^\dag}{\partial t}&= & + i\frac{\beta_2}{2}\frac{\partial^2a_x^\dag}{\partial t^2 } -i\gamma \left[a_x^\dag a_x+(1-b)a_y^\dag a_y\right ] a_x^\dag -i\gamma b ( a_y^\dag)^2a_x e^{+2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{-i\gamma \hbar \omega_0}\left[\zeta_2 a_x^\dag+\sqrt{b } \zeta_4a_y^\dag e^{+i\delta\beta_0 z}\right],\\ \frac{\partial a_y}{\partial z}+\frac{1}{v_{gy}}\frac{\partial a_y}{\partial t}&= & -i\frac{\beta_2}{2}\frac{\partial^2a_y}{\partial t^2 } + i\gamma \left[a_y^\dag a_y+(1-b)a_x^\dag a_x\right ] a_y + i\gamma b ( a_x)^2a_y^\dag e^{+2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{i\gamma \hbar \omega_0}\left[\zeta_1 a_y-\sqrt{b } \zeta_3a_xe^{+i\delta\beta_0 z}\right],\\ \frac{\partial a_y^\dag}{\partial z}+\frac{1}{v_{gy}}\frac{\partial a_y^\dag}{\partial t}&= & + i\frac{\beta_2}{2}\frac{\partial^2a_y^\dag}{\partial t^2 } -i\gamma \left[a_y^\dag a_y+(1-b)a_x^\dag a_x\right ] a_y^\dag -i\gamma b ( a_x^\dag)^2a_y e^{-2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{-i\gamma \hbar \omega_0}\left[\zeta_2 a_y^\dag-\sqrt{b } \zeta_4a_x^\dag e^{-i\delta\beta_0 z}\right],\end{aligned}\ ] ] where @xmath78 and @xmath79 are stochastic envelope fields associated to the @xmath73- and @xmath74-axis modes respectively , and @xmath80 is a parameter that measures the strength of the nonlinear coupling between the @xmath73 and @xmath74 components . its value lies between 0 and 1 and depends on the nonlinearity mechanism . for silica fiber , we can set @xmath81 because the kerr non linearity has principally an electronic origin . four independent gaussian random fields @xmath82 are needed to reproduce the effect of vacuum fluctuations . they are characterized by the second order moments ( [ moments ] ) , as in the scalar case , with @xmath83 . the demonstration of this set of equations is outlined in the appendix [ app : a ] . note that , in contrast to the scalar case , a perturbation analysis does not lead to simple analytical formulas for vacuum - fluctuations induced v - mi . for this reason eqs . ( [ scnls ] ) constitute a valuable theoretical tool . we have developed a method for integrating numerically the snlse in the case where the pump is represented by a pulse of finite duration . our method is based on the split - step fourier ( ssf ) method @xcite . we have had to generalize the method in two ways . first of all the stochastic noise is modeled by including a noise term at each propagation step . second in the case of the v - mi we must alternate not only between the time and fourier domain , but also between the linear and circular polarization basis . switching from time to fourier domain is the basics of the ssf method : it permits to handle the time - derivatives in a simple way . similarly , it turns out that , whereas the terms with time - derivatives are easier to handle in the linear polarization basis , the @xmath16-dependent terms are better managed in the circular polarization basis . the quantum noise in the snlse ( [ scnls ] ) contains four independent real zero - mean gaussian white noise functions @xmath84 characterized by eq . ( [ moments ] ) . in the numerical method , time and space are discretized with respective discretization steps @xmath85 and @xmath86 . so each family of noise functions @xmath84 becomes a finite number of random variables @xmath87 $ ] chosen according to a zero - mean gaussian law of variance @xmath88 . the variance value is imposed by the normalization condition ( [ moments ] ) . the matrixes @xmath87 $ ] ( @xmath89 ) define the stochastic path of each realization . in contrast , when we will study the effect of classical noise we will add it once , at the beginning of the pulse propagation , to the spectral distribution of the signal @xmath90 . as we have previously indicated only the expectation values of observables have a physical meaning in the stochastic equations . from a numerical point of view this means that one must average the calculated quantities over several realizations of the stochastic path , and/or the classical input noise . usually , averaging over a hundred of realizations gives an uncertainty on the numerical results less than @xmath91 db in the non - zero gain frequency - range . finally we note that including the stochastic terms do not increase significantly the numerical complexity of a single realization . some sample spectra obtained using our algorithm are presented in fig . [ fig1 ] in the cases of s - mi ( fig . [ fig1]a ) , low birefringence v - mi ( fig . [ fig1]b ) , and high birefringence v - mi ( fig . [ fig1]c ) . the energy spectral density is plotted versus the frequency detuning from the pump . the physical parameters used in these simulations are listed in table [ table1 ] . in every simulation , the pump wave has been supposed to be an unchirped linearly polarized gaussian pulse with peak power @xmath22 and full - width - half - maximum duration @xmath92 . no classical noise has been added , and an average over 50 realizations of the stochastic process has been performed . .simulation parameters for fig . [ fig1]a - c . [ cols="^,^,^,^,^",options="header " , ] the frequency detuning of the side bands agrees with linear perturbation theory . ( in order to make the comparison easier we have indicated in fig . [ fig1]a - c the angular frequency shifts @xmath93 at which maximum gain is expected on the basis of the linear perturbation theory . ) moreover the snlse predict quantitatively the effect of vacuum fluctuations on the evolution of the energy spectral density of the electromagnetic field . in sec . [ sec:4b ] we will show that this evolution is also in very good agreement with linear perturbation theory when the number of stokes and anti - stokes photons generated is small enough . when the side bands are well - developed , the perturbation theory fails to predict correct values of @xmath94 . in contrast , the snlse algorithm still gives accurate results . in this limit numerical results can be easily confronted to experimental data . in @xcite , we report an experiment on high birefringence spontaneous v - mi in the anomalous dispersion regime that shows that the theoretical spectra from the snlse model tally with the experimental ones . it is interesting to note that the curves of fig . 1a look more noisy than those of fig . 1b although we have performed the same number of stochastic realizations in both cases . this is because the number of realizations needed to achieve a given precision on the expectation value of a quantum operator is a function of the relative value of the spatial step @xmath86 and the typical distance over which the nonlinear effects act ; both are different in the simulations of fig . [ fig1]a and fig . [ fig1]b . in practice , the smaller the spatial step , the fewer the number of realizations needed to achieve a given precision on expectation values . in fig . [ fig1 ] , 50 realizations are enough to estimate @xmath94 with an accuracy of about @xmath95 db . . ( a ) spontaneous s - mi in perfectly isotropic fiber in the anomalous dispersion regime . black , dark gray and light gray curves correspond to a propagation length @xmath35 = 500 , 1000 , 1500 m , respectively . the inset exhibits the pump spectral broadening due to the self - phase - modulation ( spm ) effect . ( b ) spontaneous v - mi in a slightly birefringent fiber in the normal dispersion regime . the pump is polarized along the slow axis , stokes and anti - stokes photons appear on the orthogonal axis . black , dark gray and light gray curves correspond to a propagation length @xmath35 = 16 , 24 , 32 m , respectively . ( c ) spontaneous v - mi in a strongly birefringent fiber in the normal dispersion regime . the pump polarization axis makes an angle of 45 degrees with the slow axis . stokes ( anti - stokes ) photons appear on the slow ( fast ) axis . black , dark gray and light gray curves correspond to a propagation length @xmath35 = 10 , 20 , 30 m , respectively.,title="fig:",width=309,height=510 ] + we also point out that the noise level visible at non phase - matched frequencies has no physical meaning . it can be lowered by averaging over a higher number of realizations . however , the number of realizations needed to achieve an accurate estimation of @xmath94 in the non phase - matched part of the spectrum is usually very high . when tractable , a linear perturbation analysis will be less time - consuming . nm , @xmath96 60 ps@xmath97 km@xmath98 , @xmath99 w@xmath98 km@xmath98 , @xmath100 ps , @xmath101 400 w , and @xmath102 40 m. the pump wave is polarized along the slow axis . the fiber beat - length was varied from 10 m to 5 cm . the values of @xmath103 and @xmath104 where deduced from eq . ( [ biref ] ) . the plot shows the maximum value of @xmath94 in the side bands as a function of @xmath103 . black squares ( gray squares ) correspond to a single ( double ) peak structure ( see the text).,title="fig:",width=309 ] + because our algorithm permits to investigate intermediate birefringence , we have also studied the effect of group velocity mismatch on the transition from low to high birefringence limits . to our knowledge , this transition has never been fully investigated before . [ fig2 ] shows the results of simulations with the same parameters as in fig . [ fig1]b except that the propagation length has been set to @xmath105 m and that the fiber beat length @xmath106 was varied from 10 m to 5 cm . the value of the maximal energy spectral density @xmath94 in the side bands is plotted versus the phase mismatch parameter . by lowering @xmath106 , we increase the value of the phase mismatch @xmath103 and the group - velocity mismatch @xmath104 according to the relations @xmath107 where @xmath108 is the vacuum speed of light . when the birefringence increases , the side bands move away from the pump spectrum and their amplitude decreases . subsequently the side bands acquire a double peak structure ( see fig . [ fig2 ] ) . this behavior is due to the walk - off of the produced stokes and anti - stokes photons . one easily shows that @xmath109 , where @xmath110 and @xmath111 are the stokes and anti - stokes photons group velocities , respectively . applying this formula to our simulation and taking @xmath112 m@xmath98 , one sees that the stokes and anti - stokes photons have walked 87.6 ps away while their fwhm duration is 100 ps . stokes and anti - stokes walk - off limits the coherent exponential amplification of quantum noise . the typical length scale over which the side bands growth takes place is given by @xmath113 . we point out that this analysis also hold for a cw - pump : the coherent amplification of the side bands stops when the walk - off of stokes and anti - stokes photons exceeds the coherence length of the pump . however , in the cw - case , stokes and anti - stokes photons generated in the first coherence length act as an input noise that will be amplified in the following coherence length . the process is reproduced as many times as the number of coherence lengths in the propagation distance . one usually argues @xcite that the weak birefringence phenomenology disappears because the coherent - coupling terms in eqs . ( [ scnls ] ) ( those containing the factor @xmath114 $ ] ) average to zero when @xmath103 is high . this statement is equivalent to saying that @xmath93 tends to infinity . our analysis shows however that walk - off has an even stronger effect . until now we have not yet demonstrated that modeling vacuum - fluctuations induced mi using snlse predicts the correct values of @xmath94 . in subsections [ sec:4b ] we compare the absolute values of the energy spectral density at the maximum gain obtained using our program and the linear perturbation analysis . having clarified in this way the interpretation of the results of our numerical simulations we turn to a detailed comparison of the effect of classical and quantum noises . that is we compare the effects of classical noise in the initial conditions and the quantum noise added at each step of the integration . in our numerical simulations we have taken the pump laser to be a gaussian pulse without chirp . its instantaneous power and energy spectral density can be written [ inputsignal ] @xmath115 with @xmath116 . the numerical integration of the snlse provides us with the spectral density of energy @xmath117 at the end of the fiber , see eq . ( [ se ] ) . on the other hand the linear perturbation theory is based on small perturbation analysis around a continuous monochromatic pump . we would like to compare quantitatively the predictions of these two approaches . for definiteness we carry out this comparison in the case of scalar mi . we shall focus our investigation on the intensity of the sidebands at the peak of the mi gain ( @xmath118 ) in the two approaches when we modify the propagation length and as we modify the duration @xmath119 of the pulse . the linear perturbation theory is based on a continuous monochromatic pump . for this reason the theory predicts a _ rate _ of photon production per unit time . this suggests that if one takes the pump to be a pulse localized in time , the number of photons produced should be proportional to the pulse duration , all other parameters being kept constant . simulations based on snlse confirm this phenomenology . this is illustrated in fig . [ fig3]a , where the energy spectral density at @xmath93 is plotted as a function of @xmath120 for three different pulse durations . note however that the above argument is valid for square pulses but is not very satisfactory for gaussian ones . a better understanding of the origin of this scaling can be obtained by making appeal to the notion of mode and of heisenberg box . . in order to keep the maximum gain frequency constant we have kept the peak power @xmath22 constant in each figure . the horizontal axis is given in dimensionless units of @xmath120 where @xmath22 is either the power of the continuous pump wave in the linear approximation , or the peak power of the gaussian pump pulse . the top panel is plotted in the density of energy representation whereas the bottom panel is plotted in the number of photons - per - mode representation using the rescaling of eq . ( [ nomega ] ) . in both panels , the up - triangles , the circles and the down - triangles correspond to fwhm durations respectively equal to 4 ns , 1 ns and 0.25 ns . note that these three curves coincide perfectly in panel ( b ) , thereby showing the relevance of the rescaling ( [ nomega ] ) . in the bottom panel , the dash - dotted line results from the analytic solution eq . ( [ s - mivf ] ) . the dashed line corresponds to eq . ( [ s - mivf ] ) convoluted with the pump shape then rescaled according to eq . ( [ nomega ] ) , see the text . all other parameters are identical to those used in fig . [ fig1]a.,width=309 ] in order to introduce the notion of mode , recall that a temporal signal can be represented by a distribution in the time - frequency plane . but because of the time - frequency uncertainty relations , a point in this plane has no physical meaning . this problem is well known in signal processing where one usually thinks in terms of local time - frequency decompositions , using windowed fourier transforms ( wft ) or wavelet transforms @xcite . such a local time - frequency decomposition allows one to decompose a signal into orthogonal local functions , called _ modes_. these can be depicted as surface elements in the time - frequency plane . fourier - transform limited pulses , such as our pump pulse ( [ inputsignal ] ) , are represented by gaussian distributions on the time - frequency plane , called the wigner - ville distributions . this distribution is very similar to the wigner distribution used in quantum optics to represent a quantum state of light . in particular two different fourier - limited gaussian pulses sufficiently different in time or central frequency can be considered as quasi - orthogonal modes . a set of quasi non overlapping gaussian wigner - ville distributions can be taken as a base for the time - frequency decomposition of the field . one can visualize this modal base by imagining that the time - frequency plane is paved with adjacent elementary surface elements , called heisenberg boxes , roughly representing the area of gaussian quasi non overlapping wigner - ville distributions . the precise area of the heisenberg boxes is a matter of taste depending on how strong orthogonality is required . a usual convention is to take this area equal to @xmath121 . this set of ( quasi ) mode is convenient for our problem because , during the pulse propagation in the fiber , the uncertainty on the creation time of a photon is defined by the variance @xmath122 of the pump pulse , and implies an uncertainty on its frequency defined by the variance @xmath123 . more precisely , in the case of s - mi , the pump pulse ( [ inputsignal ] ) produces photons that occupy single heisenberg boxes located at the same time as the pump but at angular frequencies @xmath25 ( @xmath124 ) . formulas ( [ s - miqu ] ) and ( [ s - mivf ] ) of sec . [ sec:2b ] thus give the number of photons created in these stokes and anti - stokes modes . our numerical simulations provide us with the spectral distribution of energy @xmath125 . hence we need to reexpress this as the number @xmath126 of photons produced per mode of duration @xmath119 and spectral width @xmath127 : @xmath128 fig . [ fig3]b shows the effect of scaling the spectra according to eq . ( [ nomega ] ) . when expressed in terms of number of photons per modes the three curves of fig . [ fig3]a ( corresponding to three different pump durations ) come down to a single one in fig . [ fig3]b ( continuous line ) . this shows that the notion of mode helps in interpreting the results of the numerical integration . the number of produced photons does not depend of the pump duration : the pump duration just alters their time - frequency characteristics . the dash - dotted curve in fig . [ fig3]b corresponds to a direct application of eq . ( [ s - mivf ] ) . a discrepancy with the simulations based on the snlse can be noted . it is simply due to the fact that our choice of size of heisenberg boxes , hence the normalization factor in eq . ( [ nomega ] ) , is somewhat arbitrary . by taking the heisenberg boxes a bit bigger , one can put the continuous and dash - dotted curves of fig [ fig3]b in superposition . in the rest of the text , however , we maintain the normalization relation eq . ( [ nomega ] ) for clarity . there is another way to compare the results of snlse simulations to the linear perturbation theory that avoids the concept of modes . one first computes the power spectral density given by the quantum perturbation analysis on a monochromatic pump wave . then one convolutes this with the gaussian spectral distribution of the real pump pulse . this procedure gives a good approximation of the energy spectral density generated by the gaussian pump . the dashed curve of fig . [ fig3]b corresponds to the peak energy spectral density computed by this method and rescaled according to eq . ( [ nomega ] ) in order to be independent of the pulse duration . the agreement with the continuous curve is now much better . the origin of the small difference in slope in the exponential amplification regime still remains unclear . it may be due to the self - phase - modulation broadening of the pump spectrum , which is not taken into account by the linear perturbation analysis . as a conclusion we obtain a good quantitative agreement between the simulation based on snlse and the linear perturbation analysis . we have also shown that the only physical quantity that can be rigorously predicted by the quantum nonlinear propagation theory is the energy spectral density and that the concept of mode , although very useful , must be handled with care . especially formulas like ( [ s - miqu ] ) and/or ( [ s - mivf ] ) may only be used as a rough approximation tool because there is no objective way to define a time - frequency mode . from figs . [ fig3 ] it is clear that the spontaneous mi growth can be divided into two different stages . so long as the number of particles created by mode is less than one ( @xmath129 ) , one is in a quantum regime dominated by vacuum fluctuations . in contrast , for @xmath126 above 1 the modulation instability is amplified exponentially and quantum effects become negligible . we now investigate the transition between the quantum and classical regimes in presence of some _ classical _ noise . we have chosen to model this noise by modifying the initial conditions and adding a white noise to the amplitude of the pump pulse in the fourier domain . for definiteness and simplicity , we continue to focus on scalar modulation instability . the classical initial noise @xmath130 was chosen according to two criteria : ( i ) the noise must correspond to a random fluctuation of the pump amplitude in the time domain and ( ii ) the statistic of @xmath130 ( for each frequency ) must lead to a spectral density of energy @xmath131 constant as a function of the frequency . several noise definitions can meet these criteria . we have chosen to study two particular cases : the pure spectral phase noise @xmath132 and the gaussian noise @xmath133 where @xmath134 is a real constant and @xmath135 are independent real zero - mean gaussian white noise random fields . in our simulations @xmath136 where discretized and replaced with random quantities @xmath137 $ ] ( one for each discretized frequency ) drawn according a zero - mean gaussian law of variance 1 . we have compared s - mi spectra averaged over 50 realizations , obtained from both classical noises ( with quantum noise set to zero ) . both noises lead to equivalent results . the difference between the spectra is lower than 0.3 db on the full spectral span , which is less than the residual averaging noise ( see fig.[fig1]a ) , and therefore negligible . hereafter , the classical noise is set according to eq . ( [ classicnoise1 ] ) . it corresponds to a white flux of photons without any phase correlations between each frequency component . in fig . [ sanssto ] we compare the peak intensity of side bands in the case of vacuum - fluctuations induced mi and the ( unphysical ) case of mi induced by the classical noise alone . in the quantum regime where the number of particles per mode is small the two approaches differ strongly whereas in the exponential amplification regime they give similar results , although different classical noise levels give rise to different final number of photons . ) . the input noises in the classical situation correspond to the following amount of photons per mode : ( i ) @xmath138 ( dash - dotted line ) , ( ii ) @xmath139 ( dotted line ) , and ( iii ) 10 ( dashed line ) . the simulation parameters are those of fig . [ fig1]a.,width=309 ] a. the inset is a zoom corresponding to the the lowest @xmath140-parameters . it is drawn in a logarithmic - scale.,width=309 ] the simulations of fig . [ bruit1 ] take into account both classical and quantum noises . they illustrate the realistic situation when both noises compete for producing modulation instability . if the order of magnitude of the classical noise is such that there is less than one photon per mode , the quantum noise dominates and the curve is ( except for very small values of the gain ) indistinguishable from the case where there is no classical noise . on the other hand if the number of noise photons per mode is much larger than one , the classical noise dominates and the intensity of the sidebands is indistinguishable from purely classical situations depicted in fig . [ sanssto ] . these numerical results are consistent with eqs . ( [ s - miqu ] ) , in which the term 1 represents the contribution of the vacuum fluctuation . if @xmath32 or @xmath33 are higher than 1 , eqs . ( [ s - miqu ] ) tend to eqs . ( [ s - micl ] ) corresponding to the classical description . in summary we have shown that if one considers only the peak intensity of the sidebands , the quantum origin of the instability can only be seen in the regime where the number of photons per mode ( produced or initially present ) is small whereas when the number of photons per mode is large the effect of vacuum fluctuations is indistinguishable from that of classical noise . good quantitative agreement between the two approaches in the exponential amplification regime is obtained when the number of noise photons is @xmath139 per mode . note that other quantum effects , such as two mode squeezing , may be present in the regime where many photons are produced per mode , but exhibiting them requires looking at correlations between the two sidebands . we can now discuss the well - known trick which consists in introducing a half photon per mode into a classical simulation ( then removing it ) to simulate the spontaneous effects . to see how this works we compare eqs . ( [ s - micl ] ) which describe the mi induced by classical noise and eq . ( [ s - mivf ] ) which is derived from the quantum theory and gives the number of photons produced per mode by the action of vacuum fluctuations alone . now , if we introduce the same amount of stokes and anti - stokes noise photons @xmath141 in eqs . ( [ s - micl ] ) we find : @xmath142 where @xmath143 and @xmath144 denote respectively the classical approach and the quantum approach . taking @xmath145 , eq . ( [ comparison ] ) shows the agreement between the quantum predictions ( right hand side ) and the classical predictions ( left hand side ) . note that @xmath141 can be any real value ( except for 0 ) , hence the spontaneous growth of the number of photons per mode can be simulated with any number of initial classical photons if the pump depletion is neglected . we have compared , using numerical integration , the direct quantum approach based on the snlse ( without classical noise ) and the classical approach in which one first integrates the nlse with some initial classical noise then rescales the spectra according to the left side of eq . ( [ comparison ] ) . the discrepancy between both approaches is measured by the following ratio in db scale : @xmath146 applying eq . ( [ eq : ratio ] ) to the data reported in fig . [ sanssto ] , one founds that for @xmath120 higher than 0.2 , @xmath147 is constant for any classical noise amplitude : @xmath147=1.9 db @xmath1480.3 db . note that both noise definitions eq . ( [ classicnoise1 ] ) and eq . ( [ classicnoise2 ] ) lead to the same @xmath147 . moreover this results may be extended to @xmath149 lower than 0.2 by increasing the number of realizations . the origin of non - zero value of @xmath147 may come from the mode definition used in eq . ( [ nomega ] ) . indeed the numerical integration of snlse gives the physical spectral density of energy @xmath94 whereas the calculus trick left hand side of eq . ( [ comparison ] ) leads to values interpreted as a number of photons per mode which must be rescaled to give @xmath94 . in summary , in the case of pump pulses of finite duration , a full quantum treatment based on the snlse leads to a direct quantitative prediction whereas the calculus trick only gives a good approximation whose accuracy is dependent of the modes definition . we generalized to an arbitrary level of birefringence the stochastic nonlinear schrdinger equations describing the propagation of pulses through a nonlinear @xmath21 medium with linear birefringence and group - velocity dispersion , and developed numerical routines to compute them . because these stochastic equations are equivalent to quantum field operator equations , we used them to compute spontaneous ( or vacuum - fluctuations induced ) modulation instability spectra in various birefringence regimes , including weak , high but also intermediate birefringence which has not been studied so far . in particular we showed that the decline of the number of photons generated by the weak - birefringence v - mi when the birefringence increases , is attributable to the increase of the walk - off between stokes and anti - stokes photons , although the weak - birefringence v - mi gain remains constant . we then investigated the absolute values of the energy spectral density at the maximum gain in the case of scalar modulation instabilities induced by vacuum fluctuations . we obtained good quantitative agreement between the simulation based on snlse and the linear perturbation analysis . then we have carried out a detailed comparison of the effect of classical and quantum noise and shown that the quantum origin of the instability can only be seen in the regime where the number of photons per mode produced or initially present is small . finally we note that the quantum nonlinear propagation theory predicts the energy spectral density and that the concept of mode , although very useful , must be handled with care . the present work forms the basis for numerical and experimental investigation of vacuum - fluctuations induced v - mi in regimes which have been little investigated so far , and we hope to report on this in the near future @xcite . although we have not developed this aspect in this article the stochastic equations can also be used to computed intensity correlations between side bands and predict special quantum effects like squeezing . for this reason the stochastic equations ( [ scnls ] ) are a valuable tool for computing quantum effects in birefringent nonlinear @xmath21 media , especially optical fibers . the stochastic model may also be adapted to include raman and brillouin effects ( see @xcite for the scalar case ) . higher order dispersion effects can also be included in a straightforward way . we would like to thank marc haelterman , stephane coen and eric lantz for usefull discussions . we acknowledge financial support from the communaut franaise de belgique under arc 00/05- 251 , and from the iuap programme of the belgian government under grant no.v-18 . in order to derive the snlse ( [ scnls ] ) , we will proceed in three stages . first , we will establish the interaction hamiltonian that governs field evolution in a lossless , dispersive , and birefringent fiber . we will only present an heuristic derivation of this hamiltonian and put the emphasis on appropriate approximations . a rigorous derivation requires a discussion of electromagnetic field quantization in material media @xcite , which is outside the scope of this article . second , we will use this hamiltonian to find the liouville equation describing the evolution of the density operator of the field and convert it into a fokker - planck equation using the @xmath54-representation . finally , we will establish the connection between the fokker - planck equation and the stochastic equations ( [ scnls ] ) . in a dispersive birefringent medium the positive - frequency part of the @xmath150-polarized electric field component ( @xmath151 ) can be decomposed on monochromatic modes in the following way @xcite : @xmath152 eq . ( [ a1 ] ) can be seen as defining @xmath153 . the operators @xmath153 and its hermitic conjugated @xmath154 are , respectively , the annihilation and creation operators for a @xmath150-polarized photon propagating in the fiber with a propagation constant @xmath155 and having an angular frequency @xmath156 . according to the canonical quantization , they satisfy the commutation rule @xmath157=\delta(\beta-\beta').\ ] ] in eq . ( [ a1 ] ) , @xmath158 and @xmath159 are respectively the linear index of refraction and group velocity corresponding to the @xmath150-polarized monochromatic mode with frequency @xmath156 , and @xmath160 the operator representing the @xmath150-polarized electric field component is given by @xmath161 where @xmath162^\dag$ ] is the negative - frequency part of @xmath163 . the total hamiltonian @xmath164 representing the sum of the vacuum electromagnetic energy and the dielectric energy stored in the fiber can be decomposed in a linear part @xmath165 and a nonlinear one @xmath166 . the linear part , @xmath167 takes into account the free - field energy and the energy stored into the dielectric through linear interactions , including the effects of _ linear _ dispersion and _ linear _ birefringence through the dispersion relations @xmath168 . if the field bandwidth is narrow compared to the central angular frequency @xmath1 , dispersion can be neglected in the @xmath21 interactions and the nonlinear part of the hamiltonian can be written @xmath169 where @xmath170 stands for @xmath171 . this simplified hamiltonian only takes into account the kerr effect , which is the dominant one for quasi - monochromatic fields . since the medium is supposed lossless , the @xmath21 tensor has the full permutation symmetry @xcite . the degeneracy factor @xmath172 takes this symmetry into account , by counting the number of way to permute the frequency arguments and the indexes of the @xmath21 tensor . a further useful approximation is to consider that the @xmath21 process is isotropic @xcite : @xmath173 since the field bandwidth is supposed narrow , the permutation symmetry also requires that @xmath174 : @xmath175 using eq . ( [ a6 ] ) , the nonlinear hamiltonian becomes another consequence of the narrow - bandwidth assumption is that the square - rooted bracket in eq . ( [ a1 ] ) can be taken out of the integral and one can write @xmath178 where @xmath179 in eqs . ( [ a8 ] ) and ( [ a9 ] ) , @xmath180 , @xmath181 , and @xmath182 stand respectively for the index of refraction , the group - velocity , and the propagation constant at frequency @xmath1 on the @xmath150-axis . the operator @xmath183 is an envelope operator because fast oscillations in space and time have factored out . this implies that @xmath183 is explicitly time - dependent in the schrodinger picture . the operator @xmath184 represents the number of @xmath150-polarized photons in @xmath185 $ ] . one can easily check that @xmath186=\delta_{ss'}\delta(z - z').\ ] ] using eq . ( [ a8 ] ) , @xmath166 takes the following simple form : where @xmath188 let s note that we have set @xmath189 and @xmath190 . the linear hamiltonian @xmath165 can also be expressed in function of the operators @xmath191 , @xmath151 , by developing @xmath156 in a taylor expansion around @xmath182 up to the second order , @xmath192 where @xmath193 and @xmath194 . using eq . ( [ a13 ] ) and inverting eq . ( [ a9 ] ) , one finds that @xmath195 , where @xmath196 and @xmath197dz.\ ] ] in the heisenberg picture , the hamiltonian @xmath198 is responsible of a free oscillation @xmath199 of the fields @xmath200 ( @xmath151 ) . this oscillation will cancel out the explicit oscillation @xmath201 already present in the definition ( [ a9 ] ) . for this reason , we prefer to continue the discussion in the interaction picture : @xmath202 to simplify the notations we will drop the @xmath203 index in later equations . in the quantized theory , the state of the electromagnetic field is represented by the density operator @xmath204 . its evolution , in the interaction picture , is given by the liouville equation @xmath205,\ ] ] where @xmath206 is the hamiltonian defined by eq . ( [ a18 ] ) . using eqs . ( [ a16 ] ) and ( [ a12 ] ) , the calculation of the right - hand side of eq . ( [ a19 ] ) is straightforward , so we do not write it here explicitly . in order to obtain stochastic equations from the liouville equation ( [ a19 ] ) , we generalized the argument of drummond and gardiner @xcite for monomode fields and their extension to multimode scalar fields given is @xcite . we introduce the multimode coherent states @xmath207 defined as the eigenstates of the annihilation operators @xmath153 @xmath208 as a consequence , @xmath207 are also eigenstates of the envelope operator ( [ a17 ] ) @xmath209 with @xmath210 this suggest the alternative notation @xmath211 , with @xmath212 , for the multimode coherent state . the basic idea of @xmath54-representation is to expand the density operator on nondiagonal coherent state projection operators defined as @xmath213 where @xmath214 is a new set of fields different from @xmath215 . denoting @xmath216 , this expansion can be written in the following way : @xmath217 where the integration measure @xmath218 means that the integration is carried over all the possible fields @xmath219 and @xmath220 , @xmath151 . taking into account the definition ( [ a20 ] ) one can show that , the @xmath53-function always exist and is positive for any density operator . the @xmath53-function is useful for calculating normal ordered moments : @xmath224 in particular , eq . ( [ a22b ] ) shows that @xmath53 is normalized to unity : @xmath225 . it can be interpreted as a genuine probability density on the ( infinite dimensional ) space sustained by the field @xmath226 and @xmath227 . where summation over @xmath229 and @xmath230 is implied . the @xmath231 s are the components of a four - dimension drift vector @xmath232 , with @xmath233 the @xmath234 , @xmath235 , and @xmath236 components have a similar form . @xmath234 is obtained from ( [ a24 ] ) by making the substitution ( i ) @xmath237 , @xmath238 , and @xmath239 . @xmath235 is obtained by ( ii ) exchanging @xmath73- and @xmath74-indexes in ( [ a24 ] ) , and making the subtitution @xmath240 . to obtain @xmath236 , both substitutions ( i ) and ( ii ) must be performed . the @xmath241 are the elements of a symmetric diffusion matrix @xmath242 that can be written in the form @xmath243 , where @xmath244 using eqs . ( [ a22 ] ) , on can deduce from eq . ( [ a23 ] ) that the @xmath245 verifies a functional fokker - planck equation with a semi positive - definite diffusion matrix . we refer to @xcite for a demonstration since eq . ( [ a23 ] ) has the same structure as eq . ( 4.19 ) in @xcite . because of the semipositivity of the diffusion matrix , the positivity of @xmath53 is maintained during evolution . the stochastic equations equivalent to the fokker - planck equation for @xmath53 can be written in the following compact form : @xmath246 where @xmath83 , and @xmath247 are the independent zero - mean gaussian white noise random fields introduced in sec . [ sec:3 ] and characterized by the second order moments ( [ moments ] ) . the @xmath232 vector gives the deterministic evolution of the fields as predicted by the classical theory of light . the way vacuum fluctuations modify the classical evolution is determined by the structure of the @xmath248 matrix . if one discards the stochastic terms , the fields @xmath249 and @xmath250 appear to be just complex conjugated of each other . however , when vacuum fluctuations are taken into account , @xmath249 and @xmath250 must be treated as independent fields that are only complex conjugate in mean . as they stand , eqs . ( [ a26 ] ) seems to differ from eqs . ( [ scnls ] ) . actually , both writings are equivalent . to highlight the equivalence we first notice that , according to the instantaneous - power normalization of the @xmath251 fields , one has the following relations : [ a29 ] @xmath255 a_x + i\gamma b ( a_y)^2a_x^\dag e^{-2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{i\gamma \hbar \omega_0}\left[\zeta_1 a_x+\sqrt{b } \zeta_3a_ye^{-i\delta\beta_0 z}\right],\\ \frac{\partial a_x^\dag}{\partial z}+\frac{1}{v_{gx0}}\frac{\partial a_x^\dag}{\partial t}&= & -i\frac{\omega''_x}{2v_{gx0}}\frac{\partial^2a_x^\dag}{\partial z^2 } -i\gamma \left[a_x^\dag a_x+(1-b)a_y^\dag a_y\right ] a_x^\dag -i\gamma b ( a_y^\dag)^2a_x e^{+2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{-i\gamma \hbar \omega_0}\left[\zeta_2 a_x^\dag+\sqrt{b } \zeta_4a_y^\dag e^{+i\delta\beta_0 z}\right],\\ \frac{\partial a_y}{\partial z}+\frac{1}{v_{gy0}}\frac{\partial a_y}{\partial t}&= & + i\frac{\omega''_y}{2v_{gy0}}\frac{\partial^2a_y}{\partial z^2 } + i\gamma \left[a_y^\dag a_y+(1-b)a_x^\dag a_x\right ] a_y + i\gamma b ( a_x)^2a_y^\dag e^{+2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{i\gamma \hbar \omega_0}\left[\zeta_1 a_y-\sqrt{b } \zeta_3a_xe^{+i\delta\beta_0 z}\right],\\ \frac{\partial a_y^\dag}{\partial z}+\frac{1}{v_{gy0}}\frac{\partial a_y^\dag}{\partial t}&= & -i\frac{\omega''_y}{2v_{gy0}}\frac{\partial^2a_y^\dag}{\partial z^2 } -i\gamma \left[a_y^\dag a_y+(1-b)a_x^\dag a_x\right ] a_y^\dag -i\gamma b ( a_x^\dag)^2a_y e^{-2i\delta\beta_0z}\nonumber\\ & \phantom{=}&+\sqrt{-i\gamma \hbar \omega_0}\left[\zeta_2 a_y^\dag-\sqrt{b } \zeta_4a_x^\dag e^{-i\delta\beta_0 z}\right].\end{aligned}\ ] ] eqs . ( [ a29 ] ) differs from eqs . ( [ scnls ] ) only in the first term of the right member . for each axis , the group - velocity dispersion parameter is @xmath256 . when the typical pulse duration @xmath257 is such that @xmath258 is much bigger than the group - velocity @xmath181 , which is the common situation in fiber - optics , the following operator approximation holds @xmath259 inserting ( [ a30 ] ) into eqs . ( [ a29 ] ) , and noting that usually @xmath260 one obtains the stochastic equations ( [ scnls ] ) .

we have calculated the scattering lengths between the pseudoscalar meson and charmed triplet , sextet , and excited sextet baryon to the third order with the heavy baryon chiral perturbation theory . the chiral expansion of some pion and eta channels converges well .

1205.0467

up to now many heavy hadrons have been discovered experimentally . their inner structures and interactions attract much attention . some of them are speculated to be possible new hadron states beyond the traditional quark model . for example , the newly discovered @xmath0 states are treated as the @xmath1 molecular states @xcite . many other molecular states , such as those composed of @xmath2 , @xmath3 , are also proposed @xcite . whether there is attractive interaction between the particles is the most important condition to form molecular states . on the other hand , the hadron - hadron interaction may distort the conventional quark model spectrum through the coupled channel effects . for example , the bare charm - strange scalar meson lies around 2.4 gev according to the quark model calculation . experimentally the mass of the @xmath4 state was measured to be around 2.3 gev . the attractive interaction between the @xmath5 meson and kaon is essential to lower its mass through the coupling effect between the bare @xmath6 state and the @xmath7 continuum @xcite . there has been lots of research work on the strong interactions of the charmed or bottomed mesons , such as the lattice study , calculations with the chiral perturbation theory and so on @xcite . the charmed triplet ( @xmath8 ) , sextet ( @xmath9 ) and excited sextet ( @xmath10 ) baryons are relatively stable particles . the pair from the ground and corresponding excited sextet form a degenerate doublet in the heavy quark spin symmetry limit . they interact with other particles through the exchange of the pseudoscalar mesons in the low energy effective field theory . it is very important to study the strong interaction between the lightest pseudoscalar meson ( @xmath11 ) and charmed baryon . a physical observable such as the scattering amplitude can be expanded order by order with the explicit power counting in the heavy baryon chiral perturbation theory ( hb@xmath12pt ) . the inclusion of the nonanalytic corrections resulting from the loop diagrams would highly reduce the error of extraction in the lattice study @xcite , which is one of the motivations of the present investigation . in this work , we shall study the pseudoscalar meson and charmed baryon scattering lengths to the third order with hb@xmath12pt . we include the interaction of @xmath9 and @xmath10 explicitly for the @xmath13 scattering instead of absorbing their effects into the low energy constants ( lecs ) at higher order since the mass difference among the charmed baryons is small and the couplings between them can not be neglected . the situation is similar for the @xmath14 and @xmath15 cases . we express the results as power series in @xmath16 with the explicit power counting , where @xmath17 represents the mass and momentum of pseudoscalar goldstone bosons , the residual momentum of charmed baryons , the mass difference among charmed baryons , while @xmath18 is either the mass of charmed baryons or @xmath19 . this paper is organized as follows : in sec . [ sectmat ] , we list the hb@xmath12pt lagrangians of the pseudoscalar mesons and charmed baryons , with which we get the expressions of the @xmath20-matrices at thresholds . in sec . [ seclec ] , we estimate the lecs in the lagrangians . we present the numerical results and discussions in sec . [ secnum ] . [ seccon ] is a short conclusion . the average mass of the charmed triplet baryons @xmath21 , which provides the base when we refer to the mass difference in the following . the hb@xmath12pt lagrangians at the leading order read @xmath22 + \tr[\bar b_{6 } ( i v\cdot d-\delta_2 ) b_{6 } ] -\tr[\bar b_{6}^ * ( i v\cdot d-\delta_3 ) b_{6}^*]\nonumber\\ & & + 2 g_1 \tr(\bar b_6 s\cdot u b_6 ) + 2 g_2 \tr(\bar b_6 s\cdot u b_{\bar 3}+{\rm h.c . } ) + g_3 \tr(\bar b_{6 \mu}^ * u^\mu b_{6}+{\rm h.c.})\nonumber\\ & & + g_4 \tr(\bar b_{6 \mu}^ * u^\mu b_{\bar 3}+{\rm h.c . } ) + 2 g_5 \tr(\bar b_{6}^ * s\cdot u b_{6}^ * ) + 2 g_6 \tr(\bar b_{\bar 3 } s\cdot u b_{\bar 3 } ) , \label{lbphione}\end{aligned}\ ] ] where @xmath23 is the velocity of a slowly moving baryon , @xmath24 is the spin matrix , @xmath25 is the coupling of the @xmath26-vertex , and @xmath27 is the mass difference between the charmed baryons , @xmath28 the field notations are @xmath29 @xmath30,\quad u_\mu=\frac{i}{2 } \{\xi^\dagger , \partial_\mu \xi\},\quad \xi = \exp(i \frac{\phi}{2f } ) , \quad \chi_\pm = \xi^\dagger\chi\xi^\dagger\pm\xi\chi\xi,\quad \chi=\mathrm{diag}(m_\pi^2,\ , m_\pi^2,\ , 2m_k^2-m_\pi^2 ) , \label{uxidef}\ ] ] and the definition of @xmath10 is similar to that of @xmath9 . the covariant derivatives , @xmath31 , will generate the @xmath32-vertexes . the lagrangian at @xmath33 contains the counter terms and the recoil terms . the counter terms are constructed on the basis of the chiral and other symmetries and proportional to @xmath34 in eq . ( [ lbphitwo ] ) . the recoil terms are derived from the lagrangians of the leading order and proportional to @xmath35 . we list the relevant terms below , @xmath36\tr[\chi_+ ] + \bar c_1 \tr[\bar b_{\bar 3 } \tilde \chi_+ b_{\bar 3 } ] + \left(\bar c_2-\frac{2g_6 ^ 2+g_2 ^ 2}{4m_0}\right ) \tr[\bar b_{\bar 3 } v\cdot u~ v\cdot u b_{\bar 3}]\nonumber\\ & & + \left(\bar c_3-\frac{2g_6 ^ 2-g_2 ^ 2}{4m_0}\right ) \bar b_{\bar 3}^{ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{\bar 3,cd}\nonumber\\ & & + c_0 \tr[\bar b_6 b_6]\tr[\chi_+ ] + c_1 \tr[\bar b_6 \tilde \chi_+ b_6 ] + \left(c_2-\frac{2g_2 ^ 2+g_1 ^ 2}{4m_0}\right ) \tr[\bar b_6 v\cdot u~ v\cdot u b_6]\nonumber\\ & & + \left(c_3+\frac{2g_2 ^ 2-g_1 ^ 2}{4m_0}\right ) \bar b_6^{ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{6,cd } + c_4 \tr[\bar b_6 b_6 ] \tr [ v\cdot u~ v\cdot u ] \nonumber\\ & & -\tilde c_0 \tr[\bar b_6^ * b_6^*]\tr[\chi_+ ] -\tilde c_1 \tr[\bar b_6^ * \tilde \chi_+ b_6^ * ] -\left(\tilde c_2-\frac{g_5 ^ 2}{4m_0 } \right)\tr[\bar b_6^ * v\cdot u~ v\cdot u b_6^*]\nonumber\\ & & -\left(\tilde c_3-\frac{g_5 ^ 2}{4m_0 } \right)\bar b_6^{*ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{6,cd}^ * -\tilde c_4 \tr[\bar b_6^ * b_6^ * ] \tr [ v\cdot u~ v\cdot u ] , \label{lbphitwo}\end{aligned}\ ] ] where the traceless @xmath37 are defined as : @xmath38 $ ] . the lagrangian at @xmath39 also contains the recoil - term part , @xmath40 + \frac{2g_6 ^ 2-g_2 ^ 2}{8m_0 ^ 2}\bar b_{\bar 3}^{ab } v\cdot u_a^{~c}(iv\cdot d ) v\cdot u_b^{~d } b_{\bar 3,cd } + \frac{g_2 ^ 2}{8m_0 ^ 2}\tr[\bar b_{\bar 3 } v\cdot u~v\cdot u b_{\bar 3}]\delta_2\nonumber\\ & & -\frac{g_2 ^ 2}{8m_0 ^ 2}\bar b_{\bar 3}^{ab } v\cdot u_a^{~c } v\cdot u_b^{~d } b_{\bar 3,cd } \delta_2 + \frac{2g_2 ^ 2+g_1 ^ 2}{8m_0 ^ 2 } \tr[\bar b_6 v\cdot u ( iv\cdot d ) v\cdot u b_6]\nonumber\\ & & -\frac{2g_2 ^ 2-g_1 ^ 2}{8m_0 ^ 2 } \bar b_6^{ab } v\cdot u_a^{~c}(iv\cdot d ) v\cdot u_b^{~d } b_{6,cd } + \frac{g_1 ^ 2}{8m_0 ^ 2 } \tr[\bar b_6 v\cdot u~v\cdot u b_6]\delta_2 + \frac{g_1 ^ 2}{8m_0 ^ 2 } \bar b_6^{ab } v\cdot u_a^{~c } v\cdot u_b^{~d } b_{6,cd}\delta_2\nonumber\\ & & -\frac{g_5 ^ 2}{8m_0 ^ 2}\tr[\bar b_6^ * v\cdot u(iv\cdot d+\delta_3 ) v\cdot u b_6^ * ] -\frac{g_5 ^ 2}{8m_0 ^ 2}\bar b_6^{*ab } v\cdot u_a^{~c}(iv\cdot d+\delta_3 ) v\cdot u_b^{~d } b_{6,cd}^*. \label{lbphitrir}\end{aligned}\ ] ] we neglect the contributions from the finite counter - term part at @xmath39 as in refs . @xcite , and cancel the divergences of the loop diagrams with the following infinite part @xmath41 b_{\bar 3}\right ] -(\frac{5\alpha g_2 ^ 2 \delta_2}{2f^2 } -\frac{5\beta g_4 ^ 2 \delta_3}{8f^2 } ) l \tr[\bar b_{\bar 3 } \tilde \chi_+ b_{\bar 3 } ] + ( \frac{10\alpha g_2 ^ 2 \delta_2}{3f^2 } - \frac{5\beta g_4 ^ 2 \delta_3}{6f^2 } ) l \tr[\bar b_{\bar 3 } v\cdot u~ v\cdot u b_{\bar 3}]\nonumber\\ & & -(\frac{32\alpha g_2 ^ 2 \delta_2}{3f^2 } - \frac{8\beta g_4 ^ 2 \delta_3}{3f^2 } ) l \bar b_{\bar 3}^{ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{\bar 3,cd}\nonumber\\ & & + \frac{3}{2f^2 } l \tr\left[\bar b_6 [ v\cdot u , \tilde\chi_- ] b_6\right ] + ( \frac{\alpha g_2 ^ 2 \delta_2}{f^2 } + \frac{7\beta g_3 ^ 2 \delta_1}{24f^2 } ) l \tr[\bar b_6 \tilde \chi_+ b_6 ] + ( \frac{4\alpha g_2 ^ 2 \delta_2}{f^2}+\frac{5\beta g_3 ^ 2\delta_1}{2f^2})l \tr[\bar b_6 v\cdot u~ v\cdot u b_6]\nonumber\\ & & + ( \frac{8\alpha g_2 ^ 2 \delta_2}{f^2}-\frac{\beta g_3 ^ 2\delta_1}{f^2})l \bar b_6^{ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{6,cd } -(\frac{4\alpha g_2 ^ 2 \delta_2}{3f^2}+\frac{11\beta g_3 ^ 2\delta_1}{18f^2})l \tr[\bar b_6 b_6 ] \tr [ v\cdot u~ v\cdot u ] \nonumber\\ & & -\frac{3}{2f^2}l \tr\left[\bar b_6^*[v\cdot u , \tilde\chi_- ] b_6^*\right ] + ( \frac{7g_3 ^ 2\delta_1}{24f^2}+\frac{g_4 ^ 2 \delta_3}{4f^2 } ) l \tr[\bar b_6^ * \tilde \chi_+ b_6^ * ] + ( \frac{5g_3 ^ 2\delta_1}{2f^2}+ \frac{g_4 ^ 2 \delta_3}{f^2})l\tr[\bar b_6^ * v\cdot u~ v\cdot u b_6^*]\nonumber\\ & & -(\frac{g_3 ^ 2\delta_1}{f^2}-\frac{2 g_4 ^ 2 \delta_3}{f^2})l \bar b_6^{*ab } v\cdot u_a^{~c}~ v\cdot u_b^{~d } b_{6,cd}^ * -(\frac{11g_3 ^ 2\delta_1}{18f^2}+\frac{g_4 ^ 2 \delta_3}{3f^2 } ) l \tr[\bar b_6^ * b_6^ * ] \tr [ v\cdot u~ v\cdot u ] , \label{lbphitric}\end{aligned}\ ] ] where in the @xmath5 dimensional space - time , @xmath42 @xmath43 is the projection operator for the rarita - schwinger field , @xmath44 is the euler constant , @xmath45 is the energy scale . the scattering length @xmath46 is related to the threshold @xmath20-matrix @xmath47 by @xmath48 . at the leading order , only the @xmath32-vertex from the contact terms in @xmath49 contributes to the @xmath20-matrices at the threshold . at the second order , the corresponding @xmath32-vertex of @xmath50 contributes . at the third order , in addition to the contribution of @xmath51 , the @xmath20-matrices also receive the contribution from the loop diagrams consisting of the vertices in the leading order lagrangian . the nonvanishing loop diagrams are shown in fig . [ loopdiag ] . since there is no vertex like @xmath52 at the leading order , the charmed baryons in different representations do not appear in the diagrams ( i ) as intermediate states . but they appear in the diagrams ( ii ) through the axial coupling . we calculate the loop diagrams with dimensional regularization and the modified minimal subtraction scheme . we use the lecs in eq . ( [ lbphitric ] ) to cancel the divergence . at last we express the @xmath20-matrices in terms of @xmath53 rather than @xmath54 with the help of their relation in refs . @xcite . the isospin symmetry is explicitly kept throughout our calculation . so we only list the 49 isospin - independent @xmath20-matrices in the following subsections . we list the @xmath20-matrices for the pseudoscalar meson and @xmath55 charmed baryon scattering , separate them order by order with braces , and distinguish between the tree and loop diagram contribution at @xmath39 with square brackets , @xmath56 + [ -\frac{1}{16 } v\left(m_k^2,-m_\pi\right ) -\frac{1}{16 } v\left(m_k^2,m_\pi\right ) \right.\nonumber\\&&\left . -\frac{4}{9 } m_\pi^2 y_1(m_\eta)+2 m_\pi^2 w_1(m_\pi ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { \pi\xi_c}&=&\left\{\frac{m_\pi}{f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 48 \bar c_0 - 4 \bar c_1 - 3 \bar c_2)}{6 f_\pi^2}\right\}+\left\ { [ \frac{m_\pi^2 ( 16 \bar d_1 m_\pi+\bar d_2 \delta_2)}{2 f_\pi^2 m_0 ^ 2 } ] + [ \frac{1}{32 } v\left(m_k^2,-m_\pi\right)-\frac{3}{32 } v\left(m_k^2,m_\pi\right ) \right.\nonumber\\&&\left . -\frac{1}{4 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{2 } m_\pi^2 w_1(m_\eta)-\frac{1}{9 } m_\pi^2 y_1(m_\eta)+\frac{1}{2 } m_\pi^2 w_1(m_\pi)+m_\pi^2 y_1(m_\pi ) ] \right\ } , \nonumber\\ t^{(3/2 ) } _ { \pi\xi_c}&=&\left\{-\frac{m_\pi}{2 f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 48 \bar c_0 - 4 \bar c_1 - 3 \bar c_2)}{6 f_\pi^2}\right\}+\left\ { [ -\frac{m_\pi^2 ( 8 \bar d_1 m_\pi-\bar d_2 \delta_2)}{2 f_\pi^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_\pi\right ) \right.\nonumber\\&&\left . -\frac{3}{16 } v\left(m_\pi^2,-m_\pi\right)-\frac{1}{16 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{2 } m_\pi^2 w_1(m_\eta)-\frac{1}{9 } m_\pi^2 y_1(m_\eta)+\frac{1}{2 } m_\pi^2 w_1(m_\pi)+m_\pi^2 y_1(m_\pi ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { k\lambda_c}&=&\left\{-\frac{m_k}{2 f_k^2}\right\}+\left\{-\frac{m_k^2 ( 48 \bar c_0 - 4 \bar c_1 - 3 \bar c_2)}{6 f_k^2}\right\}+\left\ { [ -\frac{m_k^2 ( 8 \bar d_1 m_k-\bar d_2 \delta_2)}{2 f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{1}{16 } v\left(m_k^2,m_k\right)-\frac{3}{32 } v\left(m_\eta^2,-m_k\right)-\frac{3}{32 } v\left(m_\pi^2,-m_k\right)+\frac{8}{9 } m_k^2 y_1(m_\eta ) ] \right\ } , \nonumber\\ t^{(0 ) } _ { k\xi_c}&=&\left\{\frac{m_k}{f_k^2}\right\}+\left\{-\frac{m_k^2 ( 24 \bar c_0 - 8 \bar c_1 + 3 \bar c_3)}{3 f_k^2}\right\}+\left\ { [ \frac{m_k^2 ( 8 \bar d_1 m_k-\bar d_3 \delta_2)}{f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_k\right)-\frac{1}{4 } v\left(m_k^2,m_k\right ) \right.\nonumber\\&&\left . -\frac{3}{16 } v\left(m_\eta^2,m_k\right)+m_k^2 w_1(m_\eta)+\frac{2}{9 } m_k^2 y_1(m_\eta)+2 p_1 - 3 u_1 ] \right\ } , \nonumber\\ t^{(1 ) } _ { k\xi_c}&=&\left\{0\right\}+\left\{-\frac{m_k^2 ( 24 \bar c_0 + 4 \bar c_1 - 3 \bar c_2 - 3 \bar c_3)}{3 f_k^2}\right\}+\left\ { [ \frac{m_k^2 \delta_2 ( \bar d_2+\bar d_3)}{f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_k\right)-\frac{1}{16 } v\left(m_\pi^2,m_k\right ) \right.\nonumber\\&&\left . + m_k^2 w_1(m_\eta)+\frac{2}{9 } m_k^2 y_1(m_\eta)-\frac{2 p_1}{3}+u_1 ] \right\ } , \nonumber\\ t^{(0 ) } _ { \eta\lambda_c}&=&\left\{0\right\}+\left\{-\frac{24 \bar c_0 m_\eta^2 - 8 \bar c_1 m_\eta^2 + 4 \bar c_1 m_\pi^2-\bar c_2 m_\eta^2+\bar c_3 m_\eta^2}{3 f_\eta^2}\right\}+\left\ { [ \frac{m_\eta^2 \delta_2 ( \bar d_2-\bar d_3)}{3 f_\eta^2 m_0 ^ 2 } ] + [ -\frac{3}{16 } v\left(m_k^2,-m_\eta\right ) \right.\nonumber\\&&\left . -\frac{3}{16 } v\left(m_k^2,m_\eta\right)-\frac{64}{27 } m_k^2 y_1(m_\eta)+\frac{4}{3 } m_k^2 w_1(m_k)+\frac{8}{3 } m_k^2 y_1(m_k)+\frac{28}{27 } m_\pi^2 y_1(m_\eta)-2 m_\pi^2 w_1(m_\pi ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { \eta\xi_c}&=&\left\{0\right\}+\left\{-\frac{48 \bar c_0 m_\eta^2 + 8 \bar c_1 m_\eta^2 - 4 \bar c_1 m_\pi^2 - 5 \bar c_2 m_\eta^2 - 4 \bar c_3 m_\eta^2}{6 f_\eta^2}\right\}+\left\ { [ \frac{m_\eta^2 \delta_2 ( 5 \bar d_2 + 4 \bar d_3)}{6 f_\eta^2 m_0 ^ 2 } ] + [ -\frac{3}{32 } v\left(m_k^2,-m_\eta\right ) \right.\nonumber\\&&\left . -\frac{3}{32 } v\left(m_k^2,m_\eta\right)-\frac{8}{3 } m_k^2 w_1(m_\eta)-\frac{16}{27 } m_k^2 y_1(m_\eta)+\frac{10}{3 } m_k^2 w_1(m_k)+\frac{4}{3 } m_k^2 y_1(m_k)+\frac{7}{6 } m_\pi^2 w_1(m_\eta ) \right.\nonumber\\&&\left . + \frac{7}{27 } m_\pi^2 y_1(m_\eta)-\frac{1}{2 } m_\pi^2 w_1(m_\pi)-m_\pi^2 y_1(m_\pi ) ] \right\}\label{tphib3}\end{aligned}\ ] ] where the superscript @xmath57 in @xmath58 refers to the isospin of the channel , the functions @xmath59 , @xmath60 , @xmath61 , @xmath62 , and @xmath63 are listed in appendix [ secfunc ] , and some combination coefficients are defined as @xmath64 we have used the gell - mann - okubo mass relation @xmath65 to make the expression more concise . besides the eight @xmath20-matrices listed above , the other three isospin - independent ones can be written in terms of those in eq . ( [ tphib3 ] ) by crossing symmetry , @xmath66_{m_k\rightarrow -m_k},\quad t^{(1 ) } _ { \bar k\xi_c}=\frac12\left[t^{(1)}_{k\xi_c}+t^{(0)}_{k\xi_c}\right]_{m_k\rightarrow -m_k},\quad t^{(0 ) } _ { \bar k\xi_c}=\frac12\left[3t^{(1)}_{k\xi_c}-t^{(0)}_{k\xi_c}\right]_{m_k\rightarrow -m_k}.\quad\ ] ] there are 19 isospin independent @xmath20-matrices for the pseudoscalar meson and charmed sextet baryon scattering . @xmath67 + [ -\frac{1}{8 } v\left(m_k^2,-m_\pi\right)-\frac{1}{8 } v\left(m_k^2,m_\pi\right)-\frac{4}{9 } m_\pi^2 w_2(m_\eta ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { \pi\xi'_c}&=&\left\{\frac{m_\pi}{f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 48 c_0 - 4 c_1 - 3 c_2 - 12 c_4)}{12 f_\pi^2}\right\}+\left\ { [ \frac{m_\pi^2 ( 16 d_1 m_\pi+d_2 \delta_2)}{4 f_\pi^2 m_0 ^ 2 } ] + [ -\frac{3}{32 } v\left(m_k^2,-m_\pi\right ) \right.\nonumber\\&&\left . -\frac{7}{32 } v\left(m_k^2,m_\pi\right)-\frac{1}{4 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{36 } m_\pi^2 w_2(m_\eta)-\frac{1}{2 } m_\pi^2 y_2(m_\eta)+\frac{1}{4 } m_\pi^2 w_2(m_\pi)+\frac{1}{2 } m_\pi^2 y_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(3/2 ) } _ { \pi\xi'_c}&=&\left\{-\frac{m_\pi}{2 f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 48 c_0 - 4 c_1 - 3 c_2 - 12 c_4)}{12 f_\pi^2}\right\}+\left\ { [ -\frac{m_\pi^2 ( 8 d_1 m_\pi - d_2 \delta_2)}{4 f_\pi^2 m_0 ^ 2 } ] + [ -\frac{3}{16 } v\left(m_k^2,-m_\pi\right ) \right.\nonumber\\&&\left . -\frac{1}{8 } v\left(m_k^2,m_\pi\right)-\frac{3}{16 } v\left(m_\pi^2,-m_\pi\right)-\frac{1}{16 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{36 } m_\pi^2 w_2(m_\eta)-\frac{1}{2 } m_\pi^2 y_2(m_\eta)+\frac{1}{4 } m_\pi^2 w_2(m_\pi ) \right.\nonumber\\&&\left . + \frac{1}{2 } m_\pi^2 y_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(0 ) } _ { \pi\sigma_c}&=&\left\{\frac{2 m_\pi}{f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 24 c_0 + 4 c_1 - 3 c_2 + 9 c_3 - 6 c_4)}{6 f_\pi^2}\right\}+\left\ { [ \frac{m_\pi^2 ( 16 d_1 m_\pi+d_2 \delta_2 - 3 d_3 \delta_2)}{2 f_\pi^2 m_0 ^ 2 } ] \right.\nonumber\\&&\left . + [ \frac{1}{16 } v\left(m_k^2,-m_\pi\right)-\frac{3}{16 } v\left(m_k^2,m_\pi\right)-\frac{1}{2 } v\left(m_\pi^2,-m_\pi\right)-v\left(m_\pi^2,m_\pi\right)-\frac{1}{9 } m_\pi^2 w_2(m_\eta)+4 m_\pi^2 w_2(m_\pi ) \right.\nonumber\\&&\left . -6 m_\pi^2 y_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(1 ) } _ { \pi\sigma_c}&=&\left\{\frac{m_\pi}{f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 24 c_0 + 4 c_1 - 3 c_2 - 6 c_3 - 6 c_4)}{6 f_\pi^2}\right\}+\left\ { [ \frac{m_\pi^2 ( 8 d_1 m_\pi+d_2 \delta_2 + 2 d_3 \delta_2)}{2 f_\pi^2 m_0 ^ 2 } ] + [ -\frac{1}{8 } v\left(m_k^2,m_\pi\right ) \right.\nonumber\\&&\left . -\frac{1}{4 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{9 } m_\pi^2 w_2(m_\eta)-m_\pi^2 w_2(m_\pi)+4 m_\pi^2 y_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(2 ) } _ { \pi\sigma_c}&=&\left\{-\frac{m_\pi}{f_\pi^2}\right\}+\left\{-\frac{m_\pi^2 ( 24 c_0 + 4 c_1 - 3 c_2 - 6 c_4)}{6 f_\pi^2}\right\}+\left\ { [ -\frac{m_\pi^2 ( 8 d_1 m_\pi - d_2 \delta_2)}{2 f_\pi^2 m_0 ^ 2 } ] + [ -\frac{1}{8 } v\left(m_k^2,-m_\pi\right ) \right.\nonumber\\&&\left . -\frac{1}{2 } v\left(m_\pi^2,-m_\pi\right)-\frac{1}{4 } v\left(m_\pi^2,m_\pi\right)-\frac{1}{9 } m_\pi^2 w_2(m_\eta)+m_\pi^2 w_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { k\omega_c}&=&\left\{\frac{m_k}{f_k^2}\right\}+\left\{-\frac{m_k^2 ( 24 c_0 + 4 c_1 - 3 c_2 - 6 c_4)}{6 f_k^2}\right\}+\left\ { [ \frac{m_k^2 ( 8 d_1 m_k+d_2 \delta_2)}{2 f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{4 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{1}{4 } v\left(m_k^2,m_k\right)-\frac{3}{16 } v\left(m_\eta^2,m_k\right)-\frac{3}{16 } v\left(m_\pi^2,m_k\right)+\frac{8}{9 } m_k^2 w_2(m_\eta ) ] \right\ } , \nonumber\\ t^{(0 ) } _ { k\xi'_c}&=&\left\{\frac{m_k}{f_k^2}\right\}+\left\{-\frac{m_k^2 ( 24 c_0 - 8 c_1 + 3 c_3 - 6 c_4)}{6 f_k^2}\right\}+\left\ { [ \frac{m_k^2 ( 8 d_1 m_k - d_3 \delta_2)}{2 f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{1}{4 } v\left(m_k^2,m_k\right)+\frac{3}{16 } v\left(m_\eta^2,-m_k\right)-\frac{3}{16 } v\left(m_\pi^2,-m_k\right)-\frac{3}{16 } v\left(m_\pi^2,m_k\right)+\frac{1}{18 } m_k^2 w_2(m_\eta ) \right.\nonumber\\&&\left . + m_k^2 y_2(m_\eta)-3 p_2+\frac{u_2}{2 } ] \right\ } , \nonumber\\ t^{(1 ) } _ { k\xi'_c}&=&\left\{0\right\}+\left\{-\frac{m_k^2 ( 24 c_0 + 4 c_1 - 3 c_2 - 3 c_3 - 6 c_4)}{6 f_k^2}\right\}+\left\ { [ \frac{m_k^2 \delta_2 ( d_2+d_3)}{2 f_k^2 m_0 ^ 2 } ] + [ -\frac{1}{16 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{3}{16 } v\left(m_\eta^2,-m_k\right)-\frac{3}{16 } v\left(m_\eta^2,m_k\right)-\frac{1}{16 } v\left(m_\pi^2,-m_k\right)-\frac{1}{8 } v\left(m_\pi^2,m_k\right)+\frac{1}{18 } m_k^2 w_2(m_\eta ) \right.\nonumber\\&&\left . + m_k^2 y_2(m_\eta)+p_2-\frac{u_2}{6 } ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { k\sigma_c}&=&\left\{\frac{m_k}{2 f_k^2}\right\}+\left\{-\frac{m_k^2 ( 48 c_0 - 28 c_1 + 3 c_2 - 12 c_4)}{12 f_k^2}\right\}+\left\ { [ \frac{m_k^2 ( 8 d_1 m_k - d_2 \delta_2)}{4 f_k^2 m_0 ^ 2 } ] + [ \frac{3}{16 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{1}{16 } v\left(m_k^2,m_k\right)+\frac{3}{32 } v\left(m_\eta^2,-m_k\right)-\frac{5}{32 } v\left(m_\pi^2,-m_k\right)+\frac{2}{9 } m_k^2 w_2(m_\eta)-\frac{4 u_2}{3 } ] \right\ } , \nonumber\\ t^{(3/2 ) } _ { k\sigma_c}&=&\left\{-\frac{m_k}{f_k^2}\right\}+\left\{-\frac{m_k^2 ( 24 c_0 + 4 c_1 - 3 c_2 - 6 c_4)}{6 f_k^2}\right\}+\left\ { [ -\frac{m_k^2 ( 8 d_1 m_k - d_2 \delta_2)}{2 f_k^2 m_0 ^ 2 } ] + [ -\frac{3}{8 } v\left(m_k^2,-m_k\right ) \right.\nonumber\\&&\left . -\frac{1}{4 } v\left(m_k^2,m_k\right)-\frac{3}{16 } v\left(m_\eta^2,-m_k\right)-\frac{1}{16 } v\left(m_\pi^2,-m_k\right)+\frac{2}{9 } m_k^2 w_2(m_\eta)+\frac{2 u_2}{3 } ] \right\ } , \nonumber\\ t^{(0 ) } _ { \eta\omega_c}&=&\left\{0\right\}+\left\{-\frac{12 c_0 m_\eta^2 + 8 c_1 m_\eta^2 - 4 c_1 m_\pi^2 - 2 c_2 m_\eta^2 - 2 c_3 m_\eta^2 - 3 c_4 m_\eta^2}{3 f_\eta^2}\right\}+\left\ { [ \frac{2 m_\eta^2 \delta_2 ( d_2+d_3)}{3 f_\eta^2 m_0 ^ 2 } ] + [ \right.\nonumber\\&&\left . -\frac{3}{8 } v\left(m_k^2,-m_\eta\right)-\frac{3}{8 } v\left(m_k^2,m_\eta\right)-\frac{64}{27 } m_k^2 w_2(m_\eta)+\frac{4}{3 } m_k^2 w_2(m_k)+\frac{8}{3 } m_k^2 y_2(m_k)+\frac{28}{27 } m_\pi^2 w_2(m_\eta ) ] \right\ } , \nonumber\\ t^{(1/2 ) } _ { \eta\xi'_c}&=&\left\{0\right\}+\left\{-\frac{48 c_0 m_\eta^2 + 8 c_1 m_\eta^2 - 4 c_1 m_\pi^2 - 5 c_2 m_\eta^2 + 4 c_3 m_\eta^2 - 12 c_4 m_\eta^2}{12 f_\eta^2}\right\}+\left\ { [ \frac{m_\eta^2 \delta_2 ( 5 d_2 - 4 d_3)}{12 f_\eta^2 m_0 ^ 2 } ] + [ \right.\nonumber\\&&\left . -\frac{15}{32 } v\left(m_k^2,-m_\eta\right)-\frac{15}{32 } v\left(m_k^2,m_\eta\right)-\frac{4}{27 } m_k^2 w_2(m_\eta)-\frac{8}{3 } m_k^2 y_2(m_\eta)+\frac{5}{3 } m_k^2 w_2(m_k)+\frac{2}{3 } m_k^2 y_2(m_k ) \right.\nonumber\\&&\left . + \frac{7}{108 } m_\pi^2 w_2(m_\eta)+\frac{7}{6 } m_\pi^2 y_2(m_\eta)-\frac{1}{4 } m_\pi^2 w_2(m_\pi)-\frac{1}{2 } m_\pi^2 y_2(m_\pi ) ] \right\ } , \nonumber\\ t^{(1 ) } _ { \eta\sigma_c}&=&\left\{0\right\}+\left\{-\frac{24 c_0 m_\eta^2 - 8 c_1 m_\eta^2 + 4 c_1 m_\pi^2-c_2 m_\eta^2-c_3 m_\eta^2 - 6 c_4 m_\eta^2}{6 f_\eta^2}\right\}+\left\ { [ \frac{m_\eta^2 \delta_2 ( d_2+d_3)}{6 f_\eta^2 m_0 ^ 2 } ] + [ \right.\nonumber\\&&\left . -\frac{3}{16 } v\left(m_k^2,-m_\eta\right)-\frac{3}{16 } v\left(m_k^2,m_\eta\right)-\frac{16}{27 } m_k^2 w_2(m_\eta)+\frac{2}{3 } m_k^2 w_2(m_k)+\frac{4}{3 } m_k^2 y_2(m_k)+\frac{7}{27 } m_\pi^2 w_2(m_\eta ) \right.\nonumber\\&&\left . -\frac{2}{3 } m_\pi^2 w_2(m_\pi)-\frac{2}{3 } m_\pi^2 y_2(m_\pi ) ] \right\ } , \label{tphib6}\end{aligned}\ ] ] where @xmath68 moreover , with crossing symmetry we get @xmath69_{m_k\rightarrow -m_k},\quad t^{(1/2 ) } _ { \bar k\sigma_c}=\frac13\left[4t^{(3/2)}_{k\sigma_c}-t^{(1/2)}_{k\sigma_c}\right]_{m_k\rightarrow -m_k},\quad t^{(1/2 ) } _ { \bar k\omega_c}=\left[t^{(1/2 ) } _ { k\omega_c}\right]_{m_k\rightarrow -m_k},\nonumber\\ & & t^{(1 ) } _ { \bar k\xi^\prime_c}=\frac12\left[t^{(1)}_{k\xi^\prime_c}+t^{(0)}_{k\xi^\prime_c}\right]_{m_k\rightarrow -m_k},\quad t^{(0 ) } _ { \bar k\xi^\prime_c}=\frac12\left[3t^{(1)}_{k\xi^\prime_c}-t^{(0)}_{k\xi^\prime_c}\right]_{m_k\rightarrow -m_k}.\quad \label{tphib6c}\end{aligned}\ ] ] there are also 19 independent @xmath20-matrices for the scattering between pseudoscalar meson and the excited charmed sextet baryon . even including the loop correction , we notice that @xmath70 is the same as @xmath71 in the heavy baryon symmetry limit . taking into account of the heavy baryon symmetry breaking effect , we obtain @xmath72 after making the following replacements in the expressions of the corresponding @xmath71 in eqs ( [ tphib6],[tphib6c ] ) @xmath73 we have listed the @xmath20-matrices of the pseudoscalar meson and charmed baryon scattering in the above three subsections . we have assumed the su(3 ) flavor symmetry and taken the su(3 ) breaking effect into account perturbatively . one can also study the scattering of @xmath74 , @xmath75 , and @xmath76 with su(2 ) flavor symmetry . we can construct the relevant lagrangians with exact su(2 ) chiral symmetry from the beginning . alternatively , we can extract su(2 ) lagrangians from the su(3 ) lagrangians in eqs . ( [ lphiphi ] ) , ( [ lbphione ] ) , ( [ lbphitwo ] ) , ( [ lbphitrir ] ) , and ( [ lbphitric ] ) . now the coupling constants @xmath25 and other lecs are different from those in the su(3 ) case . then we can obtain the su(2 ) @xmath20-matrices from the su(3 ) ones . more specially , we may drop the terms proportional to @xmath77 , @xmath78 , and @xmath79 in the su(3 ) @xmath80 , and replace @xmath81 and other lecs with new independent ones @xmath82 , @xmath83 , @xmath84 etc . after that we get the su(2 ) @xmath80 . the contributions of the dropped terms actually are absorbed by the redefined lecs at @xmath39 . unfortunately , the investigation of the @xmath85 scattering with su(2 ) chiral perturbation theory introduces more independent lecs . especially the su(2 ) lecs at @xmath39 can not be neglected . we do not include the contribution from the kaon and eta explicitly , which contribute to @xmath39 lecs here . in the following we will concentrate on the su(3 ) case only . similar to the nucleon case @xcite , the chiral correction to the charmed baryon axial - vector coupling would also be @xmath33 , which contributes to the @xmath20-matrices at @xmath86 or higher order , thus can be neglected . using @xmath87 and @xmath88 obtained by fitting the decay widths of @xmath89 and @xmath90 @xcite , @xmath91 with the quark model symmetry , and @xmath92 with heavy quark spin symmetry , we have @xmath93 we also need @xcite @xmath94 since there are no available experimental data to extract the low energy constants at @xmath33 , we utilize the crude su(4 ) flavor symmetry to make a rough estimate of some of these lecs in appendix [ secsufour ] , @xmath95 we would assume @xmath96 with the heavy quark spin symmetry in the numerical calculation . as for the dimensionless lec @xmath97 , we will take it to be in the natural range of [ -1,1 ] as in ref . we list the @xmath20-matrices order by order for the pseudoscalar meson and charmed baryon scattering in tables [ tabtphib3 ] , [ tabtphib6 ] , [ tabtphib6s ] . the positive real parts of the scattering lengths indicate that there exists the attractive interaction in the following channels : @xmath98 , @xmath99 , @xmath100 , @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 , @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 , @xmath116 , @xmath117 , @xmath118 , @xmath119 , @xmath120 , @xmath121 , @xmath122 , @xmath123 , @xmath124 , @xmath125 , @xmath126 , @xmath127 , @xmath128 , @xmath129 , @xmath130 , and @xmath131 , where the superscripts refer to the isospin . .the @xmath11-@xmath132 threshold @xmath20-matrices order by order in units of fm with @xmath133 . [ cols="^,^,^,^,^,^,^ " , ] it is interesting to notice that the inclusion of the @xmath10 intermediate states does not make the convergence better . let s denote the contribution of the intermediate particle @xmath134 to the @xmath20-matrix through the axial couplings in the heavy quark symmetry limit as @xmath135 . for the @xmath136 scattering , we can get the following ratio from eqs . ( [ tphib3],[puydef ] ) @xmath137 for the @xmath138 scattering , @xmath139 and for the @xmath140 scattering , the ratio is @xmath141 one notices that @xmath142 is larger than @xmath143 for the @xmath136 and @xmath140 scattering , while it is smaller than @xmath143 for the @xmath138 scattering . the correction from the @xmath10 and @xmath9 states has the same sign as required from heavy quark symmetry . their contribution is constructive , which worsens the convergence . from tables [ tabtphib6 ] and [ tabtphib6s ] , one notices that the numerical value of @xmath144 is very close to that of the corresponding @xmath145 at every order . as can be seen in table [ tabtphib3trin ] , the contribution of the sum of all the loop diagrams to @xmath144 and @xmath145 is almost the same , which is the manifestation of the heavy flavor symmetry . sometimes the nearby resonances or possible molecular states in the pseudoscalar meson and heavy baryon scattering channel might also destroy the convergence of the chiral expansion . for example , the @xmath146(@xmath147 ) charmed baryons couple strongly to the goldstone boson and @xmath148(@xmath149 ) charmed baryons based on a unitary baryon - meson coupled - channel model in ref . the convergence of the chiral expansion might improve if the @xmath146 and @xmath147 charmed baryons are included explicitly . one may also wonder whether the recoil correction might spoil the convergence . in the past several years there has some progress in the development of the @xmath12pt in the covariant form such as the extended - on - mass - shell renormalization scheme @xcite and infrared regularization method @xcite . it will be very interesting to compare the results within the different schemes . we estimate the lecs at @xmath33 assuming the su(4 ) flavor symmetry and using the pseudoscalar meson and nucleon - octet coupling constants as input . however , the su(4 ) flavor symmetry is broken in nature . the convergence of the chiral expansion might improve if the lecs could be determined more accurately . in this work , we have studied the pseudoscalar meson and charmed baryon scattering length to @xmath39 with hb@xmath12pt . the convergence of the chiral expansion of some pion and eta channels is good . because of the large heavy baryon mass , the recoil correction is small . it is easy to get the @xmath20-matrices for the pseudoscalar meson and bottomed baryon scattering from those in sec . [ sectmat ] with the corresponding parameters replaced . the numerical results do not change much due to the heavy quark flavor symmetry . according to our convention , the scattering length with a positive real part indicates there is attraction in this channel , which provides useful information on the strong interaction between the pseudoscalar meson and heavy flavor baryon . for example , one may have a rough idea in which channels there may ( or not ) exist loosely bound molecular states composed of a heavy flavor baryon and a pseudoscalar meson . these systems are similar to the pionic hydrogen . moreover , we hope our present calculation , especially nonanalytic parts , would be useful to the chiral extrapolation of future simulation of the pseudoscalar meson and heavy baryon scattering on the lattice . this project was supported by the national natural science foundation of china under grants 11075004 , 11021092 and ministry of science and technology of china(2009cb825200 ) . we list the functions and constants in the @xmath20-matrices here , @xmath150 + \frac{\omega^3 \ln \frac{|m|}{\lambda } } { \pi ^2 f^4 } -\frac { \omega^2}{\pi^2 f^4 } \begin{cases } -\sqrt{m^2-\omega^2 } \arccos\left(-\frac{\omega}{|m| } \right ) & m^2\geq \omega^2 \\ \sqrt{\omega^2-m^2 } \ln \frac{\sqrt{\omega^2-m^2}-\omega}{|m| } & m^2<\omega^2 , \omega<0 \\ \sqrt{\omega^2-m^2 } \left(-\ln \frac{\sqrt{\omega^2-m^2}+\omega}{|m|}+i \pi \right ) & m^2<\omega^2 , \omega\geq 0 \end{cases } , \label{vdef}\end{aligned}\ ] ] where @xmath151 + \frac{2\omega^3 - 3m^2\omega}{24\pi^2 } \ln\frac{|m|}{\lambda } + \frac{m^2-\omega^2}{12\pi^2 } \left\ { \begin{array}{ll } -\sqrt{m^2-\omega^2 } \arccos\frac{-\omega}{|m| } & m^2>\omega^2\\ \sqrt{\omega^2-m^2 } \ln \frac{-\omega+\sqrt{\omega^2-m^2}}{|m| } & m^2<\omega^2 , \omega<0\\ \sqrt{\omega^2-m^2 } ( i \pi-\ln\frac{\omega+\sqrt{\omega^2-m^2}}{|m| } ) & m^2<\omega^2 , \omega>0 \end{array } \right . , \nonumber\\ & & h_2(\omega , m , m)=\frac{1}{f^4 } \left\ { \begin{array}{ll } \frac{h_1(m^2,-\omega)-h_1(m^2,-\omega)}{m^2-m^2 } \qquad\qquad & m^2\neq m^2\\ \left.\frac{\partial h_1(z,-\omega)}{\partial z}\right|_{\{z\to m^2\ } } \qquad\qquad & m^2=m^2\\ \end{array } \right.\label{hdef}\end{aligned}\ ] ] the su(4 ) flavor @xmath152-plet includes the nucleon octet , @xmath153 triplet , @xmath89 sextet , and @xmath154 triplet , which can be expressed by a 3-rank tensor @xmath155 : @xmath156 with the su(4 ) 20-representation and the isospin , hypercharge , and charm of the physical particles , one obtains the individual components @xmath157 where we have normalized @xmath155 so that the lagrangian of the self - energy is @xmath158 . with the chiral symmetry , parity , c - parity , and hermiticity , we can construct the lagrangian @xmath159 + \alpha_1 \bar t^{abc } { \tilde \chi_+}{}_a^{~d } t_{dbc } + \alpha_2 \bar t^{abc } { \tilde \chi_+}{}_a^{~d } t_{dcb } + \alpha_3 \bar t t \tr[v\cdot u~v\cdot u ] + \alpha_4 \bar t^{ibc } v\cdot u_i^{~n}~v\cdot u_n^{~j } t_{jbc}\nonumber\\ & & + \alpha_5 \bar t^{ibc } v\cdot u_i^{~n}~v\cdot u_n^{~j } t_{jcb } + \alpha_6 \bar t^{ija } v\cdot u_i^{~m}~v\cdot u_j^{~n } t_{mna } + \alpha_7 \bar t^{ija } v\cdot u_i^{~m}~v\cdot u_j^{~n } t_{nma},\end{aligned}\ ] ] where @xmath160 and @xmath161 are similar to those in eq . ( [ uxidef ] ) with the extended @xmath162 and @xmath163 @xmath164 contains @xmath165 for the nucleon octet , @xmath166 for the @xmath153 triplet , @xmath167 for the @xmath89 triplet , and so on . so comparing @xmath168 with that in ref . @xcite and using their values of lecs @xmath169 , @xmath170 , @xmath171 , @xmath172 , @xmath173 , @xmath174 , @xmath175 , we get @xmath176 with a still unknown dimensionless constant @xmath97 . then comparing @xmath177 and @xmath167 with eq . ( [ lbphitwo ] ) , one gets @xmath178 the analytic terms from loop corrections are the polynomials of @xmath179 possessing the symmetries of lagrangians . they can be absorbed by the lecs . in other words , the tree and partial loop corrections have the same chiral structure . one may divide the @xmath20-matrix into the analytic and nonanalytic part . the analytic contribution originates from both loop and tree diagrams , while the nonanalytic contribution originates only from the loop graphs . one may also use the nonanalytic part to discuss the convergence of the @xmath20-matrix since the lecs of the third order can not be determined now . because of lack of enough data , we are still unable to estimate the lecs at @xmath39 accurately . as a very crude approximation , we invoke the `` nonanalytic dominance approximation '' to check the convergence of the chiral expansion , which assumes large cancellation of the analytic terms between loop and tree diagrams . under this approximation , we list the scattering lengths with the nonanalytic approximation in the last column of tables [ tabtphib3 ] , [ tabtphib6 ] , [ tabtphib6s ] . the difference between the last two columns of the tables could be regarded as a measure of the error resulting from lecs at @xmath39 . in our present calculation , some polynomials of @xmath179 , such as @xmath184 , @xmath185 , appear like nonanalytic in quark mass @xmath186 at first sight . however , we have checked that the polynomials in our results are analytic in quark mass since the momentum of the external boson at the threshold also contributes a factor @xmath187 , which is simply a kinematical factor . we can extract the nonanalytic contribution with the new functions of @xmath61 and @xmath188 by dropping the analytic terms in the squared brackets in eqs . ( [ vdef ] , [ hdef ] ) . comparing the total loop contribution with the sole nonanalytic part in table [ tabtphib3trin ] , we notice that the chiral expansion does become better when appropriate lecs absorb the analytic contribution . there are only 13 channels where the magnitude of the loop correction is smaller than 1/4 in unit of @xmath189 . in contrast there are 19 channels where the magnitude of the nonanalytic case is smaller than 1/4 in unit of @xmath189 . there are 9 badly convergent channels where the magnitude of the total loop is larger than 2/3 . there are only 5 badly convergent channels where the nonanalytic correction is larger than 2/3 . we have also checked our previous results for the excited charmed meson and pseudoscalar meson scattering lengths . the nonanalytic terms are smaller than the total loop contributions in the most channels @xcite .

we discuss the statistical mechanics of violent relaxation in stellar systems following the pioneering work of lynden - bell ( 1967 ) . the solutions of the gravitational vlasov - poisson system develop finer and finer filaments so that a statistical description is appropriate to smooth out the small - scales and describe the `` coarse - grained '' dynamics . in a coarse - grained sense , the system is expected to reach an equilibrium state of a fermi - dirac type within a few dynamical times . we describe in detail the equilibrium phase diagram and the nature of phase transitions which occur in self - gravitating systems . then , we introduce a small - scale parametrization of the vlasov equation and propose a set of relaxation equations for the coarse - grained dynamics . these relaxation equations , of a generalized fokker - planck type , are derived from a maximum entropy production principle ( mepp ) . we make a link with the quasilinear theory of the vlasov - poisson system and derive a truncated model appropriate to collisionless systems subject to tidal forces . with the aid of this kinetic theory , we qualitatively discuss the concept of `` incomplete relaxation '' and the limitations of lynden - bell s theory . # 1eq.#1 # 10=-.025em0 - 0 -0.05em0 - 0-.025em.0233em0

astro-ph0212205

it has long been realized that galaxies , and self - gravitating systems in general , follow a kind of organization despite the diversity of their initial conditions and their environement . this organization is illustrated by morphological classification schemes such as the hubble sequence and by simple rules which govern the structure of individual self - gravitating systems . for example , elliptical galaxies display a quasi - universal luminosity profile described by de vaucouleurs @xmath1 law and most of globular clusters are well fitted by the michie - king model @xcite . the question that naturally emerges is , what determines the particular configuration to which a self - gravitating system settles . it is possible that their present configuration crucially depends on the conditions that prevail at their birth and on the details of their evolution . however , in view of their apparent regularity , it is tempting to investigate whether their organization can be favoured by some fundamental physical principles like those of thermodynamics and statistical physics . we ask therefore if the actual states of self - gravitating systems are not simply more probable than any other possible configuration , i.e. if they can not be considered as maximum entropy states . this thermodynamical approach may be particularly relevant for globular clusters and elliptical galaxies which are described by a distribution function that is almost isothermal @xcite . in the case of globular clusters , the relaxation proceeds via two - body encounters and this collisional evolution is well - described by kinetic equations of a fokker - planck - landau type for which a h - theorem is available . by contrast , for elliptical galaxies , two - body encounters are completely negligible ( the corresponding relaxation time @xmath2 exceeds the age of the universe by many orders of magnitude ) and the galaxy dynamics is described by the _ vlasov equation _ , i.e. collisionless boltzmann equation @xcite . since the vlasov equation rigorously conserves entropy , a relaxation towards an isothermal distribution looks at first sight relatively surprising . yet , the inner regions of elliptical galaxies appear to be isothermal and this fact stemed as a mystery for a long time . in a seminal paper , lynden - bell @xcite argued that the violently changing gravitational field of a newly formed galaxy leads to a redistribution of energies between stars and provides a mechanism analogous to a relaxation in a gas . the importance of this form of relaxation had previously been stressed by a number of authors including hnon and king but lynden - bell showed for the first time the relevance of a statistical description . he argued that the vlasov - poisson system develops an intricate _ mixing process _ in phase space associated with the heavily damped oscillations of a protogalaxy initially far from mechanical equilibrium and collapsing under its own gravity . as a result , the solutions of the vlasov equation are not smooth but involve intermingled filaments at smaller and smaller scales . in this sense , there is no convergence towards equilibrium but rather the formation of a fractal - like structure in phase space . however , if we introduce a macroscopic level of description and make a local average of the distribution function over the filaments , the resulting `` coarse - grained '' distribution function is smooth and is expected to reach a maximum entropy state ( i.e. most mixed state ) on a very short time scale of the order of the dynamical time @xmath3 . this process is called _ violent relaxation _ and is acknowledged to account for the regularity of elliptical galaxies or other collisionless self - gravitating systems . lynden - bell predicted that the equilibrium state should be described by a fermi - dirac distribution function or a superposition of fermi - dirac distributions . here , degeneracy is due not to quantum mechanics but to the liouville theorem that prevents the smooth distribution function from exceeding the maximum of its initial value . in the non degenerate limit , the fermi - dirac distribution functions reduce to maxwellians . the prediction of an isothermal distribution for collisionless stellar systems was considered as a triumph in the 1960 s and laid the foundation of a new type of statistical mechanics . of course , the validity of the theory is conditioned by a hypothesis of ergodicity which may not be completely fulfilled . this is the complicated problem of `` incomplete relaxation '' which limitates the power of prediction of lynden - bell s approach . however , as we shall see , these difficulties should not throw doubt on the importance of this statistical description . a similar relaxation process is at work in two - dimensional turbulence ( described by the 2d euler equation ) and can explain the organization and maintenance of coherent vortices , such as the great red spot of jupiter , which are common features of large - scale geophysical or astrophysical flows @xcite . the mathematical relevance of this statistical description has been given by robert @xcite introducing the concept of young measures . the formal analogy between two - dimensional vortices and stellar systems has been discussed by chavanis @xcite . this paper is organized as follows . in section [ vp ] , we introduce the gravitational vlasov - poisson system and list its main properties . in section [ lb ] , we present the statistical approach of lynden - bell @xcite to the problem of violent relaxation . in section [ fd ] , we show that the fermi - dirac equilibrium distribution predicted by lynden - bell is not entirely satisfactory since it has an infinite mass . we must therefore invoke _ incomplete relaxation _ and introduce truncated models . in the artificial situation in which the system is enclosed within a spherical box , we can calculate the fermi - dirac spheres explicitly and prove the existence of a global entropy maximum for each value of energy . for low energies , this equilibrium state has a degenerate core surrounded by a dilute atmosphere , as calculated by chavanis & sommeria @xcite . more generally , we determine the complete equilibrium phase diagram and discuss the nature of phase transitions in self - gravitating systems . in section [ mepp ] , we describe the coarse - grained relaxation of collisionless stellar systems towards statistical equilibrium in terms of a generalized fokker - planck equation . this relaxation equation is derived from a phenomenological maximum entropy production principle ( mepp ) and involves a diffusion in velocity space compensated by a nonlinear friction . in section [ qt ] , we present a quasilinear theory of the vlasov - poisson system and show that it leads to a kinetic equation of a landau type . when the system is close to equilibrium ( so that a thermal bath approximation can be implemented ) this equation reduces to the fokker - planck equation of the thermodynamical approach and the diffusion coefficient can be explicitly evaluated . this provides a new , self - consistent , equation for the `` coarse - grained '' dynamics of stellar systems where small scales have been smoothed - out in an optimal way . in section [ tm ] , we use this kinetic model to derive the distribution function of a tidally truncated collisionless stellar system . this truncated model preserves the main features of lynden - bell s distribution ( including degeneracy ) but has a finite mass , avoiding the artifice of a spherical container . other truncated models attempting to take into account incomplete relaxation are discussed . for most stellar systems , the encounters between stars are negligible @xcite and the galaxy dynamics is described by the self - consistent vlasov - poisson system @xmath4 @xmath5 here , @xmath6 denotes the distribution function ( defined such that @xmath7 gives the total mass of stars with position @xmath8 and velocity @xmath9 at time @xmath10 ) , @xmath11 is the gravitational force ( by unit of mass ) experienced by a star and @xmath12 is the gravitational potential related to the star density @xmath13 by the newton - poisson equation ( [ poisson ] ) . the vlasov equation ( [ vlasov ] ) simply states that , in the absence of encounters , the distribution function @xmath14 is conserved by the flow in phase space . this can be written @xmath15 where @xmath16 is the material derivative and @xmath17 is a generalized velocity field in the @xmath18-dimensional phase space @xmath19 [ by definition , @xmath20 is the generalized nabla operator ] . since the flow is incompressible , i.e. @xmath21 , the hypervolume of a `` fluid '' particle is conserved . since , in addition , a fluid particle conserves the distribution function , this implies that the total mass ( or hypervolume ) of all phase elements with phase density between @xmath14 and @xmath22 is conserved . this is equivalent to the conservation of the casimir integrals @xmath23 for any continuous function @xmath24 ( they include in particular the total mass @xmath25 ) . it is also straightforward to check that the vlasov - poisson system conserves the total energy ( kinetic @xmath26 potential ) @xmath27 the angular momentum @xmath28 and the impulse @xmath29 in the following , we shall work in the barycentric frame of reference and assume that the system is non rotating so that the conservation of @xmath30 and @xmath31 can be ignored ( see refs . @xcite for a generalization ) . when the initial condition is far from equilibrium , the vlasov - poisson system develops a complicated mixing process in phase space and generates intermingled filaments due to stirring effects . therefore , a _ deterministic _ description of the flow in phase space requires a rapidly increasing amount of information as time goes on . for that reason , it is appropriate to undertake a _ probabilistic _ description in order to smooth out the small - scales ( fine - grained ) and concentrate on the locally averaged ( coarse - grained ) quantities . this statistical analysis has been considered a long time ago by lynden - bell @xcite . the statistical mechanics of continuous systems is not as firmly established as in the usual case of n - body systems but robert @xcite has recently developed a mathematical justification of this procedure in terms of young measures . we shall describe below the argumentation of lynden - bell , which is more intuitive in a first approach . starting from some arbitrary initial condition , the distribution function is stirred in phase space but conserves its values @xmath32 ( levels of phase density ) and the corresponding hypervolumes @xmath33 as a property of the vlasov equation . let us introduce the probability density @xmath34 of finding the level of phase density @xmath32 in a small neighborhood of the position @xmath35 in phase space . this probability density can be viewed as the local area proportion occupied by the phase level @xmath32 and it must satisfy at each point the normalization condition @xmath36 the locally averaged distribution function is then expressed in terms of the probability density as @xmath37 and the associated ( macroscopic ) gravitational potential satisfies @xmath38 since the gravitational potential is expressed by space integrals of the density , it smoothes out the fluctuations of the distribution function , supposed at very fine scale , so @xmath39 has negligible fluctuations . it is then possible to express the conserved quantities of the vlasov equation as integrals of the macroscopic fields . these conserved quantities are the global probability distributions of phase density @xmath40 ( i.e. , the total hypervolume occupied by each level @xmath32 of phase density ) @xmath41 and the total energy @xmath42 as discussed above the gravitational potential can be considered as smooth , so we can express the energy in terms of the coarse - grained functions @xmath43 and @xmath44 neglecting the internal energy of the fluctuations . to determine the equilibrium distribution of the system , we need to introduce an entropy functional like in ordinary statistical mechanics . as is customary , lynden - bell defines the mixing entropy as the logarithm of the number of microscopic configurations associated with the same macroscopic state ( characterized by the probability density @xmath45 ) . after introducing a counting `` la boltzmann '' , he arrives at the expression @xcite : @xmath46 where the integral extends over phase space and over all the levels of phase elements . a mathematical justification of this entropy has been given by robert @xcite . the most likely distribution to be reached at equilibrium is then obtained by maximizing the mixing entropy ( [ e6 ] ) subject to the constraints ( [ e4])([e5 ] ) and the normalization condition ( [ e1 ] ) . this variational problem is treated by introducing lagrange multipliers so that the first variations satisfy @xmath47 where @xmath48 is the inverse temperature and @xmath49 the chemical potential " of species @xmath32 . the resulting optimal probability density is a gibbs state which has the form @xcite : @xmath50 the previous analysis gives a well defined procedure to compute the statistical equilibrium states . the gravitational field is obtained by solving the poisson equation ( [ poisson ] ) with the distribution function ( [ e2 ] ) determined by the gibbs state ( [ e8 ] ) . the solution depends on the lagrange multipliers @xmath51 and @xmath52 which must be related to the conserved quantities @xmath53 and @xmath54 by equations ( [ e5])([e4 ] ) . this precedure determines _ critical points _ of entropy . whether these critical points are maxima or not is decided by the sign of the second order variations of entropy . following lynden - bell @xcite , we consider a particular situation that presents interesting features and for which the previous problem can be studied in detail . keeping only two levels @xmath55 and @xmath56 is convenient to simplify the discussion and is probably representative of more general cases . within this `` patch '' approximation , the mixing entropy reduces to @xmath57 and the equilibrium coarse - grained distribution @xmath58 takes explicitly the form @xmath59 where @xmath60 is the energy of a star ( per unit of mass ) and @xmath61 an equivalent lagrange multiplier . equation ( [ e10 ] ) is , apart from a reinterpretation of the constants , the distribution function for the self - gravitating fermi - dirac gas . here , the exclusion principle @xmath62 is due to the incompressibility constraint ( [ e1 ] ) not to quantum mechanics . because of the averaging procedure , the coarse - grained distribution function can only _ decrease _ by internal mixing , as vaccum is incorporated into the patch , and this results in an `` effective '' exclusion principle . lynden - bell s distribution therefore corresponds to a @xmath63-th type of statistics since the `` fluid '' particles are distinghuishable but subject to an exclusion principle . however , formally , this distribution coincides with the fermi - dirac distribution . in the fully degenerate case , this equilibrium has been studied extensively in connection with white dwarf stars @xcite . the structure of equilibrium may crucially depend on the degree of degeneracy ( see section [ fd ] ) but lynden - bell @xcite gives arguments according to which stellar systems would be non degenerate . in that limit , @xmath64 , and equation ( [ e10 ] ) reduces to the maxwell - boltzmann statistics @xmath65 this was in fact the initial goal of lynden - bell in 1967 : his theory of violent relaxation " was able to justify a maxwell - boltzmann equilibrium distribution , without recourse to collisions , on a short time scale @xmath66 consistent with the age of ellipticals . in addition , the individual mass of the stars never appears in his theory based on the vlasov equation . therefore , the equilibrium state ( [ e11 ] ) does not lead to a segregation by mass in contrast with a collisional relaxation . this is in agreement with the observed light distributions in elliptical galaxies that would show greater colour differences if a marked segregation by mass was established . the first problem to be tackled is obviously the computation of self - gravitating fermi - dirac spheres . let us first consider the non degenerate limit @xmath64 corresponding to a dilute system . in that case , the mixing entropy ( [ e9 ] ) reduces to the ordinary boltzmann entropy @xmath67 whose maximization at fixed mass and energy leads to the maxwell - boltzmann statistics ( [ e11 ] ) . substituting this optimal distribution in the poisson equation ( [ poisson ] ) , we find that the gravitational potential is determined at equilibrium by the differential equation @xmath68 where the lagrange multipliers @xmath69 and @xmath51 have to be related to the total mass and total energy of the system . the boltzmann - poisson equation ( [ e13 ] ) has been studied extensively in the context of isothermal gaseous spheres @xcite and in the case of collisional stellar systems such as globular clusters @xcite . we can check by direct substitution that the distribution @xmath70 is an exact solution of equation ( [ e13 ] ) known as the _ singular isothermal sphere _ @xcite . since @xmath71 at large distances , the total mass of the system @xmath72 is infinite ! more generally , we can show that any solution of the boltzmann - poisson equation ( [ e13 ] ) behaves like the singular sphere as @xmath73 and has therefore an infinite mass . this means that no equilibrium can exist in an unbounded domain : the density can spread indefinitely while conserving energy and increasing entropy @xcite . in practice , the infinite mass problem does not arise if we realize that the relaxation of the system is necessarily _ incomplete _ @xcite . there are two major reasons for incomplete relaxation : ( i ) the mean field relaxation process is dependant on the strength of the variations in potential . as these die out , the relaxation ceases and it is likely that the system may find a stable steady state before the relaxation process is completed . therefore , the relaxation is effective only in a finite region of space ( roughly the main body of the galaxy ) and during a finite period of time ( while the galaxy is dynamically unsteady ) . orbits which lie partly outside the relaxing region and have periods longer than the time for which the galaxy is unsteady will not acquire their full quotas of stars . ( ii ) in practice , the galaxy will not be isolated but will be subject to the tides of other systems . therefore , high energy stars will escape the system being ultimately captured by the gravity of a nearby object . these two independant effects have similar consequences and will produce a modification of the distribution function at high energies . some truncated models can be obtained by developing a kinetic theory of encounterless relaxation ( sections [ mepp]-[tm ] ) . this kinetic approach will provide a precise framework to understand what limits relaxation and why complete equilibrium is not reached in general . however , for the present , we shall avoid the infinite mass problem by confining artificially the system within a box of radius @xmath74 . the calculation of finite isothermal spheres has been carried out by antonov @xcite , lynden - bell & wood @xcite , katz @xcite , padmanabhan @xcite , de vega & sanchez @xcite and chavanis @xcite ( see chavanis @xcite for an extension in general relativity ) . these studies were performed in the framework of gaseous stars or collisional stellar systems ( e.g. , globular clusters ) but they extend in principle to violently relaxed collisionless stellar systems described by the distribution function ( [ e11 ] ) . therefore , we shall give a brief summary of these classical results before considering the case of the fermi - dirac distribution ( [ e10 ] ) . the phase diagram is represented in fig . [ figure1 ] where the inverse temperature is plotted as a function of minus the energy . it is possible to prove the following results : ( i ) there is no global maximum for the boltzmann entropy . ( ii ) there are not even critical points for the boltzmann entropy if @xmath75 . ( iii ) local entropy maxima ( lem ) exist if @xmath76 ; they have a density contrast @xmath77 ( upper branch ) . ( iv ) critical points of entropy with density contrast @xmath78 are unstable saddle points sp ( spiraling curve ) . conclusions ( i ) and ( ii ) have been called gravothermal catastrophe " or `` antonov instability '' . when @xmath75 , there is no hydrostatic equilibrium and the system is expected to collapse and overheat . this is a natural evolution ( in a thermodynamical sense ) because a self - gravitating system can always increase entropy by taking a `` core - halo '' structure and by making its core denser and denser ( and hotter and hotter ) @xcite . as discussed by lynden - bell & wood @xcite , this instability is probably related to the negative specific heats of self - gravitating systems : by losing heat , the core grows hotter and evolves away from equilibrium . the `` gravothermal catastrophe '' picture has been confirmed by sophisticated numerical simulations of globular clusters that introduce a precise description of heat transfers between the `` core '' and the `` halo '' using moment equations @xcite , orbit averaged fokker - planck equation @xcite or fluid equations for a thermally conducting gas @xcite . in these studies , the collapse proceeds self - similarly ( with power law behaviours ) and the central density becomes infinite in a finite time . this singularity has been known as `` core collapse '' and many globular clusters have probably experienced core collapse . this is certainly the most exciting theoretical aspect of the collisional evolution of stellar systems . in practice , the formation of hard binaries can release sufficient energy to stop the collapse and even drive a reexpansion of the system @xcite . then , a series of gravothermal oscillations should follow but it takes times much larger than the age of globular clusters @xcite . it has to be noted that , although the central density tends to infinity , the mass contained in the core tends to zero so , in this sense , the gravothermal catastrophe is a rather unspectacular process @xcite . however , the gravothermal catastrophe may also be at work in dense clusters of compact stars ( neutron stars or stellar mass black holes ) embedded in evolved galactic nuclei and it can presumably initiate the formation of supermassive black holes via the collapse of such clusters . indeed , when the central redshift ( related to the density contrast ) exceeds a critical value @xmath79 , a dynamical instability of general - relativistic origin sets in and the star cluster undergoes a catastrophic collapse to a black hole on a dynamical time @xcite . it has been suggested that this process leads naturally to the birth of supermassive black holes of the right size to explain quasars and agns @xcite . since collisionless stellar systems undergoing violent relaxation are described by an _ isothermal _ distribution function of the fermi - dirac type ( [ e10 ] ) , it is particularly relevant to ask whether galaxies in their birth stages can undergo a form of gravothermal catastrophe and if so what they do about it . the collapse of galaxies was described qualitatively by lynden - bell & wood @xcite : `` ... the centers will then begin to separate into a core - a sort of separate body which might even be called a nucleus . this will cease to shrink when it becomes degenerate in lynden - bell s sense . the system will then have a core - halo structure which is an equilibrium of an isothermal fermi - dirac gas sphere . these will show a variety of different nuclear concentrations depending on the degeneracy parameter . it is evident that theory developed along these lines has the chance of making sense of the variety of different galactic nuclei . '' ) .,scaledwidth=90.0% ] ) .,scaledwidth=90.0% ] the computation of isothermal fermi - dirac spheres has been performed only recently by chavanis & sommeria @xcite . it can be proved that a global entropy maximum exists for all accessible values of energy @xcite . therefore , degeneracy has a stabilizing role and is able to stop the `` gravothermal catastrophe '' . the equilibrium diagram is represented in figs . [ figure2]-[figure3 ] and now depends on the degeneracy parameter @xmath80 , where @xmath81 is the maximum allowable value of the distribution function and @xmath82 its typical average value in the box of radius @xmath74 . we see that the inclusion of degeneracy has the effect of unwinding the spiral ( dashed line ) . when @xmath83 is small ( fig . [ figure2 ] ) , there is only one critical point of entropy for each value of energy and it is a global entropy maximum ( gem ) . for small values of @xmath84 ( high energies ) the solutions almost coincide with the classical , non degenerate , isothermal spheres . when @xmath84 is increased ( low energies ) the solutions take a `` core - halo '' structure with a partially degenerate core surrounded by a dilute maxwellian atmosphere . it is now possible to overcome the critical energy @xmath85 and the critical density contrast @xmath86 found by antonov for a classical isothermal gas . in that region , the specific heat @xmath87 is negative . as energy decreases further , more and more mass is concentrated into the nucleus ( which becomes more and more degenerate ) until a minimum accessible energy , corresponding to @xmath88 , at which the nucleus contains all the mass . in that case , the atmosphere has been `` swallowed '' and the system has the same structure as a cold white dwarf star @xcite . this is a relatively singular limit since the density drops to zero at a finite radius whereas for partially degenerate systems , the density decays like @xmath89 at large distances . when the degeneracy parameter @xmath83 is large ( fig . [ figure3 ] ) , there are now several critical points of entropy for each single value of energy in the range @xmath90 . the solutions on the upper branch ( points a ) are non degenerate and have a smooth density profile ; they form the `` gaseous '' phase . the solutions on the lower branch ( points c ) have a `` core - halo '' structure with a degenerate nucleus and a dilute atmosphere ; they form the `` condensed '' phase . according to the theorem of katz @xcite , they are both entropy maxima while the intermediate solutions , points b , are unstable saddle points ( sp ) . these points are similar to points a , except that they contain a small embryonic nucleus ( with small mass and energy ) which plays the role of a `` germ '' in the langage of phase transitions . for small values of @xmath84 , the non degenerate solutions ( upper branch ) are global entropy maxima ( gem ) . they become local entropy maxima ( metastable states ) at a certain @xmath91 depending on the degeneracy parameter . at that @xmath92 , the degenerate solutions ( lower branch ) that were only local entropy maxima ( lem ) suddenly become global entropy maxima . we expect therefore a phase transition to occur between the `` gaseous '' phase ( upper branch ) and the `` condensed '' phase ( lower branch ) at @xmath92 . for @xmath93 , the non degenerate solutions ( points a ) are metastable but we may suspect @xcite that they are long - lived so that they _ are _ physical . it is plausible that these metastable states will be selected by the dynamics even if states with higher entropy exist . this depends on a complicated notion of `` basin of attraction '' as studied by chavanis , rosier and sire @xcite with the aid of a simple model of gravitational dynamics . therefore , the true phase transition will occur at @xmath85 at which the gaseous phase completely disappears : at that point , the gravothermal catastrophe sets in but , for collisionless stellar systems ( or for fermions ) , the core ultimately ceases to shrink when it becomes degenerate . in that case , the system falls on to a global entropy maximum which is the true equilibrium state for these systems ( point d ) . this global entropy maximum has a `` core - halo '' structure with a degenerate core and a dilute atmosphere . a simple analytical model of these phase transitions has been proposed in ref . @xcite and provides a fairly good agreement with the full numerical solution . a particularity of self - gravitating systems , which are in essence non - extensive , is that the statistical ensembles ( microcanonical and canonical ) are not interchangeable . therefore , the description of the equilibrium diagram is different whether the system evolves at fixed energy of fixed temperature . a discussion of this interesting phenomenon can be found in the review of padmanabhan @xcite and in chavanis @xcite . for astrophysical purposes , it is still a matter of debate to decide whether collisionless stellar systems like elliptical galaxies are degenerate ( in the sense of lynden - bell ) or not . since degeneracy can stabilize the system without changing its overall structure at large distances , we have suggested that degeneracy could play a role in galactic nuclei @xcite . the recent simulations of leeuwin and athanassoula @xcite and the theoretical model of stiavelli @xcite are consistent with this idea especially if the nucleus of elliptical galaxies contains a primordial massive black hole . indeed , the effect of degeneracy on the distribution of stars surrounding the black hole can explain the cusps observed at the center of galaxies @xcite . this form of degeneracy is also relevant for massive neutrinos in dark matter models where it competes with quantum degeneracy @xcite . in fact , the fermi - dirac distribution of massive neutrinos in dark matter models ( which form a collisionless self - gravitating system ) might be justified more by the process of `` violent relaxation '' @xcite than by quantum mechanics @xcite . as shown by chavanis & sommeria @xcite , violent relaxation can lead to the formation of a dense degenerate nucleus with a small radius and a large mass . this massive degenerate nucleus could be an alternative to black holes at the center of galaxies @xcite . basically , a collisionless stellar system is described in a self - consistent mean field approximation by the vlasov - poisson system ( [ vlasov])([poisson ] ) . in principle , these coupled equations determine completely the evolution of the distribution function @xmath6 . however , as discussed in section [ lb ] , we are not interested in practice by the finely striated structure of the flow in phase space but only by its macroscopic , i.e. smoothed - out , structure . indeed , the observations and the numerical simulations are always realized with a finite resolution . moreover , the `` coarse - grained '' distribution function @xmath43 is likely to converge towards an equilibrium state contrary to the exact distribution @xmath14 which develops smaller and smaller scales . if we decompose the distribution function and the gravitational potential in a mean and fluctuating part ( @xmath94 , @xmath95 ) and take the local average of the vlasov equation ( [ vlasov ] ) , we readily obtain an equation of the form @xmath96 for the `` coarse - grained '' distribution function with a diffusion current @xmath97 related to the correlations of the `` fine - grained '' fluctuations . the problem in hand consists in determining a closed form for the diffusion current @xmath98 . its detailed expression @xmath99 depends on the subdynamics and is therefore extremely difficult to capture from first principles . indeed , the `` violent relaxation '' is a very nonlinear and very chaotic process ruling out any attempt to implement _ perturbative _ methods if we are far from equilibrium . for that reason , there is a strong incentive to explore _ variational _ methods which are considerably simpler and give a more intuitive understanding of the problem . we propose to describe the relaxation of the coarse - grained distribution function @xmath43 towards the gibbs state ( [ e8 ] ) with a maximum entropy production principle @xcite . this thermodynamical principle assumes that : `` during its evolution , the system tends to maximize its rate of entropy production @xmath100 while satisfying all the constraints imposed by the dynamics '' . there is no rigorous justification for this principle and it is important therefore to confront the mepp with kinetic theories ( when they are available ) in order to determine its range of validity . in any case , the mepp can be considered as a convenient tool to build relaxation equations which are mathematically well - behaved and which can serve as numerical algorithms to calculate maximum entropy states . it is also `` perfectible '' in the sense that any new information about the dynamics of the system can be incorporated . this principle is reminiscent of jaynes @xcite ideas and is a clear extension of the well - known principle of statistical mechanics according to which : `` at equilibrium , the system is in a maximum entropy state consistent with all the constraints '' . this is also the most natural extension of lynden - bell s theory out of equilibrium . like for equilibrium states , the evolution of the flow in phase space is described by a set of local probabilities @xmath101 and the locally averaged gravitational potential @xmath102 is obtained by the integration of equation ( [ poisson ] ) where @xmath103 is replaced by @xmath104 . the phase elements are thus transported in phase space by the corresponding averaged velocity field @xmath105 and we suppose that , in addition , they undergo a diffusion process . this diffusion occurs only in velocity space , due to the fluctuations of the gravitational field @xmath106 . there is no diffusion in position space since the velocity @xmath9 is a pure coordinate and does not fluctuate . as a result , the probability densities will satisfy a convection - diffusion equation of the form @xmath107 where @xmath108 is the diffusion current of the phase element @xmath32 . this equation conserves the total hypervolume ( in phase space ) occupied by each phase element . multiplying equation ( [ e16 ] ) by @xmath32 and integrating over the levels of phase density returns equation ( [ e15 ] ) with @xmath109 . to apply the mepp , we need first to compute the rate of change of entropy during the convection - diffusion process . taking the time derivative of equation ( [ e6 ] ) , expressing @xmath110 by equation ( [ e16 ] ) and noting that @xmath111 is conserved by the advective term , we get @xmath112 we now determine @xmath113 such that , for a given field @xmath114 at each time @xmath10 , @xmath113 maximizes the entropy production @xmath100 with appropriate dynamical constraints , which are : -the conservation of the local normalization condition ( [ e1 ] ) implying @xmath115 -the conservation of energy expressed from equations ( [ e5 ] ) and ( [ e15 ] ) as @xmath116 - a limitation on the eddy flux @xmath117 , characterized by a bound @xmath118 , which exists but is not specified @xmath119 this variational problem can be solved by introducing ( at each time @xmath10 ) lagrange multipliers @xmath120 , @xmath51 , @xmath121 for the three respective constraints . it can be shown by a convexity argument that reaching the bound ( [ e20 ] ) is always favorable for increasing @xmath100 , so that this constraint can be replaced by an equality . the condition @xmath122 yields an optimal current of the form @xcite : @xmath123 the lagrange multiplier @xmath124 has been eliminated , using the condition ( [ e18 ] ) of local normalization conservation . at equilibrium , the diffusion currents must vanish and we can check that this yields the gibbs state ( [ e8 ] ) with @xmath51 the corresponding inverse temperature @xcite . during the evolution , this quantity varies with time and is determined by the condition of energy conservation ( [ e19 ] ) . introducing ( [ e22 ] ) in the constraint ( [ e19 ] ) , we get @xmath125 where @xmath126 are the local moments of the fine - grained distribution function . we have thus obtained a complete set of relaxation equations which exactly conserve the distribution of phase levels and energy . the rate of entropy production can be put in the form @xcite : @xmath127 so the diffusion coefficient @xmath128 must be positive for entropy increase . except for its sign , the diffusion coefficient is not determined by the thermodynamical approach as it is related to the unknown bound @xmath118 . these relaxation equations ( [ e16])([e22])([e23 ] ) can be simplified in the case of a single non zero density level @xmath81 and provide a new non linear equation for the evolution of the coarse - grained distribution function @xmath129 : @xmath130 in the non degenerate limit ( @xmath131 ) , equation ( [ e25 ] ) takes the form of a fokker - planck equation @xmath132 this fokker - planck equation ( sometimes called the kramers - chandrasekhar equation ) is well - known in the case of collisional stellar systems ( without the bar on @xmath14 ! ) and is usually derived from a markov hypothesis and a stochastic langevin equation @xcite . it can also be obtained from the following argument : because of close encounters , the stars undergo brownian motion and diffuse in velocity space ( responsible for the first term in the right hand side ) . however , under the influence of this diffusion , the kinetic energy per star will diverge as @xmath133 and one is forced to introduce an ad hoc " dynamical friction ( second term in the right hand side ) with a friction coefficient @xmath134 ( einstein relation ) in order to compensate this divergence and recover the maxwellian distribution of velocities at equilibrium . from these two considerations ( which express the fluctuation - dissipation theorem ) results the ordinary fokker - planck equation ( [ e26 ] ) . the fokker - planck equation is justified here by an argument of a very wide scope that does not directly refer to the subdynamics : it appears to maximize the rate of entropy production at each time with appropriate constraints . therefore , it applies equally well to collisional or ( coarse - grained ) collisionless stellar systems . in this description , the diffusion term and the `` dynamical friction '' directly result from the variational principle and are associated with the variations of @xmath135 and @xmath53 respectively . furthermore , the inverse temperature @xmath51 appears as a lagrange multiplier associated with the conservation of energy and the einstein relation is automatically satisfied . in addition , our procedure can take into account the degeneracy of collisionless stellar systems , keeping equation ( [ e25 ] ) instead of the non degenerate limit ( [ e26 ] ) . in the degenerate case , the friction is non linear in @xmath43 so that equation ( [ e25 ] ) is _ not _ a fokker - planck equation in a strict sense . this nonlinearity is necessary to recover the fermi - dirac distribution function at equilibrium ( while a linear friction drives the system towards the maxwell - boltzmann distribution ) . this generalization is important because degeneracy is specific to collisionless systems and may be crucial for the existence of an equilibrium state ( see section [ fd ] ) . in equations ( [ e25 ] ) ( [ e26 ] ) , @xmath51 is not constant but evolves with time so as to conserve the total energy . in the non degenerate limit , and assuming that @xmath128 is constant , we get after a part integration @xmath136 this equation relates the formal lagrange multiplier @xmath137 to the inverse of the _ average _ kinetic energy . this is of course satisfying on a physical point of view . an alternative fokker - planck equation involving a _ temperature @xmath138 instead of @xmath139 has been proposed by clemmow & dougherty @xcite in the case of collisional systems . the energy is assumed to be locally conserved by the collisions , which is valid when the mean free path is much smaller than the size of the system . by contrast , this hypothesis does not seem to be justified for the violent relaxation of a collisionless system , which is rather a global process . let us briefly review the main properties of our relaxation equations . first of all , they rigorously satisfy the conservation of energy and phase space hypervolumes like the vlasov - poisson system ( the conservation of impulse and angular momentum can also be satisfied by introducing appropriate lagrange multipliers @xcite ) . moreover , they guarantee the increase of entropy at each time @xmath140 with an optimal rate . of course , this h - theorem is true for the coarse - grained entropy @xmath141 and not for the fine - grained entropy @xmath142 which is constant , as the integral of any function of @xmath14 . the source of irreversibility is due to the coarse - graining procedure that smoothes out the small - scales and erases the microscopic details of the evolution . accordingly , the relaxation equations ( [ e16])([e22])([e23 ] ) are likely to drive the system towards an equilibrium state ( the gibbs state ( [ e8 ] ) ) contrary to the vlasov equation that develops finer and finer scales . in mathematical terms , this means that the distribution function @xmath14 converges to an equilibrium state @xmath43 _ in a weak sense_. in fact , the situation is more complicated since the gibbs state does not exist in an infinite domain ( section [ fd ] ) . we shall see , however , in section [ tm ] that the diffusion coefficient is proportional to the fluctuations of the distribution function ( equation ( [ e65 ] ) ) . now , in physical situations , these fluctuations vanish before the systems had time to relax completely . as a result , the relaxation stops and the system remains frozen in a subdomain of phase space . it is only in this subdomain ( corresponding to the main body of the galaxy ) that the gibbs state is justified . this provides a physical mechanism for confining galaxies and justifying truncated models . alternatively , if the galaxy is not isolated but subject to the tides of a neighboring object , a tidally truncated model can be explicitly derived from these relaxation equations ( section [ tm ] ) . it has a finite mass while preserving the essential features of lynden - bell s distribution function . the relaxation equations ( [ e16])([e22])([e23 ] ) can be simplified in two particular limits when : ( i ) the initial condition is approximated by a single level of phase density @xmath81 surrounded by vaccum ( ii ) degeneracy , in lynden - bell s sense , is neglected . these simplifications lead to the fokker - planck equation ( [ e26 ] ) . if the galaxy has sufficiently large energy , this equation will drive the system towards a maxwell - boltzmann equilibrium state ( [ e11 ] ) . however , if the energy falls below a critical value , the fokker - planck equation does not reach any equilibrium state anymore and the system can achieve ever increasing values of entropy by developing core collapse ( antonov instability ) . for collisionless stellar systems , this `` gravothermal catastrophe '' should stop when the center of the galaxy becomes degenerate ( see section [ fd ] ) . in that case , the fokker - planck equation ( [ e26 ] ) is not valid anymore and must be replaced by the degenerate relaxation equation ( [ e25 ] ) . this equation converges towards the fermi - dirac equilibrium state ( [ e10 ] ) , which exists for all values of energy . in summary , the mepp is able to yield relevant relaxation equations for the coarse - grained dynamics of collisionless stellar systems experiencing violent relaxation . this relatively elegant and simple variational principle shows that the global structure of the relaxation equations is determined by purely thermodynamical arguments . all explicit reference to the subdynamics is encapsulated in the diffusion coefficient which can not be captured by the mepp ( it appears as a lagrange multiplier related to an unknown bound on the diffusion flux ) . it must be therefore calculated with a kinetic model such as the quasilinear theory described in the next section . the quasilinear theory of the vlasov - poisson system was first considered by kadomtsev & pogutse @xcite for a homogeneous coulombian plasma and extended by severne & luwel @xcite for an inhomogeneous gravitational system . this theory was further discussed by chavanis @xcite in an attempt to make a link with the maximum entropy production principle . we shall give here a simple account of the quasilinear theory . further details can be found in refs . @xcite . our objective is to obtain an expression for the diffusion current @xmath113 by working directly on the vlasov - poisson system , i.e. without _ assuming _ that the entropy increases as is done in the thermodynamical approach . since the diffusion current @xmath143 is related to the fine - grained fluctuations of the distribution function , any systematic calculation starting from the vlasov equation ( [ vlasov ] ) must necessarily introduce an evolution equation for @xmath144 . this equation is simply obtained by substracting equation ( [ e15 ] ) from equation ( [ vlasov ] ) . this yields @xmath145 the essence of the quasilinear theory is to assume that the fluctuations are weak and neglect the nonlinear terms in equation ( [ e29 ] ) altogether . in that case , equations ( [ e15 ] ) and ( [ e29 ] ) reduce to the coupled system @xmath146 @xmath147 where @xmath148 is the advection operator in phase space . physically , these equations describe the coupling between a subdynamics ( here the small scale fluctuations @xmath144 ) and a macrodynamics ( described by the coarse - grained distribution function @xmath43 ) . introducing the greenian @xmath149 we can immediately write down a formal solution of equation ( [ e31 ] ) , namely @xmath150 although very compact , this formal expression is in fact extremely complicated . indeed , all the difficulty is encapsulated in the greenian @xmath151 which supposes that we can solve the smoothed - out lagrangian flow @xmath152 between @xmath10 and @xmath153 . in practice , this is impossible and we will have to make some approximations . the objective now is to substitute the formal result ( [ e33 ] ) back into equation ( [ e30 ] ) and make a closure approximation in order to obtain a self - consistant equation for @xmath154 . if the fluctuating force @xmath155 were external to the system , we would simply obtain a diffusion equation @xmath156 with a diffusion coefficient given by a kubo formula @xmath157 however , in the case of the vlasov - poisson system , the gravitational force is in fact produced by the distribution of matter itself and this coupling will give rise to a friction term in addition to the pure diffusion . indeed , we have @xmath158 where @xmath159 represents the force created by a ( field ) star in @xmath160 on a ( test ) star in @xmath8 ( newton s law ) . therefore , considering equations ( [ e33 ] ) and ( [ e37 ] ) , we see that the fluctuations of the distribution function @xmath161 are given by an iterative process : @xmath162 depends on @xmath163 which itself depends on @xmath164 etc ... we shall solve this problem perturbatively in an expansion in powers of the gravitational constant @xmath165 . this is the equivalent of the `` weak coupling approximation '' in plasma physics . to order @xmath166 , we obtain after some rearrangements @xmath167 in this expression , the greenian @xmath165 refers to the fluid particle @xmath168 and the greenian @xmath169 to the fluid particle @xmath170 . to close the system , it remains for one to evaluate the correlation function @xmath171 . we shall assume that the mixing in phase space is sufficiently efficient so that the scale of the kinematic correlations is small with respect to the coarse - graining mesh size . in that case , @xmath172 where @xmath173 and @xmath174 are the resolution scales in position and velocity respectively . now , @xmath175 we shall assume for simplicity that the initial condition in phase space consists of a patch of uniform distribution function ( @xmath55 ) surrounded by vaccum ( @xmath56 ) . this is the two - levels approximation already considered in sections [ lb ] and [ mepp ] . in that case @xmath176 and , therefore , @xmath177 substituting this expression in equation ( [ e39 ] ) and carrying out the integrations on @xmath178 and @xmath179 , we obtain @xmath180 we have written @xmath181 , @xmath182 , @xmath183 and @xmath184 where @xmath185 and @xmath186 are the position and velocity at time @xmath153 of the stellar fluid particle located in @xmath187 , @xmath188 at time @xmath10 . it is determined by the characteristics ( [ e34 ] ) of the smoothed - out lagrangian flow . equation ( [ e43 ] ) is a non markovian integrodifferential equation : the value of @xmath43 in @xmath8 , @xmath9 at time @xmath10 depends on the value of the whole field @xmath189 at earlier times . if the decorrelation time @xmath190 is short , we can make a _ markov approximation _ and replace the bracket at time @xmath153 by its value taken at time @xmath10 . noting furthermore that the integral is dominated by the contribution of field stars close to the star under consideration ( i.e. when @xmath191 ) , we shall make a _ local approximation _ and replace @xmath192 and @xmath193 by their values taken at @xmath194 . in that case , the foregoing equation simplifies in @xmath195 where , now , @xmath196 . the explicit reference to the past evolution of the system is only retained in the memory function @xmath197 this function can be calculated explicitly if we assume that , between @xmath153 and @xmath10 , the stars follow linear trajectories , so that @xmath198 and @xmath199 @xcite . this leads to the generalized landau equation @xmath200 where @xmath201 is the tensor @xmath202 and @xmath203 , @xmath204 . this equation applies to inhomogeneous systems but , as a result of the local approximation , the effect of inhomogeneity is only retained in the advective term . equation ( [ e45 ] ) is very similar to the well - known landau equation of collisional self - gravitating systems and electric charges @xcite . there are nevertheless two important differences : ( i ) the friction term is non linear in @xmath43 , accounting for the degeneracy discovered by lynden - bell at equilibrium . ( ii ) the diffusion coefficient is proportional to the mass @xmath205 of a macrocell completely filled by the phase fluid instead of the mass @xmath206 of an individual star in the ordinary landau equation . in general @xmath207 so that the relaxation by phase mixing is much more rapid than the collisional relaxation . from the above theory , we find that the collisionless relaxation operates on a time scale @xmath208 , the dynamical time , whereas the collisional relaxation operates on a time scale @xmath209 where @xmath210 is the number of stars in the galaxy . therefore , the relaxation by phase mixing really corresponds to a `` violent relaxation '' @xcite . it can be shown that the generalized landau equation ( [ e45 ] ) conserves mass and energy @xcite . in fact , as a result of the local approximation , the energy is conserved locally , as if the system were homogeneous , and we have @xmath211 it is also easy to show that equation ( [ e45 ] ) satisfies a h - theorem ( @xmath212 ) for the fermi - dirac entropy ( [ e9 ] ) . when a stationary state is reached @xmath213 and the fermi - dirac distribution ( [ e10 ] ) is obtained , in agreement with lynden - bell s statistical theory @xcite . this provides therefore another way of justifying his results from a dynamical point of view which does not explicitly rely on a maximization of entropy ( the @xmath214-theorem is not assumed but _ derived _ from the kinetic theory ) . it can be noted that these properties result from the _ symmetry _ of the landau collision term and not from lagrange multipliers like in the thermodynamical approach . this is more satisfactory on a physical point of view . it is important to stress , however , that this quasilinear theory can not describe the early , very non linear and very chaotic , stages of the `` violent relaxation '' . indeed , the detailed study of severne & luwel @xcite reveals that the various approximations introduced in the theory make equation ( [ e45 ] ) applicable only for @xmath215 , where @xmath3 is the dynamical time . since the relaxation time is precisely of order @xmath3 , the quasilinear theory is only _ marginally _ applicable and can describe , at most , the late quiescent phases of the relaxation , when the fluctuations have weaken . by contrast , it is plausible that equation ( [ e25 ] ) is more general and more appropriate to the context of `` violent relaxation '' since it results from a thermodynamical approach which exploits at best the chaoticity of the system and the complete lack of information that we have to face at small scales . as a clear difference , it should be noted that the mepp takes into account only the _ global _ conservation of energy ( the lagrange multiplier @xmath216 is a kind of inverse average temperature determined by an integration over the whole system ) whereas the approximations introduced in the kinetic model lead to a _ local _ conservation of energy . this strong locality can not account for the rather collective processes which are involved in the violent relaxation and may unveil a failure of the quasilinear theory . however , in section [ tm ] , we show that the two approaches are consistant if we are close to equilibrium and we use the quasilinear theory as a model to determine an explicit expression for the diffusion coefficient that appears in equation ( [ e25 ] ) . the equations provided by the mepp and by the quasilinear theory have a very different mathematical structure . the relaxation equation ( [ e25 ] ) is a partial differential equation whereas the generalized landau equation ( [ e45 ] ) is an integrodifferential equation : the value of @xmath217 at time @xmath218 depends explicitly on the _ whole _ distribution of velocities @xmath219 at time @xmath10 through an integration over @xmath220 . the usual way to transform an integrodifferential equation into a partial differential equation is to make a guess for the function appearing in the integral and refine the initial guess by successive iterations . in practice , we simply make _ one _ sensible guess . therefore , if we are close to equilibrium ( and this is in fact dictated by the conditions of validity of the quasilinear theory ) , it seems natural to replace the distribution function @xmath221 by the fermi - dirac distribution @xmath222 in more physical terms , this amounts to a `` thermal bath approximation '' : the stars have not yet relaxed completely , but when we focus on the relaxation of a given stellar fluid particle ( described by @xmath43 ) we can consider , in a first approximation , that the rest of the system ( described by @xmath221 ) is at equilibrium . with this thermal bath approximation , the generalized landau equation ( [ e45 ] ) reduces to the nonlinear partial differential equation @xcite : @xmath223 with a diffusion coefficient @xmath224 equation ( [ e49 ] ) is precisely the equation derived from the mepp . together with the explicit expression of the diffusion coefficient ( [ e51 ] ) it provides a self - consistent model for the `` coarse - grained '' dynamics of collisionless stellar systems experiencing violent relaxation . more general equations including anisotropic effects can also be obtained from this formalism @xcite . we shall now assume that the galaxy is not isolated but subject to the tides of other systems . in that case , high energy stars that have elongated orbits are removed by the gravity of these objects . we seek therefore a stationary solution of equation ( [ e49 ] ) satisfying the boundary condition @xmath225 , where @xmath226 is the escape energy above which @xmath227 . this solution will provide a truncated distribution function with a finite mass @xcite . in fact , this problem was already tackled by lynden - bell in his seminal paper @xcite . he proposed to describe the evolution of the coarse - grained distribution function @xmath43 by the ordinary fokker - planck equation ( [ e26 ] ) with the heuristic argument that the fluctuations of the gravitational potential during violent relaxation play the same role as collisions . with the additional ( heuristic ) argument that @xmath228 for large velocities , he could obtain a stationary solution of a michie - king type @xcite : @xmath229 since stars with @xmath230 are removed by the tidal field , this distribution function provides a depletion of high energy states and solves the infinite mass problem . our present approach justifies the two phenomenological arguments of lynden - bell since equation ( [ e49 ] ) reduces to the fokker - planck equation if we assume @xmath131 ( no degeneracy ) and equation ( [ e51 ] ) gives a diffusion coefficient @xmath228 for large velocities . there is however a kind of `` gap '' in lynden - bell s approach since equation ( [ e52 ] ) does not reduce to the fermi - dirac statistics ( [ e10 ] ) in the limit of low energies . using the more general equation ( [ e49 ] ) which properly accounts for degeneracy effects , we now try to build up a `` one piece '' distribution function which makes the bridge between lynden - bell s statistics ( [ e10 ] ) for @xmath231 and the michie - king model ( [ e52 ] ) for @xmath232 . during the stage of violent relaxation , the stars extract their energy from the rapid fluctuations of the gravitational field . by this process , some stars may acquire very high energies and escape from the system ( being ultimately captured by the gravity of other systems ) . for these stars , @xmath233 is a good approximation and equation ( [ e49 ] ) reads @xmath234 we seek a stationary solution of the form @xmath235 . using the identity @xmath236 ( valid for sufficiently large @xmath237 ) , we obtain @xmath238 or , equivalently , @xmath239 where @xmath240 is a constant of integration . if @xmath241 , we recover lynden - bell s distribution function ( [ e10 ] ) . if @xmath242 , equation ( [ e55 ] ) accounts physically for an escape of stars at a constant rate @xmath240 . the system is therefore not truly static since it looses gradually stars but we may consider that the galaxy passes by a succession of quasi - stationary states which are solution of equation ( [ e55 ] ) . as stated previously , this equation is valid only for high energy stars . for lower energies , the system has settled down in a pure equilibrium state and @xmath241 , leading to lynden - bell s distribution . our goal , now , is to solve equation ( [ e55 ] ) . put under the form @xmath243 we recognize a riccatti equation @xcite . with the change of variables @xmath244 , it can be converted into a linear partial differential equation of second order @xmath245 the associated characteristic polynomial @xmath246 has a strictly positive discriminant @xmath247 . therefore , the general solution of equation ( [ e57 ] ) is @xmath248 where @xmath249 and @xmath250 , @xmath251 are constants of integration . the solution of equation ( [ e56 ] ) is therefore @xmath252 setting @xmath253 ( @xmath254 otherwise @xmath43 would be constant ) , it can be rewritten @xmath255 this distribution function vanishes at the escape energy @xmath256 defined by @xmath257 with this new variable , we obtain the result @xmath258 where @xmath259 is a solution of @xmath260 now , for the cases of physical interest @xmath261 , which means that degeneracy is negligible for energies close to the escape energy @xmath226 . in that case , @xmath262 and we obtain the final result @xcite : @xmath263 the previous calculation is justified as long as @xmath264 , corresponding to relatively strong fluctuations . when the fluctuations die down at the end of the relaxation , the diffusivity @xmath128 and therefore the diffusion current @xmath117 go to zero . there is no more evaporation but the distribution function ( [ e64 ] ) is maintained as a stationary solution of the vlasov equation . when @xmath232 , we recover the michie - king model ( [ e52 ] ) and when @xmath265 equation ( [ e64 ] ) reduces to the fermi - dirac distribution function ( [ e10 ] ) . therefore , the distribution function ( [ e64 ] ) connects _ continuously _ the two limits considered by lynden - bell @xcite and can serve as a relevant model for ( possibly degenerate ) collisionless stellar systems . in particular , this distribution function could describe galactic halos limited in extension as a consequence of tidal interactions with other systems @xcite . it could also be used to calculate realistic equilibrium states of collisionless stellar systems without the artifice of a material box . however , the main results of the box model @xcite should not be dramatically altered . the previous model describes self - gravitating systems subject to tidal forces . this form of confinement was first introduced in the case of globular clusters tidally trucated by a nearby galaxy @xcite . this model can also describe a fraction of elliptical galaxies living in a rich environement . however , for the majority of elliptical galaxies , tidal forces are weak and the galaxy can be considered as isolated . now , for isolated systems , other processes can account for `` incomplete relaxation '' and lead to different truncated models . starting from equation ( [ e36 ] ) , we can show that the diffusion coefficient is proportional to the fluctuations of the distribution function integrated over the velocity : @xmath266 since these fluctuations rapidly decay as we depart from the relatively well - mixed central region of the galaxy , the diffusion current decreases also and this results in a confinement of the distribution function . it is expected therefore that the relaxation will be effective only in a limited region of space where @xmath267 is sufficiently large . on the other hand , as the system develops finer and finer filaments during the mixing process , the diffusion coefficient is expected to decrease in time . the diffusion current takes therefore increasingly small values and the relaxation is slowed down . this decay may be quite rapid and a quantitative treatment of this effect would require a better understanding of the correlation function @xmath268 whose expression was simply postulated in section [ qt ] . the development of these ideas will lead to other truncated models probably closer to those introduced phenomenologically by , e.g. , stiavelli & bertin @xcite , tremaine @xcite and hjorth & madsen @xcite . in particular , the above discussion is quite consistent with the scenario of incomplete violent relaxation developed by hjorth & madsen . these authors introduce a two - step process : ( i ) in a first step , they assume that violent relaxation proceeds to completion in a _ finite _ spatial region , of radius @xmath269 , which represents roughly the core of the galaxy ( where the fluctuations are important ) . at this stage , the escape energy represents no special threshold so that negative as well as positive energy states are populated in that region . ( ii ) after the relaxation process is over , positive energy particles leave the system and particles with @xmath270 move in orbits beyond @xmath269 , thereby changing the distribution function to something significantly ` thinner ' than a boltzmann distribution . the crucial point to realize is that the differential energy distribution @xmath271 , where @xmath272 is the number of stars with energy between @xmath273 and @xmath274 , will be discontinuous at the escape energy @xmath275 since there is a finite number of particles within @xmath269 after the relaxation process . it can be shown that this discontinuity implies necessarily that @xmath276 for @xmath277 @xcite . therefore , the truncated model consistent with this scenario is @xcite : @xmath278 this truncated model corresponds formally to a composite configuration made of an isothermal core and a polytropic envelope with index @xmath279 . of course , if the core is degenerate , the boltzmann distribution for @xmath280 must be replaced by the fermi - dirac one . this truncated model gives very good fit with elliptical galaxies and can reproduce the @xmath1 law @xcite . the box model of section [ fd ] can be considered as a simple approximation of this more realistic model , the `` box '' playing the role of the envelope . in fact , the scenario developed by hjorth & madsen is almost equivalent to complete violent relaxation in a finite container with the container removed after the relaxation . this scenario solves the infinite mass problem and rehabilitates the use of statistical mechanics to understand the structure of elliptical galaxies . we have described in this paper the process of violent relaxation in stellar systems from the viewpoint of statistical mechanics , as originally introduced by lynden - bell @xcite . lynden - bell proposed that the structure of galaxies could result from a law of chaos : there is a total lack of information at small scales since the stars have complicated orbits , but the exciting phenomenon is that _ microscopic _ disorder leads to _ macroscopic _ order . the same ideas of statistical mechanics have been introduced in two - dimensional turbulence described by the euler equation @xcite to explain the formation and maintenance of large scale coherent vortices like the great red spot of jupiter or the cyclones and anticyclones that populate the earth atmosphere @xcite . the formation of self - organized vortices in two - dimensional turbulence can also have applications in the context of planet formation where large - scale vortices present in the solar nebula could efficiently trap dust particles to form the planetesimals and the planets ( see a complete list of references in chavanis @xcite ) . in fact , the statistical mechanics of the 2d euler equation is equivalent to the theory of lynden - bell applying to the vlasov equation . in a sense , the vlasov equation is just the euler equation for a `` fluid '' evolving in a six - dimensional phase space . in this analogy @xcite , the vorticity and the stream function in 2d turbulence play the same role as the distribution function and the gravitational field in galaxies . this analogy concerns not only the equilibrium states ( the formation of large - scale structures ) but also the relaxation towards equilibrium @xcite and the statistics of fluctuations @xcite . a kinetic theory of two - dimensional vortices can be developed in analogy with stellar dynamics @xcite . in this kinetic theory , a point vortex experiences a diffusion process and a `` systematic drift '' . this `` systematic drift '' @xcite is the counterpart of the `` dynamical friction '' @xcite experienced by a star as a result of close encounters . relaxation equations analogous to equations ( [ e25 ] ) and ( [ e45 ] ) have been derived in the context of vortex dynamics @xcite . in addition , the problem of `` incomplete relaxation '' is also encountered in 2d turbulence to explain the confinement of a vortex ( e.g. , a dipole or a tripole ) that forms after a rapid merging @xcite . it has been demonstrated explicitly in two - dimensional turbulence ( where the numerical simulations are easier to implement ) that the relaxation equations derived from the mepp and including a space dependant diffusion coefficient related to the fluctuations of the vorticity ( analogous to eq . ( [ e65 ] ) ) can account for this kinetic confinement @xcite . we believe that the relaxation equations proposed in the stellar context should work as well . the statistical mechanics of continuous vorticity fields also predicts a fermi - dirac distribution at equilibrium with the same interpretation of the degeneracy as in lynden - bell s theory . although this degeneracy is hard to evidence for galaxies ( and remains controversal ) , it has been vindicated by various numerical simulations and laboratory experiments of two - dimensional fluids . this suggests that the degenerate version of the theory should also be used in the stellar context . if all these effects are taken into account correctly , it is plausible that the statistical mechanics of 2d vortices and self - gravitating systems has a chance to account for the fascinating process of self - organization in nature .

we consider the problem of detecting clones in wireless mobile ad - hoc networks . we assume that one of the devices of the network has been cloned . everything , including certificates and secret keys . this can happen quite easily , because of a virus that immediately after sending all the content of the infected device to the adversary destroys itself , or just because the owner has left his device unattended for a few minutes in a hostile environment . the problem is to detect this attack . we propose a solution in networks of mobile devices carried by individuals . these networks are composed by nodes that have the capability of using short - range communication technology like blue - tooth or wi - fi , where nodes are carried by mobile users , and where links appear and disappear according to the social relationships between the users . our idea is to use social physical contacts , securely collected by wireless personal smart - phones , as a biometric way to authenticate the legitimate owner of the device and detect the clone attack . we introduce two mechanisms : personal marks and community certificates . personal marks is a simple cryptographic protocol that works very well when the adversary is an insider , a malicious node in the network that is part , or not very far , from the social community of the original device that has been cloned . community certificates work very well when the adversary is an outsider , a node that has the goal of using the stolen credentials when interacting with other nodes that are far in the social network from the original device . when combined , these mechanisms provide an excellent protection against this very strong attack . we prove our ideas and solutions with extensive simulations in a real world scenario with mobility traces collected in a real life experiment . [ complexity measures , performance measures ]

1105.3716

you have left your smartphone on the table at a cafeteria . you soon realize and go back to take it . fortunately , it is still there ! you feel safe while you are not safe at all . an adversary has connected your smartphone to a laptop and dumped all of its memory , including public and secret cryptographic keys . it is a matter of seconds or , at most , minutes . you do nt revoke your certificates ( you feel safe ! ) and , a month later , you discover that your credentials have been used by someone else . if you think this can not happen people take very good care of their personal devices consider that according to a fairly recent report ( wtop , 15 nov 2006 ) 478 laptops have been lost or stolen from the irs ( the internal revenue service is the united states federal government agency that collects taxes and enforces the internal revenue laws ) between 20022006 ; 112 held sensitive taxpayer data , including ssns . portable personal devices smart - phones , laptops , and pdas are more and more used in our everyday life . we use them to make phone calls , to plan our activities , to surf the web , to manage our banking account , and probably very soon to make purchases . the system we consider is a network of personal smart - phones that connects to each other by using short - range communication technology like blue - tooth . the nodes are the devices and the links appear and disappear as people move and get in physical touch . as a consequence , the network is not a collection of randomly moving objects it has a social structure that can be exploited to deliver revolutionary applications , and , we believe , new and surprisingly effective solutions to classical networking and security problems . these networks have been called with several different names_pocket switched networks _ @xcite , _ mobile social networks _ , _ opportunistic mobile ad - hoc networks _ , among others and have drawn the attention of many researchers in the community . most of the work has focused on forwarding , that is how to route messages in such a way to deliver them as quickly as possible and as cheaply as possible to destination . here , we focus on security . the attack that we have described in the first rows of this paper is the _ clone attack_. the clone attack might as well be performed by a virus that infects our device and sends all the data , including secret keys associated with the certificates , to the adversary . so , you are not safe even though you never forget your smart - phone at the coffee shop . in this paper we are looking for a solution to this problem . of course , we do not want one that consists in typing a password every time we use our certificates . people do nt like passwords and tend to forget them , or , even worse , choose trivial ones . our idea is to use social physical contacts , securely collected by wireless personal smart - phones , as a biometric way to authenticate the owner of the device . indeed , our social physical contacts our family members , the same barman everyday at the coffee shop , our colleagues at work , our old friends when we hang out and relax after work characterize ourselves in a distinctive way . of course , each day is somewhat different and we do nt always meet the same people , but surely there is a strong regularity in the communities we live in and in circle of friends that we usually meet in person . we will see how to use this regularity , complemented with other essential mechanisms , to detect the clone attack in wireless networks of smart - phones or other personal devices and to prevent the misuse of stolen certificates . in this paper we introduce two protocols : personal marks and community certificates . personal marks is a simple cryptographic protocol that proves to be an excellent way to detect the attack when the adversary is an _ insider _ , a person that belongs to the same community of the victim . community certificate is a solution based on certificates that tells how the node is expected to behave in terms of social contacts . if the clone is an _ outsider _ , a node that behaves in a different way with respect to the victim and meets different nodes and different communities , then community certificate works very well , the certificate soon expires , and the clone can not authenticate any more to the other nodes of the network . the two protocols are meant to be used at the same time and , according to a large set of experiments made with well - known traces of human contacts computed during real life experiments , they collectively prove to protect the nodes against the clone attack in an excellent way . the detection of the clone attack is one of the most investigated security issue in wireless ad - hoc networks . as far as static wireless networks are concerned , there are three main approaches to the problem : centralized , local and distributed random based techniques . the centralized techniques @xcite require nodes to send either location information or key information along with the node i d to a base station which in turn processes the information received and , in case of anomalies ( same node i d with different locations ) , trigger a revocation procedure . aside from presenting a single point of failure ( the base station ) , this approach has the drawback of incurring high communication cost to the network due to the high number of messages involved . local - based schemes @xcite make use of voting mechanisms within nodes neighborhoods to detect clones . though they do nt suffer from the drawbacks of the central based schemes , they fail on detecting replication attacks where node clones are scattered in different areas of the network . the distributed and random based techniques @xcite require nodes to send signed location information to randomly selected destinations on the network in a hop - by - hop fashion . these schemes rely on the high probability of intersection of different location - declaration routing paths started from replicas of the same node . the nodes on which such intersection occurs are called witnesses , and their task is to then trigger node revocation procedures . another distributed scheme @xcite uses a random value , distributed to the nodes by the base station , to generate independent clusters in the network . each cluster is an exclusive subset maximal independent set ( esmis ) and has a cluster head denoted with sldrs . one or more trees ( whose nodes correspond to sldrs ) are defined over the network graph , and bottom - up aggregation protocols on such trees are used to detect node replicas ( that are present in more than one esmis ) . although these distributed techniques do not present single points of failure , the overhead they incur either in message traffic or in computational terms @xcite is far from negligible . moreover , all the aforementioned techniques , by relying on fixed geographical position of the nodes in the network are not apt to be used in mobile scenarios such the one we consider @xcite . with the increasing use of cellular technology , cloning of mobile devices for the purpose of making fraudulent telephone calls become a real threat to phone carriers . while this was a real threat in the nineties when cdma was adopted @xcite , in the gsm era it started to be considered as a minor / non feasible attack , at least for fraudulent billing purposes . with the spreading of the mobile pay systems ( 72.8 million of users at the end of 2009 , expected 220 million by 2011 in china only @xcite ) , and especially those based on credit card transactions , phone cloning has become a vicious threat to the users portfolio @xcite . thus a lot of research on detecting such attacks have been done , most of which is based on the use of neural networks by the carrier to detect possible anomalies . for a good survey see @xcite . the idea of exploiting information regarding social ties between nodes is not new , actually it is common to a good part of the literature on pocket switched networks ( psn ) and similar social networks . much research has been dedicated to the analysis of the data collected during real - life experiments , to compute statistical properties of human mobility , and to uncover its structure in sub - communities @xcite . later on , most of the work in the field focused on message forwarding and to find the best strategy to relay messages in order to route them to destination as fast as possible ( see @xcite , among many others ) . also security problematics such as node capture @xcite or selfishness @xcite have been solved by making use of social relationships among nodes . biometrics have been used for centuries to authenticate people . the most common example of biometrics is the handwritten signature that we use everyday to certify our identity . with technological development , the use of biometrics to authenticate has become a well - known area of investigation in computer security . several biological measurements have been used to identify people to computer systems : voice , face , iris , keystroke dynamics ( the way we type is often sufficiently unique to identify ourselves ) , and others . in many cases these are not fully automatic . with keystroke dynamics , for example , the user has to type . as an excellent reference and starting point on computer biometrics , see @xcite . the authors in @xcite propose two intrusion detection systems that have similar biometrics ideas : the first one is build upon radio frequency fingerprinting ( rff ) , whereas the second one leverages user mobility profiles ( ump ) . however both detection systems are centralized , and rely on the fact that the intruder ( the clone ) behaves substantially differently from the real user in terms of geographical movements . thus , compared with the solutions proposed in this paper , both systems are based on a completely different idea and are not able to detect anomalies when the clone behaves similarly to the original node ( for example , when the clone attack happens in a building ) . lastly , the solutions are not distributed . to the best of our knowledge , this is the first paper that presents biometric authentication techniques based on the social contacts of the owner or the device to detect the clone attack in mobile wireless networks of smart - phones . the techniques , namely _ community certificates _ and _ personal marks _ , are _ only _ build upon the social - guided meeting patterns of users in the mobile social network . they are thoroughly orthogonal to trust or voting mechanisms , or geography based techniques to detect cellphone theft or credit card fraud , that are based on different ideas and can be used at the same time . our network setting is made of last generation smart - phones . these devices are more and more popular we can easily envision a society in which almost everybody will carry a personal computing device of this kind . commercial smart - phones are usually able to communicate by using short - range communication technology , like blue - tooth and/or wi - fi , aside classic gsm . when two people meet , their personal devices establish a link by using one of the above mentioned short - range technologies . in the network we consider , nodes are devices carried by people , and links appear and disappear as people move and meet . smart - phones are not - so - small devices that can easily handle video / audio streaming , 3d games , web surfing and ssl sessions , and other applications . therefore , we can safely assume that nodes are able to perform public key cryptography that is used to sign messages and to establish secure communication sessions among peers . the nodes are equipped with public / private key pairs , and the former is signed by a trusted authority ca . we assume that our protocols execute by setting up authenticated and encrypted sessions among the peers . nodes are loosely time synchronized . loose time synchronization is very easy to get , if a precision in the order of the second is enough , like in our protocols . we also assume that the trusted authority is able to send a message to any node in the system , for example using the cellular network . when a clone is present , the message is received by the original node and by the clone as well ( of course it is perfectly possible that the clone has turned off its interface to the cellular network ) . our protocols rely on authenticated logs of physical contacts between devices . however , if not carefully implemented , these logs can be attacked . we assume that the nodes use techniques like distance bounding @xcite to guarantee that the log of a contact between two devices can be collected only when the devices are provably in physical proximity . these mechanisms are based on measuring the delay in the communication . lastly , we assume that the users have access to an alternative way to authenticate to the authority . there are several examples of such mechanisms . one example is gmail : if you forget your password , you can still authenticate by responding to a list of personal questions that , most probably , only you can respond . in other systems , you might be able to authenticate by using a smart - card at your desktop at home . another example is the pair of pin and puk codes used in cellular phones , the first one is short and easy to type , the second one longer , more tedious , and is meant to be used only to recover from pin blocking or other uncommon events . in any case , we assume that the alternative mechanism to authenticate is secure but long , burdensome , and we definitely want to use it only in rare and exceptional circumstances like when we need to recover from a clone attack . in the rest of this paper , we use @xmath0 to denote a message @xmath1 signed by node @xmath2 . we consider the following scenario : a smart - phone is infected by a worm or by a virus . the goal of the virus is to make a perfect copy of the device of the victim , to send all the data to the adversary s device , and to destroy itself without leaving any trace . a similar attack can be performed in a different way : you left your device unattended for just a couple of minutes ; when you go back to take it , it is still there . what you do nt know is that the adversary has made a clone by copying all the memory of your device . later on , the clone can be used to interact in the network by using the victim s credentials . note that everything has been cloned , including certificates and secret keys . we assume that the clone can be switched off arbitrarily by the adversary to prevent detection the device might be turned on only when necessary ( that is , when it is used to connect to the other peers in the network with the stolen credentials ) . to start with , we assume that the adversary has been able to clone only one device . later in the paper we will discuss what is going to happen if more devices can be stolen . our goal is to build a system able to detect and/or stop as fast as possible and as accurately as possible this attack . the users of our system are humans their mobility pattern , the people they meet , and the places they visit are not random . rather , the dynamics of the network is highly influenced by the fact that almost everything in our everyday life is guided by social relationships ( e.g. friends , family ) and interests ( e.g. work , school , gym , or playing chess ) . if we think of our daytime , it usually follows a do - repeat cycle that might look like this : go to work ( or school ) ; meet with colleagues ; finish work and go out with the same old friends or do whatever we like to do during our free time ; go back home ; go to work again and so on . of course , it is not always like that ( fortunately ) , but surely there is a regularity in the things we do and the people we meet . the same cycle is performed by our mobile device , and the ones of our friends , family members , or colleagues that we meet over and over again . the people ( devices ) we meet everyday characterize ourselves in a remarkably distinctive way . one of the ideas in this paper is to use this characterization as efficiently as possible in order to make nodes prove who they are not with something they know ( passwords ) , or something they have ( certificate ) , but with the proof that they actually meet physically and regularly the people they are supposed to meet . we start by introducing personal marks . personal marks is a very simple cryptographic protocol we can use to distinguish clones . actually , personal marks is able to tell you that the person that you have before is not the same person that you met last time , though it ca nt really tell who is the impostor . let s make a step back and proceed more formally . personal marks consists in _ marking _ each node in your community with a small cryptographic object , signed by yourself , that you can check to have some level of guarantee that node @xmath3 is exactly the same node @xmath3 you have met earlier . first of all , we call _ community _ of node @xmath2 its set of friends the set of other nodes that gets frequently in physical contact with node @xmath2 . one way to compute the community of node @xmath2 is by using the distributed community detection algorithm described in @xcite . communities are a fundamental notion in social networks , people meet more often other people of the same communities and this intuitive property have been leveraged in the literature in several ways . personal marks works as follows : when two nodes of the same community @xmath2 and @xmath3 meet , they both put a mark on the other device . the mark that node @xmath2 gives to node @xmath3 contains a timestamp signed by node @xmath2 itself . same is done by node @xmath3 and given to node @xmath2 . the next time they meet , they both request the mark to check it before starting the new session . if there is an clone present in the community , say a clone @xmath4 of original node @xmath3 , and @xmath4 gets in touch with the community of @xmath3 , personal marks is going to detect the anomaly . assume that node @xmath3 has been cloned at time @xmath5 and that node @xmath2 gets in physical contact with node @xmath3 for the first time after the attack at time @xmath6 . clearly , node @xmath2 does not know if the node is the original or the clone and , by using personal marks , it freshens the mark with the node . the detection is triggered when node @xmath2 meets the _ other _ node , that can not help but exchanging the old version of the mark , a mark with a timestamp in the interval @xmath7 that does not pass the test by node @xmath2 . if the adversary is an insider , node @xmath2 will soon meet both the nodes , the clone and the original one ( that is a friend ) . while node @xmath2 is not able to tell who is the original node and who is the impostor , the clone attack is detected and node @xmath3 is revoked by the authority . the legitimate node @xmath3 has now to go through an alternative , longer procedure to re - authenticate . in our system , a personal mark is a small cryptographic object that carries some randomness from both peers . the authentication protocol in personal marks is made of two parts : the mark - check , and the mark - exchange . let us start from the mark - exchange protocol . each time two nodes @xmath2 and @xmath3 meet , they contribute some randomness @xmath8 and @xmath9 to freshen the mark ( or to establish one if this is the first time the nodes meet ) . the mark is of the form @xmath10 , where @xmath11 is a timestamp ( the timestamp is not crucial for the system to work , but it gives information about the timing of the attack in case of detection ) . the mark is signed and stored by both peers . note that , if the protocol is not completed ( may be for lack of continuos connectivity ) , the peers can safely assume that it has not happened . @xmath12{r_i } j\\ i & \xleftarrow[\hspace*{5cm}]{\langle \text{mark},t , r_i , r_j\rangle_j } j\\ i & \xrightarrow[\hspace*{5cm}]{\langle \text{mark},t , r_i , r_j\rangle_i } j\\ i & \xleftarrow[\hspace*{5cm}]{\langle \text{ok}\rangle } j\\ & \text{\ ; both nodes $ i$ and $ j$ store the mark.}\end{aligned}\ ] ] let us now look at the mark - check protocol , executed before the mark - exchange protocol when it is not the first time that nodes @xmath2 and @xmath3 meet . suppose , without loss of generality , that first node @xmath2 checks node @xmath3 and then node @xmath3 checks node @xmath2 . node @xmath2 sends a request to which @xmath3 has to reply with the last version of the mark . after receiving the mark from @xmath3 , @xmath2 checks the signature and the timestamp and notifies node @xmath3 . in the case when the test is failed , both the correct mark and the mark exchanged during this protocol can be sent to the authority by using gsm or broadcast in the network as a proof of the clone attack , and the credentials of node @xmath3 are thus revoked . of course , the legitimate node @xmath3 is also invited to re - authenticate and get new credentials . @xmath12{\langle \text{mark\_rqst}\rangle } j\\ i & \xleftarrow[\hspace*{5cm}]{\langle \text{mark\_rpl},\,m\rangle } j\\ \nonumber i & \xrightarrow[\hspace*{5cm}]{\langle \text{markcheck\_ok}\rangle } j\\ & { \hspace{5mm}\normalsize\text{or}}\\ \nonumber i & \xrightarrow[\hspace*{5cm}]{\langle \text{markcheck\_failed}\rangle } j\end{aligned}\ ] ] note that if a clone refuses to initiate or to complete either of the two protocols with the goal of avoiding detection , then it does not authenticate and its cloned certificates are going to expire soon thanks to community certificates , the other mechanism we propose in this paper that works very well when the adversary does not meet the members of the communities of the original owner . personal marks works on the assumption that both the clone and the original node get in touch with a common node in the community of the original node , sooner or later . this is often true the adversary is someone that is part of the same community of the victim : a colleague at workplace , or a fellow student at college . in this case , we will say that the adversary is an _ insider _ , and personal marks works extremely well as shown later in the experiments with real mobility traces . however , this is not always the case . it is perfectly possible that the clone uses the credentials that it has stolen far from the communities of the original node in the network . in this case , we will call the adversary an _ outsider_. this is also a very important scenario , in many real cases the attack can be performed by someone that we do nt know and that has the goal of taking our device to a different city or in a different part of the same city and use the credentials when needed . clearly , personal marks ca nt do much against outsiders . outsiders never meet someone from the community of the victim . therefore , they are never checked and they do nt need any fresh mark . to fight back outsiders , we have to develop a completely new mechanism . when a node , say node @xmath2 , agrees to enter the system , it automatically enters a training period during which it securely collects signed and timestamped logs of the physical contacts with other nodes . at the end of the period the logs are reported to the authority . the authority uses the logs to build a signed certificate @xmath13 that is sent back to node @xmath2 . clearly , all these messages are encrypted and authenticated , like all sessions in our protocols . the certificate is of the form @xmath14 . in the certificate , @xmath15 is the set of `` best friends '' of node @xmath2 ( its community ) ; @xmath16 is a mapping that tells the `` strength of the relationship '' between node @xmath2 and his best friends . more specifically , for every @xmath17 , @xmath18 is a value computed as a monotone function of the inter - contact times between nodes @xmath2 and @xmath3 observed during the training period . moreover , for every @xmath17 , @xmath19 is a timestamp signed by node @xmath3 that certifies at what time there has been the last contact between node @xmath2 and @xmath3 ( this object is similar to the mark in personal marks ) . given node @xmath20 , we say that the timestamp @xmath19 is _ fresh _ if @xmath21 . certificate @xmath22 is valid at time @xmath11 if and only if at least @xmath23 signed timestamps in set @xmath24 are fresh . in other words , the certificate is valid only if the node has been able to collect enough fresh signatures by physically meeting people in his circle of friends . when the authority generates the certificate , @xmath24 can be prepared with timestamps signed by the authority , just to make the certificate immediately valid . note that all this process is totally automatic . note as well that we assume that no attack occurs during the training period . this is quite reasonable indeed , if such an attack takes place , the authority would receive training logs from both the victim and the clone at the same time , thus revealing the attack . the community certificate is a requirement for the authentication of node @xmath2 to the other peers of the system . indeed , when @xmath2 meets another node , say node @xmath3 , it is required to show the certificate @xmath13 in case it is willing to set up a communication or to use one of the services provided by node @xmath3 . the simple protocol is described in figure [ fig : authentication ] . @xmath25{\langle \text{auth\_rqst}\rangle } e\\ i & \xrightarrow[\hspace*{5cm}]{\langle \text{auth\_rpl},\,comc_i\rangle } e\\ \nonumber i & \xleftarrow[\hspace*{5cm}]{\langle \text{auth\_ok}\rangle } e\\ & { \normalsize\text{or}}\\ \nonumber i & \xleftarrow[\hspace*{5cm}]{\langle \text{auth\_denied}\rangle } e\end{aligned}\ ] ] as mentioned , an adversary that has the goal of using the cloned smart - phone outside the communities of the legitimate owner is an outsider . community certificates proves to be an excellent mechanism to stop clone attacks by outsiders . indeed , when taken outside the communities of the owner , the certificate expires soon . the adversary is simply not able to collect enough fresh signatures to keep the certificate valid . in the experiments with real traces , we show that it is possible to choose @xmath15 , @xmath16 , and @xmath23 in such a way that the certificates are continuously and consistently valid , when carried by the legitimate owner , and expire quickly when carried by outsiders . therefore , a combined use of community certificates and personal marks is an excellent way to provide efficient and secure authentication and to thwart the clone attack in mobile wireless networks of smart - phones . in section [ sec : caveats ] we discuss a few ideas , add - ons , and solutions to extend the features of community certificates and to deal with non - frequent scenarios like the case when the legitimate owner has to use his own community certificate during a long travel far from his usual communities . so far , we have described community certificates as a static system . however , in real life it may be possible , even if it is not common , that we change our own community . in general , it is reasonable to imagine the following scenarios : ( i ) our community changes completely since , for example , we move to another town ; and ( ii ) one of our friends moves away , or a new node is our new best friend . here , we see that it is easy to design protocols to dynamically change the community certificate in a secure way . in case ( i ) , it is enough to start off a new training phase and to get a new certificate . in case ( ii ) , we can initiate a selective update of the certificate to remove one node , or to add a new one , or to update the parameter of a node that is already part of our ring . of course , the addition and/or the removal can change all the parameters of the certificate , like mapping @xmath16 or @xmath26 . the procedure can be easily secured . indeed , when the procedure starts , the authority sends a gsm message to the node . if a clone requests the procedure to change the certificate according to his own communities , than the message is received by the original owner as well , that promptly detects the attack and sends to the authority a signed request of certificate revocation . in this paper , we consider the problem of detecting a single clone attack . generally speaking , if the adversary is very powerful , it is however possible that it clones a whole set of mobile nodes . while we do nt explicitly deal with this case in this work , it is still useful to see what is going to happen with personal marks and community certificates . what we can say is that personal marks keeps working with ( approximately ) the same performance . according to our experiments , communities are not closed sets of nodes . usually nodes belong to many different communities and the structure of the network is characterized by many _ overlapping _ communities . this is in accordance with other well - known research work in the literature @xcite . in other words , it is usually impossible to select a small community of nodes that have very little contacts with nodes outside the community ( of course , if you can select the whole network than it is easy , but if the adversary is able to clone the whole network that there is little we can do about it ) . therefore , personal marks is still efficient in detecting the attack , thanks to all the interactions between the nodes of the cloned community with nodes outside the community . a similar argument is true for community certificates it is usually impossible to isolate a small community whose nodes do not have nodes that are outside the community in the certificate . however , it is in principle possible to isolate and clone a community in such a way that the cloned devices can authenticate each other . to make this attack possible only if the community to be cloned is very big , parameter @xmath23 must be set high enough , and certificates should be computed in such a way not to form cliques . while this is out of the scope of this paper , it is certainly a open and interesting research direction . in any case , that means that the adversary is able to hijack a whole community . moreover , we will see how to extend community certificate by admitting fixed devices in the certificate . for example , a fixed infrastructure in both our home and our office . while cloning a device can be easy , stealing a fixed infrastructure can be considerably harder . in this section , we present some experimental results in order to show the detection performances of our system . the results refer to the two typologies of attack we consider : insiders and outsiders . in the first case the detection relies on the personal marks sub - system . rather , in the second , the community certificates sub - system comes into play and the detection is determined by the expiring time of the certificate when in the hand of the adversary . we start off by giving detailed information on the dataset we used to evaluate our protocol . then we present the results obtained with personal marks and with community certificates . all the experiments have been performed by using a simulator driven by three type of traces : wlan ( real ) , bluetooth ( real ) , and simulated traces . for the wlan traces we used the _ dartmouth _ @xcite and _ ucsd _ @xcite datasets . the bluetooth trace we used is the _ reality _ @xcite , collected during the reality mining project at the mit media lab . finally , the simulated trace has ben generated using the _ swim _ @xcite model . in contrast with the simulated scenario , in which we have been able to select the number of nodes and the length of the experiment , in the case of wlan and bluetooth traces we had to select a period of time reasonably long that contained a set of nodes that are fairly active . the reason is that real traces often suffer from data loss and contain nodes whose activity is recorded only for a portion of the total time span covered by the trace . this might cause loss of performance of our system . the dartmouth trace consists of the smnp logs of the access points across the dartmouth college campus from april 2001 to june 2004 . to infer the contacts between the nodes we follow the assumption widely used in the literature that two nodes are able to communicate to each other whenever they are associated to the same access point ( @xcite @xcite ) . from this trace we have selected a time span of 8 weeks , from january 5 , 2004 to march 1 , 2004 , during which a set of @xmath27 nodes have recored to have at least @xmath28 contacts a day for at least the @xmath29 of the days . this ensures us that these nodes stay active during the whole period we used to evaluate our protocol . the ucsd trace is part of the wireless topology discover project ( wtd ) @xcite . the trace consists of the logs extracted from pda carried by a set of about 275 freshmen students of the university of california , in san diego . the trace spans a period of @xmath30 weeks between semptember 22 , 2002 and december 8 , 2002 , during which each pda periodically recorded the signal strength of all the aps in its range . to infer the contacts between the nodes we again used the assumption that two nodes can communicate as long as they are in the range of a the same access point . as also reported in @xcite , we have found that the trace is characterized by a steady decline of the user population that especially affects the last two weeks . for this reason we decided to restrict our tests to the first @xmath31 weeks and to use only the set of @xmath32 nodes that recorded at least 1 contact a day for at least the @xmath29 of the days . as opposed to the dartmouth and ucsd traces that used wifi radio logs , the reality trace has been collected using short ranged bluetooth radios . more in detail , the trace included the bluetooth records collected by 94 cellphones distributed to student and faculty on the mit campus during 9 months ( from september 2004 to june 2005 ) . we have chosen this particular trace since it is one of the few existing traces which encompasses a large number of nodes communicating through bluetooth technology for a long period of time . like other traces , however , many nodes recorded very few to no sightseeings for long periods of time . in order to keep the amount of nodes high we thus restricted ourselves to a period of 8 weeks going from october 18th to december 13th 2004 and discarded the nodes that did nt record at least 1 contact a day for at least the @xmath33 of the days . this selection yielded a final set of @xmath34 nodes . this trace has been generated using the swim model ( @xcite ) that has been shown to simulate well human mobility in conference and university campus environments . the swim simulator has also been shown @xcite to be able to properly scale a reference scenario by keeping the nodes density constant . thanks to this we have been able to replicate the statistical and social properties of the cambridge campus bluetooth data set @xcite ( that is only 11 days long ) increasing both the number of nodes and the time span by keeping the same dynamics of the original real trace in terms of average number of contacts per user : the generated trace contains @xmath35 nodes and is @xmath31 weeks long . since the simulator keeps track of all the contacts between the nodes , it hasnt been necessary to preprocess the trace and we ve been able to use the whole set of nodes . as we already discussed in the section [ sec : communitycertificates ] , before a node can use community certificates , it is required to go trough a _ training period _ during which it collects logs of contacts . at the same time , the personal marks system described in section [ sec : marks ] assumes that each node know its community . for this reason we decided to split all the traces in two parts : the former used as a training period for all the nodes , the latter used to test the performances of our protocols . during the training period , we both collect data about the contacts between the nodes needed by the community certificates and extract the communities of the nodes . for this , we use the @xmath23-clique algorithm @xcite , one of the most popular in the area of social mobile networking ( @xcite ) , of which there is a distributed version ( @xcite ) . since the movement of people is usually characterized by a high degree of regularity , we have seen that it is not necessary that the training period be long : in every trace we used a training period consisting of only the initial @xmath36 of the total time span of the trace . this is enough for our system to quickly detect both insiders and outsiders . a further evidence that our intuition is good is given by the results shown in table [ tab : traces ] , where it can be seen that the properties of the trace during the training period are very close to those of the rest of the trace . in particular , the last row of the table shows what we have called the average _ stability _ of the nodes . we defined the stability of a node @xmath37 as a measure of the average similarity between the set of nodes @xmath37 met during the training period and those it met during each week of the rest of the trace . more in detail , if @xmath38 is the set of nodes that @xmath37 met during the training period and @xmath39 is the set of nodes that @xmath37 met during the @xmath2-th week of the trace , the stability of @xmath37 is computed as @xmath40 where @xmath41 is the number of weeks of the trace ( training period excluded ) . as shown in the table , the average stability of the nodes is always close or greater than @xmath42 meaning that , on average , a node meets every week at least half of the nodes it met during the training period . in this section we briefly give details about the communities found in the various traces . to use the @xmath23-cliques community detection algorithm we first had to extract a social graph from the contacts happening in the training period of each trace . we put the edges in the graphs in such a way that they reflect the repetitive social behaviour of the nodes . since all the not - simulated traces we used have been collected in university campuses , we expect the social connectivity to be given by students that , for example , frequent the same classes , study in the same library or live in the same dormitory . once the social graph is built , the @xmath23-cliques community detection algorithm is run over it in order to find groups of nodes socially close to each other . however , it is important to note that according to the experiments our protocols are pretty robust with respect to different definitions of `` friend '' used in the literature , as long as it is true that friends meet regularly and frequently . to test the performance of personal marks , we run a trace - driven simulation where we make each pair of nodes of the same community check and exchange marks each time they meet . in order to simulate the attack , for every node @xmath2 we do the following : for every node @xmath3 in the community of node @xmath2 we generate a clone attack of node @xmath3 against node @xmath2 ( node @xmath3 is the insider ) and measure the time required by personal marks to detect the attack . performing the experiment is rather straightforward . as we described in section [ sec : marks ] , a node in the victim s community detects the attack in two cases : the first case is when the node first meets the adversary and then the victim , the second case is when the node first meets the victim and then the adversary . in the first case , the mark of the victim node will be found to be outdated , while , in the second case , it is the mark of the adversary that will be found to be outdated . thus , we measure the average time it takes for any of these two cases to happen after the attack happened . figure [ fig : marks ] shows the performance the personal marks sub - system on all the traces real and synthetic ones . we show , for a given number of days @xmath37 , what is the number of nodes for which the detection of the attack takes more than @xmath37 days on average . first , we can notice from the results that , consistently in all the traces , the time needed to detect an insider is very low . indeed , in the majority of the cases the attack is detected in one day , one day and a half at most . more in detail , in the dartmouth trace ( figure [ fig : marks - dartmouth ] ) , the attack gets detected in less than one day in the @xmath43 of the cases and in less than two days in the @xmath44 of the cases . in the ucsd trace the attack is detected in less than one day in the @xmath45 of the cases , but for the @xmath46 of the cases it takes less than two days . in the reality trace ( figure [ fig : marks - reality ] ) the performance of personal marks is slightly worse : in the @xmath47 of the cases the attack is detected in less than one day and a half , while it takes less than three days for the @xmath48 of the cases . the reason why it takes more time to detect the attack is that the reality trace is considerably sparser than the other traces . indeed , table [ tab : traces ] shows that the average number of contacts is the minimum among all the traces . lastly , personal marks shows the best performances in the swim trace ( figure [ fig : marks - swim ] ) : in the @xmath49 of the cases , the nodes need at most half a day to detect the attack , while in the @xmath50 of the cases it takes at most one day and a half . this is not surprising . in the real traces , in fact , all the nodes have periods of time ( that can last even a few days ) where they did nt report any contact . of course , this is not very realistic all of the traces capture only a very small fraction of the real contacts of the user ( the other people in the experiment , often a small subset of the other students at school ) . we expect that , in case of wide deployment in the real world , the number of contacts would be much higher than any of these traces both real and synthetic ones and the performance , already good , would greatly improved even compared to the synthetic traces , that have been designed to mimic the dynamics of the cambridge traces . the community certificate sub - system has the goal to fight back outsiders . this sub - system has three main components : set @xmath15 of nodes appearing in the certificate of node @xmath2 , the values in @xmath16 , and parameter @xmath26 . let us concentrate on @xmath15 and @xmath16 . the first one is the set of nodes from which node @xmath2 receives the signed timestamps it uses to prove its identity . the second one defines for how long a signed timestamp given to @xmath2 by @xmath3 is _ fresh _ ( see section [ sec : communitycertificates ] ) . clearly , the correct definition of @xmath15 and @xmath16 is crucial . for instance , if the set @xmath15 were large and the values @xmath18 for every @xmath3 in @xmath15 were big , then it would be easy for node @xmath2 to prove its identity . in fact , the probability that node @xmath2 meets at least @xmath26 nodes in the set @xmath15 would be large , and , at the same time , the signed timestamps @xmath51 that node @xmath2 would receive from them would last for long time . this would surely translate into a low number of false positives in our system ( legitimate users that can not authenticate ) . at the same time , however , this would give more chances to the adversary to keep his certificate valid for a long period of time . thus , by choosing the elements of set @xmath15 and the values of @xmath16 we trade - off the ability of the system to correctly authenticate honest nodes and the performance in blocking outsiders . given these considerations , an intuitive choice is that of putting in @xmath15 the set of nodes that @xmath2 is likely to encounter more often , and in @xmath18 the expected inter contact times between @xmath2 and each of the nodes in @xmath15 . to define @xmath15 and @xmath16 we again use the social behavior of node @xmath2 observed during the training period as an approximation of the its behaviour the rest of the trace . to compute @xmath15 set , we first consider the set of nodes @xmath39 that @xmath2 has seen at least once during the training period . then , for every node @xmath52 we calculate the number @xmath53 of days in which @xmath2 met @xmath3 , and compute the average value : @xmath54 finally , we take as @xmath15 the set of nodes @xmath3 such that @xmath55 . in figure [ fig : fs_size ] we show the distribution of the size of the sets @xmath15 in all the traces . for the dartmouth trace ( figure [ fig : fs_size - dartmouth ] ) the average size of @xmath15 is @xmath56 , for ucsd ( figure [ fig : fs_size - ucsd ] ) it is @xmath57 , for reality ( figure [ fig : fs_size - reality ] ) it is @xmath58 and for swim it is @xmath59 . it is worth to observe that the sizes of the @xmath15 sets do nt change much from trace to trace despite the fact that in dartmouth and swim the number of nodes is more than @xmath60 times bigger than that of ucsd and reality . we interpret this fact as an hint that our definition of @xmath15 is meaningful and rather stable . the values in @xmath16 are computed starting from the inter - contacts observed between node @xmath2 and the nodes in @xmath15 during the training period . more in detail , for every pair of nodes @xmath61 we computed @xmath18 as a the average number of days between contacts of nodes @xmath2 and @xmath3 . first , we tested the system in order to know if the honest nodes are always able to certificate their identity . we define a false positive as the event that a legitimate user is not able to authenticate since he is temporarily out of his community for a long time . to perform these tests for all the nodes we have set the value @xmath26 to @xmath62 and run the simulation as if no attack happens . setting @xmath26 to @xmath62 means that for a node it is sufficient to have just one valid signed update to prove its identity . this permits us to study the lower bound of the number of false positives the system can ensure given the definitions of @xmath15 and @xmath16 we provided above . in the case of the dartmouth trace , we have found that @xmath63 nodes ( that is just the @xmath64 of all the @xmath27 nodes ) show false positives : 36 of them have only one false positive , while the rest have two false positives . moreover , the periods in which these nodes are not able to prove their identity always lasts one day at most . considering the simple way in which we defined the sets @xmath15 and the values @xmath16 these results are pretty good . in the case of the ucsd trace our tests performed even better : only one node over the @xmath32 we tested ( that is , the @xmath65 of the nodes ) has false positives . in particular the node is not able to prove its identity only for three time intervals , all of them lasting less than one day . we got a similar result in the reality trace too : only one node over @xmath34 ( @xmath66 of the nodes ) had false positives and this happened only twice and for less than one day . finally , in the swim trace no node has generated false positives . we have shown that the way we defined @xmath15 and @xmath16 is good enough to ensure that only a very small percentage of honest nodes suffers of false positives . now it is time to study the effectiveness of the community certificates sub - system in limiting the power of a clone attack by an outsider . to do that , we study is the average time it takes for the community certificate of the nodes to expire in absence of contacts with nodes in the set @xmath15 . this gives us a measure of the amount of time the adversary is able to use a cloned identity . the reason is that , once the adversary cloned an identity , he / she wo nt be able to use it inside the community of the victim or to renew it withouth being quickly detected by the personal marks sub - system . in such a situation , the only option left to the adversary is that of using the identity far from the victim s community . when we studied the average duration of the community certificates , we used , for each node , the highest value @xmath26 for which the node is not affected by false positives . this gives us an idea of how much we can stress the system to get the lowest possible average certificate duration given our definitions of @xmath15 and @xmath16 . the results of our measurements are shown in figure [ fig : certificates ] . each graph shows , for a given number of days @xmath37 , what is the number of nodes whose certificate has , on average , a duration of more than @xmath37 days when cloned by an outsiders . so , the faster the better . the first thing we can notice is that in all the traces the average duration of the certificate is never more than @xmath67 or @xmath68 days in the worst cases , and roughly between @xmath42 and @xmath69 in the majority of the cases . more in detail , in the dartmouth trace ( figure [ fig : certificates - dartmouth ] ) the community certificate has an average validity of less than one day and a half for @xmath70 of the nodes , while only @xmath65 of the nodes have a certificate lasting , on average , more than two days . in the ucsd trace ( figure [ fig : certificates - ucsd ] ) , the community certificates last on average at most one day and a half for @xmath43 of the nodes . this value is slightly smaller than the corresponding one found in the dartmouth trace . however , in the worst case the average certificate duration in the ucsd trace is @xmath71 days as compared to the more than @xmath72 days of the dartmouth trace . in the case of the reality trace , the average certificate duration is at most one day and a half for the @xmath73 of the nodes and it is at most @xmath74 days long for all the nodes . this value is the minimum among the maximum average certificate duration we measured in the three real traces we have used . finally the simulated trace swim shows very good performance : @xmath50 of the nodes have an average certificate duration smaller than @xmath62 day and the maximum average certificate duration is @xmath75 days . figure [ fig : k ] shows the distributions of the values @xmath26 that we have found to be the maximum we could use on each node without generating false positives . as we can see , there is a big difference between the optimal values found for the dartmouth and swim traces ( figures [ fig : k - dartmouth ] and [ fig : k - swim ] ) and those found in the reality and ucsd traces ( figures [ fig : k - reality ] and [ fig : k - ucsd ] ) . in the first two traces there is a big number of nodes for which we have been able to select ( automatically from the training period ) high values of @xmath26 without producing false positives . for instance , in the dartmouth trace , for @xmath70 of the nodes we have been able to select a value @xmath26 greater than 1 and for @xmath76 of the nodes a value greater than @xmath30 . in swim , the sets @xmath15 are generally smaller than those found in dartmouth ( see figure [ fig : fs_size ] ) and this translates into smaller values of @xmath26 . nevertheless in swim it is possible to set a value @xmath26 greater than @xmath62 for @xmath50 of the nodes and a value greater than @xmath77 for @xmath78 of the nodes . on the other hand , in ucsd and reality we can use a @xmath26 greater than one only for @xmath79 and for the @xmath80 of the nodes respectively . this big difference can be explained by the fact that , as shown in table [ tab : traces ] , in dartmouth and swim the nodes have a much higher number of contacts than in ucsd and reality . in the case of dartmouth this happens because the number of nodes is high and the trace has been collected using wifi data rather than a short ranged bluetooth radio . on the other hand , in swim , since we are simulating a bluetooth trace , the nodes do nt have a range as high as that of dartmouth hence they tend to have contacts with smaller portions of the network . however , as already stressed in the previous sections , using the swim simulator we never miss a contact between two nodes and , moreover , we do nt have data losses typical of the real traces . overall , we can say that the definitions we used for @xmath15 and @xmath81 seem to work well in different scenarios . indeed , they make it easy to the large majority of the honest nodes to keep their community certificates up to date . at the same time , the adversary that clones an honest identity , will be able to use it for only short period of time that is in the order of one or two days at most . so far we have assumed that each node receives its signed timestamps from the other nodes in its community . however , the community certificates protocol is actually more general than that . indeed , the certificate authority may potentially put into the set @xmath15 the i d of any kind of device with a private / public key pair that is able to send signed timestamps to the nodes that enter its range of communication . for instance , we could think of a wifi access point in a university campus . in such a scenario it would be perfectly reasonable for a student to have stored in the set @xmath15 of his / her smartphone the public key of the access points of the university buildings he / she frequents the most . there would be three good reasons to do that . the first one is that there are locations , just like friends , that we `` meet '' more often than others ; it is easy to check experimentally that the pattern of meetings is similar to that of two nodes carried by humans . the second one is that the access points have fixed locations and are , usually , always on . the third one is that it is supposedly harder to tamper with an access point and to move it to another location rather than a mobile device . this would make it more difficult for the adversary to refresh the community certificate of a cloned identity @xmath2 by compromising a given number of nodes in the set @xmath15 ( see section [ subsection : multiple_attacks ] ) . for these reasons we found it interesting to test the performance of community certificates in a situation where the certificate authority puts in @xmath82 the access points of the most frequented locations of each node . to do that , we used the dartmouth and ucsd traces since they have been collected exactly recording the associations between mobile nodes and the access point of the dartmouth and university of california san diego campuses respectively ( see section [ subsection : datasets ] ) . we performed the tests with the same definitions of @xmath82 and @xmath83 we used in section [ subsec : commcert ] , though , this time , we are interested in the contacts between nodes and access points rather than between pair of nodes . the results are shown in figures [ fig : dartmouth.aps ] and [ fig : ucsd.aps ] . the first difference we can notice between these results and the previous ones is the change in the sizes of the sets @xmath82 . in the case of the dartmouth trace ( figure [ fig : dartmouth.aps - fs_size ] ) the average size of @xmath82 is now only @xmath68 , while before it was @xmath84 . on the contrary , in the ucsd ( figure [ fig : ucsd.aps - fs_size ] ) trace the size of the @xmath82 increased from an average of @xmath57 to an average of @xmath85 . this is a consequence of the fact that in dartmouth the contacts between the nodes happened in a restricted set of places , while , in ucsd , the nodes were generally more mobile . note that these values are small as compared to the total number of access points , that is about @xmath86 in both traces . the second difference is that the average certificate duration of the nodes slightly increases with respect to our previous results ( remember , however , that in these experiments we have removed the mobile friends ! ) . in the case of dartmouth ( figure [ fig : dartmouth.aps - certificates ] ) , the percentage of nodes whose community certificate has an average validity of less than one day has dropped from @xmath87 to @xmath88 . also , the community certificates have an average duration of less than @xmath69 days for @xmath89 of the nodes while before it was for @xmath70 . in the case of ucsd ( figure [ fig : ucsd.aps - certificates ] ) @xmath90 of the nodes have a community certificate lasting on average less than one day ( before it was @xmath91 ) while for @xmath92 it lasts less than one day and a half ( was @xmath43 ) . to summarize , the results are not as good as with real mobile friends . however , they are good enough to make it reasonable to complement the set @xmath82 with a small number of entities that are part of the fixed infrastructure . this is not going to reduce the performance considerably and helps build a stronger system against coordinate attacks to multiple mobile nodes . lastly , figures [ fig : dartmouth.aps - certificates ] and [ fig : ucsd.aps - certificates ] show distribution of the values of @xmath26 . community certificates can be complemented and extended in such a way to design more sophisticated and complete versions . here we discuss a few of these add - ons . [ [ travelling ] ] travelling + + + + + + + + + + if we travel , especially alone , our community certificate is going to expire soon . while this is generally true , we can easily design a few solutions : ( i ) the user can suspend the community checks on the certificate during the travel ; ( ii ) if we often travel between two cities , like rome and san diego , the user can prepare two different _ profiles _ , in the same certificate , that can be selected when changing community . of course , it is fundamental that these solutions can be used only by the the legitimate owner by performing the burdensome alternative authentication that is used in case of exceptional events . [ [ complex - patterns ] ] complex patterns + + + + + + + + + + + + + + + + it is fairly easy to extend community certificates in such a way that it is not enough to get @xmath23 fresh signed timestamps from your circle of friends . rather , the certificate needs @xmath93 fresh signed timestamps from one sub - community ( for example your family ) and @xmath94 fresh signed timestamps from another sub - community ( for example your colleagues ) to be valid . a community certificate like this can easily be more efficient and more useful in particular situations . for example when we need at least one signature from one of our bosses to be able to perform critical operations . or at least one signature from a fixed infrastructure in the building of our company , for example . [ [ electronic - handshakes ] ] electronic handshakes + + + + + + + + + + + + + + + + + + + + + in addition to standard physical contacts , community certificates can be extended to use _ electronic handshakes_. assume that , when meeting a friend , you use a particular secure `` handshake '' made with your devices ( like bumping the two together , for example ) . handshakes can be used at the place of physical proximity . in this way , contacts are more trustworthy , but , on the other hand , they need human intervention and are much fewer in number . alternatively , handshakes could be used to exchange a stronger signed timestamp . by using _ complex patterns _ , one or more of these handshakes can be required to enable a higher level of authentication to perform critical operations . in this paper we introduce personal marks and community certificates . the fundamental idea of community certificates is that , in networks of mobile people , authentication can be based on the notion of community , and nodes can authenticate by showing that they indeed meet the people that are part of their community . while the social structure of these networks has been extensively used in networking , to the best of our knowledge this is the first time that this has been used as a biometric to authenticate device . we also present personal marks , a way a community can use to protect itself against insiders performing a clone attack . the combined used of these mechanisms deliver an excellent protection of the social mobile network against the clone attack . indeed our experiments show that the detection is always correct , and fast enough considered the slow dynamics of the trace we have used . in any real system , we believe that the much faster dynamics of meetings can help deliver much faster detection times . m. barbeau , j. hall , and e. kranakis . detecting impersonation attacks in future wireless and mobile networks . in _ in proceedings of madnes 2005 - workshop on secure mobile ad - hoc networks and sensors_. svlncs , 2005 . h. cai and d. y. eun . crossing over the bounded domain : from exponential to power - law inter - meeting time in manet . in _ proceedings of the 13th annual acm international conference on mobile computing and networking _ , mobicom 07 , pages 159170 , new york , ny , usa , 2007 . h. cai and d. y. eun . toward stochastic anatomy of inter - meeting time distribution under general mobility models . in _ proceedings of the 9th acm international symposium on mobile ad hoc networking and computing _ , mobihoc 08 , pages 273282 , new york , ny , usa , 2008 . a. chaintreau , p. hui , j. crowcroft , c. diot , r. gass , and j. scott . pocket switched networks : real - world mobility and its consequences for opportunistic forwarding . technical report , computer laboratory , university of cambridge , 2006 . h. chan , a. perrig , and d. song . random key predistribution schemes for sensor networks . in _ proceedings of the 2003 ieee symposium on security and privacy _ , sp 03 , pages 197 , washington , dc , usa , 2003 . ieee computer society . h. choi , s. zhu , and t. l. porta . set : detecting node clones in sensor networks . in _ securecomm07 : ieee / createnet conference on security and privacy for emerging areas in communication networks _ , 2007 . m. conti , r. di pietro , a. gabrielli , l. v. mancini , and a. mei . the smallville effect : social ties make mobile networks more secure against node capture attack . in _ proceedings of the 8th acm international workshop on mobility management and wireless access _ , mobiwac 10 , pages 99106 , new york , ny , usa , 2010 . m. conti , r. di pietro , l. v. mancini , and a. mei . a randomized , efficient , and distributed protocol for the detection of node replication attacks in wireless sensor networks . in _ proceedings of the 8th acm international symposium on mobile ad hoc networking and computing _ , mobihoc 07 , pages 8089 , new york , ny , usa , 2007 . e. m. daly and m. haahr . social network analysis for routing in disconnected delay - tolerant manets . in _ proceedings of the 8th acm international symposium on mobile ad hoc networking and computing _ , mobihoc 07 , pages 3240 , new york , ny , usa , 2007 . acm . v. erramilli , m. crovella , a. chaintreau , and c. diot . delegation forwarding . in _ proceedings of the 9th acm international symposium on mobile ad hoc networking and computing _ , mobihoc 08 , pages 251260 , new york , ny , usa , 2008 . l. eschenauer and v. d. gligor . a key - management scheme for distributed sensor networks . in _ proceedings of the 9th acm conference on computer and communications security _ , ccs 02 , pages 4147 , new york , ny , usa , 2002 . p. hui , a. chaintreau , j. scott , r. gass , j. crowcroft , and c. diot . pocket switched networks and human mobility in conference environments . in _ proceedings of the 2005 acm sigcomm workshop on delay - tolerant networking _ , wdtn 05 , pages 244251 , new york , ny , usa , 2005 . p. hui , e. yoneki , s. y. chan , and j. crowcroft . distributed community detection in delay tolerant networks . in _ proceedings of 2nd acm / ieee international workshop on mobility in the evolving internet architecture _ , mobiarch 07 , pages 7:17:8 , new york , ny , usa , 2007 . n. james , e. shi , d. song , and a. perrig . the sybil attack in sensor networks : analysis & defenses . in _ proceedings of the 3rd international symposium on information processing in sensor networks _ , ipsn 04 , pages 259268 , new york , ny , usa , 2004 . t. karagiannis , j .- y . le boudec , and m. vojnovi . power law and exponential decay of inter contact times between mobile devices . in _ proceedings of the 13th annual acm international conference on mobile computing and networking _ , mobicom 07 , pages 183194 , new york , ny , usa , 2007 . s. kosta , a. mei , and j. stefa . small world in motion ( swim ) : modeling communities in ad - hoc mobile networking . in _ proceedings of the 7th ieee communications society conference on sensor , mesh and ad hoc communications and networks ( secon 2010 ) _ , boston , ma , , june 2010 . j. leguay , a. lindgren , j. scott , t. riedman , j. crowcroft , and p. hui . trace upmc / content / imote / cambridge ( v. 20061117 ) . downloaded from http://crawdad.cs.dartmouth.edu/upmc/content/imote/cambridge , november 2006 . q. li , s. zhu , and g. cao . routing in socially selfish delay tolerant networks . in _ proceedings of the 29th conference on information communications _ , infocom10 , pages 857865 , piscataway , nj , usa , 2010 . ieee press . b. parno , a. perrig , and v. gligor . distributed detection of node replication attacks in sensor networks . in _ proceedings of the 2005 ieee symposium on security and privacy _ , pages 4963 , washington , dc , usa , 2005 . ieee computer society . j. su , a. chin , a. popivanova , a. goel , and e. de lara . user mobility for opportunistic ad - hoc networking . in _ proceedings of the sixth ieee workshop on mobile computing systems and applications _ , pages 4150 , washington , dc , usa , 2004 . ieee computer society . s. apkun , l. buttyn , and j .- hubaux . : secure tracking of node encounters in multi - hop wireless networks . in _ proceedings of the 1st acm workshop on security of ad hoc and sensor networks _ , sasn 03 , pages 2132 , new york , ny , usa , 2003 .

consider the general scalar balance law in several space dimensions . the aim of this note is to improve the results of colombo , mercier , rosini who gave an estimate of the dependence of the solutions from the flow and from the source . the improvements are twofold : first the expression of the coefficients in these estimates are more precise ; second , we eliminate some regularity hypotheses thus extending significantly the applicability of our estimates . _ 2000 mathematics subject classification : _ 35l65 . _ keywords : _ multi - dimensional scalar conservation laws , krukov entropy solutions , estimate .

1010.5116

we consider here the cauchy problem for the general scalar balance law @xmath4 this kind of equation has already been intensively studied : a fundamental result is the one of s. n. krukov ( * ? ? ? * theorem 1 & 5 ) , stating the existence and uniqueness of a weak entropy solution for an initial data @xmath5 . furthermore , krukov describes the dependence of the solutions with respect to the initial condition : if @xmath6 and @xmath7 are two initial data , then the associated entropy solutions @xmath8 and @xmath9 satisfy @xmath10 furthermore , a huge literature on this subject is available in the special case the flow @xmath1 depend only on @xmath8 and not on the variables @xmath11 and @xmath12 and there is no source @xmath13 ( see for example @xcite ) . we are interested here in the dependence of the solution with respect to flow @xmath1 and source @xmath2 in the case these functions depend on the three variables @xmath11 , @xmath12 and @xmath8 . this dependence with respect to flow and source has already been investigated : this question was first addressed from the point of view of numerical analysis by b. lucier @xcite who studied the case of an homogeneous flow ( @xmath14 ) , without source term ( @xmath13 ) . more recently f. bouchut & b. perthame @xcite improved this result , always in the case of an homogeneous flow and without source . chen & k. karlsen @xcite also studied this dependence , for a flow depending also on @xmath12 , but the estimate they obtained was depending on an a priori ( unknown ) bound on @xmath15 . the purpose of the present paper is to improve the recent result of r. colombo , m. mercier & m. rosini @xcite , which provided an estimate of the total variation in the general case ( with flow and source depending on the three variables @xmath11 , @xmath12 and @xmath8 ) and of the @xmath16 distance between solutions . in particular , this estimate can be compared to the one of krukov ( [ eq : kru ] ) that give a bound on the @xmath16 distance between solutions with different initial data ( but with same flow and source ) . the estimates ( [ eq : kru ] ) and ( * ? ? ? * theorem 2.6 ) look similar but in @xcite , the coefficient @xmath17 given by krukov in ( [ eq : kru ] ) is replaced by @xmath18 . consequently , we do not recover ( [ eq : kru ] ) from @xcite in the case @xmath13 ( because @xmath19 whereas @xmath20 a priori ) . in the same setting as in @xcite , we provide here an estimate on the total variation of the solution to ( [ eq : probv ] ) , and on the dependence of the solutions to ( [ eq : probv ] ) on the flow @xmath1 , on the source @xmath2 , with better hypotheses and coefficients than in @xcite . the advances are twofold . firstly , we relax hypotheses , and thus widely extend the usability of our results . more precisely , we require here less regularity in time than in @xcite , which is very useful for applications ( see @xcite ) . furthermore , we recover the same estimate as krukov when we consider the dependence toward initial conditions only . this note is organized as follows . in section [ sec : thm ] we state the main results and compare them to those in @xcite . in section [ sec : prelim ] , we give some tools on functions with bounded variations ; in sections [ sec : prooftv ] and [ sec : proofcomparison ] we prove theorems [ teo : tv ] and [ teo : estimates ] ; finally section [ sec : tech ] contains some technical lemmas used in the preceding sections . we shall use the notations @xmath21 and @xmath22 . below , @xmath23 is a positive integer , @xmath24 ; for any positive @xmath25 , @xmath26 we denote @xmath27\times{{\mathbb{r}}}^n\times [ -u , u]$ ] ; @xmath28 stands for the ball in @xmath29 with center @xmath30 and radius @xmath31 and @xmath32 stands for the support of @xmath8 . the volume of the unit ball @xmath33 is @xmath34 . for notational simplicity , we set @xmath35 . the following induction formula gives @xmath34 in terms of the wallis integral @xmath36 : @xmath37 in the present work , @xmath38 is the characteristic function of the set @xmath39 , and @xmath40 is the dirac measure centered at @xmath11 . besides , for a vector valued function @xmath41 with @xmath42 , @xmath43 stands for the total divergence . on the other hand , @xmath44 , respectively @xmath45 , denotes the partial divergence , respectively gradient , with respect to the space variables . moreover , @xmath46 and @xmath47 are the usual partial derivatives . thus , @xmath48 . the following sets of assumptions on @xmath1 and @xmath2 will be of use below . @xmath49 f , f \textrm { have continuous derivatives } { \partial}_u f\,,\ ; { \partial}_u\nabla f\,,\ ; \nabla^2 f\,,\ ; { \partial}_u f\,,\ ; \nabla f\,;\\[5pt ] \textrm{for all } u , \,t>0 , \qquad{\partial}_u f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}}^n),\qquad f-{\mathinner{\rm div}}f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}})\ , . \end{array } \right . \\[25pt ] \mathbf{(h2^*)}\!\ ! & \left\ { \begin{array}{l } \textrm{for all } u,\ ; t>0\,,\qquad\nabla{\partial}_u f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}}^{n\times n})\,,\qquad { \partial}_u f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}})\,,\\[5pt ] \displaystyle \int_0^t\!\!\int_{{{\mathbb{r}}}^n}{{\left\|\nabla(f-{\mathinner{\rm div}}f)(t , x,\cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-u , u];{{\mathbb{r}}}^n ) } { \mathinner{\mathrm{d}{x}}}\,{\mathinner{\mathrm{d}{t } } } < \infty\ , . \end{array } \right . \\[30pt ] \mathbf{(h3^*)}\!\ ! & \left\ { \begin{array}{l } \textrm{for all } u,\ , t>0 \qquad { \partial}_u f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}}^n)\,,\qquad { \partial}_u f\in { { \mathbf{l}^\infty}}(\omega_t^u ; { { \mathbb{r}}})\,,\\[5pt ] \displaystyle \int_{0}^t \!\ ! \int_{{{\mathbb{r}}}^n } \ ! { { \left\|(f- { \mathinner{\rm div}}f)(t , x , \cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-u , u];{{\mathbb{r } } } ) } \ , \mathrm{d}x\ , \mathrm{d}t < + \infty\ , . \end{array } \right . \end{array}\ ] ] comparing these sets of hypotheses to * ( h1)*,*(h2 ) * and * ( h3 ) * in @xcite , we note that * no derivatives in time are now needed ; * the @xmath50 norm are now taken on the domain @xmath51\times { { \mathbb{r}}}^n\times [ -u , u]$ ] that is smaller than @xmath52 , that was the domain considered in @xcite . let us recall the fundamental theorem [ teo : kruzkov ] assume @xmath53 hold . then , for any @xmath54 , there exists a unique weak entropy solution @xmath8 to ( [ eq : probv ] ) in @xmath55 continuous from the right . moreover , if a sequence @xmath56 converges to @xmath6 in @xmath57 , then for all @xmath58 the corresponding solutions @xmath59 converge to @xmath60 in @xmath57 . we give here a result similar to the one obtained by colombo , mercier and rosini ( * ? ? ? * theorem 2.5 ) , but under weaker assumptions . [ teo : tv ] assume that @xmath53 and @xmath61 hold . let @xmath62 . then , the weak entropy solution @xmath8 of ( [ eq : probv ] ) satisfies @xmath63 for all @xmath64 . let @xmath65 be real positive . let us denote @xmath66\times { { \mathbb{r}}}^n)}$ ] , @xmath67 , @xmath68 } { \mathop{\rm supp}}(u(t))$ ] and @xmath69\times \mathcal{s}_{t_0}(u)\times[-\mathcal{u } , \mathcal{u}]\,,\label{eq : ktu}\\ \label{eq : kappao } \kappa^*_0 & = & ( 2n+1 ) \ , { { \left\|\nabla \ , \partial_u f\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_{t_0}^u ; { { \mathbb{r}}}^{n\times n } ) } + { { \left\|\partial_u f\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_{t_0}^u ; { { \mathbb{r } } } ) } \end{aligned}\ ] ] then for all @xmath70 $ ] , with @xmath36 as in ( [ eq : w ] ) , @xmath71;{{\mathbb{r } } } ) } \mathrm{d}x\ , \mathrm{d}t \,.\ ] ] note that , with @xmath72 , we have @xmath73 . consequently , @xmath74 we can note here several improvements with respect to ( * ? ? ? * theorem 2.5 ) . first , as we already noted , the set of hypotheses is weaker since we do not require @xmath1 to be @xmath75 and @xmath2 to be @xmath76 with respect to the time variable : they only have to be continuous in time , which is useful in applications , see for example @xcite . a second improvement stands in the @xmath50 norms , that are taken on smaller domains than in @xcite . last , the expression of the coefficient @xmath77 that does not content any longer the constant @xmath78 . indeed , in ( * ? ? ? * theorem 2.5 ) it was given by @xmath79 besides , it does not seem possible to erase the coefficient @xmath80 completely from the expression ( [ result ] ) , except in the case @xmath2 and @xmath1 do not depend on @xmath8 , see remark [ rem : opt ] . we want now to estimate the difference @xmath81 , where * @xmath8 is the solution of ( [ eq : probv ] ) with flow @xmath1 , source @xmath2 and initial condition @xmath6 , * @xmath9 is the solution of ( [ eq : probv ] ) with flow @xmath82 , source @xmath83 and initial condition @xmath7 . we search for an estimate of @xmath81 in term of @xmath84 , @xmath85 and @xmath86 . f. bouchut & b. perthame in @xcite obtained such an estimate in the particular case @xmath1 , @xmath82 depend only on @xmath8 and @xmath87 . the following result is an improvement of the result of r. colombo , m. mercier and m. rosini ( * ? ? ? * theorem 2.6 ) , in which we gave a similar result under stronger assumptions and with a coefficient @xmath88 that was not compatible with the result of krukov ( [ eq : kru ] ) . [ teo : estimates ] let @xmath89 , @xmath90 satisfy @xmath53 , @xmath91 satisfy @xmath61 and @xmath92 satisfy @xmath93 . let @xmath94 . let @xmath95 , we denote @xmath96\times { { \mathbb{r}}}^n)},{{\left\|v\right\|}}_{{{\mathbf{l}^\infty}}([0,t]\times{{\mathbb{r}}}^n)})$ ] , @xmath97 . let also introduce @xmath98 } \left ( { \mathop{\rm supp}}u(t ) \cup { \mathop{\rm supp}}v(t)\right ) $ ] and @xmath99\times\mathcal{s}_t(u , v)\times [ -\mathcal{v},\mathcal{v}]\,.\ ] ] furthermore , we define and @xmath77 , @xmath100 , @xmath101 as in ( [ eq : kappao ] ) and @xmath102 then , for any @xmath103 and @xmath104 , the following estimate holds : @xmath105 ) } \mathrm{d}x \ , \mathrm{d}t \right)\\ & & \qquad \times { { \left\|{\partial}_u(f - g)\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_t^u;{{\mathbb{r}}}^n ) } \\ & + & \int_0^t e^{\kappa^ * ( t - t ) } \int_{{{\left\|x - x_0\right\|}}{\leqslant}r+m(t - t ) } { { \left\|\left((f - g ) - { \mathinner{\rm div}}(f - g ) \right)(t , x,\cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-{v}_t,{v}_t ] ) } \ , \mathrm{d}x\ , \mathrm{d}t \ , . \end{aligned}\ ] ] note as above that , with @xmath106 , we have @xmath107 and @xmath108 . consequently , @xmath109 as above , we can note some improvements with respect to ( * ? ? ? * theorem 2.6 ) : * the hypotheses are weaker : no derivative in time is needed for @xmath1 and @xmath2 . * the @xmath50 norms are taken on smaller domains . * the coefficient @xmath88 is better than the @xmath110 given in ( * ? ? ? * theorem 2.6 ) by @xmath111 indeed , @xmath88 only depends on @xmath2 , which is consistent with the previous krukov s result ( [ eq : kru ] ) , whereas @xmath110 was also depending on @xmath1 . note that , denoting @xmath112 , we have @xmath113 and @xmath114 in the case @xmath115 , we conlude by @xmath116 . + in the case , @xmath117 , we can replace @xmath118 by @xmath119 then we have @xmath120 and @xmath121 . + in all cases , we obtain @xmath122 , and the estimate of theorem [ teo : estimates ] becomes @xmath123 ) } \mathrm{d}x \ , \mathrm{d}t \right)\\ & & \qquad \times { { \left\|{\partial}_u(f - g)\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_t^u;{{\mathbb{r}}}^n ) } \\ & + & \int_0^t e^{\kappa^ * ( t - t ) } \int_{{{\left\|x - x_0\right\|}}{\leqslant}r+m(t - t ) } { { \left\|\left((f - g ) - { \mathinner{\rm div}}(f - g ) \right)(t , x,\cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-{v}_t,{v}_t ] ) } \ , \mathrm{d}x\ , \mathrm{d}t \ , . \end{aligned}\ ] ] recall the following theorem ( see ( * ? ? ? * theorem 3.9 and remark 3.10 ) ) : [ thm : afp ] let @xmath124 . then @xmath125 if and only if there exists a sequence @xmath126 in @xmath127 converging to @xmath8 in @xmath57 and satisfying @xmath128 moreover , @xmath129 is the smallest constant @xmath130 for which there exists a sequence as above . let us also recall the following property of any function @xmath131 : @xmath132 for a proof , see ( * ? ? ? * remark 3.25 ) . now , in a similar way as j. dvila @xcite , we prove the following proposition , which is an improvement of ( * ? ? ? * proposition 4.3 ) . indeed , in ( * ? ? ? * proposition 4.3 ) , the equality ( [ eq : limm ] ) is valid only for @xmath133 . in the present proposition we extend this result to all @xmath134 . [ prop : normtv2 ] let @xmath135 with @xmath136 $ ] . let @xmath137 . for all @xmath138 , we introduce @xmath139 such that @xmath140 . assume that there exists a constant @xmath141 such that for all @xmath142 , @xmath143 positive , @xmath144 then @xmath145 and @xmath146 where @xmath147 note that the first part of the proof is the same as the first part of the proof of ( * ? ? ? * proposition 4.3 ) . we introduce a regularisation of @xmath8 : @xmath148 , with @xmath149 , where @xmath150 is defined as in ( [ eq : mu ] ) . we note that @xmath151 and that @xmath152 tends to @xmath8 in @xmath57 when @xmath153 . furthermore , for @xmath143 and @xmath154 positive , by change of variables we get @xmath155 making @xmath156 and using the dominated convergence theorem when @xmath157 , we obtain @xmath158 remark that for fixed @xmath159 , when @xmath160 , the scalar product @xmath161 is positive ( respectively , negative ) when @xmath162 is in a half - space , say @xmath163 ( respectively , @xmath164 ) . we can write @xmath165 , with @xmath166 and @xmath167 in the hyperplane @xmath168 . hence @xmath169 so we obtain @xmath170 now , let @xmath126 be a sequence of functions in @xmath171 converging to @xmath8 in @xmath57 and such that @xmath172 converges to @xmath129 when @xmath173 . then , doing the same computation as above , we obtain @xmath174 taking @xmath175 and then @xmath173 , we have consequently @xmath176 then , we take the supremum limit when @xmath142 goes to 0 . we obtain @xmath177 we conclude the proof by reassembling ( [ eq : liminf ] ) and ( [ eq : limsup ] ) . the following proof is quite similar to the one of ( * ? ? ? * theorem 2.5 ) . the differences come from the use of proposition [ prop : normtv2 ] instead of ( * ? ? ? * proposition 4.3 ) and from avoiding the derivatives in time to appear . in order to be clear , we rewrite here most of the steps of the proof . in particular , the beginning of the proof is similar to ( * ? ? ? * proof of theorem 2.5 ) up to ( [ eq : nu ] ) . theorem [ teo : tv ] first , we assume that @xmath178 . the general case will be considered only at the end of this proof . by krukov theorem ( * theorem 5 & section 5 remark 4 ) , the set of hypotheses * ( h1 * ) * gives us existence and uniqueness of a weak entropy solution for any initial condition @xmath179 . let @xmath8 be the weak entropy solution to ( [ eq : probv ] ) associated to @xmath180 . let us denote @xmath181 and @xmath182 for @xmath183 . then , for all @xmath184 and for all test functions @xmath185 in @xmath186 , we have @xmath187 & \\ \times \mathrm{sign}(u - k ) \ , \mathrm{d}x \ , \mathrm{d}t & { \geqslant}0 \end{array}\ ] ] for all @xmath188 , and @xmath189 & \\ \times \mathrm{sign}(v - l ) \ , \mathrm{d}y \ , \mathrm{d}s & { \geqslant}0 \end{array}\ ] ] for all @xmath190 . let @xmath191 , @xmath192 and set @xmath193 observe that @xmath194 , @xmath195 , @xmath196 . choose @xmath197 in ( [ eq : utv ] ) and integrate with respect to @xmath198 . analogously , take @xmath199 in ( [ eq : vtv ] ) and integrate with respect to @xmath200 . summing the obtained inequalities , we obtain @xmath201 \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s & { \geqslant}0 . \end{array}\ ] ] introduce a family of functions @xmath202 such that for any @xmath203 : + , left , and of @xmath204 , right.,title="fig : " ] , left , and of @xmath204 , right.,title="fig : " ] @xmath2050,1\right[\ , , & y ' & { \geqslant}0\ , , & \displaystyle \int_{{\mathbb{r}}}y'(s ) \ , \mathrm{d}s & = 1 \,.\nonumber \end{aligned}\ ] ] let @xmath206 , @xmath66\times { { \mathbb{r}}}^n;{{\mathbb{r}}})}$ ] and @xmath207 which is bounded by * ( h1*)*. let us also define , for @xmath208 , @xmath104 , ( see figure [ fig : chipsi ] ) : @xmath209 where we also need the compatibility conditions @xmath210 and @xmath211 . , left , and of @xmath212 , right.,title="fig : " ] , left , and of @xmath212 , right.,title="fig : " ] observe that @xmath213}$ ] and @xmath214 as @xmath215 tends to @xmath216 . on @xmath217 and @xmath212 we use the bounds @xmath218 } \quad \mbox { and } \quad \mathbf{1}_{b(x_0 , r+m(t_0-t ) ) } { \leqslant}\psi { \leqslant}\mathbf{1}_{b(x_0 , r+m(t_0-t)+\theta ) } \,.\ ] ] in ( [ eq : sumtv ] ) , choose @xmath219 . with this choice , we have @xmath220 setting @xmath221 , the first line in ( [ eq : sumtv ] ) becomes @xmath222\ ! \mathrm{sign}(u - v ) \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \\ \!\ ! = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left ( { { \left|u - v\right|}}\ , \chi ' \ , \psi - b(t , x , u , v ) \chi \ , y_\theta ' \right ) \ , \psi \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \\ \!\ ! { \leqslant } & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } { { \left|u - v\right| } } \ , \chi ' \ , \psi \ , \psi \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s\ , , \end{aligned}\ ] ] since @xmath223 is positive for all @xmath224 . due to the above estimate and to ( [ eq : sumtv ] ) , we have @xmath225 \\ & \displaystyle \times \mathrm{sign}(u - v ) \ , \mathrm{d}x\,\mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s & { \geqslant } & 0 . \end{array}\ ] ] now , we aim at bounds for each term of this sum . introduce the following notations : @xmath226\!\\ & \qquad \qquad\qquad\qquad { \varphi}\ , \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s \\ \nonumber l_2 = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \ ! \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left [ \int_0 ^ 1 \ ! \nabla ( f - { \mathinner{\rm div}}f ) \left(t , rx+(1-r)y , u\right)\cdot ( x - y ) \ , \mathrm{d}r \right ] { \varphi}\\ \label{eq : l2 } & \qquad \qquad \qquad \qquad \times \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s\ , . \\ \nonumber l_t = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left ( f(t , y , v ) - f(s , y , v ) - { \mathinner{\rm div}}f(t , y , u ) + { \mathinner{\rm div}}f(s , y , u ) \right ) \ , \varphi \\ & \qquad\qquad\qquad\qquad \times \ , \mathrm{sign}(u - v ) \ , \mathrm{d}x\,\mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \ , . \end{aligned}\ ] ] then , the above inequality is rewritten as @xmath227 . choose @xmath228 where , for @xmath229 , @xmath230 satisfies ( [ eq : mu])([eq : mu ] ) and @xmath231 - 1,0\right [ \,.\ ] ] now , we want to estimate separately @xmath232 , @xmath233 , @xmath234 , @xmath235 , @xmath236 and @xmath237 . note first that if @xmath238 } { \mathop{\rm supp}}u(t)\ } $ ] , the integrand in @xmath233 and @xmath235 vanishes , so denoting @xmath239 } { \mathop{\rm supp}}u(t)\,,\ ] ] the space of integration of @xmath233 and @xmath235 is in fact @xmath240 . the main differences with respect to the proof of ( * ? ? ? * theorem 2.5 ) are the following : * the @xmath50 norm that we took on @xmath241 , are now taken on @xmath242\times\mathcal{s}_t(u)\times[-\mathcal{u},\mathcal{u}]$ ] , where @xmath243)$ ] . * for @xmath234 and @xmath237 , by dominated convergence theorem , we get when @xmath244 @xmath245 which avoids the use of time derivatives . * the @xmath50 norm of @xmath8 in @xmath236 is now taken on @xmath246 $ ] where @xmath247 . we do not rewrite the estimates on @xmath232 , @xmath233 , @xmath235 , @xmath236 , that are the same as in ( * ? ? ? * theorem 2.5 ) , up to the space in the @xmath50 norm . see remark [ rem : opt ] for precisions on the estimate of @xmath236 . letting @xmath248 we get @xmath249 ) } \ , \mathrm{d}y\,\mathrm{d}t\ , , \\ \limsup_{{\varepsilon } , { { \eta } } , \theta \to 0 } l_{t } & = & 0 \ , , \end{aligned}\ ] ] where @xmath250 above , the right hand sides are bounded thanks to * ( h2*)*. collating all the obtained results and using the equality , @xmath251 @xmath252 & { \leqslant } & \displaystyle \int_{{{\mathbb{r}}}^n } \int_{{{\left\|x - x_0\right\|}}{\leqslant}r + m t_0 } { { \left|u(0,x)-u(0,y)\right| } } \ , \frac{1}{{{\lambda}}^n } \mu_1\left(\frac{{{\left\|x - y\right\|}}}{{{\lambda}}}\right ) \ , \mathrm{d}x \ , \mathrm{d}y \\[10pt ] & & \displaystyle - { { \left\|\nabla{\partial}_u f \right\|}}_{{{\mathbf{l}^\infty}}(\sigma_{t_0}^u ) } \int_0^{t } \int_{{{\mathbb{r}}}^n } \int_{{{\left\|x - x_0\right\|}}{\leqslant}r + m(t_0-t ) } \!\!\ ! { { \left|u(t , x)-u(t , y)\right| } } \\[10pt ] & & \displaystyle \qquad \qquad \qquad \times \frac{1}{{{\lambda}}^{n+1 } } \ , \mu_1'\left ( \frac{{{\left\|x - y\right\|}}}{{{\lambda } } } \right ) \ , { { \left\|x - y\right\| } } \,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t \\[10pt ] & & \displaystyle + \big ( n { { \left\|\nabla{\partial}_u f\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_{t_0}^u ) } + { { \left\|{\partial}_u f\right\|}}_{{{\mathbf{l}^\infty}}(\sigma_{t_0}^u ) } \big ) \ ! \int_0^{t } \ ! \int_{{{\mathbb{r}}}^n } \ ! \int_{{{\left\|x - x_0\right\|}}{\leqslant}r + m(t_0-t ) } \ ! { { \left|u(t , x)-u(t , y)\right| } } \\[10pt ] & & \displaystyle \qquad \qquad \qquad \times \frac{1}{{{\lambda}}^n } \mu_1\left(\frac{{{\left\|x - y\right\|}}}{{{\lambda}}}\right ) \,\mathrm{d}x\,\mathrm{d}y\,\mathrm{d}t \\[10pt ] & & \displaystyle + { { \lambda}}m_1 \int_0^{t}\int_{{{\mathbb{r}}}^n } { { \left\|\nabla(f - { \mathinner{\rm div}}f)(t , y,\cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-{u}_t,{u}_t ] ) } \ , \mathrm{d}y\ , \mathrm{d}t \ , . \end{array}\ ] ] if @xmath253 and under the present assumption that @xmath254 , using proposition [ prop : normtv2 ] , ( [ eq : cst0 ] ) and ( [ eq : m1 ] ) , we directly obtain that @xmath255 ) } \mathrm{d}y\ , \mathrm{d}t \,.\ ] ] the same procedure at the end of this proof allows to extend ( [ eq : ck ] ) to more general initial data , providing an estimate of @xmath256 in the situation studied in @xcite . now , it remains to treat the case when @xmath257 . as in ( * theorem 2.5 ) , a direct use of gronwall lemma is not possible , but we can first obtain an estimate of the function : @xmath258 indeed , we get that if @xmath25 is such that @xmath259 then we obtain , with @xmath260 , @xmath261 furthermore , by ( [ eq : mu ] ) and ( [ eq : mu ] ) there exists a constant @xmath262 such that for all @xmath263 @xmath264 divide both sides in ( [ inegalite ] ) by @xmath142 , rewrite them using ( [ eq : fdelta ] ) , ( [ eq : boundmu ] ) , apply ( [ eq : afp ] ) and obtain @xmath265 ) } \mathrm{d}y\ , \mathrm{d}t\ , . \end{aligned}\ ] ] an application of ( [ eq : fdelta ] ) yields an estimate of the type @xmath266 where the positive constant @xmath267 is independent from @xmath143 and @xmath142 . applying proposition [ prop : normtv2 ] we obtain that @xmath268 for @xmath269 , where @xmath270 the next step is to obtain a general estimate of the @xmath271 norm . the starting point is ( [ inegalite ] ) . recall the definitions ( [ eq : m1 ] ) of @xmath272 and ( [ eq : t1 ] ) of @xmath273 . moreover , by integration by part we obtain @xmath274 the following step is _ not _ similar to ( * ? ? ? * proof of theorem 2.5 ) : we divide both terms in ( [ inegalite ] ) by @xmath142 , apply ( [ eq : limm ] ) on the first , second and third terms in the right hand side , with @xmath275 in the second and third case , and with @xmath276 in the second case . we obtain for all @xmath277 $ ] with @xmath278 @xmath279 ) } \mathrm{d}x \ , \mathrm{d}t\ , . \end{aligned}\ ] ] next , an application of the gronwall lemma shows that @xmath280 is bounded on @xmath281 $ ] @xmath282 ) } \ , \mathrm{d}x\ , \mathrm{d}t\ ] ] for @xmath277 $ ] , @xmath283 as in ( [ eq : m1 ] ) , ( [ eq : cst0 ] ) and @xmath284 . now , it remains only to relax assumption on the regularity of @xmath6 and to note that the bound ( [ eq : tv ] ) is additive in time . these steps are the same as in ( * ? ? ? * theorem 2.5 ) , so we do not write them . [ rem : opt ] the constant @xmath80 in front of @xmath285)}{\mathinner{\mathrm{d}{x}}}\,{\mathinner{\mathrm{d}{t}}}$ ] in theorem [ teo : tv ] comes from the estimate of the term @xmath236 defined by ( [ eq : l2 ] ) . we have indeed @xmath286 if @xmath287 does not depend on @xmath8 , then , with the same computations as in the proof of proposition [ prop : normtv2 ] , considering @xmath288 as a linear application , we get : @xmath289 which allows us to get rid of the constant @xmath78 into the bound of @xmath236 . however , in the general case , because of the dependence of @xmath8 in @xmath162 , we are led to take the supremum of @xmath60 . we obtain the following : @xmath290 we can no longer do the same computations as in the proof of proposition [ prop : normtv2 ] . indeed , it is not allowed to permute @xmath291 and @xmath292 , consequently , if we want to isolate the variable @xmath162 from the other variables , we use the cauchy - schwartz inequality to obtain : @xmath293 the constant @xmath80 appears here when we divide by @xmath294 , since , by lemma [ lem : mu ] , @xmath295 . in the general case , we were consequently not able , using this method , to erase the constant @xmath80 on the right hand side of ( [ result ] ) . we give now the proof of theorem [ teo : estimates ] . the beginning of this proof is similar , up to ( [ eq : i ] ) , to the proof of theorem 2.6 in @xcite . we rewrite it in order to be complete and clear . theorem [ teo : estimates ] let @xmath296 , @xmath297 , and set @xmath298 as in ( [ eq : phi ] ) . by krukov theorem ( * theorem 5 & section 5 , remark 4 ) , the set of hypotheses * ( h1 * ) * gives us existence and uniqueness of a weak entropy solution for any initial condition in @xmath299 . let @xmath8 be the krukov solution associated to @xmath6 and @xmath9 be the krukov solution associated to @xmath7 . by definition of krukov weak entropy solution , we have for all @xmath300 , for all @xmath301 @xmath302 \\ \times \mathrm{sign}(u - l ) \ , \mathrm{d}y \ , \mathrm{d}s { \geqslant}0 \end{array}\ ] ] and for all @xmath303 , for all @xmath304 @xmath305 \\ \times \mathrm{sign}(v - k ) \ , \mathrm{d}x \ , \mathrm{d}t { \geqslant}0 . \end{array}\ ] ] choose @xmath306 in ( [ eq : v ] ) and integrate with respect to @xmath198 . analogously , take @xmath307 in ( [ eq : u ] ) and integrate with respect to @xmath200 . by summing the obtained equations , we get , denoting @xmath308 and @xmath309 : @xmath310 \\ & \times \mathrm{sign}(u - v ) \ , \mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s & { \geqslant } & 0 \ , . \end{array}\ ] ] we introduce a family of functions @xmath202 as in ( [ eq : y ] ) . let @xmath206 and denote @xmath311 with @xmath312\times{{\mathbb{r}}}^n ) } , { { \left\|v\right\|}}_{{{\mathbf{l}^\infty}}([0,t_0]\times{{\mathbb{r}}}^n)})$ ] . we also define @xmath313 as in ( [ eq : chipsi ] ) , for @xmath208 , @xmath314 ( see also figure [ fig : chipsi ] ) . note that with these choices , equalities ( [ eq : derivatives ] ) still hold . note that here the definition of the test function @xmath315 is essentially the same as in the preceding proof ; the only change is the definition of the constant @xmath316 , which is now defined with reference to @xmath82 . we also introduce as above the function @xmath317 that is positive for all @xmath318 , and we have : @xmath319 \psi \ , \mathrm{sign}(u - v ) \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \\ { \leqslant } & \int_{{{\mathbb{r}_+^ * } } } \ ! \int_{{{\mathbb{r}}}^n } \ ! \int_{{{\mathbb{r}_+^ * } } } \ ! \int_{{{\mathbb{r}}}^n } \left [ { { \left|u - v\right| } } \chi ' \psi - b(t , x , u , v ) \chi y_\theta ' \right ] \psi \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \\ { \leqslant } & \int_{{{\mathbb{r}_+^ * } } } \ \int_{{{\mathbb{r}}}^n } \ ! \int_{{{\mathbb{r}_+^ * } } } \ ! \int_{{{\mathbb{r}}}^n } { { \left|u - v\right| } } \ , \chi ' \ , \psi \ , \psi \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \ , . \end{aligned}\ ] ] due to the above estimate and ( [ eq : sum ] ) , we obtain @xmath320 & & \\ & \displaystyle \times \mathrm{sign}(u - v ) \,\mathrm{d}x \,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s & { \geqslant } & 0 \ , , \end{array}\ ] ] i.e. @xmath321 , where @xmath322 \mathrm{sign}(u - v ) \ , \mathrm{d}x \ , \mathrm{d}t \ , \mathrm{d}y \ , \mathrm{d}s \ , , \\ \nonumber j_t & = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \big[\left ( f(s , y , v ) - f(s , y , u ) + f(t , y , u ) - f(t , y , v ) \right ) \cdot ( \nabla \psi ) \phi \\ \nonumber & & \qquad \qquad \qquad \qquad + ( { \mathinner{\rm div}}f(t , y , v)-{\mathinner{\rm div}}f ( s , y , v))\big ] \times \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s\ , , \\ \label{eq : k } k & = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left ( ( g - f)(t , y , u ) - ( g - f)(t , y , v ) \right ) \cdot ( \nabla\psi ) \phi \\ \nonumber & & \qquad \qquad \qquad \qquad \times \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s\ , , \\ \nonumber l_x & = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left ( f(t , y , u ) - g(t , x , v ) + { \mathinner{\rm div}}g(t , x , v ) - { \mathinner{\rm div}}f(t , y , v ) \right ) \varphi \\ \label{eq : lx } & & \qquad \qquad \qquad \qquad \times \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s\ , , \\ \nonumber l_t & = & \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \int_{{{\mathbb{r}_+^ * } } } \int_{{{\mathbb{r}}}^n } \left ( f(s , y , u ) - f(t , y , u ) \right ) \varphi \\ \nonumber & & \qquad \qquad \qquad \qquad \times \mathrm{sign}(u - v ) \,\mathrm{d}x\,\mathrm{d}t\,\mathrm{d}y\,\mathrm{d}s \ , . \end{aligned}\ ] ] now , we choose @xmath323 as in ( [ eq : nu ] ) , ( [ eq : mu ] ) , ( [ eq : mu ] ) . the estimate on @xmath232 is the same as in the proof of ( * ? ? ? * theorem 2.6 ) : thanks to lemma [ lem : esti ] , we obtain @xmath324 however , for @xmath325 , we derive a new estimate with respect to ( * ? ? * theorem 2.6 ) . indeed , as @xmath326 is @xmath75 in space , we can use the following taylor expansion : @xmath327 besides , we note that @xmath328 in the same way , we have @xmath329 so that finally @xmath330\cdot\nabla \mu \end{aligned}\ ] ] after a change of variable , we obtain @xmath331\\ & \qquad\qquad\qquad\qquad\qquad\cdot \frac{z}{{{\left\|z\right\|}}}\mu_1'({{\left\|z\right\|}})\,\mathrm{sign}(u - v ) \big\ } { \mathinner{\mathrm{d}{z}}}{\mathinner{\mathrm{d}{x}}}{\mathinner{\mathrm{d}{t}}}\,.\end{aligned}\ ] ] when @xmath332 goes to 0 , we obtain by the dominated convergence theorem @xmath333 as @xmath334 , we finally get @xmath335 for @xmath234 and @xmath237 , we avoid now the use of the derivatives in time thanks to an application of the dominated convergence theorem . we obtain @xmath336 for @xmath337 , we have @xmath338{\varphi}{\mathinner{\mathrm{d}{y}}}\,{\mathinner{\mathrm{d}{s}}}\,{\mathinner{\mathrm{d}{x}}}\,{\mathinner{\mathrm{d}{t}}}\,.\end{aligned}\ ] ] note that @xmath339 vanishes for @xmath340 . consequently , with @xmath341}({{\left\|u(t)\right\|}}_{{{\mathbf{l}^\infty}}({{\mathbb{r}}}^n ) } , { { \left\|v(t)\right\|}}_{{{\mathbf{l}^\infty}}({{\mathbb{r}}}^n)})$ ] and @xmath342\times \mathcal{s}_t(u , v ) \times [ -\mathcal{v},\mathcal{v}]\,,\ ] ] we obtain @xmath343)}{\mathinner{\mathrm{d}{x}}}{\mathinner{\mathrm{d}{t } } } \nonumber \\ & + { { \left\|{\partial}_u f\right\|}}_{{{\mathbf{l}^\infty } } ( \sigma_{t_0}^{u , v})}\int_0^t\int_{b(x_0,r+m(t_0-t ) ) } { { \left|u(t , x)-v(t , x)\right|}}{\mathinner{\mathrm{d}{x}}}{\mathinner{\mathrm{d}{t}}}\,.\label{eq : est_lx}\end{aligned}\ ] ] in order to estimate @xmath344 as given in ( [ eq : k ] ) , we follow the same procedure as in ( * ? ? ? * theorem 2.6 ) : let us introduce a regularisation of the @xmath345 dependent functions . in fact , let @xmath346 and @xmath347 , where @xmath348 and @xmath349 are such that @xmath350 and @xmath351 - 1,1\right[$ ] , @xmath352 . we introduce @xmath353 so that we obtain @xmath354{\partial}_{y_i}{\varphi}{\mathinner{\mathrm{d}{y}}}\\ & = & \!\!\int_{{{\mathbb{r}}}^n}\int_u^{u_\beta } \left({\partial}_u(s_\alpha(u - v)p_i(u))-{\partial}_u s_\alpha(u - v)p_i(v)\right){\partial}_{y_i}{\varphi}{\mathinner{\mathrm{d}{y}}}\\ & = & \!\ ! \int_{{{\mathbb{r}}}^n } \int_{u}^{u_\beta}\left ( s_\alpha'(u - v)(p_i(u)-p_i(v))+s_\alpha(u - v)p_i'(u)\right ) { \partial}_{y_i}{\varphi}{\mathinner{\mathrm{d}{y}}}\ , . \end{aligned}\ ] ] now , we use the relation @xmath355 to obtain @xmath356 when @xmath357 tends to @xmath216 , using the dominated convergence theorem we obtain @xmath358 applying the dominated convergence theorem again , we see that @xmath359 consequently , it is sufficient to find a bound independent of @xmath357 and @xmath360 on @xmath361 , where @xmath362 integrating by parts , we obtain @xmath363 we now search for a bound for each term of the sum above . * for @xmath364 , recall that @xmath365 . hence , by the dominated convergence theorem , we get that @xmath366 when @xmath367 . indeed , @xmath368 * for @xmath369 , denoting @xmath370\times\{\mathcal{s}_{t_0}(u)+b(0,\beta)\}\times[-{{\left\|u\right\|}}_{{{\mathbf{l}^\infty}}},{{\left\|u\right\|}}_{{{\mathbf{l}^\infty}}}]$ ] , we get @xmath371 we note besides that @xmath372 when @xmath373 . * for @xmath374 , we just pass to the limit in @xmath375 and then make @xmath332 goes to 0 . we see that : @xmath376 finally , letting @xmath377 and @xmath378 , due to ( * ? ? ? * proposition 3.7 ) , we obtain @xmath379 now , we collate the estimates obtained in ( [ eq : esti ] ) , ( [ eq : estj ] ) , ( [ eq : est_lx ] ) , and ( [ eq : estk ] ) . remark the order in which we pass to the various limits : first @xmath380 and , after , @xmath157 . therefore , we get @xmath381 ) } \mathrm{d}y\ , \mathrm{d}t \biggr ] \end{aligned}\ ] ] or equivalently @xmath382 where @xmath383 ) } \ ! \mathrm{d}y \ , \mathrm{d}t . \end{aligned}\ ] ] the bound ( [ result ] ) on @xmath384 gives : @xmath385 where @xmath77 is defined in ( [ eq : kappao ] ) and @xmath386 ) } \ , \mathrm{d}x \ , , \\ c(t ) & = & \int_{b(x_0,r+m(t_0-t ) ) } { { \left\|\left((f - g)-{\mathinner{\rm div}}(f - g)\right)(t , y,\cdot)\right\|}}_{{{\mathbf{l}^\infty}}([-{v}_t,{v}_t ] ) } \,\mathrm{d}y \ , , \end{aligned}\ ] ] since @xmath387 . consequently @xmath388 by a gronwall type argument , we obtain @xmath389 taking @xmath390 , we finally obtain the result . in the preceding proof , the main changes comparing to @xcite are essentially in the bound of @xmath233 . furthermore , we also gain some regularity hypotheses by avoiding the use of the derivative in time . we give below a lemma that was used in the previous proof . let us recall from @xcite the following useful technical results : [ lem : esti ] let @xmath232 be defined as in ( [ eq : i ] ) . then , @xmath395t , t+{{\eta}}\right [ } \int_{{{\left\|y - x_0\right\|}}\le r + { { \lambda}}+m(t_0-t)+\theta } { { \left|u(t , y)-u(s , y)\right| } } \ , \mathrm{d}y \ , . \end{aligned}\ ] ]

protoplanetary disks are believed to accrete onto their central t tauri star because of magnetic stresses . recently published shearing box simulations indicate that ohmic resistivity , ambipolar diffusion and the hall effect all play important roles in disk evolution . in the presence of a vertical magnetic field , the disk remains laminar between 1 , and a magnetocentrifugal disk wind forms that provides an important mechanism for removing angular momentum . questions remain , however , about the establishment of a true physical wind solution in the shearing box simulations because of the symmetries inherent in the local approximation . we present global mhd simulations of protoplanetary disks that include ohmic resistivity and ambipolar diffusion , where the time - dependent gas - phase electron and ion fractions are computed under fuv and x - ray ionization with a simplified recombination chemistry . our results show that the disk remains laminar , and that a physical wind solution arises naturally in global disk models . the wind is sufficiently efficient to explain the observed accretion rates . furthermore , the ionization fraction at intermediate disk heights is large enough for magneto - rotational channel modes to grow and subsequently develop into belts of horizontal field . depending on the ionization fraction , these can remain quasi - global , or break - up into discrete islands of coherent field polarity . the disk models we present here show a dramatic departure from our earlier models including ohmic resistivity only . it will be important to examine how the hall effect modifies the evolution , and to explore the influence this has on the observational appearance of such systems , and on planet formation and migration .

1501.05431

understanding the complex dynamical evolution of protoplanetary disks ( ppds ) is of key interest both for building a comprehensive theory of planet formation , and as a means of explaining the observationally inferred properties of these objects ( see * ? ? ? * for a recent review ) . for example , ppds are known to accrete gas onto their host stars at a typical rate of @xmath1 @xcite and have life times in the range @xmath2@xmath3 @xcite . more recent observations have indicated the presence of turbulence in the disks surrounding the young stars tw hydrae and hd 163296 , based on analysis of their molecular line emission @xcite . further evidence for turbulent broadening comes , for instance , from the co rovibrational band head shape measured by @xcite in v1331 cygni . during the t tauri ( class ii ) phase , self - gravity is expected to provide a negligible contribution to angular momentum transport ( due to the low disk - to - star mass ratio ) , and disk accretion is instead believed to be driven by magnetic effects . two possible sources of angular momentum transport are through a magnetocentrifugally driven wind @xcite , or through the operation of the magnetorotational instability ( mri , * ? ? ? * ) , whose nonlinear outcome in the ideal - mhd limit is magnetohydrodynamic turbulence @xcite . except for the innermost regions of ppds , where the temperature @xmath4 , and the ionization of alkali metals such as sodium and potassium provides strong coupling between the gas and magnetic field @xcite , non - ideal mhd effects such as ohmic resistivity , ambipolar diffusion ( ad ) and hall drift are expected to be important due to the low ionization levels of the gas ( e.g. * ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? external sources of ionization such as stellar x - rays , fuv photons and galactic cosmic rays are expected to ionize the disk surface layers , providing good coupling there even though recent results suggest that cosmic rays may be highly attenuated by the stellar wind @xcite . nevertheless , deep within the disk the evolution should be controlled by the non - ideal effects . the recognition of this led @xcite to propose what has now become the traditional dead - zone model in which the disk surface layers accrete by sustaining mri turbulence , with the shielded interior maintaining an inert and magnetically decoupled `` dead zone '' , where the mri is quenched by the action of ohmic resistivity only . the potential importance of the other non - ideal effects has long been recognized ( e.g. * ? ? ? * ; * ? ? ? * ) , but it is only recently that nonlinear shearing box simulations including ambipolar diffusion and the hall term have been performed in relevant parameter regimes , leading to a modified picture of how disks accrete that deviates significantly from the traditional dead zone one . generally speaking , it is expected that in disk regions between 1@xmath0 , ohmic resistivity will be dominant near the midplane , the hall effect will be dominant at intermediate disk heights and ambipolar diffusion will be most important in low density regions higher up in the disk @xcite . the highest altitude surface layers are expected to be ionized by stellar fuv photons , and as such will evolve in the ideal mhd limit @xcite . shearing box simulations presented by @xcite that include ohmic resistivity and ambipolar diffusion for models computed at @xmath5 demonstrate that ad quenches mri turbulence . they conclude that disks will be completely laminar between 1@xmath0 with angular momentum transport occurring because a magnetocentrifugal wind is launched from near the disk surface . for reasonable assumptions about the magnetic field strength and geometry , accretion rates in agreement with the observations are obtained . the inclusion of the hall term in the presence of a vertical magnetic field introduces dichotomous behavior , arising from the fact that the coupled dynamics depends on the sign of @xmath6 , where @xmath7 and @xmath8 signify the disk angular momentum vector and the ambient magnetic field direction , respectively . shearing box simulations presented by and @xcite show that when @xmath9 , the hall effect leads to the amplification of the horizontal magnetic fields within approximately two scale heights of the disk midplane , and the generation of a large scale maxwell stress through magnetic braking . in addition , the magnetocentrifugally driven wind is also seen to strengthen . when @xmath10 , however , magnetic stresses and winds are seen to weaken relative to the opposite case , and relative to the evolution observed when hall effects are neglected . quite how this _ hall dichotomy _ maps onto observations of ppds remains an open question . the internal dynamics of protoplanetary disks in the region 1@xmath0 are clearly of great significance for planet formation . the growth and settling of grains depends on the level of turbulence , with a laminar disk or one displaying weak turbulence providing the most favorable conditions although it should be noted that some turbulent mixing is required to maintain a population of small grains in the surface regions of ppds so that their spectral energy distribution can be reproduced by radiative transfer models . the collisional growth of planetesimals has been shown to be affected strongly by the level of turbulence @xcite , and the migration of low mass planets is also highly sensitive to the presence or otherwise of turbulence , with significant stress levels being required to unsaturate corotation torques . this paper is the latest in a series in which we are exploring the influence of magnetic fields and non ideal mhd effects on the formation of planets , with the eventual goal of producing comprehensive models of ppds that will be used to study planet building and evolution . in earlier work @xcite we have used a combination of global and shearing box simulations to examine the dynamical evolution of dust grains , boulders and planetesimals in turbulent disks , with and without dead zones . more recently , we have studied the influence of magnetic fields on gap formation and gas accretion onto a forming giant planet using global simulations that also included ohmic resistivity @xcite . in this paper , we include both ohmic resistivity and ambipolar diffusion in global disk simulations , and follow the dynamical evolution of the resulting disk models as a precursor to examining how ambipolar diffusion affects gas accretion onto a giant planet . of particular interest is the nature of the accretion flow in the disk , the nature of any magnetocentrifugal wind that is launched , and how these vary as a function of small changes in model parameters . these are the first quasi - global simulations to be conducted of ppds that are threaded by vertical magnetic fields and which include this combination of non ideal mhd effects , and as such are most comparable with the shearing box simulations presented by @xcite , hereafter . a fundamental question raised by the shearing box simulations is whether or not a physical wind solution with the correct field geometry emerges spontaneously in global disk simulations in a way that is not generally observed in shearing boxes because of their radial symmetry . probably the most important result contained in this paper is that physical wind solutions do indeed arise spontaneously in our global simulations , demonstrating the robustness of many of the features obtained in the shearing box simulations of . this paper is organized as follows : in section [ sec : meth ] we describe the numerical method , the disk model , as well as the ionization chemistry . our results are presented in section [ sec : res ] , where we mainly focus on the emerging wind solutions and the dynamical stability and evolution of forming current layers . in section [ sec : corona ] , we will moreover report an instability that is driven by the sharp transition into the fuv dominated layer , and in section [ sec:3d ] we assess whether secondary instabilities can drive non - axisymmetric evolution . we summarize our results in section [ sec : summary ] and discuss implications for planet formation theory in section [ sec : concl ] . we present magnetohydrodynamic ( mhd ) simulations of protoplanetary accretion disks employing 2d axisymmetric and 3d uniformly - spaced spherical - polar meshes . in the following , the coordinates @xmath11 denote spherical radius , co - latitude and azimuth , respectively . we moreover ignore explicit factors of the magnetic permeability @xmath12 in our notation . simulations were run using the single - fluid nirvana - iiicode , which implements a second - order - accurate godunov scheme @xcite . the code is formulated on orthogonal curvilinear meshes @xcite and employs the constrained transport method @xcite to maintain the divergence - free constraint of the magnetic field . as an alteration to the publicly available distribution of the code , we here adopt the upwind reconstruction technique of @xcite to obtain the edge - centered electric field components needed for the magnetic field update ( also see * ? ? ? this modification became necessary to take advantage of the more accurate hlld approximate riemann solver of @xcite , which offers advanced numerical accuracy for flows in the subsonic regime . our numerical model is based on solving the standard resistive mhd equations including ohmic resistivity as well as an electromotive force resulting from the mutual collision of ions and neutrals . given the typical number densities within ppds , the applicability of the strong - coupling approximation is warranted by detailed chemical modeling @xcite . accordingly the gas dynamics can be modeled by a single - fluid representing the motion of the neutral species , and the additional term can simply be evolved in a time - explicit fashion @xcite . the total - energy formulation of the nirvana - iiicode is expressed in conserved variables @xmath13 , @xmath14 , and @xmath15 , and if we define the total pressure , @xmath16 , as the sum of the gas and magnetic pressures , we can write our equation system as @xmath17 \partial_t(\rho{\mathbf v } ) + \nabla{\!\cdot\!}[\rho\mathbf{vv}+p^{\star}i-\mathbf{bb } ] & = & - \rho \nabla\phi \ , , \nonumber\\[4pt ] \partial_t e + \nabla{\!\cdot\!}[(e + p^{\star}){\mathbf v}- ( { \mathbf v}{\!\cdot\!}{\mathbf b}){\mathbf b } ] & = & \nabla \cdot \left[\ , ( { \mathbf e_{\rm o}}\!+\!{\mathbf e_{\rm ad}}){\!\times\!}{\mathbf b}\right ] \nonumber\\ & & - \rho ( \nabla\phi){\!\cdot\!}{\mathbf v}+ \gamma \ , , \nonumber\\[4pt ] \partial_t { \mathbf b}-\!\nabla{\!\times\!}(\ , { \mathbf v}{\!\times\!}{\mathbf b}+ { \mathbf e_{\rm o}}+ { \mathbf e_{\rm ad}}\ , ) & = & 0 \ , , \label{eq : mhd}\end{aligned}\ ] ] where the electromotive forces stemming from the ohmic and ambipolar diffusion terms are given by @xmath18 @xmath19\times{\hat{{\mathbf b}}}\ , , \label{eq : emf_ad}\ ] ] ( with @xmath20 the unit vector along @xmath21 ) , respectively . the gravitational potential @xmath22 is a simple point - mass potential of a solar - mass star . we moreover obtain the gas pressure as @xmath23 , where @xmath24 is appropriate for a diatomic molecular gas . the source term @xmath25 , mimicking radiative heating and cooling , is included to relax the specific thermal energy density @xmath26 to its initial radial temperature profile . the relaxation is done on a fraction of the local dynamical time scale , which is short enough to avoid strong deviations from the equilibrium model but long enough to suppress vertical corrugation of the disk due to the goldreich - schubert - fricke instability ( see * ? ? ? * for a detailed study of this phenomenon ) . we remark that , while this instability is physical in nature , its appearance may be exaggerated in a strictly isothermal simulation , and more realistic modeling including radiative transfer should be employed to study its relevance . because we use a time - explicit method , large peak values in the dissipation coefficients @xmath27 and @xmath28 impose severe constraints on the numerically permissible integration time step . we address this by using a state - of - the - art second order super - time - stepping scheme recently proposed by @xcite . in comparison with conventional first - order methods , this runge - kutta - legendre scheme is free of adjustable parameters and has been proven superior in terms of both accuracy and robustness . to maintain the second - order accuracy of our time integration , we use strang - splitting for the diffusive terms . in our previous local simulations ( cf . appendix b1 in * ? ? ? * ) , we have found that applying a constant cap on @xmath27 can _ qualitatively _ change the way in which the top and bottom disk layers are coupled to each other . for a typical wind configuration , the horizontal field components generally change sign at the midplane . we imagine that in this situation a clipped @xmath27 would affect the amount of field diffusing into the midplane region and conversely the strength of the current sheets above and below the magnetically decoupled layer . we therefore choose not to apply a cap on @xmath27 . however , we limit @xmath28 such that @xmath29 ( also see ) , where @xmath30 is the plasma parameter , and @xmath31 is the equivalent of a magnetic reynolds number for ad . this we find greatly reduces the numerical cost without noticeably altering the obtained solution . the ohmic and ambipolar diffusivities are evaluated dynamically for each grid - cell and are based on a precomputed look - up table , that has been derived self - consistently from a detailed chemical model accounting for grain - charging and gas - phase chemistry . the ionization rates that enter this equilibrium chemistry are governed by ionizing radiation entering the disk . the column densities that govern the attenuation of the external ionizing radiation are also computed on - the - fly , that is , they reflect the instantaneous state of the disc . in principle , this allows for a self - limiting of the emerging wind via shielding of ionizing radiation @xcite . in addition to some minor modifications to the grain charging prescription ( as detailed below ) , we have improved the realism of our ionization model compared to our previous work . the main alterations concern the inclusion of two additional radiation sources , an un - scattered direct x - ray component , and hard uv photons . these have in common a shallow penetration depth but comparatively high flux levels , thus mainly affecting the ionization level of the disk s surface layers . adopting a very simple prescription based on the recent study by @xcite , we have added the effect of an fuv ionization layer based on an assumed ionization fraction of @xmath32 ( representing full ionization of the gas - phase carbon and sulfur atoms susceptible to losing electrons when struck by fuv photons ) , and a collision rate coefficient of @xmath33 . we further assume that fuv photons penetrate to a gas column density of @xmath34 . due to their lower amplitude , we ignore any scattered , diffuse or ambient fuv illumination of the disk and only evaluate sight - lines pointing directly towards the central star . note that this deviates from the local simulations of , where the vertical gas column was used for attenuation of the incident flux . our treatment is motivated by the assumption that a large fraction of the fuv photons originate directly from the central star ( see upper right panel of figure 9 in * ? ? ? * who study the forward scattering of fuv photons in detail ) . a similar treatment is adopted for the unscattered x - ray component , for which we use the fit to the monte - carlo radiative - transfer results of @xcite given by eq . ( 21 ) in @xcite . deviating from our previous local simulations ( cf . eq ( 1 ) in * ? ? ? * ) , our new x - ray illumination is @xmath35\;{\it r}^{-2}\nonumber \\[2pt ] & + & 6{\!\times\!}10^{-12}{\,\rm s}^{-1}\;\ ; \exp{\,-\left(\frac{\sigma_c}{\sigma_{\rm ab}}\right)^{\,\beta}}\;{\it r}^{-2 } \label{eq : xray}\end{aligned}\ ] ] where @xmath36 ( @xmath37 ) is the gas column to the top ( bottom ) of the disk surface , and @xmath38 is the column density along radial rays towards the star . the radial column contains a contribution from the inner disk ( which is not part of the computational domain ) and reaches in to an inner truncation radius of @xmath39 . adopting values from @xcite , the shape coefficients are @xmath40 , and @xmath41 for scattered x - rays , and @xmath42 , and @xmath43 , for absorption of the direct component , respectively . both contributions additionally decay with the squared radius to account for the decrease in x - ray luminosity . to account for typical median values of @xmath44 in observational data of luminosities in young star clusters such as in the orion nebula @xcite , we enhance the incident x - ray flux by a factor of @xmath45 compared to the above stated values . in view of future work , we envisage additional improvements to the x - ray ionization model by adopting the new results of @xcite , who consider the enhanced ionizing effects of x - rays due to the presence of heavier elements ( assuming solar abundance ) . these alterations have been found to lead to enhanced ionizing fluxes at intermediate column densities compared to the original results of @xcite . our chemical network is based on the one used by , labeled ` model4 ` , with grain surface reactions and a simplified gas - phase reaction set involving one representative metal and one molecular ion . small changes in recent years include correcting the electron sticking probabilities for the grain charge @xcite and consistently treating the molecular ion such as hco@xmath46 ( with a mass of 29 protons ) since this is the long - lived species in a series of several reactions . further details on the reaction network and diffusivities can be found in section 2.2 . of @xcite as well as section 4.2 of @xcite . ) for ohmic resistivity ( light orange ) , and ambipolar diffusion ( red ) . solid lines represent the fiducial case of a dust - to - gas mass ratio of @xmath47 , whereas dash - dotted lines show profiles for a value of @xmath48 . we also plot the initial profile of the plasma @xmath49 parameter ( solid black line ) . ] [ fig : elsa ] the resulting _ initial _ ionization profile expressed in elsasser numbers @xmath50 with @xmath51 is plotted in fig . [ fig : elsa ] at a radius of @xmath5 , together with the height - dependent plasma @xmath49 parameter . the implications of this figure for our simulations are discussed in sect [ sec : prelim ] . the region with @xmath52 , where the plasma parameter becomes constant with height is caused by applying a lower limit on the gas density to avoid excessive alfvn speeds in the disk halo . following , we chose this limit at @xmath53 ( in our global model , this value is with respect to the initial midplane value at any given cylindrical radius ) . we remark that the equilibrium density profile is artificially steep in our isothermal disk model with temperature constant on cylinders . in this respect , the density floor acts to mimic a disk with realistic radiation thermodynamics , where the irradiation heating of the upper disk layers naturally creates an increased pressure scale height and hence a much shallower density profile . because @xmath54 ( and hence @xmath55 ) depends on the magnetization of the plasma , the actual density profile in the disk halo has an influence on how well the gas is coupled to the magnetic field lines , which in turn affects the efficiency of the magneto - centrifugal wind launching mechanism . this caveat should be kept in mind when interpreting the mass loading of the wind , which we expect to be a function of the thermodynamics in real protoplanetary disks . we now describe the underlying protostellar disk model , which is largely identical to the one used in @xcite . the equilibrium disk model is based on a locally - isothermal temperature , @xmath56 , decreasing with cylindrical radius as @xmath57 . for @xmath58 , such a profile leads to a constant opening angle throughout the disk as it is commonly used for global numerical simulations of ppds , and which provides us with the reference disk model here . to study what effect disk flaring has on the absorption of direct fuv photons , we also produce a moderately flaring disk with @xmath59 . for this value , the resulting power - law index for the local opening angle , @xmath60 , is @xmath61 , which is in agreement with observational constraints for this value based on sub - mm observations @xcite , and from multi - wavelength imaging . the radial power - law exponents for the gas surface density , @xmath62 , are @xmath63 for our fiducial model , and @xmath64 for the flaring disk model , respectively . our temperature distribution is complemented by a power - law for the midplane density , @xmath65 , with @xmath66 . the equilibrium disk solution is given by @xmath67\,\right)}\,,\;\textrm{and}\quad \label{eqn : rho}\ ] ] @xmath68^\frac{1}{2}\!\ ! , \label{eqn : omega}\ ] ] which can be derived from the requirement of hydrostatic/-dynamic force balance in the vertical and radial directions , and where the keplerian angular velocity @xmath69 . in the above equations , the square of the isothermal sound speed results from the prescribed temperature profile as @xmath70 , with a parameter @xmath71 corresponding to a value of @xmath72 . guided by previous results of local 3d simulations that exhibit quasi - axisymmetric structure , we mainly focus on 2d - axisymmetric simulations . our computational domain covers @xmath73{\,\rm au}$ ] . in the latitudinal direction , the grid covers eight pressure scale heights on each side of the disk midplane , that is , @xmath74 $ ] , with @xmath75 . in the case of the flaring disk , the coverage of scale heights varies from @xmath76 at @xmath77 to @xmath78 at @xmath79 . the grid resolution is @xmath80 , which means that we are able to resolve relevant mri wave lengths . in the axisymmetric models we use @xmath81 cells , corresponding to @xmath82 grid points per pressure scale height , matching the resolution of the three - dimensional box simulations of . for the two non - axisymmetric simulations , the azimuthal extent is restricted to @xmath83 $ ] , with @xmath84 ( a quarter wedge ) , or @xmath85 to limit the computational expense . we furthermore note that in the ideal - mhd case , where turbulence develops , the azimuthal domain size has been found to have an effect on the final saturated state@xcite , at least for the case of zero net flux . because our ionization model depends on real physical values of the gas column density , we have to chose a meaningful normalization factor @xmath86 , which we fix according to a surface density of @xmath87 at the location @xmath88 , compatible with the typical minimum - mass solar nebula ( mmns * ? ? ? * ) . in physical units , the disk temperature is @xmath89 at a radius of @xmath90 , and @xmath91 at @xmath0 , resembling typical expected conditions within the protosolar disk . while our values for @xmath13 , @xmath56 and @xmath92 are similar to the mmsn values at @xmath0 , our adoption of different values for the radial power - law indices means that these values are different at @xmath5 compared to those adopted by . our model disk is initially threaded by a weak vertical magnetic field @xmath93 corresponding to a midplane @xmath94 independent of radius for our fiducial disk . to achieve this , the vertical magnetic flux has to vary as a power - law in radius , taking into account the radial distribution of the midplane gas pressure that itself depends on the density and temperature . in our disk model , the gas pressure , @xmath95 decreases outward , implying that @xmath93 falls off with radius , too . while such a configuration is generally expected when accounting for inward advection and outward diffusion of the embedded vertical flux @xcite , our particular choice of keeping the _ relative _ field strength constant is of course arbitrary . to preserve the solenoidal constraint to machine accuracy , we compute the poloidal field components from a suitably defined vector potential @xmath96 . the initial disk model is perturbed with random white - noise fluctuations in the three velocity components . the rms amplitude of the perturbation is chosen at one percent of the local sound speed . we furthermore add a white - noise perturbation in the magnetic field with an rms amplitude of a few @xmath97 , which is on the sub - percent level compared with the net - vertical field of typically a few ten @xmath98 . to complete the description of our numerical setup , we need to specify boundary conditions ( bcs ) . for the fluid variables , we employ standard `` outflow '' conditions at the inner and outer radial domain edges , @xmath99 and @xmath100 . this type is a combination of a zero - gradient condition in the case of @xmath101 ( where @xmath102 denotes the outward - pointing normal vector ) , and reflective bcs in the opposite case , thus preventing inflow of material from outside the domain . at the upper and lower boundaries ( that is , in the @xmath103 direction ) , we furthermore balance the ghost zone values such that a hydrostatic equilibrium is obtained . this helps to control artificial jumps of the fluid variables adjacent to the domain boundary as they are frequently encountered with finite - volume schemes . in contrast to our earlier global simulations of layered accretion disks subject to ohmic resistivity and containing an embedded gas - giant planet @xcite , we here make a different choice for the vertical magnetic - field boundary condition . whereas we previously applied magnetic bcs of the _ perfect conductor _ type ( that is , forcing the normal field component to zero and allowing non - zero parallel field ) , we here use _ pseudo vacuum _ conditions , conversely enforcing the perturbed parallel field to vanish at the surface and only allowing a perpendicular perturbed field . note that we exclude the initial net - vertical field from the procedure such that only deviations from the background field are subject to the normal - field condition . the vertical flux threading the disk is thus preserved . since the disk s upper layers are magnetically dominated , and the lorentz force acts to align the flow lines with the magnetic field , letting the field lines cross the domain boundary is of course a prerequisite for launching an outflow . while we realize that forcing the radial and azimuthal components of the perturbed field to vanish at the boundary may unduly restrict the magnetic topology of the emerging wind solution , we postpone the study of less - restrictive but more cumbersome boundaries to future work . unlike in a radially - periodic local shearing - box simulation , our global model is critically affected by the inner radial boundary condition . in a real protoplanetary disk , we can expect that alkali metals will be thermally ionized in this region , leading to the development of the mri on timescales short compared with the orbital timescale at the inner domain boundary of our model . this poses the question how to best interface the mri - turbulent inner disk with the magnetically decoupled midplane of the outer disk that we model . it is likely that mri - generated fields can efficiently leak into the outer disk via magnetic diffusion . in a first attempt to account for the mri - active inner disk , we gradually taper the diffusivity coefficients to zero within the inner @xmath104 of our disk model . studying the inner edge of the dead zone will however require dedicated simulations including this transition region . the main aim of our paper is to establish the laminar wind solutions that previously found in local shearing box simulations in the context of global disk simulations . in the interest of economic use of computational resources and to guarantee adequate numerical resolution of our global models , we primarily focus on 2d axisymmetric simulations , but have also run three - dimensional simulations to check for non - axisymmetric solutions . since all our runs include a net - vertical flux , axisymmetry is warranted to obtain a reading on the development of the mri . in cases where the solution proves laminar , axisymmetry is likely to produce a reasonably accurate picture . we address the possibility of non - axisymmetric secondary instabilities using a limited set of three - dimensional calculations , described in section [ sec:3d ] . .summary of simulation runs . [ tab : models ] [ cols="<,^,^,^,^,^,^,^",options="header " , ] the emerging current layers are directly related to resistively - modified vertically - global mri channel modes see @xcite , and fig . [ fig : mri_mode ] in section [ sec : mri_current ] . with mri channels potentially spanning large radial extents , as in the ohmic run shown in fig . [ fig : vis2d_o - b6 ] , such layers may be fed diffusively with magnetic field from the mri - active inner disk . this is supported by our present models , where current layers appear to emerge from inner disk regions . what determines the exact shape of the surviving field configuration at late times is presently unclear . we speculate that the particular realization seen in the lower panel of fig . [ fig : vis2d_oa - b5 ] is not necessarily a unique solution given the specific disk geometry and ionization model , but may to some extend depend on the early evolution history . however , simulations that were initiated with different random seeds , but were otherwise identical , only showed minor variations in the final field appearance . the configurations observed at late times are moreover found to remain quasi - stationary , at least during the evolution time of @xmath105 that we have investigated . perhaps the most noteworthy result from run oa - b5 is the spontaneous emergence of the expected `` hourglass '' geometry for the magnetic field , allowing magnetic tension forces to accelerate the wind . for the adopted vertical field polarity , the field must bend radially outward near the upper and lower disk surfaces , and the azimuthal field should point in the positive direction in the lower hemisphere and reverse direction above the midplane , as observed in the upper panel of figure [ fig : vis2d_oa - b5 ] . the shearing box simulations of have reflectional symmetry in the radial direction , that is , with respect to mirroring about @xmath106 , and hence contain no information about the location of the star . instead of giving rise to a physical wind solution , their fiducial run produces a radial field that possesses odd - symmetry about the midplane , with an inward pointing field in one hemisphere and an outward pointing one in the other . to test the robustness of the emerging solution in our setup , we initiate several instances of run oa - b5 , using different random number seeds when perturbing the initial velocity field , and in each case we recover the correct wind solution with outward bending field lines . our initial model has a net - vertical field with a radial dependence such that the midplane value of @xmath107 remains constant . because the gas pressure itself decreases with radius , the corresponding @xmath93 ( with @xmath108 ) results in an _ outward _ magnetic pressure force , which moreover has a stronger acceleration effect in the low - density upper disk layers . while a vertical flux distribution that has the flux centrally concentrated can be expected when accounting for inward advection and outward diffusion of flux @xcite , our particular choice is of course arbitrary . radial magnetic gradients are furthermore known to affect the outflow s mass loading in axisymmetric calculations where the disk is a boundary condition @xcite . in our simulations , the pressure gradient related to @xmath93 may potentially affect the direction toward which the field lines first bend from their starting configuration . we have tested this by running reference models with different magnetic pressure gradients . if @xmath109 , the initial condition has an unbalanced inward pressure force , and we do indeed find the field lines to spontaneously bend _ inward _ ; this happens simultaneously in the upper and lower hemisphere of the disk , such that the overall symmetry that is obtained is again even . the launching of the inward wind is restricted to the inner disk , probably because the vertical field lines become too stiff ( owing to @xmath110 increasing with radius ) to be suitable for wind launching in the outer disk . for a neutral situation with @xmath111 , we still observe _ outward _ bending of the field lines . starting from the inner radial domain , the outward propagating establishment of the wind region is found to stall at some radius , whereas the wind was quickly established throughout the entire domain in model oa - b5 . as before ( for the case of an outward magnetic pressure gradient ) , this is probably related to the field lines becoming too strong to support the wind launching mechanism at large radii . the wind indeed stalls further out in a run with weaker overall net - vertical field . this poses the question whether the vertical profiles of @xmath112 , and @xmath113 that we obtain from our ionization model put strong constraints on permissible vertical field amplitudes as a requirement for wind launching . a possible reason for the outward bending , even in the case of neutral magnetic pressure forces , may be that the vertical shear present in the global models ( because of the baroclinic conditions ) naturally bends the azimuthal component of the field in the correct direction required by the physical wind solution . we have checked that the outward bending of the vertical field lines is however equally seen in ( strongly flaring ) _ globally _ isothermal models that do not have any vertical gradients in the rotation velocity or any radial gradients in the vertical magnetic field . the outward pointing configuration is energetically favored because setting it up involves diluting rather than concentrating magnetic flux . our simulations demonstrate in any case that global models spontaneously develop the correct field geometry , although we caution that this conclusion needs to be tested in future simulations that moreover adopt improved boundary conditions for the magnetic field at the upper and lower ( and potentially the inner radial ) surfaces of the simulation domain . ultimately , the emergence of the wind geometry will have to be studied in simulations that do not start from well - motivated ( but nevertheless arbitrary ) initial conditions , but do account for the formation of the ppd from a collapsing molecular cloud @xcite . lastly , for the parameters that we have considered here , we do not find a self - limiting of the wind via shielding of fuv photons @xcite . this is consistent with the t tauri scenario of , who performed disk wind chemical modeling for various protostellar evolution stages . the authors find the shielding of fuv photons to be important for the class i and class 0 cases . their t tauri case , with mass flow rates comparable to our fiducial model , is sufficiently fuv - ionized to reduce ambipolar diffusion enough so that the neutrals are swept up in the wind out to at least a radius of @xmath114 . ( see the inner dashed line in fig . [ fig : vis2d_oa - b5]b ) . the alfvnic point with respect to @xmath115 is marked by filled circles . dotted vertical lines indicate the base of the wind , defined as the point where @xmath116 becomes super - keplerian @xcite . ] [ fig : wind_oa - b5_r1 ] except for the reversal of the horizontal magnetic field components ( which at this stage appears to occur rather gently instead of within a narrow current layer ) , the field structure observed in the upper panel of fig . [ fig : vis2d_oa - b5 ] is similar to that described by and shown in their figure 10 . near the midplane , where the dominance of the ohmic resistivity retards the growth of currents , the field is dominated by the vertical component . moving to higher altitudes , where the magnetic coupling increases , the field bends outwards because a radial velocity is generated through the azimuthal force balance between the coriolis force and magnetic tension . we enter the fuv layer at disk altitudes @xmath117 , coinciding with the base of the wind and the wind itself , as illustrated by the velocity vectors . at later times , the lower panel of fig . [ fig : vis2d_oa - b5 ] shows that the varying polarity of the strong azimuthal field belts gives rise to two distinct field line geometries . in the inner region ( @xmath118 ) the orientation of the field belts is the same as the one in the wind , and the vertical field lines bend smoothly at the transition into the wind base . because some negative @xmath119 has diffused into the ohmic - dominated midplane , there is a weak current layer forming at @xmath120 . in contrast , further out at @xmath121 , the horizontal - field belts are anti - aligned with the magnetic field direction of the wind . this can also be seen in the curvature of the field lines , which are concave ( towards the star ) in this region . to connect the different layers , the azimuthal and radial field components have to change their direction within two thin current layers located at @xmath122 . to study these configurations in more detail , we plot separate vertical profiles for @xmath77 , and @xmath123 in figs . [ fig : wind_oa - b5_r1 ] and [ fig : wind_oa - b5_r2 ] , respectively . for easy reference , the radial positions are furthermore indicated in fig . [ fig : vis2d_oa - b5 ] by means of dashed lines . the isothermal sound speeds , for the initial disk model , are @xmath124 , and @xmath125 at the two respective radii . the wind profile in the inner disk ( see fig . [ fig : wind_oa - b5_r1 ] ) , is qualitatively similar to the wind solutions previously found in the quasi-1d simulations with `` even - z '' symmetry , plotted in fig . the current layer is much weaker in our case , and the dip in @xmath126 is barely visible in the left inset of fig . [ fig : wind_oa - b5_r1 ] . this may be related to differences in the ionization model resulting in @xmath127 in the region around @xmath128 , in our disk model . differences in the particular diffusivity profiles also explain the pronounced peaks in @xmath119 in this `` undead '' transition layer , which are not seen in the simulations of . but for @xmath123 , where the horizontal field belts have opposite parity compared to the field within the wind layer . inlays magnify the velocity profiles near the strong current layers , which reach a fair fraction of the local isothermal sound speed , @xmath129 . ] the profile at @xmath123 , with double field reversal in @xmath119 , is illustrated in fig . [ fig : wind_oa - b5_r2 ] . because of the strong current layers that are emerging in this case , this configuration resembles more closely the situation seen in the local models ( again cf . fig . 10 in ) . unlike in their simulation , we however observe _ two _ current layers , one on each side of the disk . these arise because the development of mri channel modes , and their subsequent evolution into `` undead '' azimuthal field belts with opposite polarity to the background wind , forces the horizontal field to change direction three times . strong current layers occur at altitudes where the field and gas couple more strongly , namely at @xmath130 . although the occurrence of strong azimuthal field belts is not precluded in local simulations ( as their existence may depend on the details of the ionization model ) , it seems likely that the alternating field belts we observe will only be a robust feature of global simulations . apart from their multiplicity , the current layers appear very similar to the ones presented in . the radial velocity @xmath126 shows a dip , while @xmath116 is modulated as a sine ( the slight offset of @xmath116 in our simulation is due to the radial pressure support ) . the characteristic modulation ( reflecting the abrupt kink in the field lines ) is clearly seen in @xmath131 and @xmath119 , which agree very closely with the @xmath132 , and @xmath133 obtained by . the emergence of such strong current layers naturally begs the question of their stability . before we however discuss this issue in section [ sec : current ] , we first want to take a look at the current sheets likely origin as remnants of mri eigenmodes . [ fig : mri_mode ] to draw a closer connection between localized mri channel modes and the current layers observed in our simulations , we compare the structure of the fastest growing mri eigenmode with the initial evolution of our standard setup with @xmath134 . the mode structure we find is qualitatively similar to that shown for a spatially constant ohmic resistivity , and resembles the eigenmodes shown in figure 4 of , who performed hall - mhd calculations that also included variable ohmic resistivity . dissipation modifies the mri eigenmodes predominantly near the midplane , where the relative magnetization is lowest and the mri wavenumbers are the highest , explaining the similarity of the solutions for constant @xmath135 , and spatially varying @xmath136 . here , instead of a stratified shearing box framework as previously employed by @xcite , we used the framework introduced by @xcite to capture the full vertical structure of the baroclinic initial condition . the eigenmode analysis uses the linear equations described in ( * ? ? ? * section 5 ) with the addition of terms for spatially varying ohmic resistivity , and using the functional forms and coefficients appropriate for the cylindrical temperature structure of the disk initial condition ( * ? ? ? * section 3.3 ) . the equations were discretized in real space with chebyshev cardinal functions on a gauss - lobatto grid with 1000 points with the same vertical extent as the global simulation ( @xmath137 scale heights ) . boundary conditions were enforced by the `` boundary bordering '' method @xcite forcing the magnetic field perturbations to zero at the vertical edges of the domain . the analysis was performed at radial position @xmath138 where the ohmic resistivity profile , @xmath139 , used was extracted from the simulation initial condition and interpolated to the grid used for the eigenmode calculation . we moreover obtained eigenmode solutions that additionally included a fixed @xmath140 profile , and these showed a nearly indistinguishable mode structure . because of the extra non - linearity introduced by the @xmath107 dependence of @xmath140 , we however restrict our comparison to the ohmic - only case . in figure [ fig : mri_mode ] , we compare the early stage of our fully nonlinear global simulation with the fastest growing eigenmode . the localized field reversals are also clearly seen in the upper panel of fig . [ fig : vis2d_oa - b5 ] , which shows the early evolution of model oa - b5 ( however including ad ) . to isolate the linear mri mode from the overlaid wind solution , we subtract respective field profiles from two distinct snapshots during the linear growth phase of the mri . this is possible because the wind configuration emerges quickly , and then provides an approximately stationary background in which the mri grows . differencing two snapshots in time hence allows us to remove the wind part and extract the exponentially - growing mri eigenmode . while growing perturbations near the midplane are high - wavenumber and thus efficiently damped , the relative field strength in the disk halo is too strong to allow for the mri modes to fit within the available domain . this only leaves an intermediate disk region near @xmath141 for mri channel modes . in the ohmic+ad simulations , this corresponds to the position of the `` undead '' field belts ( see lower panel of fig . [ fig : vis2d_oa - b5 ] ) with adjacent current layers . in those simulations , the back - reaction of the growing eigenmodes on the @xmath140 profile makes the mri self - quenching , leaving the current layers as a remnant of the initial mri channel mode . . _ bottom : _ azimuthal magnetic field strength , @xmath119 , and poloidal magnetic field lines . the wavenumber of the perturbations is found to depend non - trivially on radius . only for @xmath121 , the nonlinear stage is clearly identified . the kh pattern is furthermore imprinted onto the mass - loading of the wind.,title="fig : " ] + . _ bottom : _ azimuthal magnetic field strength , @xmath119 , and poloidal magnetic field lines . the wavenumber of the perturbations is found to depend non - trivially on radius . only for @xmath121 , the nonlinear stage is clearly identified . the kh pattern is furthermore imprinted onto the mass - loading of the wind.,title="fig : " ] [ fig : kh_oa - b5-flr ] while we find the current layers to remain stable and long - lived in model oa - b5 , some of the other simulations which we will discuss below also show signs of instability , leading to less regular fields . [ fig : vis2d_oa - b5-d4 ] an example of this is illustrated in fig . [ fig : kh_oa - b5-flr ] , where we show the vertical velocity , @xmath142 ( upper panel ) , and the azimuthal magnetic field , @xmath119 ( lower panel ) , at the end of the flaring disk simulation oa - b5-flr . both color scales have been restricted to highlight features in the current layer . in this model , the surface density interior to @xmath5 is smaller than in oa - b5 , and beyond @xmath5 it is larger , altering the ionization structure sufficiently to put the current layer in a different regime . in particular , the increased ionization in the inner disk apparently allows the azimuthal field belts that form at intermediate heights to reach larger amplitudes . this , combined with the fact that these field belts are of opposite polarity relative to the background wind field , causes the two current layers located at @xmath143 to be stronger and to support narrow , radially directed flows , or `` jets '' , similar to those in run oa - b5 , as illustrated in figure [ fig : wind_oa - b5_r2 ] . inside @xmath144 , the velocity near the current layer shows a distinct radial wavenumber , suggesting that the shear associated with the radial jets leads to growth of a kelvin - helmholtz ( kh ) instability . the pattern , however , remains more or less stationary , indicating that the instability is not growing significantly into the nonlinear stage . although not obvious in fig.[fig : kh_oa - b5-flr ] , wave - like perturbations to the magnetic field are also present in the current layer . the possibility of current - driven tearing modes being present also arises , but analysis of the shear rate associated with the radial jets suggests that the observed instability is driven by kh modes . the velocity associated with the shear is super - alfvnic , and is therefore expected to stabilize the tearing modes that may in principle grow in a shearing background @xcite . the wavelength of the velocity perturbations and magnetic field changes abruptly for @xmath145 , where now a characteristic vortex - pair pattern emerges . the exact reason for this abrupt transition is unclear , since most characteristic quantities , such as the elsasser numbers @xmath146 , and @xmath55 , only depend weakly and continuously on radius . the most likely explanation may be the radial variation of the azimuthal field ( inside the `` undead '' belts ) , which drops in strength around this radial location . this is accompanied by a weakening and widening of the current layer , and a significant reduction in the inflow velocity of the radial accretion flow associated with the current layer , and a corresponding reduction in the shear relative to the background fluid . this suggests that the abrupt change in the nature of the disturbance observed at radius @xmath147 is unlikely to be driven by a kh instability , but the weakening of the shear may allow a tearing - mode instability to develop instead . the appearance of disturbances that have the character of growing magnetic islands provides some support for this interpretation , but this is clearly a tentative explanation that will require further investigation in future work . the presence of a background wind , keplerian shear , ambipolar diffusion and a radially - directed inflow associated with the current layer make an analysis of this problem somewhat complicated , and beyond the scope of this paper . we note , however , that the local studies on the development of parasitic modes in resistive mhd @xcite may provide some insight , even though ambipolar diffusion is neglected in those studies . @xcite has analyzed mri parasitic modes in the presence of finite ohmic resistivity and viscosity . in this paper , @xmath146 , is identified as the relevant dimensionless number ( see their figure 11 ) , with tearing modes being the dominant parasitic modes for @xmath148 , and kelvin - helmholtz modes dominating otherwise . considering both @xmath146 , and @xmath55 , we assess that our simulations are situated near this transition point . figure [ fig : vis2d_oa - b5-d4 ] shows the final appearance of the entire disk for model oa - b5-d4 , which has a reduced dust fraction of @xmath48 compared to a value of @xmath47 in model oa - b5 . on one hand , as the reader may check from table [ tab : results ] , the two models are rather similar in their bulk properties , especially those related to the wind . this is unsurprising , since the reduced cross - section for recombination on grains affects the ionization fraction significantly only at large gas columns ( see fig . [ fig : elsa ] ) , below the base of the wind . the grains are less important for the ionization state in the fuv layer , where the electrons are so abundant that many remain free once the grains charge to the coulomb limit @xcite . in contrast to our fiducial model , the transition between the base of the wind and the ohmic - dominated disk interior is very chaotic in model oa - b5-d4 , whereas it is much more stable in model oa - b5 . it appears that the larger values of @xmath146 and @xmath149 in this model at heights @xmath150 allow the channel modes to grow further into the nonlinear regime , such that parasitic modes are able to develop and cause the field belts to break up into regions with locally coherent field . the field amplification during this phase means ambipolar diffusion strengthens , as in run oa - b5 . eventually the local evolution is driven by a combination of field winding and diffusion . the end result is the creation of local `` undead '' patches that evolve on timescales that are comparable to the run duration . the different evolution of the horizontal field belts observed in runs oa - b5-flr and oa - b5-d4 compared with oa - b5 show that the evolution of the disk interior depends critically on the precise disk model that determines the elsasser number profiles . in particular , the ionization state in the region between the ohmic - dominated disk midplane and the ad - dominated intermediate disk layers depends sensitively on various model parameters . this suggests that a careful parameter study is required to obtain a reliable picture of how ppds evolve over their life - times especially since their properties change substantially over these time scales . such a parameter study would appear to be most warranted when additional physics is included in the model , such as the hall effect and more realistic thermodynamics . in a situation where the gas is allowed to be heated by resistive effects , the ionization balance may be shifted towards higher conductivity in the vicinity of the current layers , potentially leading to intermittent behavior . a distinct instability emerges in the magnetically dominated corona of the laminar disk ( see fig . [ fig : fishbone ] ) . the instability is found to grow close to the smallest available wavenumber , and is only present in the highly - resolved axisymmetric simulations . the perturbations are roughly aligned with the outward - pointing field lines of the laminar wind solution . similar patterns have been observed in 2d - axisymmetric shearing box simulations with strong vertical fields performed by , who also finds field - aligned structures on the shortest length scales resolved on the grid . the instability also bears resemblance with the `` striped wind '' seen by . similar to the instability seen by , we find a moderate modulation of the vertical mass flux by the instability pattern . we however do not find material falling down along the field lines as reported there for certain parameters . we identify three candidate mechanisms for creating such disturbances : ( i ) the vertical - shear instability investigated recently by @xcite , potentially modified by anisotropy from field - line tension , ( ii ) a kelvin - helmholtz type instability related to the vertical gradient of the horizontal velocity near the base of the wind , and ( iii ) a buoyant interchange instability of the azimuthal field . it is important to note that the kh - type instability can create the elongated aspect ratio seen in fig . [ fig : fishbone ] , when perturbations are carried - away by the overlaid laminar wind velocity and we indeed find this to be the case in situations where the instability occurred before the wind was established . the conclusive hint that a kelvin - helmholtz instability causes the pattern in fig . [ fig : fishbone ] is a strong velocity gradient at the exact location of the transition into the fuv layer . this gradient can be clearly identified in fig . [ fig : wind_oa - b5_r2 ] ( situated just above the wind base at @xmath151 ) . the shear layer manifests itself as an additional steepening of the wind profiles , and can also be seen in the corresponding plot of ( see their figure 10 ) . the physical origin of this sharp velocity gradient can be traced back to a peak in the ambipolar part of the electromotive force in the induction equation . because the coefficient @xmath152 sharply drops at the position of the fuv interface , the term @xmath153\times{\hat{{\mathbf b}}})$ ] can become significant in the induction equation even for smoothly varying ( or constant ) @xmath21 . in the steady state , this curl of the ambipolar electromotive force needs to be balanced by the induction term , @xmath154 , requiring as a consequence a significant shear gradient in the horizontal velocity ( again assuming a smooth @xmath21 ) . given its strong dynamical effect , this raises the question whether the coincidence of the wind base and the fuv interface is in fact accidental as claimed by . in our disk model , the fuv layer is imposed _ ad hoc _ , with a prescribed column density ( section [ sec : fuv ] ) . the resulting interface is smeared out to a prescribed width . reflecting the predominant _ direct _ origin of the fuv photons , while scattering is expected to remain negligible , the tapering is done in the radial direction , and as a function of column density ( rather than position ) . as a consequence , the resulting @xmath152 may develop sharp steps on the numerical grid in the @xmath103 direction in turn resulting in sharp velocity gradients . we accordingly ran a reference model , where the fuv transition was softer . in this model , we did not observe a strong velocity gradient near the wind base , and also found the instability to be largely suppressed . also the flaring - disk model , oa - b5-flr , did not show any signs of the instability . this can potentially be explained by the fuv transition following a _ concave _ line in this case , producing a different aliasing of the tapering function on the spherical - polar mesh . we conclude that the instability seen in fig . [ fig : fishbone ] is likely caused by the particular numerical parametrization of the fuv layer in our model , and is entirely unrelated to the `` striping '' instability seen in and . the instability may be be triggered by numerical effects at the grid scale , in a manner analogous to how physical secondary kelvin - helmholtz instabilities can be triggered by purely numerical effects as a shear layer thins to the grid scale @xcite . we remark that in the case of this wind instability too the mechanism behind the instability is genuinely physical , and may occur in real ppds where there is a sharp local change in the ionization state . magnetic field . _ top : _ model oa - b5-flr with a laminar wind . _ bottom : _ a run with mri active surface layers in a domain of slightly different radial size , from @xcite . owing to the restricted azimuthal extent in both runs , the only wavenumbers present are multiples of four.,title="fig : " ] + magnetic field . _ top : _ model oa - b5-flr with a laminar wind . _ bottom : _ a run with mri active surface layers in a domain of slightly different radial size , from @xcite . owing to the restricted azimuthal extent in both runs , the only wavenumbers present are multiples of four.,title="fig : " ] the presence of secondary instability mandates investigating the non - axisymmetric evolution . this is done exemplarily for the two models oa - b5 , and oa - b5-flr , whose 3d counterparts oa - b5-nx , and oa - b5-flr - nx are also listed in table [ tab : results ] , showing only minor deviations from their respective 2d runs . given the moderate magnetic reynolds numbers in these regions , it is not _ per se _ obvious whether non - axisymmetric instabilities can transfer significant amounts of energy into modes with higher azimuthal wavenumber . to address this question , in fig . [ fig : mpower ] , we show vertically - averaged azimuthal power spectra of the magnetic field . the color - coding indicates the logarithm of the power spectrum of @xmath155 as function of wavenumber , @xmath156 , and radial position in the disk . in the upper panel , we show the result for model oa - b5-flr . with the exception of the kelvin - helmholtz features ( between @xmath157 ) also seen in fig . [ fig : kh_oa - b5-flr ] , almost all power resides in @xmath158 . even in the regions of the secondary instabilities , the power in modes with @xmath159 remains very small . the power seen adjacent to the inner radial domain boundary is caused by reducing the diffusion coefficients there , which is an attempt to mimic the mri - active inner disk ( see section [ sec : bcs ] for details ) . this can be contrasted to the case of the classic layered ppd with mri - active surface layers ( see model m1 in * ? ? ? * ) , which we plot in the lower panel of figure [ fig : mpower ] . even with the laminar midplane layer , a significant amount of power is found for high-@xmath156 modes ( note the different color bar in this figure ) . the lack of power near the radial boundaries is owing to a dissipative buffer region . @xcite focused on the effects of a planet , which they inserted after the time shown here . we hence conclude that in the simulations that are dominated by axisymmetric configurations of ad - laminar winds , secondary instabilities do not lead to significant power in non - axisymmetric perturbations . this may potentially change in the presence of stronger current layers that are prone to nonlinear stage of parasitic modes as , for instance , can be seen for model oa - b5-d4 with a stronger dust depletion . when combined with stronger net - vertical fields and flaring , these dust - depleted disks may develop intermediate layers that are on the verge of becoming turbulent . we have presented the first quasi - global mhd simulations of ppds that include ohmic resistivity and ambipolar diffusion . in agreement with the shearing box simulations of , we find that the inclusion of ambipolar diffusion has a dramatic effect on the evolution . our main results may be summarized as follows . 1 . in simulations where we include a weak vertical magnetic field and ohmic resistivity , but neglect ambipolar diffusion , we obtain evolution that is consistent with the traditional dead zone picture : surface layers that sustain mri - driven turbulence and a quasi - laminar midplane region . 2 . the inclusion of ambipolar diffusion in models with weak magnetic fields , where the midplane value of @xmath160 , results in quenching of the surface layer turbulence . such disks display very weak internal motions and negligible angular momentum transport . our most important result arises when we increase the strength of the magnetic field so that @xmath161 . this leads to a disk which is entirely laminar between 1@xmath0 , and , despite the still relatively weak field , to the launching of a magnetocentrifugal wind from the upper ionized layers , resulting in a significant wind stress being exerted on the disk that drives inward mass flow at a rate @xmath162 . we have tested the robustness of the wind solution by running multiple simulations with modified initial conditions . because of the outward pressure force from the initial @xmath93 , and the vertical differential rotation , we generally obtain magnetic fields that point outwards in the radial direction in both hemispheres , as required for a physical wind solution that extracts angular momentum from the disk . this is with the exception of a model with @xmath109 , where an inward pointing wind was found . the calculations treat the time - varying absorption of the fuv radiation from the star and its vicinity in the intervening material , yet there is no sign that absorption by the wind limits the ionization of the launching layer for mass loss rates in the t tauri range . strong horizontal field belts are formed as remnants of stagnated mri modes at intermediate heights in our disk models , and are likely maintained through a balance between field winding and ambipolar diffusion . these are found to have field polarities that may be aligned or anti - aligned with the background field associated with the wind , affecting the number of current layers that arise in the disk because of local field reversals . the belts polarity may depend sensitively on the early evolution of the mri . the current layers drive substantial accretion flows that are narrowly localized in height . the associated shear can be strong enough to allow small - scale kelvin - helmholtz instabilities to grow . the global evolution of the horizontal field belts is found to depend on the ohmic and ambipolar elsasser numbers ( and hence the ionization fraction ) . in our fiducial model , the belts alternate in polarity over large radial extents , but otherwise appear to be stable and only evolve on time scales that are long compared to the dynamical time scales . 9 . reducing the dust grain abundance by an order of magnitude , or allowing the disk to have a flaring structure , leads to more complex evolution in which the belts break - up into smaller - scale islands of locally coherent horizontal field polarity . we hypothesize that this evolution results from the increased ionization fraction that allows mri channel modes to grow further into the nonlinear regime , where they are prone to break up _ via _ parasitic modes . our global simulations are most comparable to the shearing box simulations of , and so it is worth briefly comparing our results with theirs . our results for weakly magnetized disks ( @xmath163 ) with ambipolar diffusion and ohmic resistivity compare well with their results for disks with no net vertical magnetic field , that is , no sustained wind and very weak turbulence and angular momentum transport . this indicates that there is a clear lower limit for the net vertical magnetic field required to drive an accretion flow that is in agreement with observations ( given our restricted assumptions about the disk microphysics ) . our results are in complete qualitative agreement with for higher magnetic field strength : disks are laminar and a magnetocentrifugal wind is launched from the highly ionized surface layers . the associated wind stress is sufficient to drive an accretion flow that matches typical rates measured to be reaching the surfaces of young stars . unlike earlier studies of magnetocentrifugal wind launching from deeper towards the midplane @xcite , there is no requirement for a strong magnetic field to be present in the disk when the wind is launched from high altitudes , since the requirements for field rigidity are met for globally weak fields in these low density regions . the appearance of long - lived horizontal field belts is not observed in the shearing box simulations of , but this is because of modest differences in the disk models and assumed ionization physics ( their model has a somewhat larger surface density at @xmath5 , and we include a direct component to the disk illumination by stellar x - ray and fuv photons ) . the models agree qualitatively on the appearance of strong current layers , but the stability of these could not be examined by as their models in which the field geometry gave rise to these current layers used a quasi-1d approximation . this latter point illustrates the primary difference between the global and local shearing box simulations . the global simulations appear to produce and sustain the correct physical wind geometry spontaneously , whereas the symmetries associated with the shearing box normally lead to a radial magnetic field that points inwards in one hemisphere and outwards in the other , such that the correct geometry needs to be enforced . the potential role of the rather arbitrary initial conditions , and the vertical boundaries in setting the magnetic field topology in global models should be investigated . our choice to impose a vertical field is motivated by the typical `` hourglass '' morphology ( e.g. figures 2 and 3 in * ? ? ? * and references therein ) emerging also in zoom - in simulations of ppd formation @xcite , and observed in molecular cores @xcite ; even though the topologies of ppds magnetic fields have not yet been unambiguously determined @xcite . the results presented in this paper , combined with those by , indicate that the traditional picture of a ppd hosting a dead zone in the shielded disk interior , with turbulent surface layers driven by the mri , no longer holds . when ambipolar diffusion is treated , the flow becomes laminar , and accretion is driven by the magnetic stresses produced in launching a wind from the disk surface . adding the hall effect appears to further modify the flow by allowing a significant maxwell stress to develop near the disk midplane through the combined action of field winding and hall drift , but does not lead to turbulence . we are then left with a picture of ppds in which the innermost regions sustain vigorous mri - induced turbulence because of collisional ionization of alkali metals , the intermediate regions between @xmath164 are laminar with angular momentum transport occurring through a disk wind and perhaps through maxwell stresses near the midplane ( an effect that only arises if @xmath165 and when including the hall term ) , and an outer region where the hall effect and ohmic resistivity are sub - dominant and weak turbulence arises @xcite . the picture painted above has important implications for dust evolution and planet formation . in principle , dust coagulation and settling in the laminar region should occur rapidly , and in the absence of a source of stirring the loss of small grains should affect the spectral energy distribution ( sed ) . this was examined by , who showed that this rapid evolution of the sed is not in agreement with observations , and for many years this important result has provided indirect evidence for the existence of turbulent stirring in the planet forming regions of protoplanetary disks . in the absence of a magnetic origin for local turbulence this suggests that either another source of turbulence exists , such as the vertical shear instability ( e.g. * ? ? ? * ) , or that small grains are continuously delivered to the laminar region from the outer and/or inner turbulent regions . the results in this paper are consistent with the remanent magnetic field measurements in chondrules by @xcite in that the minimum midplane field strength for a wind driven disk at the presumed chondrule forming location ( the midplane at @xmath166 to @xmath167 ) we find is smaller than the upper bound on the ambient field strength in the chondrule cooling region found in that work . this allows the chondrules studied by @xcite to have formed in the midplane of a wind driven laminar disk , while also allowing for the chondrule formation process to locally concentrate the magnetic field . the influence of the traditional dead - zone picture for protoplanetary disks during various stages in planet formation has been examined in some detail . for example , @xcite showed that the stirring of planetesimals by the propagation of waves from the active region into the dead zone can stir up small planetesimals ( @xmath168 ) and cause them to undergo collisional destruction , or prevent their runaway growth . in recent work ( nelson et al . 2014 , in prep . ) , it has furthermore been shown that low mass planets orbiting in disk models with traditional dead zones are likely to undergo very rapid inward migration because the weak stresses there are unable to prevent saturation of the corotation torque . in principle , the new emerging picture of ppds that is , assuming that maxwell stresses operate near the midplane because of the hall effect can ameliorate these problems by removing the turbulent stirring , and by providing a significant stress that can prevent corotation torque saturation . full simulations are clearly required to examine whether or not the latter effect can be realized during full nonlinear evolution . finally , @xcite examined gap formation and gas accretion onto a giant planet within a disk with a traditional dead zone , and observed that the gas in the gap region became enlivened because of enhanced penetration of x - rays and cosmic rays . this created a turbulent and chaotic gas flow into the planet hill sphere , and the enhanced magnetic coupling also led to the launching of a magnetocentrifugally accelerated jet from the circumplanetary disk . it remains unclear whether or not these effects will arise when ambipolar diffusion is included , given the low density of the gap region . we thank martin pessah , anna tenerani and marco velli for useful advice on plasma instabilities , and the referee for a helpful report . the research was carried out in part at the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration and with the support of the nasa origins of solar systems program via grant 13-oss13 - 0114 . the research leading to these results has received funding from the danish council for independent research ( dff ) and fp7 marie curie actions cofund under the grant - id : dff 1325 - 00111 , and from the people programme ( marie curie actions ) of the european union s seventh framework programme ( fp7/2007 - 2013 ) under rea grant agreement 327995 . this work used the nirvana - iiicode developed by udo ziegler at the leibniz institute for astrophysics ( aip ) . we acknowledge that the results of this research have been partly achieved using the prace-3ip project ( fp7 ri-312763 ) resource ` fionn ` based in ireland at the irish centre for high - end computing ( ichec ) . computations were also performed on the ` astro2 ` node at the danish center for supercomputing ( dcsc ) , and on the u.k . stfc - funded ` dirac ` facility .

the contribution to the nuclear transverse response function arising from two particle - two hole ( 2p-2h ) states excited through the action of electromagnetic meson exchange currents ( mec ) is computed in a fully relativistic framework . the mec considered are those carried by the pion and by degrees of freedom , the latter being viewed as a virtual nucleonic resonance . the calculation is performed in the relativistic fermi gas model in which lorentz covariance can be maintained . all 2p-2h many - body diagrams containing two pionic lines that contribute to are taken into account and the relative impact of the various components of the mec on is addressed . the non - relativistic limit of the mec contributions is also discussed and compared with the relativistic results to explore the role played by relativity in obtaining the 2p-2h nuclear response .

nucl-th0304084

two particle two hole states in nuclei can be excited through the action of two - body electromagnetic ( em ) currents that enter via noether s theorem and the requirement of minimal coupling once an effective lagrangian ( embodying baryonic and mesonic degrees of freedom ) is invoked . the currents reflect the dual role played by mesons in nuclear structure studies : mesons carry both the force which accounts for the binding of nuclei and the currents ( in particular the em current ) which respond to external fields impinging on the nucleus . accordingly , the natural classification into mec and correlation currents follows . in particular , both substantially contribute to the inclusive , inelastic scattering of electrons from nuclei in the excitation energy domain extending through the quasielastic peak ( qep ) to high inelasticity , the region we shall explore in the present paper . a long - standing issue of hadronic modeling is the difficulty of treating currents and interactions consistently , that is , of respecting gauge invariance by fulfilling the continuity equation . the model employed in the present study , the relativistic fermi gas ( rfg ) model , has special advantages over most others : gauge invariance can be respected and fully relativistic modeling , while not easy , can be undertaken . given the focus of modern electron scattering experiments on kinematics where the energies and momenta transferred to the nucleus are large , and thus where relativity is expected to be important , one has strong motivation to explore such models in which fundamental symmetries can be maintained , even at the expense of accepting their obvious dynamical limitations . given the necessity to break the problem down into tractable pieces , in the present work we forgo the issue of gauge invariance and focus on a fully relativistic description of the mec contribution to the em excitation of the 2p-2h states of the rfg . this is already a non - trivial computational problem and thus the issue of how gauge invariance can be maintained ( which is presently also being explored in other work ) is left for future presentation . furthermore , we shall limit our attention in this work to the roles played by the pion and @xmath1 in obtaining the mec contributions specifically to the transverse response @xmath0 . the relevance of the @xmath2 and @xmath1 in @xmath0 for the physics of the quasielastic regime is well - established and , moreover , we wish to compare our results with the only existing fully relativistic computation of the 2p-2h mec contribution to @xmath0 , namely that of @xcite ( quoted as dbt in the following ) . that this is to our knowledge the only previous study of this type is not surprising . indeed , to carry out our project it has been necessary to compute a very large number of terms . specifically , to get the direct contribution ( see below ) we have had to compute about 3000 terms , whereas to get the exchange contribution ( which , as we shall see , comes into play only via the @xmath1-isobar ) it has been necessary to compute more than 100,000 terms . the analytic manipulation of the traces of the relativistic currents has been done using the algebraic computer program form @xcite , and the numerical computation of the individual relativistic contributions requires up to seven - dimensional monte carlo integrations . to make direct comparison between our results and those of dbt we have used exactly the same form factors and @xmath1-width of the latter , although more modern versions of these quantities can straightforwardly be incorporated in future work we will update our predictions by doing so . in order to appreciate the importance of lorentz covariance , we have calculated @xmath0 for the rfg not only fully relativistically , but also in the non - relativistic limit . for the latter a few calculations are available , specifically those carried out in @xcite . as we shall see , in the current work some important differences between the dbt results and our results in the high energy domain are found , notably in the non - relativistic approximations obtained in the two studies . in fig . [ fig : currents ] we display the free - space two - body isovector mec entering in our calculation . using the labelling in the figure and defining the four - momenta @xmath3 and @xmath4 ( the four - momentum carried by the virtual photon is then @xmath5 ) their relativistic expression is . ] @xmath6 for the pion - in - flight current ( diagram _ ( g ) _ ) and @xmath7\ ] ] for the seagull ( or contact ) current ( diagrams _ ( e ) _ and _ ( f ) _ ) . in the above @xmath8 @xmath9 and @xmath10 are the pion and nucleon masses , respectively , @xmath11 is the ( large ) volume enclosing the fermi gas and @xmath12 ( @xmath13 ) the pseudo - vector pion - nucleon coupling constant . the other coupling constants are @xmath14 and @xmath15 ; for simplicity , here we omit the form factors ( but they are included in the calculations of the response ) . the index @xmath16 attached to the vertex operators distinguishes between the two interacting nucleons . the two - body meson exchange currents in free - space . the thick lines in the diagrams _ ( a ) _ to _ ( d ) _ represent the @xmath1 propagation.,height=302 ] the @xmath1 current is derived from the peccei pseudo - vector lagrangian and reads @xmath17 \pi(k_2)_{(2 ) } + ( 1\leftrightarrow2)\right\}.\end{aligned}\ ] ] it corresponds to diagrams _ ( a)-(d ) _ of fig . [ fig : currents ] . in the above equations @xmath18 denotes the isobar mass , @xmath19 , @xmath20 and @xmath21 , @xmath22 yield the strength of the coupling of the @xmath1 to the em and pionic fields , respectively . in eq . ( [ eq : jmud ] ) the following definitions have been introduced : [ eq : jcurr ] @xmath23 and @xmath24 the analogous expressions associated with the diagrams ( c ) and ( d ) are simply obtained through the interchange ( @xmath25 ) as indicated in eq . ( [ eq : jmud ] ) , which also implies @xmath26 and @xmath27 ( @xmath28 , @xmath29 ) . furthermore , the rarita - schwinger ( rs ) @xmath1 propagator , namely @xmath30 is used ( we employ here the metric @xmath31 ) . in the appendix , the explicit expressions of eqs . ( [ eq : jcurr]a - b ) are reported . the above given currents are the basic ingredients required in modeling @xmath0 . since one of our goals is to assess the quality of a given non - relativistic approximation for @xmath0 , we also need the corresponding matrix elements of the currents in that limit . under the assumption that the momenta of the nucleons involved in the response are small in comparison to their rest mass ( which may or may not be valid ) one obtains ) and isospin ( @xmath32 ) quantum numbers . ] @xmath33\end{aligned}\ ] ] and @xmath34 \chi_{\sigma_1}\right ] \tau_2 \frac{\chi^\dagger_{\sigma_2'}\bm{\sigma}\cdot\bm{k}_2\chi_{\sigma_2 } } { \bm{k}_2 ^ 2+\mu_\pi^2 } + ( 1\leftrightarrow 2 ) \right\ } , \nonumber \\ \label{eq : nrdcurr}\end{aligned}\ ] ] where @xmath35 and @xmath36 [ eq : coeffab ] it should be noted , however , that this is not the only possible choice for a non - relativistic @xmath1 current . for instance the one employed in @xcite and @xcite indeed has the same structure as eq . ( [ eq : nrdcurr ] ) , but occurs with coefficients given by the following expressions @xmath37 which are obtained by disregarding the last two terms in the numerator of eq . ( [ eq : coeffab]a - b ) . note that these terms have opposite sign and affect the magnitude of the coefficients @xmath38 and @xmath39 , in opposite direction , by less than 10% ) can be obtained by using the simplified form for the rs @xmath1 propagator , @xmath40 , the other terms giving contributions which are either zero or proportional to @xmath41 and @xmath42 . note that the second term , namely @xmath43 , never contributes to em processes . ] . the non - relativistic @xmath1-current employed in dbt is again given by eq . ( [ eq : nrdcurr ] ) , however with coefficients @xmath44 derived by setting @xmath45 in the numerator of the right hand side of eqs . ( [ eq : coeffab]a - b ) . the 2p-2h response function is stable with respect to these different definitions of the coefficients @xmath38 and @xmath39 ( the changes amounting to a few percent ) . we turn now to the 2p-2h transverse response function , which can be obtained through the hadronic tensor . the latter is defined according to and @xmath46 were arbitrary . from now on , we will indicate with @xmath47 and @xmath46 , respectively , momenta below ( holes ) and above ( particles ) the fermi momentum @xmath48 . the pion momenta have , as before , @xmath49 . ] @xmath50 where @xmath51 represents the fermi sphere , @xmath52 , @xmath53 and the factor 1/4 accounts for the identity of the 2 particles and 2 holes in the final states . being concerned with @xmath0 in the present work , actually we only need the spatial components of @xmath54 ( in fact only the @xmath55 and 2 components in the combination @xmath56 , since the unpolarized transverse response is the focus ) . that is , we have @xmath57 \delta^{(3)}(\bm{q}+\bm{p}_1'+\bm{p}_2'-\bm{p}_1-\bm{p}_2 ) \nonumber \\ & & \times 2 \sum_{\sigma \tau } \sum_{i , j=1}^3 \left(\delta_{ij}-\frac{q_i q_j}{\bm{q}^2}\right ) \left[j_i^\dagger(\bm{p}_1'-\bm{p}_1,\bm{p}_2'-\bm{p}_2 ) j_j(\bm{p}_1'-\bm{p}_1,\bm{p}_2'-\bm{p}_2 ) \nonumber \right . \\ & & \qquad\qquad\qquad\qquad\qquad\left . - j_i^\dagger(\bm{p}_1'-\bm{p}_1,\bm{p}_2'-\bm{p}_2 ) j_j(\bm{p}_1'-\bm{p}_2,\bm{p}_2'-\bm{p}_1)\right].\end{aligned}\ ] ] this expression explicitly displays the direct and exchange contribution stemming from the antisymmetrization of the 2p-2h final states in eq . ( [ eq : htens ] ) . in eq . ( [ eq : rt ] ) the @xmath58 are the matrix elements of the currents in eqs . ( [ eq : jmupi ] ) , ( [ eq : jmus ] ) and ( [ eq : jmud ] ) in momentum space . note that we define the excitation energy as @xmath59 . by eliminating one of the integrations via the @xmath60 , a simpler expression for the transverse response emerges : @xmath61 \sum_{\sigma \tau } \sum_{i , j=1}^3 \frac{v^4}{16 e_{\bm{p}_1'}e_{\bm{p}_2'}e_{\bm{p}_1}e_{\bm{p}_2 } } \left(\delta_{ij}-\frac{q_i q_j}{\bm{q}^2}\right ) \nonumber \\ & & \times \left[j_i^\dagger(\bm{k}_1,\bm{k}_2 ) j_j(\bm{k}_1,\bm{k}_2 ) -j_i^\dagger(\bm{k}_1,\bm{k}_2 ) j_j(\bm{k}_1',\bm{k}_2')\right],\end{aligned}\ ] ] where now @xmath62 . the exchange term in eq . ( [ eq : rtdek ] ) depends on the pionic momenta @xmath63 defined as @xmath64 , @xmath65 . by exploiting the remaining @xmath66-function and with some algebra , the integral defining the transverse response can be reduced to seven dimensions . in previous work where only the direct term and the non - relativistic limit were considered , it turned out to be possible to reduce the integration to a bi - dimensional one ; but now for the relativistic expression , this is no longer possible , even for the direct term . as anticipated in the introduction the traces of the relativistic currents typically generate a huge number of terms , so that it is not practical to report explicitly the expressions used in the numerical calculations of the relativistic responses . however , in the appendix we give the formulae for the direct pion and pion/@xmath1 contributions , the only ones having sizes suitable for publication . numerical integrations have been performed using monte carlo techniques , varying the sample size until a standard deviation better than 1% is reached . the number of configurations required for such an accuracy is @xmath67 . a more stringent test of the accuracy in the numerical calculations can be obtained in the case of the direct non - relativistic contribution , since this can be reduced to a two - dimensional integral @xcite and done through standard quadrature . also here the agreement is at the 1% level . before addressing the issue of the numerical evaluation of the fully relativistic problem , below we report the non - relativistic limit of the integrands in eq . ( [ eq : rtdek ] ) , to be referred to as @xmath68 and @xmath69 , for the direct and exchange contributions , respectively . this will allow us to connect the various pieces contributing to the transverse response to specific feynman diagrams and to ascertain from whence the major contributions arise . while the relativistic expressions for the transverse response function are quite cumbersome , the non - relativistic limit provides a relatively simple and controllable environment in which to check the correctness of the calculations . we also remark that our expression for @xmath70 will turn out to differ somewhat from the one derived in dbt ( no explicit formula is given there for the exchange part ) . the direct @xmath1 contributions to the mec 2p-2h response function.,width=604 ] the direct @xmath1 contributions to the mec 2p-2h response function.,width=604 ] the direct @xmath1 contributions to the mec 2p-2h response function.,width=604 ] for the purely pionic contribution and for the interference between the pionic and @xmath1 current we get , respectively , @xmath71 \right . \nonumber \\ & & \left . \hspace*{-15 mm } - \frac{f_{\pi nn}^3f_{\gamma\pi nn}f_{\gamma\pi\pi}}{\mu_\pi^4 } \left ( \frac{\bm{k}_{1t}^2\bm{k}_1 ^ 2f_{\pi nn}^3(\bm{k}_1 ^ 2 ) f_{\pi nn}(\bm{k}_2 ^ 2)}{(\bm{k}_1 ^ 2+\mu_\pi^2)^2(\bm{k}_2 ^ 2+\mu_\pi^2 ) } + \frac{\bm{k}_{2t}^2\bm{k}_2 ^ 2f_{\pi nn}^3(\bm{k}_2 ^ 2 ) f_{\pi nn}(\bm{k}_1 ^ 2)}{(\bm{k}_1 ^ 2+\mu_\pi^2)(\bm{k}_2 ^ 2+\mu_\pi^2)^2 } \right ) \right\ } \nonumber \\\end{aligned}\ ] ] and @xmath72 \nonumber \\ & = & 32 a \frac{f_{\pi nn}^2f_{\pi n\delta}f_{\gamma n\delta } } { 3\mu_\pi^2 m } f_{\gamma nn}(q^2)f_{\gamma n\delta}(q^2 ) \nonumber \\ & & \times \left\ { \frac{f_{\pi nn}f_{\gamma\pi\pi}}{\mu_\pi^2 } f_{\pi nn}^2(\bm{k}_1 ^ 2)f_{\pi nn}(\bm{k}_2 ^ 2 ) f_{\pi n\delta}(\bm{k}_1 ^ 2 ) \frac{2\bm{q}^2\bm{k}_1 ^ 2\bm{k}_{1t}^2 } { ( \bm{k}_1 ^ 2+\mu_\pi^2)^2(\bm{k}_2 ^ 2+\mu_\pi^2 ) } \right . \nonumber \\ & & + \frac{f_{\gamma\pi nn}}{\mu_\pi^2}\left [ f_{\pi nn}^3(\bm{k}_1 ^ 2)f_{\pi n\delta}(\bm{k}_1 ^ 2 ) \frac{2\bm{k}_1 ^ 2 ( \bm{k}_{1}\!\cdot\!\bm{q } ) } { ( \bm{k}_1 ^ 2+\mu_\pi^2)^2}\right . \nonumber \\ & & \left.\left . + f_{\pi nn}^2(\bm{k}_1 ^ 2)f_{\pi nn}(\bm{k}_2 ^ 2 ) f_{\pi n\delta}(\bm{k}_2 ^ 2 ) \frac{\bm{q}^2\bm{k}_{1t}^2}{(\bm{k}_1 ^ 2+\mu_\pi^2)(\bm{k}_2 ^ 2+\mu_\pi^2 ) } \right ] + ( 1\leftrightarrow2)\right\}. \nonumber \\\end{aligned}\ ] ] from the different coupling constants in eqs . ( [ eq : mecpion ] ) and ( [ eq : mecint ] ) it is easy to identify the contributions arising from the pion - in - flight and seagull currents . the topologically different diagrams are shown in figs . [ fig : mecpion ] and [ fig : mecint ] . finally , for the contribution to @xmath70 of the @xmath1 current alone , one has @xmath73 \right . \nonumber \\ & & \left.\phantom{\frac{\bm{k}_{1t}^2\bm{q}^2}{(\bm{k}_1 ^ 2+\mu_\pi^2 ) } } + ( 1\leftrightarrow2 ) \right\ } , \end{aligned}\ ] ] where the first two terms on the right - hand side correspond to the diagrams _ ( a)(c ) _ of fig . [ fig : mecdelta ] , and the last one to the diagrams _ ( d)(f)_. in this case six distinct diagrams contribute . in eqs . ( [ eq : mecpion ] ) , ( [ eq : mecint ] ) and ( [ eq : mecdelta ] ) @xmath74 and @xmath75 indicate the longitudinal and transverse components of the vector @xmath76 with respect to the direction fixed by @xmath77 . furthermore , in the appropriate places , the hadronic monopole form factors @xmath78 and the em ones @xmath79 have been introduced . in the non - relativistic expressions the hadronic form factors have been taken in the static limit . the cut - offs have been chosen as in dbt , namely @xmath80 mev , @xmath81 mev , @xmath82 gev@xmath83 , @xmath84 and @xmath85 gev@xmath83 . this choice clearly makes it possible a direct comparison between our results for @xmath0 and those of dbt . the exchange @xmath1 contributions to the mec 2p-2h response function.,width=604 ] the exchange @xmath1 contributions to the mec 2p-2h response function.,width=604 ] for completeness , we give also the formulae of the ( smaller ) exchange contributions to the integrand of eq . ( [ eq : rtdek ] ) , @xmath86 , in the non - relativistic limit . the purely pionic contribution is identically zero , as a consequence of charge conservation and of the fact that the photon does not couple to a neutral pion . for the interference between pion and @xmath1 ( fig . [ fig : dpex ] ) we have @xmath87 \nonumber\\ & & = \frac{16 f_{\pi nn}^3f_{\gamma\pi\pi}f_{\gamma n \delta } f_{\pi n\delta}}{3\mu_\pi^4 m } b \bm{q}^2 \left\ { \frac{(\bm{k}_2\times \bm{k}_2')_l^2 } { ( \bm{k}_2 ^ 2+\mu_\pi^2)(\bm{k}_2'^2+\mu_\pi^2 ) } \left[\frac{1}{\bm{k}_1 ^ 2+\mu_\pi^2}+\frac{1}{\bm{k}_1'^2+\mu_\pi^2}\right ] + ( 1\leftrightarrow 2 ) \right\ } \nonumber \\ & & + \frac{8 f_{\pi nn}^3f_{\gamma\pi nn}f_{\gamma n \delta } f_{\pi n\delta}}{3\mu_\pi^4 m } b \left\ { \frac{(\bm{q}\cdot\bm{k}_{2})\bm{k}_2'^2+(\bm{q}\cdot\bm{k}_{2}')\bm{k}_2 ^ 2 -(\bm{q}\cdot\bm{k}_{2}')(\bm{k}_2\cdot\bm{k}_2 ' ) -(\bm{q}\cdot\bm{k}_{2})(\bm{k}_2\cdot\bm{k}_2 ' ) } { ( \bm{k}_2 ^ 2+\mu_\pi^2)(\bm{k}_2'^2+\mu_\pi^2 ) } \right . \nonumber\\ & & \qquad \left . + \frac{(\bm{q}\!\cdot\!\bm{k}_{1})\bm{k}_2'^2- ( \bm{q}\cdot\bm{k}_{2}')(\bm{k}_1\cdot\bm{k}_2 ' ) } { ( \bm{k}_1 ^ 2+\mu_\pi^2)(\bm{k}_2'^2+\mu_\pi^2 ) } + \frac{(\bm{q}\cdot\bm{k}_{1}')\bm{k}_2 ^ 2 -(\bm{q}\cdot\bm{k}_{2})(\bm{k}_1'\cdot\bm{k}_2 ) } { ( \bm{k}_1'^2+\mu_\pi^2)(\bm{k}_2 ^ 2+\mu_\pi^2 ) } + ( 1\leftrightarrow 2)\right\}.\end{aligned}\ ] ] the contribution of the @xmath1 alone ( fig . [ fig : ddex ] ) is instead @xmath88\right . \nonumber \\ & & \left . \qquad\qquad + ab \left[\frac{2(\bm{k}_1\times\bm{k}_1')_l^2 -2 k_{1l}k_{1l}'(\bm{k}_1\cdot\bm{k}_1 ' ) + k_{1l}'^2\bm{k}_1 ^ 2+k_{1l}^2\bm{k}_1'^2 } { ( \bm{k}_1 ^ 2+\mu_\pi^2)(\bm{k}_1'^2+\mu_\pi^2 ) } \right.\right . \nonumber \\ & & \left . \qquad\qquad \frac{2(\bm{k}_1\times\bm{k}_2')_l^2 -2 k_{1l}{k}_{2l}'(\bm{k}_1\cdot\bm{k}_2 ' ) + k_{2l}'^2\bm{k}_1 ^ 2+k_{1l}^2\bm{k}_2'^2 } { ( \bm{k}_1 ^ 2+\mu_\pi^2)(\bm{k}_2'^2+\mu_\pi^2)}+ ( 1\leftrightarrow 2 ) \right ] \right\}.\end{aligned}\ ] ] equations ( [ eq : mecpion ] ) , ( [ eq : mecint ] ) and ( [ eq : mecdelta ] ) could in principle be compared with eq . ( 5.11 ) of dbt ; however , the overall normalization of the latter is not correct , since its dimension is not consistent with its definition ( namely of being the transverse part of the amplitude @xmath89 given in eq . ( 4.8 ) of dbt ) ; moreover , the relative weights of the interference and @xmath1 contributions with respect to the pionic one differ , in our calculations , by a factor 2 and 4 , respectively , from those of eq . ( 5.11 ) of dbt . these factors , however , are not able to explain the marked difference between our results and those in that paper . note that although the authors of dbt write down exactly the same expressions as we do for the non - relativistic mec currents , actually they state that the non - relativistic procedure to get their eq . ( 5.11 ) is applied at the level of the hadronic tensor , that is by reducing the ( cumbersome ) exact relativistic response . in fig . [ fig : rtnr ] we now compare our results with those of dbt , where the non - relativistic @xmath0 ( without the exchange contribution ) is shown for @xmath90 mev / c ( left ) and for @xmath91 mev / c ( right ) , with an atomic mass number of 56 and utilizing a fermi momentum @xmath92 @xmath93 . the latter value is employed for the sake of comparison with dbt , although in fact it is more appropriate for heavier nuclei . the non - relativistic transverse response function @xmath94 according to the present calculation ( solid ) and to the one of dbt ( dashed ) for @xmath90 mev / c and @xmath91 mev / c ; in both instances @xmath92 @xmath93 and @xmath95 mev . only the direct contribution is shown.,height=188 ] it is clearly apparent in the figure that our predictions differ significantly from those of dbt : while the discrepancy is mild for moderate values of @xmath96 ( roughly , those encompassing the qep ) , it becomes striking at higher energies , namely in the region of the so - called dip and of the @xmath1-peak . here our transverse response function in the proximity of the lightcone turns out to be larger by about a factor two at @xmath90 mev / c and by over a factor three at @xmath91 mev / c . note that , in order to conform as closely as possible with the dbt approach , we have accounted for the initial state binding of the two holes by phenomenologically inserting a 70 mev energy shift @xmath97 in the energy conserving @xmath66-function appearing in the response function by @xmath98 . ] . in a non relativistic framework this binding energy merely produces a shift of @xmath0 toward higher @xmath96 values . in the same spirit and to comply with gauge invariance , we have added ( as in dbt ) to the pion - in - flight current in eq . ( [ eq : piinflight ] ) two terms , to be viewed as the coupling of the virtual photon to a fictitious particle of mass equal to the @xmath99 cutoff , @xmath100 , in spite of the very minor role these terms play in shaping the transverse response . furthermore , it is worth also remarking that our present calculations agree with the previous findings of @xcite . on the other hand , the computation of dbt _ does not _ agree with a previous one presented by the same authors in @xcite , notwithstanding that both of them employ the same set of values for the parameters and the kinematic conditions . indeed , the former is lower than the latter by about 25% . in the light of these new results , we anticipate contributions from mec effects in the region beyond the qep of a magnitude not previously foreseen in most previous work . because the studies of @xcite were applied to moderate momenta and to a restricted range of transferred energies ( not exceeding 300 mev , in fact ) owing to their non - relativistic nature , it is only in the studies of dbt that suggestions for such large effects can be found . the relativistic transverse response function @xmath94 at @xmath90 mev / c and @xmath91 mev / c calculated with @xmath95 mev ( solid ) and with @xmath101 ( dot - dashed ) . only the direct contribution is shown . the non - relativistic results are also displayed in order to shed light on the role of relativity in the response ( dotted ) . for the sake of comparison the relativistic results obtained in dbt are displayed ( dashed ) . in all instances @xmath92 @xmath93.,height=188 ] in the next section we shall explore whether or not the fully relativistic treatment and the inclusion of the exchange terms modify this finding . in fig . [ fig : rtr ] we display our relativistic results both without and with the energy shift @xmath97 , for the same conditions as those of fig . [ fig : rtnr ] . the exchange contribution is neglected here as well as in all the other figures of this subsection and considered separately below . in fig . [ fig : rtr ] we also compare our relativistic predictions both with those of dbt and with our non - relativistic results . from the figure we observe the following . \a ) the fully relativistic calculation differs considerably under some circumstances from the fully non - relativistic approximation ( i. e. , the solid and dotted curves in fig . [ fig : rtr ] , respectively ) . specifically relativity implies an increase of @xmath0 over the non - relativistic results in the region up to the @xmath1 peak . specifically , the overall effects are small in the domain of the qep , are modest in the dip region and are substantial in the region of the @xmath1-peak . in the region beyond the @xmath1 peak , relativity instead yields a substantial reduction of the response with respect to the non - relativistic predictions . of course , a hybrid approach can also be adopted in which spinor matrix elements and kinematics are kept non - relativistic , but the dynamic @xmath1 propagator is used this is discussed below . \b ) the present relativistic and dbt relativistic calculations are in essential agreement at @xmath90 mev / c and up to the @xmath1 peak for @xmath91 mev / c . \c ) in the dip following the @xmath1 peak and above it up to the light cone , our transverse response is larger by about a factor two - to - three with respect to the dbt results . \d ) at variance with the non - relativistic case , the binding energy not only shifts the relativistic @xmath0 toward higher energies , but also _ increases _ it . \e ) roughly speaking two effects are especially important in determining the characteristic behavior seen in fig . [ fig : rtr ] . one is the peaking produced by the dynamic @xmath1 propagation and the second is the basic phase space available to the two - nucleon ejection process . for instance , the strong rise of @xmath0 close to the light - cone at @xmath91 mev / c comes from the growth of the available phase space . we have checked this by setting all of the current matrix elements entering in eq . ( [ eq : rtdek ] ) to unity and finding qualitatively the same behavior as in the full calculation . to gain deeper insight into the transverse response , we display its individual contributions in fig . [ fig : rtnrsep ] ( for the non - relativistic case ) and in fig . [ fig : rtrsep ] ( for the relativistic one ) . in both instances the contribution of the @xmath1 to @xmath0 is seen to be overwhelming ( although this @xmath1-dominance is mildly reduced by relativity ) , the more so at the larger momentum transfer . this feature strongly suggests the need for an appropriate treatment of the @xmath1 degrees of freedom in obtaining em responses over the whole kinematical regime under study . in particular the dynamical propagation of the latter appears to be essential , as discussed below . this in turn implies that a realistic accounting of the width of the @xmath1 should be undertaken : here , again for ease of comparison with dbt , we have treated this problem as they did . separate contributions to the transverse response function @xmath94 in the non - relativistic limit at @xmath90 mev / c and @xmath91 mev / c : pionic ( dotted ) , pionic-@xmath1 interference ( dash - dotted ) , @xmath1 ( dashed ) and total ( solid ) ; @xmath92 @xmath93 . the exchange contribution is disregarded here.,height=188 ] as in fig . , but in the relativistic case.,height=188 ] specifically , we have included the width through the replacement @xmath102 in the energy denominator of eq . ( [ eq : deltaprop ] ) , _ but not _ in the spinor factor . as far as @xmath103 is concerned , @xmath104 being the square of the energy of the decaying @xmath1 in its rest frame , we have adopted the same expression as in dbt . to assess the crucial role played in @xmath0 by the @xmath1 propagator more fully , in fig . [ fig : rtrprop ] we display the relativistic transverse response ( hence computed with the dynamical @xmath1 propagator ) together with those obtained by neglecting the frequency dependence of @xmath105 ( the width of the @xmath1 being neglected as well ) in the two possible choices of a static ( @xmath106 ) and of a constant @xmath1 propagator . the extreme sensitivity of @xmath0 to @xmath105 is clearly shown by the results in the figure . one can understand even better the importance of treating correctly the @xmath1 propagation by comparing the full relativistic results with a calculation employing the non - relativistic @xmath107 and @xmath108 vertices , but retaining the relativistic , dynamical energy dependence in the denominator of the @xmath1 propagator , @xmath109 . in other words , one can invoke a hybrid model in which everything is kept non - relativistic , but where the dynamic dependence in the @xmath1 propagator is retained . the results are mixed . if one wishes to study only the region extending through the peak seen in fig . [ fig : rtrprop ] out to the light - cone _ and _ wishes only to work at relatively small momentum transfers such as those in the figure , then this hybrid approach is reasonably good , incurring errors typically of @xmath110 . however , this is not a general statement . first , if one addresses higher @xmath111-values , much larger relativistic effects are apparent and the hybrid approach is not very successful . for instance , at @xmath112(@xmath113 ) gev / c the typical error made at the peak is @xmath114 . secondly , if one focuses on the scaling region as we shall in the follow - up study that is presently in progress then even at relatively low values of @xmath111 the effects of relativity ( i. e. , effects other than the nature of the @xmath1 propagator ) are large . for instance , even at @xmath91 mev / c one finds effects exceeding 100% in the scaling region . this is not apparent in the figures presented here , since the cross sections are so small in that region . however , this does not mean that the scaling region is irrelevant quite the opposite , the total cross section is also very small ( but measurable ) and the 2p-2h mec effects are far from negligible . the details will be presented in the near future in a separate paper . the relativistic transverse response function @xmath94 at @xmath90 mev / c and @xmath91 mev / c computed with various versions of the @xmath1 propagator to illustrate the sensitivity to the latter : exact propagator ( solid ) , static propagator ( dashed ) and constant propagator ( dotted ) ( see text for the related definitions ) . in all instances @xmath95 mev and @xmath92 @xmath93.,height=188 ] next let us comment on several general features that characterize the response in the 2p-2h sector in contradistinction with the 1p-1h case . in particular , the clean theoretical separation between the 1p-1h and the 2p-2h contributions holds only in the pure rfg framework . here , moreover , the 1p-1h response is strictly confined to the well - known kinematical domain . in finite nuclei , or even in a correlated infinite fermi system , neither of the above statements is valid any longer . we stick however to the pure rfg because this is the only model that allows us to maintain the fundamental principles of lorentz covariance , gauge invariance and translational invariance . the 2p-2h response function resides on the entire spacelike domain of the @xmath115-plane , in contrast to the 1p-1h case , which , as mentioned above , is constrained to a limited region for recent fully relativistic studies of the 1p-1h response see @xcite . since momentum conservation involves only the total momentum there is no limitation on the momentum of a particle taking part in the 2p-2h excitations and hence the energy of the latter will vary over a broad range extending from @xmath116 to the light - cone . in summary , naively the 2p-2h response function @xmath117 might not be expected to display any particular structure , but rather to appear as a broad background . as a matter of fact , as we have seen , @xmath117 indeed displays structure , but related not to the nuclear spectrum involved but to the role played by the @xmath1 resonance . furthermore , we have found that it grows with @xmath96 , reflecting the associated growth of the phase space . indeed the density per unit energy of the 2p-2h excitations goes as @xmath118 . considering the purely pionic mec currents , it is well known @xcite that the 1p-1h excited states of the rfg are reached only through the _ exchange _ term of the related mec diagrams . in contrast the 2p-2h states are reached only through the _ direct _ term . the origin of these kinds of `` selection rules '' stems , in the first case , from the spin - isospin saturation of the rfg , in the second from the impossibility of fulfilling charge conservation in the exchange contribution for the diagrams displayed in fig . [ fig : mecpion ] . the @xmath1-current , however , eludes the above constraints . it provides a direct contribution in the 1p-1h sector of the rfg spectrum , as well as an exchange contribution in the 2p-2h sector , both alone and through the interference with the pionic currents . the transverse response function @xmath94 at @xmath90 mev / c and @xmath91 mev / c including the exchange contributions : non - relativistic direct ( positive dotted ) , non - relativistic exchange ( negative dotted ) , non - relativistic total ( light solid ) , relativistic direct ( positive dashed ) , relativistic exchange ( negative dashed ) and relativistic total ( heavy solid ) . in all instances @xmath95 mev and @xmath92 @xmath93.,height=188 ] the exchange contributions are included together with the ( larger ) direct ones in fig . [ fig : rtex ] . two features are immediately apparent from the figure . first , the impact of relativity on the exchange terms is as substantial as it is for the direct ones . moreover , as for the direct case , the exchange response is first enhanced in the region of the @xmath1 and then dampened by the relativistic effects as one approaches the light cone . second , the relative magnitude of the exchange terms is seen to increase both with the energy transfer and with the momentum transfer . the fraction of the total relativistic response due to exchange goes from roughly 4% at low @xmath96 to 29% at the light cone for @xmath90 mev / c and from 7% to 39% for @xmath91 mev / c . thus , the exchange contribution , while not the dominant one in the kinematical situations we have explored , appears to us not only far from negligible , but also potentially interesting in connection with scaling ( see below ) . finally , as for the direct case , note that the present calculation of the non - relativistic exchange contribution agrees with the calculation at lower momenta reported in @xcite , whereas does not agree with that of dbt . the relativistic response , on the other hand , roughly agrees with dbt . in this paper we have carried out a fully relativistic calculation of the mec contribution of 2p-2h excitations to the transverse inclusive response within the context of the rfg . these excitations are reached through the action of the mec including the part associated with the @xmath1 degrees of freedom . we have compared our results with those of dbt , the only existing relativistic study before the present one and we have pointed out similarities and differences between the two studies . to gauge the difficulty of this project it might help to stress again , as anticipated in the introduction , that the number of terms we have computed to get @xmath0 amounts to over @xmath119 . several new aspects contained in the present study bear highlighting . in particular , while the fully relativistic results obtained here roughly agree with those reported by dbt , some substantial differences occur , especially as the lightcone is approached . furthermore , our non - relativistic results are completely in accord with our own previous work ; however , they are not in agreement with dbt in many circumstances . indeed , the strongest disagreements occur at high inelasticity and suggest to us that one should exercise extreme caution when invoking traditional non - relativistic approximations in such a regime . going beyond the extreme non - relativistic approximation @xcite , one can invoke various classes of hybrid modeling where some , but not all , of the features of the fully relativistic model are retained @xcite . we have pursued such an approach in which the dynamic @xmath1 propagator is kept , but otherwise everything is taken to be non - relativistic . the results are interesting : at modest @xmath111 ( say 1 gev / c or less ) and at high @xmath96 the hybrid approach is quite good , incurring only @xmath110 error . however , at higher @xmath111-values , even in the high @xmath96 region , and at low @xmath96 , even for modest @xmath111 , the hybrid approach is not very successful and relativistic effects are important . the high @xmath96 region , including the peak produced by the dynamic @xmath1 propagator and the rise seen as the light - cone is approached when @xmath111 is large , might pose some problems for high - energy photoreactions . the latter are related in the familiar way ( see , for example , @xcite ) to the inclusive electron scattering transverse response by @xmath120 @xmath121 being the fine - structure constant . in context , it is important to note that , while one can not go from these results directly to studies of ( e , enn ) or ( @xmath122,nn ) reactions ( see , e. g. , refs . @xcite ) , since the rfg is not well - suited to modeling such semi - inclusive processes , the understanding gained in the present work concerning relativistic effects , @xmath1 propagator effects , etc . , likely can have an impact , for instance via improved descriptions of current operators for use in other approaches . the @xmath1 propagator provides just one example : it produces a prominent peak in the 2p-2h transverse response and apparently is an important contribution in the total cross section . it is also very dependent on the specifics assumed , such as what model is taken for the propagator , how medium effects are or are not incorporated , how off - shellness is handled , etc . in particular , in view of continuing discussions about the off - shell behavior of the @xmath1 inside the nuclear medium @xcite and of our findings in this work , the 2p-2h transverse response may provide an ideal place in which to explore such issues . clearly , this is not the intent of the present work . here we wish only to make contact with the work of dbt , using exactly their ingredients and only in future work will we return to address some of these interesting issues . in particular , in the near future our intent is to address the difficult issue of the gauge invariance as was explored in the 1p-1h sector in @xcite . this requires the introduction of 2p-2h correlation currents in addition to the mec . only when this task is accomplished will a consistent ( namely , gauge invariant ) , lorentz covariant and translational invariant description of the inclusive responses ( both longitudinal and transverse ) within the rfg framework be achieved . furthermore , we are now in a position to study the degree to which these responses fulfill or violate @xmath123-scaling of first type ( namely in the momentum transfer @xmath111 ) and of second type ( namely in @xmath48 ) see @xcite . an exploration of these issues is currently being undertaken for the 2p-2h mec and will be presented in a forthcoming paper , which will also focus on the important theme of the interplay between the scaling violations arising from the mec action in both 1p-1h and 2p-2h sectors . this work has been supported in part by the infn - mit bruno rossi exchange program and by murst and in part ( twd ) by funds provided by the u.s . department of energy under cooperative research agreement no . de - dc02 - 94er40818 . next , in the following formulae , we quote explicitly the integrand entering in the direct contributions to the transverse response , limiting ourselves to the full pionic contribution and to the pionic-@xmath1 interference , since the pure @xmath1 contribution corresponds to a far too lengthy expression to be reported here . we use the notation @xmath129 . the hadronic form factors are not explicitly displayed . pionic contribution : @xmath130 \right . \nonumber \\ & & \left . - { f_{\pi nn}^3 f_{\gamma\pi nn}f_{\gamma\pi\pi } } \left [ \frac{4\bm{k}_{1t}^2 \ , k_1 ^ 2}{(k_1 ^ 2-\mu_\pi^2)^2(k_2 ^ 2-\mu_\pi^2 ) } + ( 1\leftrightarrow2 ) \right ] \right\ } ; \label{eq : rpdir}\end{aligned}\ ] ] @xmath1-pion interference contribution : @xmath131 \nonumber \\ & = & \frac{16m^2 f_{\pi n\delta}f_{\gamma n\delta}}{3\mu_\pi^4{\cal e } } \left\ { f_{\pi nn}^3 f_{\gamma\pi\pi } \left [ \frac{-k_2 ^ 2\sum\limits_m\bigl[k_1^m \bigl({\mathcal{f}}_{a}^m(p_1,p_1';q)- { \mathcal{f}}_{b}^m(p_1,p_1';q)\bigr ) \bigr ] } { ( k_1 ^ 2-\mu_\pi^2)(k_2 ^ 2-\mu_\pi^2)^2 } + ( 1\leftrightarrow 2 ) \right ] \right . \nonumber\\ & + & \left . f_{\pi nn}^2f_{\gamma\pi nn } \left [ \vphantom{\frac{2m\sum\limits_m k_{2}^m \bigl(\bigr)}{(k_1 ^ 2-\mu_\pi^2 ) } } \frac{k_2 ^ 2 \bigl ( { \mathcal{s}}_{a}(p_1,p_1';q ) + { \mathcal{s}}_{b}(p_1,p_1';q ) \bigr ) } { 4m(k_2 ^ 2-\mu_\pi^2)^2 } - \frac { \sum\limits_m k_{2}^m \bigl ( { \mathcal{f}}_{a}^m(p_1,p_1';q ) + { \mathcal{f}}_{b}^m(p_1,p_1';q)\bigr)}{2(k_1 ^ 2-\mu_\pi^2)(k_2 ^ 2-\mu_\pi^2 ) } \right.\right . \nonumber \\ & & \left.\left . \hspace*{12 cm } + ( 1\leftrightarrow 2 ) \vphantom{\frac{\sum\limits_m k_{2}^m \bigl({\mathcal{f}}_{a}^m(p_1,p_1';q)\bigr)}{2(k_1 ^ 2-\mu_\pi^2 ) } } \right ] \right\ } ; \label{eq : rdpdir}\end{aligned}\ ] ] pure @xmath1 contribution : @xmath132\right ) + ( 1\leftrightarrow 2)\right ] \nonumber \\ & & \left.-\frac{{\mathcal{f}}_{a}^m(p_1,p_1';q){\mathcal{f}}_{a}^m(p_2,p_2';q ) + { \mathcal{f}}_{b}^m(p_1,p_1';q){\mathcal{f}}_{b}^m(p_2,p_2';q ) } { 3(k_1 ^ 2-\mu_\pi^2)(k_2 ^ 2-\mu_\pi^2)}\right\}.\end{aligned}\ ] ] in the previous equations the following definitions have been introduced : @xmath133 \bigl [ p_\delta^2 - 4(k_\pi\cdot p_\delta ) \bigr ] \right . \nonumber \\ & & + \frac{4m^2}{3m_\delta^2 } \left\ { 4(k_\pi\cdot p_\delta ) \bigl [ p^m(p'\cdot q)-p'^m ( p\cdot q ) + 2p'^m q^2 \bigr ] -q^2 p_\delta^2 ( p^m+p'^m)\right\ } \nonumber \\ & & + \frac{8m}{3m_\delta } \left\ { k_\pi^m \bigl [ 4(k_\pi\cdot p_\delta)(q\cdot p_\delta ) -p_\delta^2 q^2 + 2(q\cdot p_\delta)^2 \bigr ] \right.\nonumber \\ & & \left . \hspace*{15 mm } -2p^m\bigl[q^2(k_\pi\cdot p_\delta)+(k_\pi\cdot q)(q\cdotp_\delta ) + 2 ( k_\pi\cdot q ) ( k_\pi\cdot p_\delta ) \bigr]\right\ } \nonumber \\ & & + 4 \left\ { q^2\bigl[p'^m ( q\cdot p)-p^m(q\cdot p')\bigr ] + 2 ( p\cdot p')\bigl[p^m q^2 + 2 p^m ( q\cdot p')-2p'^m ( q\cdot p_\delta)\bigr ] \right\ } \nonumber \\ & & \left . + 4m^2 \left\ { 4 p'^m \bigl[(q\cdot p_\delta ) -p^m(p'\cdot q)\bigr ] -q^2(p'^m+p^m ) \right\ } + 8 m m_\delta k_\pi^m q^2 \vphantom{\begin{array}{c}a \\ a \end{array } } \right\ } , \\ \label{eq : rdbpfdir } { \mathcal{f}}_{b}^m(p , p';q ) & \equiv & \bm{tr}\left\ { ( \rlap/{p}+m)(\rlap/{p}'-m ) j_{(b)}^m(p'+q , -q - p'+p , q ) \right\ } \nonumber \\ & = & \bm{tr}\left\ { ( \rlap/{p}'-m)(\rlap/{p}+m ) j_{(a)}^m(p'+q , -q - p'+p , q ) \right\ } = { \mathcal{f}}_{a}^m(p',p ;- q).\end{aligned}\ ] ] @xmath134 + p_\delta^2 \bigl[q^2(k_\pi\cdot p ) -(q\cdot k_\pi)(p'\cdot p_\delta ) \right.\right.\bigr . \nonumber \\ & & \left.\bigl . -3(q\cdot p')(k_\pi\cdot p_\delta ) -(q\cdot p_\delta)(k_\pi\cdot p ' ) + \bm{k}_{\pi t}\cdot\bm{p}_t \bigl(q^2+(q\cdot p')\bigr ) \right . \nonumber \\ & & \left . - \bm{p}_t\cdot\bm{p}_t'(q\cdot k_\pi)\bigr ] \right\ } + \frac{8m^3}{3m_\delta^2 } \bigl\{8(q\cdot p_\delta)(k_\pi\cdot p_\delta ) -p_\delta^2 ( q\cdot k_\pi)\bigr\ } \nonumber \\ & & + \frac{16q^2}{3m_\delta } \bigl\{(q\cdot p')(k_\pi\cdot p ) - ( q\cdot p)(k_\pi\cdot p ' ) - ( q\cdot k_\pi)(p\cdot p ' ) -4 ( p_\delta\cdot k_\pi)(p\cdot p ' ) \bigr . \nonumber \\ & & \bigl . + \bm{p}_t\cdot\bm{p}_t ' \bigl[p_\delta^2-(p_\delta\cdot k_\pi)\bigr ] \bigr\ } + \frac{16m^2}{3m_\delta } \bigl\ { ( q\cdot k_\pi)\bigl[(p\cdot p')+p_\delta^2\bigr ] -(q\cdot p)(k_\pi\cdot p ' ) \bigr . \nonumber \\ & & \left . + 2q^2(k_\pi\cdot p ' ) + 3(q\cdot p')(k_\pi\cdot p ) -4(q\cdot p')(k_\pi\cdot q ) + 2(q\cdot p_\delta)(k_\pi\cdot p_\delta ) \right.\nonumber \\ & & \left . -\bm{p}_t^2\bigl(q^2+(q\cdot p')\bigr)+ \bm{p}_t\cdot\bm{p}_t'(q\cdot p_\delta)\bigr\ } + 8 m \bigl\ { q^2 ( k_\pi\cdot p ) -(q\cdot k_\pi)(p'\cdot p_\delta ) \right.\bigr . \nonumber \\ & & \bigl.\left . + ( q\cdot p')(k_\pi\cdot p_\delta ) - ( q\cdot p_\delta ) ( k_\pi\cdot p ' ) + \bm{k}_{\pi t}\cdot\bm{p}_t(q\cdot p ' ) -2\bm{k}_{\pi t}\cdot\bm{p}_t'(q\cdot p_\delta ) \right.\bigr . \nonumber \\ & & \bigl.\left . + \bm{p}_t^2\bigl[q^2 + 2(q\cdot p')\bigr ] -\bm{p}_t\cdot\bm{p}_t'\bigl[(q\cdot p)+(q\cdot p')\bigr ] \bigr\}-8m^3\bigl\{(q\cdot k_\pi)\bigr\ } \right.\nonumber \\ & & \left . + 16m_\delta \bigl\{(p\cdot p')\bigl[q^2 + 2(q\cdot p')\bigr ] + \bm{p}_t\cdot\bm{p}_t'\bigl[q^2+(q\cdot p')\bigr ] -\bm{p}_t'^2(q\cdot p)\bigr\ } \right . \nonumber \\ & & \left . + 16m_\delta m^2\bigl\{q^2 - 2(q\cdot p)\bigr\ } \vphantom{\begin{array}{c}a \\ a \end{array } } \right\ } \frac{1}{p_\delta^2-m_\delta^2 } \nonumber \\ \label{eq : rdasdir}\end{aligned}\ ] ] in eqs . ( [ eq : rdapfdir ] ) and ( [ eq : rdasdir ] ) we used @xmath136 and @xmath137 to simplify the expressions . eqs . ( [ eq : rdbpfdir ] ) and ( [ eq : rdbsdir ] ) are consequences of the identities : @xmath138 and eq . ( [ eq : jmub ] ) . j. ryckebusch , phys . c * 64 * ( 2001 ) 044606 . l. machenil , m. vanderhaeghen , j. ryckebusch , and m. waroquier , phys . b * 316 * ( 1993 ) 17 . s. boffi and m.m . giannini , nucl . phys . a * 533 * ( 1991 ) 441 .

we study relationships between the colored jones polynomial and the a - polynomial of a knot . the aj conjecture ( of garoufalidis ) that relates the colored jones polynomial and the a - polynomial is established for a large class of two - bridge knots , including all twist knots . we formulate a weaker conjecture and prove that it holds for all two - bridge knots . along the way we also calculate the kauffman bracket skein module of the complements of two - bridge knots . some properties of the colored jones polynomial are established .

math0407521

the jones polynomial was discovered by jones in 1984 @xcite and has made a revolution in knot theory . despite many efforts little is known about the relationship between the jones polynomial and classical topology invariants like the fundamental group . the @xmath0-polynomial of a knot , introduced in @xcite , describes more or less the representation space of the knot group into @xmath1 , and has been fundamental in geometric topology . in the present paper we study relationships between the jones polynomial and the a - polynomial . one main goal of the paper is to establish for a large class of two - bridge knots the aj conjecture ( made by garoufalidis ) that relates the colored jones polynomial and the a - polynomial . this class of knots contains for example all twist knots , and much more . another main result is the calculation of the kauffman bracket skein module of all two - bridge knots . this generalizes the work @xcite where the calculations were carried out for @xmath2-torus knots and twists knots , a special class of two - bridge knots . our method is more geometric and we hope that it can be generalized to all knots . in a previous paper @xcite garoufalidis and the author proved that for every knot , the colored jones polynomial satisfies a recurrence relation . the aj conjecture states that when reducing the quantum parameter to 1 , the recurrence polynomial is essentially equal to the @xmath0-polynomial ( for details see below ) . the present paper is independent of @xcite , since we will prove the existence of recurrence relations for two - bridge knots in another way . we also formulate a weaker version of the aj conjecture ( see conjecture [ weakconjecture ] ) that we believe reflects more accurately the algebra / topology relations between the jones polynomial and the a - polynomial . we prove that the weaker conjecture holds true for all two - bridge knots . some properties of the colored jones polynomial are established . we also show that for an arbitrary alternating knot , the degree of the recurrence polynomial must be at least 2 . for a knot @xmath3 in the 3-space @xmath4 the colored jones function ( see for example @xcite ) @xmath5\ ] ] is defined for integers @xmath6 ; its value @xmath7 is known as the colored jones polynomial of the knot @xmath3 with color @xmath8 . we will recall the definition of @xmath7 in section [ jonescolor ] . in our joint work in s. garoufalidis @xcite we showed that the function @xmath9 always satisfies a non - trivial recurrence relation as described in the next subsection . partial results were obtained earlier by frohman , gelca , and lofaro through their theory of non - commutative a - ideal @xcite , which also plays an important role in the present paper . consider a function with domain the set of integers , @xmath11 , and define the linear operators @xmath12 and @xmath13 acting on such functions by : @xmath14 it is easy to see that @xmath15 , and that @xmath16 generate the _ quantum torus _ @xmath17 , a non - commutative ring with presentation @xmath18 we also use the notation @xmath19 for the subring of @xmath17 which consists only of polynomials with non - negative powers of @xmath13 and @xmath12 . traditionally @xmath19 is called the _ quantum plane_. the _ recurrence ideal _ of the discrete function @xmath20 is the left ideal @xmath21 in @xmath17 that annihilates @xmath20 : @xmath22 we say that @xmath20 is @xmath10-_holonomic _ , or @xmath20 satisfies a linear recurrence relation , if @xmath23 . in @xcite we proved that for every knot @xmath3 , the function @xmath9 is @xmath10-holonomic . denote by @xmath24 the recurrence ideal of @xmath9 . for the right - handed trefoil , one has @xmath25 the function @xmath9 satisfies @xmath26 , where @xmath27 together with the initial conditions @xmath28 , this recurrence relation determines @xmath7 uniquely . the quantum torus @xmath17 is not a principal ideal domain , and @xmath24 might not be generated by a single element . garoufalidis @xcite noticed that by adding to @xmath17 all the inverses of polynomials in @xmath13 one gets a principal ideal domain @xmath29 , and hence from the ideal @xmath24 one can define a polynomial invariant . formally one can proceed as follows . let @xmath30 be the fractional field of the polynomial ring @xmath31 $ ] . let @xmath32 be the set of all laurent polynomials in the variable @xmath12 with coefficients in @xmath30 : @xmath33 and define the product in @xmath32 by @xmath34 then it is known that every left ideal in @xmath29 is principal , and @xmath17 embeds as a subring of @xmath32 . the extension @xmath35 of @xmath24 in @xmath32 is then generated by a single polynomial @xmath36 where the degree in @xmath12 is assumed to be minimal and all the coefficients @xmath37 $ ] are assumed to be co - prime . that @xmath38 can be chosen to have integer coefficients follows from the fact that @xmath39 $ ] . it is clear that @xmath40 annihilates @xmath9 , and hence it is in the recurrence ideal @xmath41 . note that @xmath40 is defined up to a factor @xmath42 . we will call @xmath38 the _ recurrence polynomial _ of @xmath3 . for example , the polynomial @xmath43 in the previous subsection is the recurrence polynomial of the right - handed trefoil . if @xmath43 is a polynomial in @xmath44 and @xmath13 ( no @xmath12 ) , and @xmath45 then @xmath46 . hence adding all the inverses of polynomials in @xmath13 does not affect the recurrence relations . let @xmath47 be the map reducing @xmath48 . formally , if @xmath49 is an @xmath50-module , then let @xmath51 , where @xmath52 is considered as an @xmath50-module by setting @xmath48 . also if @xmath53 then @xmath54 is the image of @xmath55 in @xmath56 . thus @xmath57 is the polynomial obtained from @xmath40 by putting @xmath48 . for example , when @xmath3 is the right - handed trefoil , @xmath58 . for non - zero @xmath59 $ ] , we say that @xmath20 is _ @xmath13-essentially equal to _ @xmath60 , and write @xmath61 if the quotient @xmath62 does not depend on @xmath12 . we say that two algebraic subsets of @xmath63 with parameters @xmath64 are _ @xmath13-essentially equal _ if they are the same up to adding some lines parallel to the @xmath12-axis . it s clear that if @xmath20 is @xmath13-essentially equal to @xmath60 , then @xmath65 and @xmath66 are @xmath13-essentially equal . here @xmath65 is the algebraic set of zero points of @xmath20 . let @xmath67 $ ] be the @xmath0-polynomial of @xmath3 ( see @xcite ) ; we will review its definition in section [ apolynomial ] . garoufalidis @xcite made the following conjecture . ( * the aj conjecture * ) the polynomials @xmath57 and @xmath68 are @xmath13-essentially equal . actually , this is the strong version . the weak version of the conjecture says that @xmath69 and @xmath70 are @xmath13-essentially equal . the algebraic set @xmath70 is known as the _ deformation variety _ of the knot group , with the component @xmath71 corresponding to abelian representations of the knot group into @xmath72 , and @xmath73 to non - abelian ones . garoufalidis @xcite verified the conjecture for the trefoil and the figure 8 knot . takata @xcite gave some evidence to support the conjecture for twist knots , but did not prove it . both works are based heavily on the computer programs of wilf and zeilberger . hikami @xcite verifies the conjecture for torus knots . in all these works direct calculations with explicit formulas are used . in the present paper we prove the conjecture for a large class of two - bridge knots , using a more conceptual approach . two - bridge knots @xmath74 are parametrized by a pair of odd positive integers @xmath75 , with @xmath76 if @xmath77 ( see @xcite and section [ twobridge ] below ) . suppose @xmath78 is a two - bridge knot . \a ) the recurrence polynomial @xmath38 has @xmath12-degree less than or equal to @xmath79 . \b ) the algebraic set @xmath69 is @xmath13-essentially equal to an algebraic subset of @xmath80 . \c ) the aj conjecture holds true if ( * ) the @xmath0-polynomial is @xmath81-irreducible and has @xmath12-degree @xmath82 . [ main1 ] here @xmath81-irreducibility means irreducibility in @xmath83 $ ] . there are many two - bridge knots that satisfy condition ( * ) . for example in a recent work @xcite hoste and shanahan proved that all the twist knots satisfy the condition ( * ) . hence we have the following corollary . the aj conjecture holds true for twist knots . in a separate paper @xcite we will prove that if both @xmath84 and @xmath82 are prime , then @xmath74 satisfies the condition ( * ) . also knot tables show that many two - bridge knots with small @xmath85 satisfy the condition ( * ) . our proof of the main theorem is more or less based on the ideology that the kauffman bracket skein module ( kbsm ) is a quantization of the @xmath72-character variety ( see @xcite and section [ peripheral ] below ) , which has been exploited in the work of frohman , gelca , and lofaro @xcite where they defined the non - commutative @xmath0-ideal . the calculation of the kbsm of a knot complement is a difficult task . bullock @xcite and recently bullock and lofaro @xcite calculated the kbsm for the complements of @xmath86 torus knots and twist knots . another main result of this paper is a generalization of these works : we calculate explicitly the kbsm for complements of all two - bridge knots . we will use another , more geometric approach that allows us to get the results for all two - bridge knots . we also prove that the growth of degree ( or breadth ) of the colored jones polynomial of an arbitrary knot is at most quadratic with respect to the color , and if the knot is alternating , then the growth is exactly quadratic , given by explicit formula . this is based on the exact estimate of the crossing number , used in the proof ( of kauffman , murasugi , and thistlethwaite ) of the tait conjecture on the crossing number of alternating knots . as a corollary , we show that the @xmath12-degree of the recurrence polynomial of an alternating knot must be at least 2 . in section 1 we review the theory of skein modules the colored jones polynomial . in section 2 we study the growth of the degree of the colored jones polynomial and the @xmath12-degree of the recurrence polynomial . in section 3 we review the a - polynomial and introduce a closely related polynomial , @xmath87 . section 4 is devoted to the quantum " version of @xmath87 , the peripheral polynomial . we will formulate another weaker version of the aj conjecture and prove it holds true for two - bridge knots ( in section [ proof1 ] ) . in section 5 we calculate the skein module of the complement of two - bridge knots . the last section contains a proof of the theorem [ main1 ] . i would like to thank s. garoufalidis , r. gelca , t. ohtsuki , a. referee , p. shanahan , a. sikora , t. takata , and x. zhang for valuable helps and discussions . we recall the definition and known facts about the colored jones polynomial through the theory of kauffman bracket skein modules which was introduced by przytycki and turaev , see the survey @xcite . [ preliminaries ] recall that @xmath88 $ ] . framed link _ in an oriented 3-manifold @xmath89 is a disjoint union of embedded circles , equipped with a non - zero normal vector field . framed links are considered up to isotopy . in all figures we will draw framed links , or part of them , by lines as usual , with the convention that the framing is blackboard . let @xmath90 be the set of isotopy classes of framed links in the manifold @xmath89 , including the empty link . consider the free @xmath50-module with basis @xmath90 , and factor it by the smallest submodule containing all expressions of the form @xmath91 and @xmath92 , where the links in each expression are identical except in a ball in which they look like depicted . this quotient is denoted by @xmath93 and is called the kauffman bracket skein module , or just skein module , of @xmath89 . when @xmath94 $ ] , the cylinder over the surface @xmath95 , we also use the notation @xmath96 for @xmath93 . in this case @xmath96 has an algebra structure induced by the operation of gluing one cylinder on top of the other . the operation of gluing the cylinder over @xmath97 to @xmath89 induces a @xmath98-left module structure on @xmath99 . when @xmath89 is the 3-space @xmath101 or the 3-sphere @xmath102 , the skein module @xmath93 is free over @xmath50 of rank one , and is spanned by the empty link . thus if @xmath103 is a framed link in @xmath101 , then its value in the skein module @xmath100 is @xmath104 times the empty link , where @xmath105 , known as the kauffman bracket of @xmath103 ( see @xcite ) , and is just the jones polynomial of _ framed links _ in a suitable normalization . the solid torus @xmath106 is the cylinder over an annulus , and hence its skein module @xmath107 has an algebra structure . the algebra @xmath107 is the polynomial algebra @xmath108 $ ] in the variable @xmath109 , which is a knot representing the core of the solid torus . instead of the @xmath50-basis @xmath110 , two other bases are often useful . the first basis consists of the chebyshev polynomials @xmath111 , @xmath112 , defined by @xmath113 , @xmath114 , and @xmath115 . the second basis consists of polynomials @xmath116 , @xmath112 , satisfying the same recurrence relation , but with @xmath117 and @xmath118 . extend both polynomials by the recurrence relation to all indices @xmath119 . note that @xmath120 , while @xmath121 . for a framed knot @xmath3 in a 3-manifold @xmath89 we define the @xmath8-th power @xmath122 as the link consisting of @xmath8 parallel copies of @xmath3 . using these powers of a knot , @xmath123 is defined as an element of @xmath93 . in particular , if @xmath124 one can calculate the bracket @xmath125 , and it is essentially the colored jones polynomial . more precisely , we will define the colored jones polynomial @xmath7 by the equation @xmath126 the @xmath127 sign is added so that when @xmath3 is the trivial knot , @xmath128 : = \frac{t^{2n } -t^{-2n}}{t^2-t^{-2}}.\ ] ] then @xmath129 , @xmath130 . we extend the definition for all integers @xmath8 by @xmath131 and @xmath132 . in the framework of quantum invariants , @xmath7 is the @xmath133-quantum invariant of @xmath3 colored by the @xmath8-dimensional simple representation of @xmath133 . we will always assume @xmath3 has 0 framing . in this case @xmath7 contains only even powers of @xmath44 , i.e. @xmath134 $ ] . hence the recurrence polynomial @xmath38 can be assumed to have only even powers of @xmath44 . a pair of _ oriented _ meridian and longitude on the torus @xmath135 will define an _ algebra _ isomorphism @xmath136 between @xmath137 and the symmetric part of the quantum torus @xmath17 as follows . for a pair of integers @xmath138 let @xmath139 , where @xmath140 is the greatest common divisor of @xmath141 and @xmath142 , with @xmath143 , and @xmath144 is the closed curve without self - intersection on the torus that is homotopic to @xmath145 times the meridian plus @xmath146 times the longitude . here @xmath147 is the chebyshev polynomial defined above ; and the framing of a curve on @xmath148 is supposed to be parallel to the surface @xmath135 . note that in the definition of skein modules we use _ non - oriented links _ , hence @xmath149 . as an @xmath50-module , @xmath137 is the quotient of the free @xmath50-module spanned by @xmath150 modulo the relations @xmath151 . recall that the quantum torus @xmath17 is defined as @xmath152 . let @xmath153 be the subalgebra of @xmath17 invariant under the involution @xmath154 , where @xmath155 . frohman and gelca in @xcite showed that the map @xmath156 is an isomorphism of _ algebras_. [ qtorus ] let @xmath157 be the rectangle ( in @xmath158 ) @xmath159 without 2 interior points @xmath160 and @xmath161 . then @xmath162 is the polynomial algebra @xmath163 $ ] , where @xmath164 is a small loop around @xmath165 , @xmath166 a small loop around @xmath167 , and @xmath168 a loop circling both @xmath169 ( see figure [ disk ] ) . note that in this case @xmath162 is a commutative algebra , which is not true in general when @xmath157 is replaced by an arbitrary surface , see @xcite . @xmath170 we will identify @xmath157 with the section @xmath171 in @xmath172 $ ] . the lines @xmath173 and @xmath174 divide @xmath157 into 3 parts : the left part @xmath175 ( containing @xmath165 ) , the right @xmath176 ( containing @xmath167 ) , and the middle part @xmath177 . let @xmath178 be a braid on @xmath179 strands in @xmath180 $ ] with boundary points on the lines @xmath173 and @xmath174 ( on @xmath157 ) , i.e. @xmath178 consists of @xmath179 connected paths , each begins at a point on @xmath173 ( on @xmath157 ) and goes monotonously to the right until ending at a point on @xmath174 , see figure [ closure ] . let the closure @xmath181 be the link in @xmath182 $ ] obtained by closing @xmath178 using @xmath183 parallel rainbow arcs in @xmath175 and @xmath183 parallel rainbow arcs in @xmath176 , as shown in figure [ closure ] . @xmath184 as an element of the skein module @xmath185 $ ] the closure link @xmath181 of a braid on @xmath179 strands is a polynomial having @xmath168-degree @xmath183 , and the coefficient of @xmath186 is invertible and of the form @xmath187 . [ disk21 ] consider the diagram @xmath188 of @xmath181 on @xmath157 . at every crossing point there are 2 ways to smooth the diagram , one positive that gives coefficient @xmath44 in the skein relation , and the other is negative . a _ state _ is the result of smoothing all the crossing ; what one has is a bunch of non - intersecting circles , each is one of @xmath189 , or the trivial loop . the link @xmath103 , considered as an element of @xmath190 $ ] , is the sum over all states , each with coefficient a power of @xmath44 . it is clear that for the closure @xmath181 of a braid , there is only one state ( the one with horizontal resolutions everywhere ) that gives the maximal power @xmath186 , and its coefficient is a power of @xmath44 . [ disk2 ] [ [ section ] ] one of the best known applications of the jones polynomial is a proof ( kauffman , murasugi , and thistlethwaite ) of the tait conjecture on the crossing number of alternating links , based on an exact estimate of the crossing number using the _ breadth _ of the jones polynomial . we will need a generalization of this estimate for the colored jones polynomial . for a laurent polynomial @xmath191 $ ] let @xmath192 and @xmath193 be respectively the maximal and minimal degree of @xmath44 in @xmath43 . the difference @xmath194 is called the _ _ breadth of @xmath43 . for a link diagram @xmath188 , let @xmath195 and @xmath196 be the number of circles obtained by positively ( respectively , negatively ) smoothing all the double points . \a ) suppose @xmath3 is a knot with a knot diagram @xmath188 having @xmath183 crossings . assuming the framing is black - board . then @xmath197 hence the breadth of @xmath7 grows at most as a quadratic function in @xmath8 . \b ) if @xmath3 is a non - trivial alternating knot with @xmath183 crossings . then the breadth of @xmath198 $ ] is a quadratic polynomial in @xmath8 . moreover @xmath199 [ degree10 ] \a ) the @xmath8-parallel @xmath200 of @xmath188 will have @xmath201 double points . in addition , it is easy to see that @xmath202 . hence lemma 5.4 of @xcite says that @xmath203 note that here we have taken into account the fact that we shift the index @xmath204 , and use the normalization in which the unknot takes value @xmath205 . note that @xmath206 is a strictly increasing function , @xmath207 . recall that @xmath208 . hence one has @xmath209 the proof for @xmath210 is similar . \b ) for an alternating knot diagram one has @xmath211 . note that the diagram @xmath200 is adequate in the sense of ( * ? ? ? * chapter 5 ) . for adequate diagrams the estimates of @xmath212 in part a ) are exact ( see the proof of lemma 5.4 of @xcite ) and the result follows immediately . suppose @xmath3 is a non - trivial alternating knot . then the recurrence polynomial @xmath38 has @xmath12-degree greater than 1 . [ deg2 ] assume the contrary that @xmath213 , where @xmath214 $ ] . garoufalidis @xcite showed that the polynomial @xmath215 is also in the recurrence ideal . since @xmath38 is _ the _ generator , it follows that for some @xmath216 @xmath217 one can then easily show that , after normalizing both @xmath218 by a same power of @xmath13 , one has @xmath219 the equation @xmath220 can now be rewritten as @xmath221 it is easy to see that for @xmath8 big enough , the difference of the breadths @xmath222 is a constant depending only on the polynomial @xmath223 , but not on @xmath8 . from the above equation it follows that the breadth of @xmath7 , for @xmath8 big enough , is a linear function on @xmath8 . this contradicts proposition [ degree10 ] . the following proposition was known to garoufalidis . when reduced by @xmath48 , the recurrence polynomial @xmath38 is divisible by @xmath224 . in other words , @xmath225 $ ] . [ divi ] for a function @xmath226 of two variables @xmath227 and @xmath6 let @xmath228 be the limit of @xmath229 when ( @xmath230 ) @xmath231 and @xmath232 is kept equal to @xmath109 all the time . the melvin - morton conjecture @xcite , proved by bar - natan and garoufalidis @xcite , showed that @xmath233 is the inverse of the alexander polynomial . in particular , @xmath234 . lemma [ new ] below shows that @xmath235 for any fixed @xmath183 . hence the operator @xmath12 becomes the identity after taking the limit ( @xmath230 ) . thus applying the limit ( @xmath230 ) to the equation @xmath236 we see that @xmath237 since @xmath238 , one has @xmath239 , which is equivalent to the lemma . for every fixed integer @xmath183 , one has @xmath235 . [ new ] for a knot @xmath3 of framing 0 , @xmath240 $ ] is a laurent polynomial in @xmath241 , and ( see @xcite ) @xmath242}|_{t^4= \exp\hbar}= \sum_{l=0}^\infty p_{l}(n ) \hbar^l,\ ] ] where @xmath243 is a polynomial in @xmath8 of degree at most @xmath244 : @xmath245 the limit ( @xmath230 ) is the same as the limit @xmath246 , with @xmath247 . under this limit , @xmath248 which does not depend on @xmath183 . we briefly recall here the definition of the a - polynomial and introduce a sibling of it . we will say that @xmath20 is _ @xmath13-essentially divisible by _ @xmath60 if @xmath20 is @xmath13-essentially equal to a polynomial divisible by @xmath60 . the set of representations of a finitely presented group @xmath249 into @xmath250 is an algebraic set defined over @xmath52 , on which @xmath250 acts by conjugation . the naive quotient space , i.e. the set of orbits , does not have a good topology / geometry . two representations in the same orbit ( i.e. conjugate ) have the same character , but the converse is not true in general . a better quotient , the algebro - geometric quotient denoted by @xmath251 ( see @xcite ) , has the structure of an algebraic set . there is a bijection between @xmath251 and the set of all characters of representations of @xmath249 into @xmath72 , hence @xmath251 is usually called the _ character variety _ of @xmath249 . for a manifold @xmath89 we use @xmath252 also to denote @xmath253 . suppose @xmath254 , the free abelian group with 2 generators . every pair of generators @xmath255 will define an isomorphism between @xmath251 and @xmath256 , where @xmath257 is the set of non - zero complex pairs @xmath258 and @xmath259 is the involution @xmath260 , as follows : every representation is conjugate to an upper diagonal one , with @xmath12 and @xmath13 being the upper left entry of @xmath261 and @xmath262 respectively . the isomorphism does not change if one replaces @xmath263 with @xmath264 . let @xmath265 be the closure of @xmath102 minus a tubular neighborhood @xmath266 of a knot @xmath3 . the boundary of @xmath265 is a torus whose fundamental group is free abelian of rank two . an orientation of @xmath3 will define a unique pair of an oriented meridian and an oriented longitude such that the linking number between the longitude and the knot is 0 . the pair provides an identification of @xmath267 and @xmath256 which actually does not depend on the orientation of @xmath3 . the inclusion @xmath268 induces the restriction map @xmath269 let @xmath270 be the image of @xmath271 and @xmath272 the lift of @xmath270 under the projection @xmath273 . the zariski closure of @xmath274 in @xmath63 is an algebraic set consisting of components of dimension 0 or 1 . the union of all the 1-dimension components is defined by a single polynomial @xmath275 $ ] , whose coefficients are co - prime . note that @xmath276 is defined up to @xmath277 . it is known that @xmath276 is divisible by @xmath224 , hence @xmath278 , where @xmath279 $ ] is called the @xmath0-polynomial of @xmath3 . it is known that @xmath280 $ ] . by definition , @xmath281 does not have repeated factor , and is not divisible by @xmath224 . can @xmath282 have a factor a non - constant polynomial depending on @xmath13 only ? [ question ] it s also instructive and convenient to see the dual picture in the construction of the a - polynomial . for an algebraic set @xmath89 let @xmath283 $ ] denote the ring of regular functions on @xmath89 . for example , @xmath284= \gt^\sigma$ ] , the @xmath154-invariant subspace of @xmath285 $ ] , where @xmath286 . the map @xmath271 in the previous subsection induces an algebra homomorphism @xmath287 \equiv \gt^\sigma \longrightarrow r[{{\chi}}(x)].\ ] ] we will call the kernel @xmath288 of @xmath289 the _ classical peripheral ideal _ ; it is an ideal of @xmath290 . let @xmath291 be the ideal extension of @xmath288 in @xmath292 . the set of zero points of @xmath293 is the closure of @xmath294 in @xmath63 . the ring @xmath295 $ ] embeds naturally into the principal ideal domain @xmath296 $ ] , where @xmath297 is the fractional field of @xmath298 $ ] . the ideal extension of @xmath293 in @xmath299 , which is @xmath300 , is thus generated by a single polynomial @xmath301 $ ] which has co - prime coefficients and is defined up to a factor @xmath302 with @xmath303 . again @xmath87 can be chosen to have integer coefficients because everything can be defined over @xmath81 . from the definitions one has immediately the polynomial @xmath87 is @xmath13-essentially divisible by @xmath304 . the two algebraic sets @xmath305 and @xmath306 are @xmath13-essentially equal . [ 500 ] note that @xmath87 might not be @xmath13-essentially equal to @xmath276 because @xmath87 might contain repeated factors . if the answer to question [ question ] is negative , then the two algebraic sets @xmath305 and @xmath306 are equal , and we have a natural way to define the multiplcity of factors of the @xmath0-polynomial , using the @xmath307-polynomial . recall that @xmath47 is the map reducing @xmath48 . one important result ( bullock , przytycki , and sikora @xcite ) in the theory of skein modules is that @xmath308 has a natural algebra structure and , when factored by its nilradical , is canonically isomorphic to @xmath309 $ ] , the ring of regular functions on the character variety of @xmath310 . the product of 2 links in @xmath311 is their union . using the skein relation with @xmath48 , it is easy to see that the product is well - defined , and that the value of a knot in the skein module depends only on the homotopy class of the knot in @xmath89 . the isomorphism between @xmath312 and @xmath309 $ ] is given by @xmath313 , where @xmath3 is a homotopy class of a knot in @xmath89 , represented by an element , also denoted by @xmath3 , of @xmath310 , and @xmath314 is a representation of @xmath310 . in many cases the nilradical of @xmath308 is trivial , and hence @xmath308 is exactly equal to the ring of regular functions on the character variety of @xmath310 . for example , this is the case when @xmath89 is a torus , or when @xmath89 is the complement of a two - bridge knots ( see section [ twobridge ] ) . in light of this fact , one can consider @xmath93 as a quantization of the character variety . [ peripheral ] recall that @xmath265 is the closure of the complement of a tubular neighborhood @xmath266 in @xmath102 . the boundary @xmath315 is a torus , and using the preferred meridian and longitude we will identify @xmath316 with @xmath153 , see subsection [ qtorus ] . the embedding of @xmath317 into @xmath265 gives us a map @xmath318 , which can be considered as a quantum analog of @xmath289 . one has the following commutative diagram @xmath319 \end{cd}\ ] ] the kernel of @xmath320 , denoted by @xmath321 , is called the _ quantum peripheral ideal _ ; it is a left ideal of @xmath153 and can be considered as a quantum analog of the classical peripheral ideal @xmath322 . the ideal @xmath321 was introduced by frohman , gelca , and lofaro in @xcite and there it is called simply the peripheral ideal . from the commutative diagram it is clear that @xmath323 . the following question is important . is it true that @xmath324 can never be 0 ? let us adapt the construction of the @xmath87 polynomial to the quantum setting . recall that @xmath32 ( see introduction ) is a principal left - ideal domain that contains @xmath153 as a subring . the left - ideal extension @xmath325 in @xmath32 is generated by a polynomial @xmath326 where @xmath327 is assumed to be minimum and all the coefficients @xmath328 $ ] are co - prime . we call @xmath329 the _ peripheral polynomial _ of @xmath3 , which is defined up to @xmath330 with @xmath331 . @xmath332 is @xmath13-essentially divisible by @xmath87 , and hence is @xmath13-essentially divisible by @xmath278 . [ 323 ] the proposition follows the fact that @xmath333 and proposition [ 500 ] . . then @xmath335 . [ t1 ] for the extensions @xmath336 and @xmath337 we also have @xmath338 , since @xmath339 . from the definition we have that @xmath340 for some non - zero polynomial @xmath341 $ ] . hence @xmath342 is not empty . take an element @xmath343 ; it is @xmath13-essentially divisible by @xmath329 , the generator . applying the map @xmath47 , one gets @xmath87 is @xmath13-essentially divisible by @xmath332 . combining with proposition [ 323 ] one has @xmath344 . for every knot we have @xmath334 , and hence @xmath335 . [ weakconjecture ] later we will show that for all two - bridge knots holds true the conjecture , which is closely related to the aj conjecture . a sufficient condition for the conjecture to hold true is given in section [ proof1 ] . there is a bilinear pairing @xmath345 where @xmath103 and @xmath346 are framed links in @xmath266 and @xmath265 respectively . the _ orthogonal ideal _ @xmath347 is defined by @xmath348 it s clear that @xmath347 is a left ideal of @xmath349 and @xmath350 . in @xcite , where @xmath347 was first introduced , @xmath347 was called the formal ideal . what is important for us is the following the orthogonal ideal is in the recurrence ideal of a knot , @xmath351 . as a consequence , @xmath352 . [ inclu ] this was proved by garoufalidis @xcite . frohman , gelca , and lofaro @xcite proved that every element @xmath346 in the orthogonal ideal @xmath347 gives rise to a linear recurrence relation for the colored jones polynomial . the idea is simple and beautiful : @xmath346 annihilates everything in @xmath353 , in particular , @xmath354 ; but this equation , after some calculation , can be rewritten as a linear recurrence relation for the colored jones polynomial . garoufalidis , using the weyl symmetry , further simplified the recurrence relation , and obtained that @xmath355 , which is stronger than the proposition . the right kernel of the bilinear form ( [ pair ] ) is trivial . this conjecture implies that @xmath356 , from which , due to an argument of a. sikora and the author , one can show that the colored jones polynomial distinguishes the unknot from other knots . the peripheral polynomial @xmath329 is divisible by the recurrence polynomial @xmath38 in the sense that there are polynomials @xmath357 $ ] and @xmath358 such that @xmath359 moreover @xmath360 and @xmath361 can be chosen so that @xmath362 . from proposition [ inclu ] we have that @xmath363 . hence the left - ideal extension @xmath364 is also a subset of @xmath21 , since both are left ideals of @xmath17 . it follows that @xmath329 , as the generator of the extension of @xmath365 in @xmath29 , is divisible by the generator of the extension of @xmath21 , and ( [ sao ] ) follows . we can assume that @xmath366 does not divide both @xmath360 and @xmath361 simultaneously . if @xmath367 then @xmath60 is divisible by @xmath366 , and hence @xmath368 is not . but then from the equality @xmath369 it follows that @xmath38 is divisible by @xmath366 , which is impossible , since all the coefficients of powers of @xmath12 in @xmath38 are supposed to be co - prime . a two - bridge knot is a knot @xmath370 such that there is a 2-sphere @xmath371 that separates @xmath102 into 2 balls @xmath372 and @xmath373 , and the intersection of @xmath3 and each ball is isotopic to 2 trivial arcs in the ball . the branched double covering of @xmath102 along a two - bridge knot is a lens space @xmath374 , which is obtained by doing a @xmath375 surgery on the unknot . such a two - bridge knot is denoted by @xmath74 . it is known that both @xmath85 are odd . one can always assume that @xmath376 . it is known that @xmath377 if @xmath378 . we will present the ball @xmath372 as the rectangular parallelepiped , see figure [ x1 ] : @xmath379 @xmath380 we suppose that the knot intersects the interior of @xmath372 in two vertical ( i.e. parallel to the @xmath381-axis ) straight intervals @xmath382 and @xmath383 , where @xmath384 , @xmath385 , and @xmath386 . after an isotopy , we assume that the part of @xmath3 outside the interior of @xmath372 are 2 non - intersecting arcs @xmath387 and @xmath388 on @xmath389 , where @xmath387 connects @xmath165 and @xmath167 , and @xmath388 connects @xmath49 and @xmath390 . later we will describe explicitly the arc @xmath387 . if one cuts @xmath391 along the arc @xmath387 , then one obtains a disk , hence the other arc @xmath388 is uniquely determined by @xmath387 , up to isotopy . let @xmath392 be the top rectangle of @xmath372 , @xmath393 . note that @xmath394 is the cylinder over a two - punctured disk @xmath395 . hence @xmath396 is isomorphic to the commutative algebra @xmath397 $ ] , as described in subsection [ disk2 ] . here @xmath164 is a small loop circling @xmath165 , @xmath166 a small loop circling @xmath167 , and @xmath398 . one of our main results is the skein module @xmath399 is free over @xmath50 with basis @xmath400 . [ main2 ] the remaining part of the section is devoted to a proof of this theorem . moreover , we will present more explicit structures of the skein module @xmath399 . let @xmath401 be the rotation by @xmath402 about the axis @xmath403 ( which is parallel to the @xmath381-axis and passing through the center of the rectangle @xmath392 ) . one has @xmath404 . for a set @xmath405 let @xmath406 $ ] be the part of @xmath270 in the strip @xmath407 , i.e. @xmath406 : = z \cap \{(x_1,x_2,x_3 ) \mid a\le x_1 \le b\}$ ] . we will consider 3 pieces @xmath408 , b_1[1,5]$ ] , and @xmath409 $ ] of @xmath372 . recall that @xmath392 is the top rectangle of the rectangular parallelepiped @xmath410 . on @xmath411 $ ] let @xmath412 be the collection of @xmath82 half - circles centered at @xmath165 , with radii @xmath413 , where @xmath414 , see figure [ ul ] . the end points of the half - circles , together with the center @xmath165 , are on a straight line ; there are exactly @xmath84 of them , including @xmath165 . mark them from bottom to top by @xmath415 . ( thus @xmath416 . ) @xmath417 similarly , on @xmath418 $ ] let @xmath419 be the collection of @xmath82 half - circles centered at @xmath167 , with radii @xmath420 , where @xmath414 . mark the end - points of the half - circles and @xmath167 from bottom to top by @xmath421 . ( thus @xmath422 . ) recall that @xmath423 . on @xmath424 $ ] we construct @xmath425 as follows . note that @xmath424 $ ] is a cylinder over @xmath426 $ ] : @xmath424= [ 1,5]\times s^2[1,1]$ ] . informally @xmath425 is the braid on the cylinder @xmath424= [ 1,5]\times s^2[1,1]$ ] representing the rotation by @xmath427 . formally , we first connect @xmath428 with @xmath429 by a straight interval . then for all @xmath430 connect @xmath431 with @xmath432 by non - intersecting arcs on @xmath424 $ ] ; up to isotopy there is a unique way to do so , see figure [ ul ] . there are in total @xmath84 arcs ; denote them by @xmath425 . we can assume that each arc in @xmath425 always travel from left to right ( no backwards traverse , just like in the case of braids ) , and moreover , @xmath425 is invariant under @xmath401 . let @xmath387 be the arc on @xmath391 obtained by combining @xmath433 and @xmath419 ; it connects @xmath165 and @xmath167 and is invariant under @xmath401 , see figure [ ull ] . up to isotopy there is a unique arc @xmath388 on @xmath391 connecting @xmath49 and @xmath390 . @xmath434 the knot @xmath3 consisting of the arcs @xmath435 , the straight intervals @xmath382 and @xmath383 is the two - bridge knot @xmath74 . it is clear that @xmath3 is a two - bridge knot . let @xmath436 be the 2-fold covering of @xmath423 branched along the 4 points @xmath437 . one can recognize the knot type by looking at the homology class of the lift of @xmath387 . note that @xmath436 is a torus , with the following preferred meridian and longitude . the plane passing through @xmath437 intersects @xmath438 $ ] in an arc @xmath439 that connect @xmath165 and @xmath49 . the total lift @xmath440 of @xmath439 is a closed curve on the the torus @xmath436 which will serve as the meridian . the total lift @xmath441 of the straight interval @xmath442 is another closed curve serving as the longitude . it is is easy to see that @xmath440 and @xmath441 form a basis of @xmath443 , and that the total lift of the curve @xmath387 , as a homology class , is equal to @xmath444 . according to ( * ? ? ? * chapter 12 ) , @xmath3 is a two - bridge knot of type @xmath445 . let @xmath446 be the boundary curve of a small normal neighborhood of the arc @xmath387 in @xmath447 . we can assume that @xmath446 is invariant under @xmath401 . then @xmath448 is obtained from @xmath372 by gluing a 2-handle to along @xmath446 . recall that on @xmath424 $ ] the curve @xmath387 consists of @xmath84 arcs , and we assume that the part of @xmath446 on @xmath424 $ ] consists of @xmath449 arcs parallel and close to those of @xmath387 . for example , the 2 arcs of @xmath387 on @xmath424 $ ] containing @xmath165 and @xmath167 are drawn in figure [ ulll ] by bold line , and the part of @xmath446 near them are drawn by lighter lines . let the plane @xmath450 intersect these lighter lines ( which are parts of @xmath446 ) at the points @xmath451 ( order from top to bottom ) , as in figure [ ulll ] . we have @xmath452 . @xmath453 let us recall the definition of the relative skein module @xmath454 ( see @xcite ) . a _ type 1 tangle _ is the disjoint union of a framed link and a framed arc in @xmath410 such that the parts of the arc near the two end points are on the boundary @xmath455 , and the framing on these parts are given by vectors normal to @xmath455 . type 1 tangles are considered up to isotopy relative the endpoints . then @xmath454 is the @xmath456-vector space generated by type 1 tangles with endpoints at @xmath457 modulo the usual skein relations , like in the definition of @xmath458 . one defines in a similar way the relative kauffman bracket skein module @xmath459;p , q')$ ] , where we identify @xmath460 $ ] with a collar of @xmath95 in @xmath410 . since @xmath461 , there is a natural bilinear map @xmath462 , where @xmath463 , which is the disjoint union of @xmath103 and @xmath346 . the pair @xmath464 divide @xmath446 into two arcs , the one that is fully drawn in figure [ ulll ] ( and that goes around point point @xmath165 ) is denoted by @xmath465 . similarly , the pair @xmath466 divide @xmath446 into two arcs , the one that is fully drawn in figure [ ulll ] ( and that goes around point point @xmath167 ) is denoted by @xmath467 . for @xmath468 among @xmath469 let @xmath470 be the straight interval connecting @xmath471 and @xmath472 , whose interior is slightly pushed inside the interior of @xmath372 ( to avoid intersections with other arcs on the boundary @xmath473 ) and whose framing is given by vectors normal to @xmath473 . let @xmath474 be @xmath475 ; @xmath476 be @xmath465 followed by @xmath477 ; @xmath478 be @xmath479 followed by @xmath467 ; and @xmath480 be @xmath465 followed by @xmath481 then followed by @xmath467 . in all cases the framing is given by vectors normal to @xmath473 . using the skein relations one can simplify the arc part of elements in @xmath482 , showing that the arc part is one of the four @xmath483 . more precisely one has the following lemma , which is a slightly different version of lemma 3.1 in @xcite . the simple proof in @xcite works for our version as well . the relative skein @xmath482 is equal to the union @xmath484 . [ aa23 ] [ [ section-1 ] ] let @xmath485 . then @xmath486 is an algebra , and @xmath487 $ ] is a left @xmath486-module , since @xmath488 : for @xmath489 and @xmath490 the action @xmath491 is the disjoint union of @xmath103 and @xmath346 . it is easy to see that the action descends to an action of @xmath486 on @xmath492\equiv { { \mathcal r}}[x , x',y]/(x = x')$ ] . recall that @xmath265 is obtained from @xmath410 by attaching a 2-handle along the curve @xmath446 . the embedding of @xmath410 into @xmath265 gives rise to a linear map from @xmath493 $ ] to @xmath458 . it is known that the map is surjective , and its kernel @xmath494 , see @xcite , can be described through slides as follows . suppose @xmath495 is a type 1 tangle whose 2 endpoints are on @xmath446 such that outside a small neighborhood of the 2 endpoints @xmath495 is in the interior of @xmath410 and in a small neighborhood of the endpoints @xmath495 is on the boundary @xmath496 . the two end points of @xmath495 divide @xmath446 into 2 arcs @xmath497 and @xmath498 . the loop @xmath446 partitions @xmath423 , which is a sphere , into 2 parts ; the one not containing @xmath169 is called the _ outside one_. let us isotope @xmath495 ( relatively to the endpoints ) to @xmath499 so that in a small neighborhood of the endpoints @xmath499 is in the outside part of @xmath446 . let @xmath500 be @xmath501 , considered as an element of the skein module @xmath396 . here @xmath502 is the framed link obtained by combining @xmath499 and @xmath497 . note that @xmath500 is defined up to a factor @xmath503 : the exchange @xmath504 changes the sign , and isotopies in neighborhoods of the endpoints change the framing , which results in a factor equal to a power of @xmath505 . it is clear that as framed links in @xmath265 , @xmath502 is isotopic to @xmath506 , since one is obtained from the other by sliding over the 2-handle attached to the curve @xmath446 . hence we always have @xmath507 . it was known that @xmath494 is spanned by all possible @xmath500 , where @xmath495 can be chosen among all type 1 tangles with pre - given two endpoints on @xmath446 . now we choose and fix the two end points of @xmath495 : @xmath43 and @xmath508 ; i.e. @xmath509 . we will assume that @xmath497 contains @xmath465 . the kernel @xmath494 is spanned over @xmath50 by @xmath510 . from the description of @xmath482 in lemma [ aa23 ] we have the kernel @xmath494 is equal to the @xmath50-span of @xmath511 . we will call an element in @xmath512 a _ relation of type @xmath513_. here @xmath514 . [ e3 ] one has @xmath515 the first identity is a tautology . the last three follows from trivially a simple isotopy of the links involved . let @xmath516 be the @xmath50-span of @xmath517 and @xmath518 . for every @xmath519 one has @xmath520 . [ e4 ] note that @xmath521 , and @xmath522 . hence for any link @xmath103 , the skein @xmath523 , as an element in @xmath524 $ ] , is obtained from the skein of @xmath103 by the involution @xmath525 . it follows that for any framed link @xmath103 , and one has @xmath526 . one has @xmath527 . first we prove that @xmath528 . since @xmath529 is already in @xmath494 , we need to show @xmath530 is in @xmath494 . note that @xmath531 is exactly @xmath532 . moreover , for any link @xmath103 in the interior of @xmath410 one has @xmath533 . hence @xmath534 , and hence @xmath528 . now we prove @xmath535 . by lemma [ e3 ] one has @xmath536 . since both @xmath479 and @xmath446 is invariant under @xmath401 , we have @xmath537 , where @xmath538 and @xmath539 are the two arcs of @xmath540 obtained by dividing @xmath446 using the two points @xmath541 . hence @xmath542 is in @xmath543 by lemma [ e4 ] . the proof that all relations of type 2 belongs to @xmath516 is similar . for type 4 , by lemma [ e3 ] we have @xmath544 . since @xmath452 , one has @xmath545 . hence @xmath546\star { { \mathcal s}}(x_1)$ ] belongs to @xmath547 by lemma [ e4 ] . thus @xmath548 is in @xmath516 . we have @xmath549/{\mathcal{k}}'$ ] . note that @xmath163/ ( x - x'){{\mathcal s}}(x_1)={{\mathcal r}}[x , y]$ ] . hence @xmath550/{\mathcal{k}}''$ ] , where @xmath551 is the @xmath50-span of @xmath552 $ ] . note that there is a natural @xmath553$]-module structure on @xmath458 : here @xmath164 is a meridian , thus belongs to the boundary of @xmath265 . over @xmath553 $ ] , @xmath492 $ ] is spanned by @xmath554 . hence @xmath551 , as a @xmath553 $ ] module , is spanned by @xmath555 . note that @xmath556 is the closure in the sense of section [ d3 ] of a braid on @xmath557 strands , while @xmath558 is the closure of a braid on @xmath559 strands . moreover , @xmath560 is the closure of a braid on @xmath561 strands , while @xmath562 is the closure of of a braid on @xmath563 strands . lemma [ disk21 ] shows that @xmath564 , as an element of @xmath492 $ ] , has @xmath168-degree @xmath565 , with highest coefficient invertible and of the form a power of @xmath44 . hence when we factor out @xmath492 $ ] by @xmath551 , we get a free @xmath553$]-module with representatives @xmath566 as a basis . this completes the proof of theorem [ main2 ] . the same proof shows that the theorem still holds true if we replace the ground ring @xmath567 $ ] by @xmath568 $ ] . for two - bridge knots one has @xmath569 , the ring of regular functions on the character variety . [ phu ] by the result of @xcite , the ring @xmath570 is @xmath571/(\varphi(\bar x,\bar y))$ ] , where @xmath572 is a polynomial of @xmath573-degree @xmath79 , with leading coefficient 1 . here @xmath574 , @xmath573 are respectively the traces of the loop @xmath575 . the corollary follows immediately . when @xmath577 , the field @xmath30 of rational functions in @xmath13 is a flat @xmath553$]-module , since @xmath30 contains the fractional field of @xmath553 $ ] as a subfield . hence the following sequence , which is obtained from ( [ sao200 ] ) by tensoring with @xmath30 , is exact @xmath578}{{\mathcal p}}\ , \hookrightarrow\ , { { \mathcal r}}(m)\otimes_{{{\mathcal r}}[x]}{{\mathcal t}}^\sigma \,\overset{id \otimes \theta}{\longrightarrow}{{\mathcal r}}(m)\otimes_{{{\mathcal r}}[x ] } { { \mathcal s}}(x ) \label{sao201}\ ] ] note that the first module @xmath579}{{\mathcal p}}$ ] is exactly @xmath580 , the left - ideal extension of @xmath321 from @xmath153 to @xmath32 . it s easy to check that the second module @xmath579}{{\mathcal t}}^\sigma$ ] is @xmath32 . one can now rewrite ( [ sao201 ] ) as the third module is a finite - dimensional @xmath30-vector space ; in fact , its basis is @xmath582 , since @xmath458 is @xmath553$]-free with the same basis , by theorem [ main2 ] . the middle module @xmath32 is an @xmath30-vector space of infinite dimension ; in fact , its basis is @xmath583 . thus the kernel @xmath580 is never 0 , and hence its generator @xmath584 is not 0 . moreover the image of @xmath585 elements @xmath586 are linearly dependent . hence there must be a non - trivial element in the kernel of @xmath12-degree less than or equal to @xmath79 . this is because @xmath329 is divisible by the recurrence polynomial @xmath38 . note that the existence of recurrence relations for _ arbitrary _ knots was established in @xcite by another method . but in @xcite the @xmath12-degree is much larger . consider again the exact sequence ( [ sao200 ] ) , but now as sequence of modules over @xmath567 $ ] , a principle ideal domain . by assumption , the last module @xmath458 is free over @xmath50 . hence when tensoring ( [ sao200 ] ) with any @xmath50-module , one gets an exact sequence . in particular , tensoring with @xmath52 , considered as @xmath50-module by putting @xmath48 , one has the exact sequence \(c ) suppose @xmath281 has @xmath12-degree @xmath82 . then @xmath304 has @xmath12-degree @xmath79 . by proposition [ degree ] , @xmath329 has @xmath12-degree less than or equal to @xmath79 . but @xmath332is @xmath13-essentially divisible by @xmath276 . hence the @xmath12-degree of @xmath329 must be exactly @xmath79 , and also recall that @xmath603 is a polynomial by proposition [ divi ] . from ( [ sao ] ) and the fact that the @xmath12-degree of @xmath38 is bigger than 1 ( proposition [ deg2 ] ) it follows that the @xmath12-degree of @xmath368 is less than @xmath82 , which is the @xmath12-degree of @xmath281 . hence if @xmath281 is @xmath81-irreducible , from ( [ 600 ] ) one must have @xmath604 and @xmath605 . this completes the proof of theorem [ main1 ] . t. t. q. l , _ varieties of representations and their subvarieties of cohomology jumps for knot groups _ , ( russian ) mat . * 184 * ( 1993 ) , no . 2 , 5782 ; translation in russian acad . sb . math . * 78 * ( 1994 ) , no . 1 , 187209 . t. t. q. l , _ on the skein modules and the a - polynomial of two - bridge knots _ , to appear .

the emission mechanism and the origin and structure of magnetic fields in gamma - ray burst ( grb ) jets are among the most important open questions concerning the nature of the central engine of grbs . in spite of extensive observational efforts , these questions remain to be answered and are difficult or even impossible to infer with the spectral and lightcurve information currently collected . polarization measurements will lead to unambiguous answers to several of these questions . recent developments in x - ray and-ray polarimetry techniques have demonstrated a significant increase in sensitivity enabling several new mission concepts , e.g. _ poet _ ( polarimeters for energetic transients ) , providing wide field of view and broadband polarimetry measurements . if launched , missions of this kind would finally provide definitive measurements of grb polarizations . we perform monte carlo simulations to derive the distribution of grb polarizations in three emission models ; the synchrotron model with a globally ordered magnetic field ( so model ) , the synchrotron model with a small - scale random magnetic field ( sr model ) , and the compton drag model ( cd model ) . the results show that _ poet _ , or other polarimeters with similar capabilities , can constrain the grb emission models by using the statistical properties of grb polarizations . in particular , the ratio of the number of grbs for which the polarization degrees can be measured to the number of grbs that are detected ( ) and the distributions of the polarization degrees ( ) can be used as the criteria . if and is clustered between 0.2 and 0.7 , the so model will be favored . if instead , then the sr or cd model will be favored . if several events with are observed , then the cd model will be favored .

0812.2483

gamma - ray bursts ( grbs ) are brief , intense flashes of @xmath0-rays originating at cosmological distances , and they are the most luminous objects in the universe . they also have broadband afterglows long - lasting after the @xmath0-ray radiation has ceased . it has been established that the bursts and afterglows are emitted from outflows moving towards us at highly relativistic speeds @xcite , and at least some grbs are associated with the collapse of massive stars ( e.g. , * ? ? ? * ; * ? ? ? observations suggest that the burst is produced by internal dissipation within the relativistic jet that is launched from the center of the explosion , and the afterglow is the synchrotron emission of electrons accelerated in a collisionless shock driven by the interaction of the jet with the surrounding medium ( for recent reviews , * ? ? ? * ; * ? ? ? * ; * ? ? ? in spite of extensive observational and theoretical efforts , several key questions concerning the nature of the central engines of the relativistic jets and the jets themselves remain poorly understood . in fact , some of these questions are very difficult or even impossible to answer with the spectral and lightcurve information currently collected . on the other hand , polarization information , if retrieved , would lead to unambiguous answers to these questions . in particular , polarimetric observations of grbs can address the following : _ magnetic composition of grb jets _ it is highly speculated that strong magnetic fields are generated at the grb central engine , and may play an essential role in the launch of the relativistic jets . however , it is unclear whether the burst emission region is penetrated by a globally structured , dynamically important magnetic field , and whether the burst is due to shock dissipation or magnetic reconnection ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? . _ emission mechanisms of the bursts _ the leading model for the emission mechanism of the prompt burst emission is synchrotron emission from relativistic electrons in a globally ordered magnetic field carried from the central engine , or random magnetic fields generated in - situ in the shock dissipation region @xcite . other suggestions include compton drag of ambient soft photons @xcite , synchrotron self - compton emission @xcite , and the combination of a thermal component from the photosphere and a non - thermal component ( e.g. , synchrotron ) @xcite . _ geometric structure of grb jets _ although it is generally believed that grb outflows are collimated , the distribution of the jet opening angles , the observer s viewing direction , and whether there are small - scale structures within the global jet are not well understood @xcite . to date , robust positive detections of grb polarization have been made only in the optical band in the afterglow phase . varying linear polarizations have been observed in several optical afterglows several hours after the burst trigger , with a level of @xmath6 , which is consistent with the synchrotron emission mechanism of grb afterglow ( for reviews , see * ? ? ? * ; * ? ? ? an upper limit @xmath7 has been obtained for the early @xmath8 optical afterglow of grb 060418 @xcite . also for radio afterglows , we have several upper limits for the polarization degree @xcite ( for some implications , see * ? ? ? as for the prompt burst emission , strong linear polarization of the @xmath0-ray emission at a level of @xmath9 was claimed for grb 021206 based on an analysis of _ rhessi _ data @xcite , although this claim remains controversial because of large systematic uncertainties @xcite . several other reports of high levels of polarization in the prompt burst emission are also statistically inconclusive @xcite . recently , more sensitive observational techniques for x - ray and @xmath0-ray polarimetry have been developed , and there are several polarimeter mission concepts . these include polarimeters for energetic transients ( _ poet _ , * ? ? ? * ; * ? ? ? * ) , polarimeter of gamma - ray observer ( _ pogo _ , * ? ? ? * ) , _ polar _ @xcite , advanced compton telescope ( _ act _ , * ? ? ? * ) , gravity and extreme magnetism ( _ gems _ , * ? ? ? * ) _ xpol _ @xcite , gamma - ray burst investigation via polarimetry and spectroscopy ( _ grips _ , * ? ? ? * ) , and so on . several of these missions , if launched , would provide definitive detections of the burst polarizations and enable us to discuss the statistical properties of the polarization degrees and polarization spectra . although there are several polarimetry mission concepts described in the literature , _ poet _ is the only one to date that incorporates a broadband capability for measuring the prompt emission from grbs , and for this reason it provides a good case study for our simulations . _ poet _ will make measurements with two different polarimeters , both with wide fields of view . the gamma - ray polarimeter experiment ( grape ; 60 - 500 kev ) and the low energy polarimeter ( lep ; 2 - 15 kev ) provide a broad energy range for the observations . suborbital versions of both _ poet _ instruments are currently being prepared for flight within the next few years . grape will fly on a sub - orbital balloon in 2011 , and the gamma - ray burst polarimeter ( grbp , a smaller version of lep ) will fly on a sounding rocket . theoretically , it has been shown that similarly high levels of linear polarization can be obtained in several grb prompt emission models ; the synchrotron model with a globally ordered magnetic field , the synchrotron model with a small - scale random magnetic field @xcite , and the compton drag model @xcite . thus the detections of grb prompt emission polarization would support these three models . in this paper , we show that these models can be distinguished by their statistical properties of observed polarizations . we performed detailed calculations of the distribution of polarization degrees by including realistic spectra of grb prompt emission and assuming realistic distributions of the physical parameters of grb jets , and show that _ poet _ , or other polarimeters with similar capabilities , can constrain the grb emission models . we use the limits of _ poet _ for grb detection and polarization measurements as realistic and fiducial limits . this paper is organized as follows . we first introduce the _ poet _ mission concept in [ sec : poet ] . in [ sec : theory ] , we summarize the properties of the observed linear polarization from uniform jets within the three emission models . based on these models , we perform monte carlo simulations of observed linear polarizations and show how the statistical properties of observed polarization may constrain grb emission mechanisms in [ sec : simulation ] . a summary and discussion are given in [ sec : summary ] . _ poet _ ( polarimeters for energetic transients ) is a small explorer ( smex ) mission concept , that will provide highly sensitive polarimetric observations of grbs and can also make polarimetry measurements of solar flares , pulsars , soft gamma - ray repeaters , and slow transients . the payload consists of two wide field of view ( fov ) instruments : a low energy polarimeter ( lep ) capable of polarization measurements in the 2 - 15 kev energy range and a high energy polarimeter ( gamma - ray polarimeter experiment ; grape ) that will measure polarization in the 60 - 500 kev energy range . _ poet _ can measure grb spectra from 2 kev up to 1 mev . the _ poet _ spacecraft provides a zenith - pointed platform for maximizing the exposure to deep space and spacecraft rotation provides a means of effectively dealing with systematics in the polarization response . _ poet _ provides sufficient sensitivity and sky coverage to detect up to 200 grbs in a two - year mission . lep and grape determine polarization by measuring the number of events versus the event azimuth angle ( eaa ) as projected onto the sky . this is referred to as a modulation profile and represents a measure of the polarization magnitude and direction of polarization for the incident beam . depending on the type of polarimeter , the eaa is either the direction of the ejected photoelectron ( lep ) or the direction of the scattered photon ( grape ) . the response of a polarimeter to @xmath10 polarized photons can be quantified in terms of the modulation factor , @xmath11 , which is given by : @xmath12 where @xmath13 and @xmath14 are the maximum and minimum of the modulation profile , respectively . the polarization fraction ( @xmath2 ) of the incident flux is obtained by dividing the measured modulation by that expected for @xmath10 polarized flux . the polarization angle ( @xmath15 ) corresponds either to the maximum of the modulation profile ( lep ) or the minimum of the modulation profile ( grape ) . to extract these parameters from the data , the modulation histograms are fit to the functional form : @xmath16 the sensitivity of a polarimeter is defined in terms of the minimum detectable polarization ( mdp ) , which refers to the minimum level of polarization that is detectable with a given observation ( or , equivalently , the apparent polarization arising from statistical fluctuations in unpolarized data ) . the precise value of the mdp will depend on the source parameters ( fluence , spectrum , etc . ) and the polarimeter characteristics . at the @xmath17 confidence level , the mdp can be expressed as , @xmath18 where @xmath19 is the observed source strength ( @xmath20 ) , @xmath21 is the total observed background rate ( @xmath20 ) , and @xmath22 is the observing time ( s ) . the ultimate sensitivity , however , may not be limited by statistics but by systematic errors created by false modulations that arise from azimuthal asymmetries in the instrument . lcc polarimetry & 60500 kev & 215 kev + detectors & bgo / plastic scintillator ( 62 ) & @xmath23 gas ( 8) + spectroscopy & 15 kev 1 mev & 2 15 kev + detectors & nai(ti ) scintillator ( 2 ) & as above + field - of - view & @xmath24 & @xmath25 + at energies from @xmath26 kev up to several mev , photon interactions are dominated by compton scattering . the operational concept for grape is based on the fact that , in compton scattering , photons are preferentially scattered at a right angle to the incident electric field vector ( the polarization vector ) @xcite . if the incident beam of photons is polarized , the azimuthal distribution of scattered photons will be asymmetric . the direction of the polarization vector is defined by the minimum of the scatter angle distribution . the grape performance characteristics are shown in table [ tabparam ] . the design of the grape instrument is very modular , with 62 independent polarimeter modules and 2 spectroscopy modules . each polarimeter module incorporates an array of optically independent 5x5x50 @xmath27 non - hygroscopic scintillator elements aligned with and optically coupled to the 8x8 scintillation light sensors of a 64-channel mapmt . two types of scintillators are employed . low - z plastic scintillator is used as an effective medium for compton scattering . high - z inorganic scintillator ( bismuth germanate , bgo ) is used as a calorimeter , for absorbing the full energy of the scattered photon . the arrangement of scintillator elements within a module has 28 bgo calorimeter elements surrounding 32 plastic scintillator scattering elements . valid polarimeter events are those in which a photon compton scatters in one of the plastic elements and is subsequently absorbed in one of the bgo elements . these events can be identified as a coincident detection between one plastic scintillator element and one bgo calorimeter element . the azimuthal scatter angle is determined for each valid event by the relative locations of hit scintillator elements . it is not necessary to know where within each element the interaction takes place ( e.g. , the depth of interaction ) . it is sufficient to know only the lateral location of each element to generate a histogram of photon scatter angles . at energies below @xmath28 kev , the most sensitive technique for broadband polarimetry is the photoelectric effect . the lep measures the polarization of incident photons with the innovative operation of a time projection chamber ( tpc ) @xcite . the lep polarimeter enclosure consists of four dual - readout detector modules each with an isolated gas volume contained by a be x - ray window . each detector module contains two 6 x 12 x 24 @xmath29 ( lxwxh ) tpcs that share a single x - ray transparent drift electrode . each tpc is comprised of a micropattern proportional counter , consisting of a shared drift electrode and a high - field gas electron multiplier ( gem ) positioned 1 mm from a strip readout plane . when an x - ray is absorbed in the gas between the drift electrode and the gem , a photoelectron is ejected in a preferential direction with a @xmath30 distribution , where @xmath31 is the azimuthal angle measured from the x - ray polarization vector . as the photoelectron travels through the gas it creates a path of ionization that drifts in a moderate , uniform field to the gem where an avalanche occurs . the charge finally drifts to the strip detector where it is read out . to estimate realistic mdp values for grbs detected by grape and lep , we perform an analytical calculation for lep and a monte carlo simulation for grape using the current instrument configuration ( table [ tabparam ] ) . the input spectrum in the calculation and the simulation is a typical grb spectrum which can be described as a smoothly broken power - law spectra characterized by photon energy at the @xmath32 spectral peak , @xmath33 , and lower and higher indices of the @xmath34 spectrum , @xmath35 and @xmath36 , respectively @xcite . ( we treat the spectral indices of the specific energy flux @xmath34 , while @xcite define @xmath37 and @xmath38 as the indices of the photon number flux , i.e. , @xmath39 and @xmath40 , since we will calculate the net polarizations by using specific energy fluxes ( equation ( [ eq : band_spec ] ) ) . ) the various @xmath33 and time - averaged flux in 2 - 400 kev , @xmath41 , are investigated with fixed @xmath42 , and a burst duration of @xmath43 s. we also assume the incident angle of bursts to be 30 degree off - axis . we interpret simulated events with @xmath44 as @xmath2-measurable events. figure 1 shows the contour of the mdp values in the @xmath45 plane for grape and lep . as can be seen in the figure , with the combination of lep and grape , it is possible to measure the polarization of grbs with @xmath46 ranging from a few kev to mev with reasonable sensitivity . we calculate the linear polarization for instantaneous emission from a thin spherical shell moving radially outward with a bulk lorentz factor @xmath47 and an opening angle @xmath48 . the comoving - frame emissivity has the functional form of @xmath49 , where @xmath50 is the normalization which may depend on direction in the comoving frame and other physical quantities of the shell and @xmath51 represents the spectral shape . a prime represents the physical quantities in the comoving frame . the delta functions describe the instantaneous emission at @xmath52 and @xmath53 . the normalization , @xmath50 , has units of @xmath54 . using the spherical coordinate system @xmath55 in the lab frame , where @xmath56 is the line of sight , we obtain the spectral fluence @xcite : @xmath57 where @xmath58 and @xmath59 are the redshift and the luminosity distance of the source , respectively , and @xmath60 . the integration is performed within the jet cone , so that it depends on the viewing angle @xmath61 , i.e. , the angle between the jet axis and the line of sight . the corresponding stokes parameters of the local emission ( i.e. , the emission from a given point on the shell ) are given by @xmath62 and @xmath63 , where @xmath64 and @xmath65 are the polarization degree and position angle of the local emission measured in the comoving frame , respectively . the stokes parameters of the emission from the whole shell can be obtained by integrating those of the local emission similarly to the intensity @xmath66 : @xmath67 the polarization degree is lorentz invariant , i.e. , @xmath68 . the position angle @xmath69 is calculated by taking account of the lorentz transformation of the electromagnetic waves , and it is measured from a fixed direction , which we choose to be the direction from the line of sight to the jet axis . then by calculating @xmath70 , we obtain the time - averaged linear polarization in the given wavebands @xmath71 $ ] : @xmath72 we consider synchrotron and compton drag ( cd ) mechanisms for the grb prompt emission . in the synchrotron case , the magnetic field consists of a globally ordered field , @xmath73 , and small - scale random field , @xmath74 , i.e. , @xmath75 . the field @xmath73 may originate from the central engine , while @xmath74 may be produced in the emission region itself . here we consider two extreme cases ; synchrotron model with an ordered field ( so ) , in which @xmath76 , and a synchrotron model with a random field ( sr ) , in which @xmath77 . for the so model , in particular , we assume a toroidal magnetic field . in the following sub - sections , we describe @xmath50 , @xmath51 , @xmath78 , and @xmath69 as functions of @xmath79 for each model , and calculate the linear polarization for given parameters @xmath0 , @xmath48 , @xmath61 , and @xmath58 . the prompt emission of grbs could be explained by synchrotron emission from accelerated electrons that have a non - thermal energy spectra by some dissipation process within the jet , e.g , internal shocks . synchrotron emission from the relativistically moving shell within a globally ordered magnetic field results in a net observed linear polarization , reflecting the direction of the field @xcite . let us assume that the jet is permeated by a toroidal field . this is a likely configuration if a magnetic field is advected by the jet with a constant speed from the central engine ( e.g. , * ? ? ? * ; * ? ? ? a general formula for calculating the observed linear polarization for synchrotron emission from a uniform jet , in which the electrons have a single power - law energy spectrum and an isotropic pitch angle distribution and the magnetic field is ordered globally , is derived by @xcite and @xcite . here we adopt their formulation and extend it for the electrons having a broken power - law energy spectrum in order to reproduce the typical observed spectra of grbs @xcite . we adopt the following form for the radiation spectrum : @xmath80 where @xmath81 and @xmath82 where @xmath83 , @xmath35 , and @xmath36 are the break frequency and low - energy and high - energy spectral indices of the comoving spectrum , respectively . ) is thought to be produced by the broken power - law energy spectrum of electrons : @xmath84 for @xmath85 and @xmath86 for @xmath87 , where @xmath88 and @xmath89 . this formulation also includes the case of @xmath90 , in which @xmath91 , @xmath92 , and @xmath93 for @xmath94 @xcite . ] if we assume that the energy spectrum of the electrons and the strength of the magnetic field are uniform in the emitting shell , then we may write @xmath95 , where @xmath96 is the angle between the direction of the emitted radiation and the local direction of the magnetic field @xcite . the local polarization degree is given by : @xmath97 for a globally ordered magnetic field , the faraday depolarization effect may be strong within the emitting region ( e.g. , * ? ? * ; * ? ? ? * ; * ? ? ? * ) , but we neglect it here for simplicity . by using a new variable @xmath98 , we obtain ( see appendix [ subsec : app_so ] ) : @xmath99^{1/2 } , \label{eq : tsinthetab}\ ] ] @xmath100 where @xmath101 . then the formulation of the net polarization degree in the observed frequency region @xmath71 $ ] becomes : @xmath102^{-1 } , \end{array } \label{eq : atorob}\ ] ] where @xmath103 , and @xmath104 @xmath105 & { \rm otherwise}. \end{array } \right . \label{eq : delta_phi}\ ] ] the polarization degree , @xmath2 , in the waveband [ @xmath106 can be calculated if the geometrical parameters , @xmath107 , the spectral parameters , @xmath108 , and the redshift , @xmath58 , are given . figure [ fig : atorob ] shows the polarization degree in the @xmath109 kev band as a function of @xmath110 for several values of @xmath111 . the other parameters are @xmath112 kev , @xmath113 , @xmath114 , and @xmath115 . the polarization degree is negligible for @xmath116 , because in this case the local polarization vectors are axisymmetric around the line of sight , i.e. , @xmath117 ( see appendix [ subsec : app_so ] ) , and the local polarizations are canceled out . for @xmath118 , a high level of polarization is obtained for @xmath119 ( i.e. , @xmath120 ) . in this case , only a fraction of the emitting shell ( i.e. , @xmath121 ) is bright because of the relativistic beaming effect , and the direction of the magnetic field is quite ordered in the bright region . the contribution of the emission from high latitude , @xmath122 , is negligible , especially for @xmath123 , so that the net polarization degree is determined only by the emission from the bright region with @xmath121 and then it is nearly constant . the results of our calculations for the case of @xmath124 and @xmath123 are consistent with the results of @xcite and @xcite . for @xmath125 , a high level of polarization is obtained for @xmath126 ( i.e. , @xmath127 ) . in this case , the bright region on the emitting shell is small , also . the polarization is higher for softer spectra ( i.e. , larger @xmath35 and @xmath36 ) . for example , for @xmath128 , @xmath112 kev , and @xmath115 , the polarization degree at the plateau for @xmath129 is @xmath130 for @xmath131 and @xmath132 , while it is @xmath133 for @xmath134 and @xmath135 . this is caused mainly by the dependence of the synchrotron polarization on the spectral indices ( equation [ eq : synch_pi ] ) . the maximum polarization degree obtained in the so model is @xmath136 for @xmath137 , @xmath138 , and @xmath139 if the magnetic field is produced at the shock itself within the jet , the directions of the field would be random on a scale as small as the plasma skin depth @xcite . it is quite plausible that the directions of the field are not completely random , but have symmetry around the direction normal to the shock . the less isotropic the magnetic field directions behind the shock , the higher the local polarization . we consider the extreme case in which the field is random strictly within the plane of the shock . in this model , the directions of the local polarization vectors on the shell are axisymmetric around the line of sight ( see below ) , so that no net polarization remains if the visible region , @xmath121 , is wholly within the jet cone . however , if the observer views the jet from an off - axis angle and the symmetry is broken a high level of polarization remains @xcite . similarly to the so model , we adopt the broken power - law form of the spectrum : @xmath140 where @xmath141 and @xmath142 is given by equation ( [ eq : band_spec ] ) . we assume that the energy distribution of the electrons and the strength of the magnetic field are uniform in the emitting shell . the local stokes parameters are given by averaging them with respect to the magnetic field directions within the shock plane ( see appendix [ subsec : app_sr ] ) . thus we may write @xmath143 , where @xmath144 represents the average . the local polarization degree is given by @xmath145 , where : @xmath146^{(\alpha+1)/2},\ ] ] @xmath147^{(\alpha-1)/2 } \right . \\ \times \left . \left[\sin^2\eta ' - \left(\frac{1-y}{1+y}\right)^2 \cos^2\eta ' \right ] \right\}. \end{array}\ ] ] the local polarization position angle measured in the lab frame is given by @xmath117 , therefore we obtain the formulation for the net polarization in the observed frequency region @xmath71 $ ] : @xmath148^{-1 } , \end{array } \label{eq : arndb}\ ] ] where @xmath149 , @xmath150 , @xmath151 , and @xmath152 and @xmath153 are given by equations ( [ eq : synch_pi ] ) and ( [ eq : delta_phi ] ) , respectively . figure [ fig : arndb ] shows the polarization degree in the @xmath109 kev band as a function of @xmath110 for several values of @xmath111 . the other parameters are @xmath154 kev , @xmath155 , and @xmath115 . the results of our calculations for the case of @xmath124 are consistent with those of @xcite and @xcite . a high level of polarization is obtained for @xmath126 ( i.e. , @xmath127 ) for each value of @xmath111 . since the local polarization vectors are axisymmetric around the line of sight , the local polarizations are canceled out if the line of sight is within the jet cone . if the jet is observed from an off - axis angle , the net polarization remains . the local polarization degree is highest for emission where @xmath156 , so that the net polarization has a maximum value . the maximum @xmath2 is higher for smaller @xmath111 , because the contribution of the emission from high latitude points ( @xmath122 ) , with a low level of local polarization , is smaller . similarly to the so model , the polarization is higher for softer spectra , mainly because of the dependence of the local polarization degree on frequency ( equation [ eq : synch_pi ] ) . for example , for @xmath157 , @xmath158 kev , and @xmath115 , the maximum polarization is @xmath159 for @xmath160 and @xmath161 , while it is @xmath162 for @xmath134 and @xmath135 . for @xmath137 , @xmath138 , and @xmath163 the maximum polarization degree in the sr model is @xmath136 . the prompt emission from grbs could be produced by bulk inverse comptonization of soft photons from the relativistic jet @xcite . the local polarization position angles are symmetric around the line of sight , similarly to the sr model . therefore this model also requires an off - axis observation of the jet to achieve a high level of polarization . however , the cd model is different from the sr model in the fact that the cd model can in principle achieve @xmath164 under the most optimistic geometric configurations , whereas the maximum @xmath2 is @xmath165 in the sr model . we assume that the seed radiation is unpolarized and has a non - thermal , isotropic spectrum , and the scattered radiation has the broken power - law spectrum @xmath166 , where @xmath81 and @xmath142 is given by equation ( [ eq : band_spec ] ) . if the intensity of the seed radiation and the electron number density of the shell are assumed to be uniform then we may write @xmath167 , and @xmath168 @xcite . the polarization vectors in the comoving frame are perpendicular to both incident and scattering directions of photons , so that we obtain @xmath169 in the lab frame . therefore we achieve the formulation for the net linear polarization in the observed frequency region @xmath71 $ ] : @xmath170^{-1 } , \end{array } \label{eq : acomp}\ ] ] where @xmath149 , @xmath150 , @xmath151 , and @xmath153 is given by equation ( [ eq : delta_phi ] ) . figure [ fig : acomp ] shows the polarization degree in the @xmath109 kev band as a function of @xmath110 for several values of @xmath111 . the other parameters are @xmath154 kev , @xmath171 and @xmath115 . the results of our calculations for the case of @xmath124 are consistent with those of @xcite . the results are similar to those of the sr model , but the polarization degree is higher than in the sr model . the polarization is higher for softer spectra , although the local polarization degree is not dependent on the frequency in this model . for instance , for @xmath172 , @xmath173 kev , and @xmath115 , the maximum polarization is @xmath174 for @xmath131 and @xmath161 , while it is @xmath175 for @xmath134 and @xmath135 , but the variation is smaller than for the synchrotron models ( see [ subsec : so ] and [ subsec : sr ] ) . this variation is caused by the kinematic effect . the local polarization degree is a maximum for @xmath156 ( i.e. , @xmath176 ) . thus the net polarization is higher when the contribution of the emission from higher latitude with @xmath122 is smaller . the high latitude emission is dimmer as the radiation spectrum is softer . therefore the net polarization is higher when the spectrum is softer . this effect also arises in the so and sr models , although in those models the intrinsic dependence of polarization on the spectrum ( equation [ eq : synch_pi ] ) is rather strong ( see [ subsec : so ] and [ subsec : sr ] ) . for @xmath137 , @xmath138 , and @xmath163 the maximum polarization degree for the cd model is @xmath177 . in this section we show the results of our monte carlo simulation of the grb prompt emission polarization . first , in [ subsec : flu ] , we give the values of the model parameters so that the observed fluences and peak energies of simulated bursts are consistent with the data obtained with the _ hete-2 _ satellite . in [ subsec : sum ] , we examine the properties of the polarization distribution of bursts detectable by the _ poet _ satellite , regardless of instrument mdp . next , in [ subsec : cum ] , we show the distribution of polarizations that can be measured by _ poet _ , and discuss how we may constrain the emission models . we performed monte carlo simulations to obtain the distribution of the observed spectral energies and fluences in the three emission models . such simulations have been developed to discuss the empirical correlation between spectral peak energies in the cosmological rest frame and isotropic @xmath0-ray energies among grbs and x - ray flashes in several models of geometrical structure of grb jets @xcite . we generated 10,000 grb jets with lorentz factor , @xmath0 , and opening angle , @xmath48 , and a random viewing angle for each jet according to the probability distribution of @xmath178 with @xmath179 rad . rad in our simulation are not detected by _ hete-2 _ or _ poet _ with the parameters we adopt in this paper . we can therefore discuss the distribution of several quantities of the detectable bursts and the event rate ratio of bursts for which polarizations can be measured to the detectable bursts without considering the bursts with @xmath180 rad . ] for each burst generated we calculate the @xmath181 spectrum to obtain the spectral peak energy , @xmath33 , and the fluence , @xmath182 , in the @xmath183 kev range by using equation ( [ eq : flu ] ) . since @xmath33 s and @xmath182 s calculated for each @xmath149 in the three models are different only by factors less than 2 , @xmath33 s and @xmath182 s of the simulated bursts may be calculated using just one model , for which we chose the cd model . the distributions of @xmath0 and @xmath48 for grb jets are highly uncertain . we make a simple assumption for the distribution and in [ subsec : cum ] we perform some simulations for different assumptions . we fix @xmath184 . we assume the distribution of @xmath48 as @xmath185 where @xmath186 and @xmath187 . the value of @xmath188 is inferred from the observations of the steepening breaks ( i.e. , jet breaks ) of some optical afterglows @xcite and from analysis of _ batse _ data using some empirical relations @xcite . there are several suggestions of events with very small @xmath48 ( e.g. , * ? ? ? * ; * ? ? ? * ) , although the value of @xmath189 is highly uncertain . the spectral parameters @xmath190 and @xmath36 are assumed as follows the first two parameters are given so that the rest - frame spectral peak energies and isotropic @xmath0-ray energies calculated for a jet viewed with @xmath191 are consistent with those of typical grbs . such an on - axis emission has approximately @xmath192 and @xmath193 . the parameters @xmath194 and @xmath195 are given through the empirical relations @xmath196 erg and @xmath197 kev ( e.g. , * ? ? ? * ; * ? ? ? we assume that the coefficients @xmath198 and @xmath199 obey the log - normal distribution @xcite with averages of @xmath200 and logarithmic variances of @xmath201 and @xmath202 , respectively . the last two parameters are fixed by @xmath203 and @xmath114 , which are typical values for grb prompt emission @xcite . the distribution of the source redshift , @xmath58 , is assumed to be in proportional to the cosmic star formation rate . we adopt the model sf2 in @xcite , i.e. , the comoving grb rate density is assumed to be proportional to @xmath204 we take the standard cosmological parameters of @xmath205 and @xmath206 . figure [ fig : ep_flu ] shows the results of @xmath207 and time - averaged flux , @xmath41 . the time - averaged flux is calculated by @xmath208 , where @xmath209 is the duration of a burst . we fix @xmath43 s , which is a typical value for long grbs ( e.g. , * ? ? ? we show only the simulated bursts that have fluxes above the detectable limit of the _ hete-2 _ satellite . they are consistent with the data obtained by _ hete-2 _ the scatter of the simulated bursts is due both to the scatter of the assumed jet parameters and the viewing angle effect @xcite . ) for @xmath210 kev in the uniform jet model , but the @xmath211 diagram we derive is still consistent with the observed dataset . ] we calculated the linear polarization , @xmath2 , by using equations ( [ eq : atorob ] ) , ( [ eq : arndb ] ) , and ( [ eq : acomp ] ) to obtain the polarization distribution of the simulated bursts that can be detected by grape and lep . the detection limits of grape and lep are set to be the mdp contours of 1.0 . ( see figure [ fig : mdp ] ) . are similar but not identical to the mdp contours of 1.0 . thus our setting for the detection limits is just for simplicity . ] figures [ fig : ep_pi_1_g ] and [ fig : ep_pi_1_l ] show the @xmath212 diagrams of all the simulated bursts that can be detected by grape and lep , respectively , in the so ( _ red open circles _ ) , sr ( _ green filled circles _ ) , and cd ( _ blue plus signs _ ) models . in the so model , most of the detectable bursts have @xmath213 in the grape band ( 60 - 500 kev ) , while they have @xmath214 in the lep band ( 2 - 15 kev ) . in the sr and cd models , most of the detectable bursts have @xmath215 in both grape and lep bands . the events with @xmath216 are distributed uniformly with @xmath217 and @xmath218 for the sr and cd models , respectively . this result can be roughly explained by the polarizations calculated as functions of @xmath111 and @xmath149 for @xmath112 kev and @xmath115 ( see figures [ fig : atorob ] , [ fig : arndb ] , [ fig : acomp ] ) and the distribution of @xmath48 and @xmath110 for the detectable bursts in this simulation , shown in figure [ fig : thj_q ] . the detectable events are dominated by the events with @xmath129 , since events with @xmath129 are much brighter than those with @xmath219 because of the relativistic beaming effect . for events with @xmath219 , narrower jets are easier to detect since they have intrinsically higher emissivities by our assumption . most of the detectable events have @xmath129 and @xmath220 ( i.e. , @xmath221 ) . for these events the so model gives @xmath213 in most cases , while the sr and cd models give @xmath215 , for the grape band as shown in figures [ fig : atorob ] , [ fig : arndb ] , and [ fig : acomp ] . the remaining detectable events mainly have @xmath219 and @xmath222 ( i.e. , @xmath223 ) . these events have @xmath224 in the so model , @xmath225 in the sr model , and @xmath218 in the cd model , for the grape band as shown in figures [ fig : atorob ] , [ fig : arndb ] , and [ fig : acomp ] . the results for the lep band can be explained similarly . in all the three models , the results show @xmath226 for almost all the detectable bursts with @xmath227 . this is due to the fact that typically the contribution of the high - energy photons with spectral index @xmath36 is larger in the grape band than in the lep band . the emission with softer spectrum has higher polarization because of the intrinsic property of the synchrotron polarization ( equation [ eq : synch_pi ] ) for the so and sr models and the kinematic effect for the cd model ( see [ subsec : compton ] ) , respectively . in the so model , the polarization of grbs with @xmath129 is higher for lower @xmath33 for the grape band . this is because the contribution from high - energy photons , with energy spectral index @xmath36 , is larger . in the sr and cd models , the higher @xmath2 grbs can be obtained for smaller @xmath48 . the maximum @xmath2 is obtained for @xmath228 . we obtain the distribution of polarization that can be measured , by using the mdp values we derived for @xmath203 , @xmath114 , and @xmath43 s ( see [ sec : poet ] ) . we interpret the simulated events with @xmath44 as ` @xmath2-measurable events ' . figure [ fig : cum_1 ] shows the cumulative distribution of @xmath2 that can be measured by grape and lep in the so , sr , and cd models . we have set the number of detectable events @xmath229 . since the polarization in the lep band is lower than in the grape band for almost all the cases as discussed in [ subsec : sum ] , the number of events for which polarization can be measured by lep is smaller than for grape . in the so model , the number of @xmath2-measurable bursts is @xmath230 , and the cumulative distribution of measurable @xmath2 varies rapidly at @xmath231 for the grape band . in the sr model , @xmath232 , and the maximum polarization is @xmath233 . in the cd model , @xmath234 , and @xmath235 . to investigate general properties of the cumulative distribution that do not depend on the model parameters , we performed simulations for other values of @xmath236 @xmath237 and @xmath238 , the lorentz factor of the jets and the power - law indices of the distribution of the opening angles of the jets , respectively . we refer to the parameters adopted for the above simulation as ` typical ' parameters . we now consider a range of parameters : @xmath239 , @xmath240 and @xmath241 , which are quite reasonable for grbs ( e.g. , * ? ? ? * ; * ? ? ? . within these parameter ranges we obtain the lower ( upper ) limit of @xmath1 for the so model ( the sr / cd models ) . figure [ fig : cum_g ] shows the results for @xmath242 and the same ` typical ' values for the other parameters . the number @xmath243 is larger in the so model and smaller in the sr and cd models than the case of @xmath184 . as @xmath0 is larger , the beaming effect is stronger and the ratio of the bursts with @xmath129 for detectable bursts is larger . thus the number of bursts with a high degree of polarization is larger in the so model and smaller in the sr and cd models . figure [ fig : cum_q1 ] shows the results for @xmath244 and the same ` typical ' values for the other parameters . since the ratio of the number of the bursts with smaller @xmath111 to that of detectable bursts is smaller , @xmath243 is slightly smaller than that for the ` typical ' parameters in the sr and cd models . figure [ fig : cum_q2 ] shows the results for @xmath245 and the ` typical ' values for the other parameters . in this case @xmath243 is slightly larger than that for the ` typical ' parameters in the sr and cd models . the number @xmath243 in the so model is similar for figure [ fig : cum_1 ] , [ fig : cum_q1 ] , and figure [ fig : cum_q2 ] in the grape band . to summarize , for the parameters @xmath246 , @xmath247 , @xmath241 , @xmath203 and @xmath114 , we can say that @xmath3 for grape and the cumulative distribution of measurable @xmath2 varies rapidly from @xmath231 in the so model . for the sr model , @xmath248 for grape , with a maximum polarization @xmath233 . for the cd model , @xmath4 for grape , and @xmath235 . since the dependence of the polarization degree on the spectral indices is relatively large in the so and sr models , we should take account of the distribution of @xmath35 and @xmath36 . the observed spectral parameters @xmath35 and @xmath36 are distributed roughly as @xmath249 and @xmath250 @xcite . within these ranges of @xmath35 and @xmath36 , the polarization degree for @xmath251 , @xmath129 , and @xmath252 kev is @xmath253 in the so model . thus the measurable polarizations are clustered at @xmath253 . the maximum polarization obtained in the so model for @xmath254 and @xmath255 is @xmath136 ( see [ subsec : so ] ) . in this case @xmath1 will be larger than 30% . in the cd model , the result will not be significantly different from the case of fixed @xmath35 and @xmath36 . in the sr model , the polarization degree does not exceed those calculated in the cd model , and thus @xmath4 . the maximum polarization obtained in the sr model for @xmath254 and @xmath255 is @xmath136 ( see [ subsec : sr ] ) . in conclusion , we can constrain the emission mechanism of grbs by using the cumulative distribution obtained by grape . if @xmath3 , the sr and cd models may be ruled out , and in this case if the measured polarizations are clustered at @xmath253 , the so model will be favored . if @xmath4 , the so model may be ruled out , but we can not distinguish between the sr and cd models with different distributions of @xmath111 , @xmath35 , and @xmath36 . if several bursts with @xmath5 are detected , however , the cd model which includes adequate number of small @xmath111 bursts will be favored . recently there has been an increasing interest in the measurement of x - ray and @xmath0-ray polarization , and the observational techniques can now achieve significant sensitivity in the relevant energy bands . several polarimetry mission concepts , such as _ poet _ , are being planned . the _ poet _ concept has two polarimeters , grape ( 60 - 500 kev ) and lep ( 2 - 15 kev ) both of which have wide fields of view . if launched , missions of this type would provide the first definitive detection of the polarization of grb prompt emission . this would enable the discussion of the statistical properties of the polarization degree and polarization spectra , which will give us diagnostic information on the emission mechanism of grbs and the nature of the grb jets that can not be obtained from current spectra and lightcurve observations . we have performed monte carlo simulations of the linear polarization from grb jets for three major emission models : synchrotron model with globally ordered magnetic field ( so model ) , synchrotron model with small - scale random magnetic field ( sr model ) , and compton drag model ( cd model ) . we assumed that the physical quantities for the emission of the jets are uniform on the emitting surface and that the jets have sharp edges . our jet angle distribution allows the detections of grbs with very small opening angles ( i.e. , smaller than 1 degree ) as suggested by several _ bursts @xcite . we have shown that the _ poet _ mission or other polarimeters with similar capabilities , i.e. , broadband spectral capabilities for the determination of @xmath33 and sensitive broadband polarimetric capabilities to minimize mdp , can constrain the emission models of grbs . furthermore , these simulations indicate that an increase in the lep effective area would be beneficial to compensate for the lower expected polarization at lower energies . as shown in figures [ fig : atorob ] , [ fig : arndb ] , and [ fig : acomp ] , the sr and cd models require off - axis observations of the jets to achieve a high level of polarization , while the so model does not . in this sense the sr and cd models are categorized as _ geometric _ models , and the so model as an _ intrinsic _ model @xcite . the distribution of observed polarizations obtained by our simulations show that the geometric sr / cd models will be ruled out if the number ratio of the @xmath2-measurable bursts to detected bursts is larger than @xmath256 , and in this case the so model will be favored if the measurable polarizations are clustered at @xmath253 . if the number ratio is smaller than @xmath257 , the so model may be ruled out , but we can not distinguish between the sr and cd models with different distributions of @xmath150 , @xmath35 , and @xmath36 , where @xmath0 and @xmath48 are the bulk lorentz factor and the opening angle of the grb jet , respectively , and @xmath35 and @xmath36 are lower and higher indices of the energy spectrum . however , if several bursts with @xmath5 are detected , the cd model which includes an adequate number of small @xmath111 bursts will be favored . if the cumulative distribution of the measurable polarizations favors the so model , the globally ordered magnetic field would be advected from the central engine . if we understand the strength of the magnetic field in the emitting region from the luminosity and the spectrum of the emission , we can constrain the strength of the field at the central engine . if the geometric sr / cd models are favored from the observations , it will be established , independently of the afterglow observations , that grb outflows are not spherical but highly collimated . if the cd model is favored by the observations , we may constrain the distribution of the parameter @xmath258 of grb jets . the cd model needs a dense optical / uv photon field interacting within the relativistic jets @xcite . we have made some simplifications in our simulations , and there are some caveats . we have assumed that the jets are uniform on the emitting surfaces and have sharp edges . to compare the simulations and the observations further , more sophisticated modeling is required ( e.g. , * ? ? ? * ; * ? ? ? we have interpreted bursts as a simple combination of pulses , without taking account of the temporal variation of the lorentz factor @xmath0 of the jet . if this is accounted for , each pulse may have different @xmath150 but the same @xmath149 . we should then average the polarization with respect to fluence of each pulse having different @xmath111 @xcite . however , in the so model , the cumulative distribution of measurable @xmath2 will not be changed significantly as long as @xmath259 , because @xmath2 is clustered into a small range for @xmath129 and @xmath251 . to average the polarization in the case of @xmath219 , the relation between the luminosity and the lorentz factor for each pulse is required to predict the polarization distribution . for the sr model we have assumed that the directions of the magnetic field are confined within the shock plane . they may be more isotropic in reality , in which case the polarization degree in the sr model will be reduced . in the synchrotron model with a combination of the globally ordered magnetic field and the locally random field , @xmath260 , the linear polarization can be calculated by @xmath261 , where @xmath262 and @xmath263 are the stokes parameters from the ordered and random fields , respectively . @xmath264 and @xmath265 are described by equations ( [ eq : atorob ] ) and ( [ eq : arndb ] ) , and @xmath266 . this model will reduce the number ratio of @xmath2-measurable bursts to detected bursts to less than 30% and the clustering of measurable polarizations will be at @xmath267 . this work is supported in part by the grant - in - aid from the ministry of education , culture , sports , science and technology ( mext ) of japan , no.19047004 ( ry , ki , and tn ) , no.18740153 ( ry ) , no.18740147 ( ki ) , and in part by the grant - in - aid for the global coe program _ the next generation of physics , spun from universality and emergence _ from the ministry of education , culture , sports , science and technology ( mext ) of japan . bz acknowledges nasa nng05gb67 g and nnx08ae57a ( nevada nasa epscor program ) and kt acknowledges nasa nnx08al40 g for partial supports . amati , l. 2002 , a&a , 390 , 81 band , d. l. , et al . 1993 , apj , 413 , 281 begelman , m. c. , & sikora , m. 1987 , apj , 322 , 650 black , j. k. , et al . 2007 , nima , 581 , 755 bloser , p. f. , et al . 2006 , chjas , 6 , 393 bloser , p. f. , et al . 2008 , accepted in nima ( arxiv:0812.0782 ) boggs , s. e. , et al . 2006 , arxiv : astro - ph/0608532 coburn w. , & boggs , s. e. 2003 , nature , 423 , 415 costa , e. , et al . 2007 , in proc . of spie vol.6686 , 30 covino , s. , ghisellini , g. , lazzati , d. , & malesani , d. 2004 , in asp conf . 312 , 169 dai , x. , & zhang , b. 2005 , apj , 621 , 875 donaghy , t. q. 2006 , apj , 645 , 436 eichler , d. , & levinson , a. 2003 , apj , 596 , l147 fendt , c. , & ouyed , r. 2004 , apj , 608 , 378 frail , d. a. , et al . 2001 , apj , 562 , l55 genet , f. , daigne , f. , & mochkovitch , r. 2007 , mnras , 381 , 732 ghisellini , g. , & lazzati , d. 1999 , mnras , 309 , l7 ghisellini , g. , ghirlanda , g. , nava , l. , & firmani , c. 2007 , apj , 658 , l75 granot , j. , piran t. , & sari , r. 1999 , apj , 513 , 679 granot , j. 2003 , apj , 596 , l17 granot , j. , & taylor , g. b. 2005 , apj , 625 , 263 greiner , j. , et al . 2008 , in aip conference proceedings , volume 1000 , 620 . gruzinov , a. , & waxman , e. 1999 , apj , 511 , 852 hill , j. e. , et al . 2008 , in proc . of 2008 nanjing grb conference ( arxiv : 0810.2499 ) hjorth , j. 2003 , nature , 423 , 847 ioka , k. , & nakamura , t. 2001 , apj , 554 , l163 ioka , k. , & nakamura , t. 2002 , apj , 570 , l21 ioka , k. , et al . 2007 , apj , 670 , l77 jahoda , k. , et al . 2007 , arxiv : astro - ph/0701090 jason , l. , et al . 2005 , in proc . of spie vol.5898 , 413 kalemci , e. , et al . 2007 , apjs , 169 , 75 lazzati , d. , ghisellini , g. , celotti , a. , & rees , m. j. 2000 , apj , 529 , l17 lazzati , d. , rossi , e. , ghisellini , g. , rees , m. j. 2004 , mnras , 347 , l1 lazzati , d. 2006 , new journal of physics , 8 , 131 levinson , a. , & eichler , d. 2004 , apj , 613 , 1079 lithwick , y. , & sari , r. 2001 , apj , 555 , 540 lyutikov , m. , pariev , v. i. , & blandford , r. d. 2003 , apj , 597 , 998 matsumiya , m. , & ioka , k. 2003 , apj , 595 , l25 mcglynn , s. , et al . 2007 , a&a , 466 , 895 medvedev , m. v. , & loeb , a. 1999 , apj , 526 , 697 mszros , p. 2006 , rep . , 69 , 2259 mizuno , t. , et al . 2005 , nucl . instrum . methods . a , 540 , 158 mundell , c. g. , et al . 2007 , science , 315 , 1822 nakar , e. , piran , t. , & waxman , e. 2003 , jcap , 10 , 5 panaitescu , a. , & mszros , p. 2000 , apj , 544 , l17 piran , t. 2005 , rev . , 76 , 1143 porciani , c. , & madau , p. 2001 , apj , 548 , 522 preece , r. d. , et al . 2000 , apjs , 126 , 19 produit , n. , et al . 2005 , nucl . instrum . methods . res . sec . a , 550 , 616 racusin , j. l. , et al . 2008 , nature , 455 , 183 rees , m. j. , & mszros , p. 1994 , apj , 430 , l93 rutledge , r. e. , & fox , d. b. 2004 , mnras , 350 , 1288 rybicki , g. b. , & lightman , a. p. 1979 , radiative processes in astrophysics ( new york : wiley interscience ) ryde , f. , et al . 2006 , apj , 652 , 1400 sagiv , a. , waxman , e. , & loeb , a. 2004 , apj , 615 , 366 sakamoto , t. , et al . 2005 , apj , 629 , 311 sari , r. 1999 , apj , 524 , l43 schady , p. , et al . 2007 , mnras , 380 , 1041 shaviv , n. j. , & dar , a. 1995 , apj , 447 , 863 spruit , h. c. , daigne , f. , & drenkhahn , g. 2001 , a&a , 369 , 694 stanek , k. z. , et al . 2003 , apj , 591 , l17 taylor , g. b. , frail , d. a. , berger , e. , & kulkarni , s. r. 2004 , apj , 609 , l1 taylor , g. b. , momjian , e. , philstrom , y. , ghosh , t. , & salter , c. 2005 , apj , 622 , 986 thompson , c. , mszros , p. , & rees , m. j. 2007 , apj , 666 , 1012 toma , k. , yamazaki , r. , nakamura , t. 2005 , apj , 635 , 481 toma , k. , ioka , k. , & nakamura , t. 2008 , apj , 673 , l123 uhm , z. l. , & beloborodov , a. m. 2007 , apj , 665 , l93 waxman , e. 2003 , nature , 423 , 388 wigger , c. , et al . 2004 , apj , 613 , 1088 willis , d. r. , et al . 2005 , a&a , 439 , 245 woods , e. , & loeb , a. 1999 , apj , 523 , 187 yamazaki , r. , ioka , k. , & nakamura , t. 2004 , apj , 606 , l33 yonetoku , d. , yamazaki , r. , nakamura , t. , & murakami , t. 2005 , mnras , 362 , 1114 zeh , a. , klose , s. , & kann , d. a. 2006 , apj , 637 , 889 zhang , b. , & mszros , p. 2002 , apj , 581 , 1236 zhang , b. , dai , x. , lloyd - ronning , n. m. , & mszros , p. 2004 , apj , 601 , l119 zhang , b. 2007 , chin . j. astron . astrophys . , we consider the synchrotron radiation from the shell moving radially outward with a bulk lorentz factor @xmath47 , and the magnetic field in the shell is globally ordered within the plane parallel to the shock plane . if the matter of the shell expands with a constant speed , the strength of magnetic field with radial direction scales as @xmath268 while that with transverse direction scales as @xmath269 . thus the field advected with the shell is likely to have the direction parallel to the shock plane . we set the line of sight ( i.e. , the direction from the central engine to the earth ) in the lab frame to be @xmath58 axis , and the direction of the magnetic field on a given point of the shell , projected onto the plane perpendicular to @xmath58 axis , to be @xmath270 axis . the given point can be described by spherical coordinates @xmath271 . then the components of the velocity vector of the given point and the unit wave vector can be described by the right - handed coordinate system @xmath272 as @xmath273 and @xmath274 , respectively . the unit wave vector in the comoving frame is @xmath275.\ ] ] since the direction of the magnetic field in the comoving frame is perpendicular to the velocity vector of the fluid , @xmath276 . then we may calculate @xmath277 , and we obtain @xmath278 in the limit @xmath47 . the direction of the polarization vector of the synchrotron radiation is calculated by @xmath279 . then we obtain the direction of the polarization vector in the lab frame by @xmath280 the results are @xmath281 and @xmath282 the angle @xmath283 is the polarization position angle measured from the axis @xmath270 ( i.e. , the direction of the local magnetic field ) . the above equation can be rewritten as @xmath284 $ ] . this result is consistent with that of @xcite . based on the above results , we consider the case that the magnetic field is axisymmetric around the jet and has a toroidal configuration . we set the direction from the line of sight to the jet axis to be @xmath285 axis . below we will rewrite the above results by using the azimuthal angle @xmath286 measured from @xmath285 axis . in the coordinate system of @xmath287 , the jet axis and the coordinates of a given point on the shell are described as @xmath288 , and @xmath289 , respectively . the magnetic field at the given point is given by @xmath290 . let the unit vectors of the directions of @xmath291 and @xmath292 projected onto @xmath293 plane be @xmath294 and @xmath295 , and @xmath296 . then we obtain @xmath297 where @xmath298 . equation ( [ eq : tsinthetab ] ) is given by inserting equation ( [ eq : cos2phi ] ) into equation ( [ eq : sinthetab ] ) . if we measure the position angle from the @xmath285 axis , we obtain equation ( [ eq : tchi ] ) , i.e. , @xmath299 . these results are consistent with those of @xcite . here we consider that the directions of the magnetic fields are confined within the plane parallel to the shock plane and that they are completely random . this field configuration is possible if the field is generated by the shock . in the comoving frame of the shell , we set the direction of @xmath300 to be axis @xmath301 , and set a right - handed coordinate system @xmath302 . let the polar and azimuthal angles of @xmath292 be @xmath96 and @xmath303 , respectively . in this coordinate system , the stokes parameters of synchrotron emissivity are given by @xmath304 next we set another right - handed coordinate system @xmath287 of which @xmath58 axis is along the velocity vector of the fluid and @xmath305 plane includes @xmath306 . then the angle between @xmath306 and @xmath58 axis is @xmath307 . here the magnetic field @xmath308 is confined within @xmath293 plane . let the azimuthal angle of @xmath308 be @xmath309 , and we obtain the relations between the components of @xmath308 in the systems @xmath302 and @xmath287 . @xmath310 then we obtain @xmath311^{1/2 } , \\ \cos(2\phi'_b ) = \frac{1}{\sin^2\theta'_b } \left [ \sin^2\eta ' - \left(\frac{1-\gamma^2\theta^2}{1+\gamma^2\theta^2 } \right)^2 \cos^2\eta ' \right ] . \end{array}\ ] ] to obtain the polarization degree of synchrotron radiation from the random field , we average the stokes parameters with respect to @xmath309 . this leads to @xmath312 . then we can calculate the polarization degree by @xmath313 , and the polarization vector is along axis 1 , i.e. , the direction perpendicular to @xmath306 and within the plane including @xmath306 and @xmath314 .

we obtained new optical spectra of 13 regions in wlm with efosc2 ; oxygen abundances are derived for nine regions . the temperature - sensitive emission line was measured in two bright regions hm 7 and hm 9 . the direct oxygen abundances for hm 7 and hm 9 are 12log(o / h ) = and , respectively . we adopt a mean oxygen abundance of 12log(o / h ) = . this corresponds to [ o / h ] = dex , or 15% of the solar value . in regions where was not measured , oxygen abundances derived with bright - line methods are in general agreement with direct values of the oxygen abundance to an accuracy of about 0.2 dex . in general , the present measurements show that the region oxygen abundances agree with previous values in the literature . the nebular oxygen abundances are marginally consistent with the mean stellar magnesium abundance ( [ mg / h ] = ) . however , there is still a 0.62 dex discrepancy in oxygen abundance between the nebular result and the a - type supergiant star wlm 15 ( [ o / h ] = ) . non - zero reddening values derived from balmer line ratios were found in regions near a second peak . there may be a connection between the location of the second peak , regions of higher extinction , and the position of wlm 15 on the eastern side of the galaxy .

astro-ph0410640

dwarf galaxies are thought to be the building blocks in the assembly of more massive galaxies within the hierarchical picture of structure formation . these galaxies are also very important venues in which questions about cosmology , galaxy evolution , and star formation may be answered . dwarf irregular galaxies are relatively low - mass , gas - rich , metal - poor , and are presently forming stars as shown by their regions , whereas low - mass dwarf spheroidal galaxies are gas - poor and no longer host present - day star - forming events . the properties of these galaxies may be similar to those found in the early universe , and dwarf irregulars may possibly be sites out of which damped lyman-@xmath7 absorber systems form at high redshift ( e.g. , @xcite ) . an important question which has yet to be fully explained is the relationship between dwarf irregular and dwarf spheroidal galaxies ( e.g. , @xcite , and references therein ) . that streams have been observed within the galaxy and m 31 ( e.g. , @xcite ) has been taken as evidence of ongoing accretion and of representing past merging of dwarfs by the more massive galaxies . however , work presented by @xcite and @xcite have shown that stars in present - day dwarf spheroidals can not make up the dominant stellar populations in the halo , bulge , or the thick disk of the galaxy , although the merging of dwarf galaxies at very early times can not be ruled out . the measurements of element abundances provide important clues to understanding the chemical history and evolution of galaxies . in star - forming dwarf galaxies , the analysis of bright nebular emission lines from the spectra of regions is used to derive abundances of @xmath7-elements ( i.e. , oxygen ) in the ionized gas ( see e.g. , @xcite ) . however , a limited number of elements can be studied by comparison to the number of elements found in the absorption spectra of stars . for a more complete picture , additional elements should be included , since various elements arise from different sites and involve different timescales . oxygen and other @xmath7-elements are created in very massive progenitor stars before being returned to the interstellar medium ( ism ) on short timescales , when these stars explode as type ii supernovae . iron is an element produced by explosive nucleosynthesis in type i supernovae from low - mass progenitor stars on longer timescales , and is also produced in type ii supernovae . because of the varying timescales for stars of different masses , the @xmath7 element - to - iron abundance ratio , [ @xmath7/fe ] , = log(x / y ) @xmath8 log(x / y)@xmath9 . ] is tied very strongly with the star formation history ( e.g. , @xcite ) . interestingly , [ @xmath7/fe ] values for three dwarf irregular galaxies are near or at solar , which indicates that stars have been forming at a very low rate and/or the last burst of star formation occurred long ago @xcite . @xcite claim that o / fe is elevated in low metallicity blue compact dwarf galaxies ( [ o / fe ] = @xmath10 ) . however , their analysis does not account for potential depletion of fe onto dust grains , and the fe abundance is only measured in fe@xmath11 , requiring very large and uncertain ionization correction factors ( icfs ) . @xcite finds that the adopted icfs underestimate the total fe abundance by factors larger than the elevated abundance ratio claimed by @xcite . thus , it is prudent to assume that the nebular fe abundances in these galaxies , and thus the nebular o / fe ratios , are quite uncertain @xcite . at present , reliable o / fe ratios will need to be obtained from stellar abundances . while a complete discussion of @xmath7/fe values is beyond the scope of the present work , brief reviews of stellar abundances in external galaxies have recently been presented by @xcite and @xcite . high efficiency spectrographs on 8- and 10-m telescopes have made possible the spectroscopic measurements of individual stars in extragalactic systems . in particular , bright blue supergiants have been observed in galaxies at distances of about 1 mpc . these hot young massive stars allow us to measure simultaneously present - day @xmath7- and iron - group elements . the important advantage of these measurements also allow for the direct comparison of stellar @xmath7-element abundances with nebular measurements , as massive stars and nebulae are similar in age and have similar formation sites . oxygen abundances derived from the spectroscopy of blue supergiants have been obtained in nearby dwarf irregular galaxies ngc 6822 , wlm , and sextans a @xcite . the relative ease with which spectra of regions have been obtained in dwarf irregular galaxies has led to establishing : ( 1 ) the metallicity - luminosity relation , thought to be representative of a mass - metallicity relation for dwarf irregular galaxies ( e.g. , @xcite ) ; and ( 2 ) the metallicity - gas fraction relation , which represents the relative conversion of gas into stars , and may be strongly affected by the galaxies surrounding environment ( e.g. , @xcite ) . it is assumed that nebular oxygen abundances are representative of the present - day ism metallicity for an entire dwarf galaxy , where there is often only a single region present . in fact , spatial inhomogeneities or radial gradients in oxygen abundances have been found to be very small or negligible in nearby dwarf irregular galaxies ( e.g. , @xcite ) , although recent observations have cast uncertainty about the assumption in ngc 6822 and wlm @xcite . here we will focus on oxygen abundances , and the comparison between stellar and nebular determinations . for the remainder of this paper , we adopt 12@xmath0log(o / h ) = 8.66 as the solar value for the oxygen abundance @xcite . wlm ( wolf - lundmark - melotte ) is a dwarf irregular galaxy at a distance of 0.95 mpc @xcite and is located in the local group . the galaxy was discovered by @xcite , and independently rediscovered by lundmark and melotte @xcite . wlm is relatively isolated , as the nearest neighbor about 175 kpc distant is the recently discovered cetus dwarf spheroidal galaxy @xcite . basic properties of the galaxy are listed in table [ table_wlm ] . a number of observations are summarized here . @xcite identified two planetary nebulae in the galaxy , and @xcite identified the brightest blue and red supergiant stars , including over 30 variable stars . ground - based optical photometry of stars were obtained by @xcite , and @xcite . the presence of a single globular cluster was established , and @xcite showed that the properties of the globular cluster are similar to those of galactic globular clusters . in independent imaging programs , @xcite detected two small shell - like features , and @xcite cataloged and measured fluxes for 21 regions in the galaxy . @xcite presented velocity fields for the brightest regions in wlm , and showed that the southern ring is expanding at a speed of 20 km s@xmath12 and that the kinetic age of the bubble is 4.5 myr . recent studies of the resolved stellar populations with the hubble space telescope ( _ hst _ ) have been carried out by @xcite with the wide field planetary camera 2 and by @xcite with the space telescope imaging spectrograph . @xcite found that over half of the stars were formed about 9 gyr ago , and that a recent burst of star formation has mostly occurred in the central bar of the galaxy . @xcite identified the horizontal branch , also confirming the presence of a very old stellar population . in the carbon star survey by @xcite , they found that wlm contained the largest fraction of carbon - to - m stars for the dwarf galaxies surveyed , and showed that wlm is an inclined disk galaxy with no evidence of an extended spherical stellar halo . @xcite searched for molecular gas in wlm , but only upper limits to the co intensity and subsequent h@xmath13 column densities were determined . recent 21-cm measurements with the australia telescope compact array have shown that there are two peaks in the distribution , and that the measured rotation curve is typical for a disk @xcite . the spectroscopy of the brightest regions were reported by @xcite and @xcite . the resulting nebular oxygen abundances were found to be 12@xmath0log(o / h ) @xmath14 , or [ o / h ] @xmath15 . @xcite measured the chemical composition of two a - type supergiant stars in wlm , and showed that the mean stellar magnesium abundance was [ mg / h ] = @xmath5 . however , the oxygen abundance in one of the stars was [ o / h ] = @xmath6 , which is about 0.7 dex or almost five times larger than the nebular abundance . this presents a vexing question : how can the young supergiant be significantly more metal - rich than the surrounding ism from which the star was born ? the research reported here is part of a program to understand the chemical evolution from the youngest stellar populations in the nearest dwarf irregular galaxies ( e.g. , @xcite ) . the motivations are : ( 1 ) to obtain a homogeneous sample of abundance measurements for regions presently known in wlm ; ( 2 ) to measure the temperature - sensitive emission line , derive direct oxygen abundances , and compare the present set of measurements with those in the literature ; and ( 3 ) to examine whether the present measurements show any inhomogeneities in oxygen abundances across the galaxy . this is the first of two papers of our study ; the measurements and analyses for regions in ngc 6822 will be discussed in the next paper ( h. lee et al . , in preparation ) . the outline of this paper is as follows . observations and reductions of the data are presented in sect . [ sec_obs ] . measurements and analyses are discussed in sect . [ sec_analysis ] , and nebular abundances are presented in sect . [ sec_abund ] . our results are discussed in sect . [ sec_discuss ] , and a summary is given in sect . [ sec_concl ] . long - slit spectroscopic observations of regions in wlm were carried out on 2003 aug . 2628 and 31 ( ut ) with the eso faint object spectrograph and camera ( efosc2 ) instrument on the 3.6-m telescope at eso la silla observatory . details of the instrumentation employed and the log of observations are listed in tables [ table_obsprops ] and [ table_obslog ] , respectively . observing conditions were obtained during new moon phase . conditions varied from photometric ( 26 aug ut ) to cloudy ( 31 aug ut ) . two - minute acquisition images were obtained in order to set an optimal position angle of the slit , so that the slit could cover as many regions possible . thirteen regions for which spectra were obtained are listed in table [ table_obslog ] and shown in fig . [ fig_h2rs ] . identifications for the regions follow from the imaging compiled by @xcite . for completeness , we provide here coordinates ( epoch j2000 ) for regions which were `` newly '' resolved in images obtained by the local group survey . hm 16 was resolved into two separate regions , which we have called hm 16 nw ( @xmath16 00@xmath1701@xmath18594 ; @xmath19 ) , and hm 16 se ( @xmath20 00@xmath1701@xmath18596 ; @xmath21 ) . to the east of region hm 18 , we took spectra of two additional compact regions : hm 18a ( @xmath20 00@xmath1701@xmath18597 ; @xmath22 ) and hm 18b ( @xmath20 00@xmath1701@xmath18599 ; @xmath23 ) . data reductions were carried out in the standard manner using iraf routines . data obtained for a given night were reduced independently . the raw two - dimensional images were trimmed and the bias level was subtracted . dome flat exposures were used to remove pixel - to - pixel variations in response . twilight flats were acquired at dusk each night to correct for variations over larger spatial scales . to correct for the `` slit function '' in the spatial direction , the variation of illumination along the slit was taken into account using dome and twilight flats . cosmic rays were removed in the addition of multiple exposures for a given region . wavelength calibration was obtained using helium - argon ( he - ar ) arc lamp exposures taken throughout each night . exposures of standard stars feige 110 , g138@xmath831 , ltt 1788 , ltt 7379 , and ltt 9491 were used for flux calibration . the flux accuracy is listed in table [ table_obslog ] . final one - dimensional spectra for each region were obtained via unweighted summed extractions . emission - line strengths were measured using software developed by m. l. mccall and l. mundy ; see @xcite . was detected in regions hm 7 and hm 9 ; these spectra are shown in fig . [ fig_specall ] . corrections for reddening and for underlying absorption and abundance analyses were performed with snap ( spreadsheet nebular analysis package , @xcite ) . balmer fluxes were first corrected for underlying balmer absorption with an equivalent width 2 @xcite . and fluxes were used to derive reddening values , @xmath24 ) , using the equation @xmath25 \label{eqn_corrthesis}\ ] ] @xcite . @xmath26 and @xmath27 are the observed flux and corrected intensity ratios , respectively . intrinsic case - b balmer line ratios determined by @xcite were assumed . @xmath28 is the extinction in magnitudes for @xmath29 , i.e. , @xmath30 , where @xmath31 is the monochromatic extinction in magnitudes . values of @xmath32 were obtained from the @xcite reddening law as defined by a ratio of the total to selective extinction , @xmath33 = @xmath34 = 3.07 , which in the limit of zero reddening is the value for an a0v star ( e.g. , vega ) with intrinsic color @xmath35 . because lines were generally unresolved , @xmath36 = 100 @xmath37 was adopted for the electron density . errors in the derived @xmath38 were computed from the maximum and minimum values of the reddening based upon @xmath39 errors in the fits to emission lines . observed flux @xmath40 and corrected intensity @xmath41 ratios are listed in tables [ table_data1 ] to [ table_data4 ] inclusive . the listed errors for the observed flux ratios at each wavelength @xmath42 account for the errors in the fits to the line profiles , their surrounding continua , and the relative error in the sensitivity function stated in table [ table_obslog ] . errors in the corrected intensity ratios account for maximum and minimum errors in the flux of the specified line and of the reference line . at the reference line , errors for both observed and corrected ratios do not include the error in the flux . also given for each region are : the observed flux , the equivalent width of the line in emission , and the derived reddening from snap . where is measured , we also have performed the additional computations to check the consistency of our results . equation ( [ eqn_corrthesis ] ) can be generalized and rewritten as @xmath43 where @xmath44 is the logarithmic extinction at , and @xmath45 is the wavelength - dependent reddening function @xcite . from equations ( [ eqn_corrthesis ] ) and ( [ eqn_corr ] ) , we obtain @xmath46 the reddening function normalized to is derived from the @xcite reddening law , assuming @xmath33 = 3.07 . as described in @xcite , values of @xmath44 were derived from the error weighted average of values for @xmath47 , @xmath48 , and @xmath49 ratios , while simultaneously solving for the effects of underlying balmer absorption with equivalent width ew@xmath50 . we assumed that ew@xmath50 was the same for , , , and . uncertainties in @xmath44 and ew@xmath50 were determined from monte carlo simulations @xcite . errors derived from these simulations are larger than errors quoted in the literature by either assuming a constant value for the underlying absorption or derived from a @xmath51 analysis in the absence of monte carlo simulations for the errors ; fig . [ fig_monte ] shows an example of these simulations for region hm 9 . in tables [ table_data1 ] to [ table_data3 ] , we included the logarithmic reddening and the equivalent width of the underlying balmer absorption , which were solved simultaneously . values for the logarithmic reddening are consistent with values of the reddening determined with snap . where negative values were derived , the reddening was set to zero in correcting line ratios and in abundance calculations . oxygen abundances in regions were derived using three methods : ( 1 ) the direct method ( e.g. , @xcite ) ; and two bright - line methods discussed by ( 2 ) @xcite , which is based on photoionization models ; and ( 3 ) @xcite , which is purely empirical . for the `` direct '' conversion of emission - line intensities into ionic abundances , a reliable estimate of the electron temperature in the ionized gas is required . to describe the ionization structure of regions , we adopt a two - zone model , with a low- and a high - ionization zone characterized by temperatures @xmath52o@xmath53 and @xmath52o@xmath54 , respectively . the temperature in the o@xmath11 zone is measured with the emission - line ratio @xmath27()/@xmath27 ( ) @xcite . the temperature in the o@xmath55 zone is given by @xmath56 ) = 0.7 \ , t_e([{\rm o\;iii } ] ) + 0.3 , \label{eqn_toplus}\ ] ] where @xmath57 k @xcite . the uncertainty in @xmath52o@xmath54 is computed from the maximum and minimum values derived from the uncertainties in corrected emission line ratios . the computation does not include uncertainties in the reddening ( if any ) , the uncertainties in the atomic data , or the presence of temperature fluctuations . the uncertainty in @xmath58(o@xmath55 ) is assumed to be the same as the uncertainty in @xmath58(o@xmath11 ) . these temperature uncertainties are conservative estimates , and are likely overestimates of the actual uncertainties . for subsequent calculations of ionic abundances , we assume the following electron temperatures for specific ions @xcite : @xmath59(n@xmath55 ) = @xmath59(o@xmath55 ) , @xmath59(ne@xmath11 ) = @xmath59(o@xmath11 ) , @xmath59(ar@xmath11 ) = 0.83 @xmath59(o@xmath11 ) @xmath0 0.17 , and @xmath59(ar@xmath60 ) = @xmath59(o@xmath11 ) . the total oxygen abundance by number is given by o / h = o@xmath61/h @xmath0 o@xmath55/h @xmath0 o@xmath11/h @xmath0 o@xmath60/h . for conditions found in typical regions and those presented here , very little oxygen in the form of o@xmath61 is expected , and is not included here . in the absence of he ii emission , the o@xmath60 contribution is considered to be negligible . ionic abundances for o@xmath55/h and o@xmath11/h were computed using o@xmath55 and o@xmath11 temperatures , respectively , as described above . measurements of the line were obtained and subsequent electron temperatures were derived in hm 7 and hm 9 . ionic abundances and total abundances are computed using the method described by @xcite . with snap , oxygen abundances were derived using the five - level atom approximation @xcite , and transition probabilities and collision strengths for oxygen from @xcite , @xcite , @xcite , and @xcite ; balmer line emissivities from @xcite were used . derived ionic and total abundances are listed in tables [ table_abund1 ] and [ table_abund2 ] . these tables include derived o@xmath55 and o@xmath11 electron temperatures , o@xmath55 and o@xmath11 ionic abundances , and the total oxygen abundances . errors in direct oxygen abundances computed with snap have two contributions : the maximum and minimum values for abundances from errors in the temperature , and the maximum and minimum possible values for the abundances from propagated errors in the intensity ratios . these uncertainties in oxygen abundances are also conservative estimates . using the method described by @xcite , we recompute oxygen abundances in regions with detections . abundances are computed using the emissivities from the five - level atom program by @xcite . as described above , we use the same two - temperature zone model and temperatures for the remaining ions . the error in @xmath58(o@xmath11 ) is derived from the uncertainties in the corrected emission - line ratios , and does not include any uncertainties in the atomic data , or the possibility of temperature variations within the o@xmath11 zone . the fractional error in @xmath58(o@xmath11 ) is applied similarly to @xmath58(o@xmath55 ) to compute the uncertainty in the latter . uncertainties in the resulting ionic abundances are combined in quadrature for the final uncertainty in the total linear ( summed ) abundance . the adopted abundances and their uncertainties computed in this manner are listed in tables [ table_abund1 ] and [ table_abund2 ] . direct oxygen abundances computed with snap are in excellent agreement with direct oxygen abundances computed with the method described by @xcite ; abundances from the two methods agree to within 0.02 dex . direct oxygen abundances for regions hm 7 and hm 9 are 12@xmath0log(o / h ) = @xmath1 , and @xmath2 , respectively ; the latter is a weighted mean of the three measured values shown in table [ table_abund2 ] . the mean oxygen abundance for these two regions is ( o / h ) = @xmath62 , or 12@xmath0log(o / h ) = @xmath3 . this value corresponds to [ o / h ] = @xmath4 , or 15% of the solar value . for historical completeness , we note that our derived nebular oxygen abundance would correspond to [ o / h ] = @xmath63 for the @xcite value of the solar oxygen abundance . for regions without measurements , secondary methods are necessary to derive oxygen abundances . the bright - line method is so called because the oxygen abundance is given in terms of the bright [ o ii ] and [ o iii ] lines . @xcite suggested using @xmath64$\lambda 3727 $ } } } ) + i({{\hbox{[\ion{o}{3}]$\lambda\lambda 4959,5007$}}})}{i({{\hbox{h$\beta$ } } } ) } \label{eqn_r23_def}\ ] ] as an abundance indicator . using photoionization models , @xcite showed that bright [ o ii ] and [ o iii ] line intensities can be combined to determine uniquely the ionization parameter and an `` empirical '' oxygen abundance in low - metallicity regions . @xcite developed a grid of photoionization models and suggested using @xmath65 and @xmath66 = @xmath27()/@xmath27 ( ) to estimate the oxygen abundance . however , the calibration is degenerate such that for a given value of @xmath65 , two values of the oxygen abundance are possible . the ratio was suggested ( e.g. , @xcite ) as the discriminant to choose between the `` upper branch '' ( high oxygen abundance ) or the `` lower branch '' ( low oxygen abundance ) . in the present set of spectra , [ ] line strengths are generally small , and has been found to be less than the threshold value of 0.1 . @xcite proposed a new calibration of the oxygen abundances using bright oxygen lines . at low abundances ( 12@xmath0log(o / h ) @xmath67 8.2 ) , his calibration is expressed as @xmath68 where @xmath65 is given by equation ( [ eqn_r23_def ] ) and @xmath69 = @xmath27()/@xmath27 ( ) . in some instances , oxygen abundances with the mcgaugh method could not be computed , because the @xmath65 values were outside of the effective range for the models . @xcite have shown that the pilyugin calibration covers the highest values of @xmath65 . oxygen abundances derived using the mcgaugh and pilyugin bright - line calibrations are listed in tables [ table_abund1 ] and [ table_abund2 ] . for each region , differences between direct and bright - line abundances are shown as a function of @xmath66 and @xmath65 in fig . [ fig_oxydiff ] . the difference between the mcgaugh and pilyugin calibrations ( indicated by asterisks ) appears to correlate with log @xmath66 , which has been previously noticed by @xcite , @xcite , and @xcite . despite the small number of detections , we find that bright - line abundances with the mcgaugh and the pilyugin calibrations are @xmath70 0.10 to 0.15 dex larger and @xmath70 0.05 to 0.10 dex smaller , respectively , than the corresponding direct abundances . we note also that region hm 19 exhibits the lowest values of @xmath65 and @xmath66 ( log @xmath65 = 0.357 , log @xmath66 = @xmath71 ) , and is the outlier in the lower left corner of both panels in fig . [ fig_oxydiff ] . generally , in the absence of , an estimate of the oxygen abundance from the bright - line calibration is good to within @xmath70 0.2 dex . we consider next argon - to - oxygen , nitrogen - to - oxygen , and neon - to - oxygen ratios , which are listed in tables [ table_abund1 ] and [ table_abund2 ] . for metal - poor galaxies , it is assumed that n / o @xmath70 n@xmath55/o@xmath55 @xcite and n@xmath55/o@xmath55 values were derived . nitrogen abundances were computed as n / h = icf(n ) @xmath72 ( n@xmath55/h ) . the ionization correction factor , icf(n ) = o / o@xmath55 , accounts for missing ions . the resulting nitrogen - to - oxygen abundance ratios were found to be the same as the n@xmath55/o@xmath55 values . the mean value of log(n / o ) = @xmath73 is in agreement with the average for metal - poor blue compact dwarf galaxies @xcite . neon abundances are derived as ne / h = icf(ne ) @xmath72 ( ne@xmath11/h ) . the ionization correction factor for neon is icf(ne ) = o / o@xmath11 . the mean log(ne / o ) is @xmath74 , which is marginally consistent with the range of ne / o values at this metallicity found by @xcite and @xcite . however , it is 0.16 dex higher than the mean value of @xmath75 found for blue compact galaxies by @xcite . since the / ratio is sensitive to the reddening correction , and our reddening corrections are only based on the h@xmath7/h@xmath76 ratio , we revisited this correction . in the spectrum for region hm 7 , higher - order balmer lines ( h9 , h10 , and h11 ) were detected ( table [ table_data1 ] ) . their intensity ratios with respect to were found to be consistent with expected values for regions at a temperature of @xmath77 = 15000 k. in the spectra ( grating # 11 ) for region hm 9 , the closest unblended balmer line to is , because h8 is blended with an adjacent helium line , and h@xmath78 is blended with adjacent [ ] and helium lines . we find that corrected / and / ratios are consistent with expected balmer ratios for @xmath77 between 13000 and 14000 k. argon is more complex , because the dominant ion is not found in just one zone . ar@xmath11 is likely to be found in an intermediate area between the o@xmath55 and o@xmath11 zones . following the prescription by @xcite , the argon abundance , was derived as ar / h = icf(ar ) @xmath72 ar@xmath11/h . the ionization correction factor is given by icf(ar ) = ar / ar@xmath11 = @xmath79^{-1}$ ] , where @xmath80 = o@xmath55/o . our mean value of log(ar / o ) = @xmath81 is in agreement with the average for metal - poor blue compact dwarf galaxies @xcite . we note here the recent work by @xcite , who suggested that the direct modeling of photoionized nebulae should be used to infer elemental abundances with accuracies similar to the observations . they showed that abundances derived from model - based ionization correction factors exceeded the range of expected errors from the original data . a comparison of the present data with published spectroscopy ( detections ) is shown in table [ table_compare ] . @xcite obtained spectra of hm 9 and hm 2 ( their regions `` # 1 '' and `` # 2 '' , respectively ) , while @xcite reported spectra for regions hm 7 and hm 9 . we have recomputed and added uncertainties to their published abundances in table [ table_compare ] . from our measurements presented above , was detected in regions hm 7 and hm 9 at a signal - to - noise of about @xmath82 and @xmath83 , respectively . our derived direct oxygen abundance for hm 7 is in agreement with the value reported by @xcite . the direct oxygen abundance for hm 9 is about 0.1 dex higher but consistent within errors with the values reported by @xcite and @xcite . when other measurements of regions are included , oxygen abundances derived with bright - line methods agree with the direct values to within 0.2 dex ( see fig . [ fig_oxydiff ] ) . the present set of measurements have shown that the nebular oxygen abundances are in agreement with values published in the literature . [ fig_oxymb ] shows the metallicity - luminosity relation for dwarf irregular galaxies . we have followed fig . 10 from @xcite and plotted the present nebular result , the mean magnesium abundance from the two a - type supergiant stars , and the derived oxygen abundance for the supergiant wlm 15 . the upper end of the range of nebular oxygen abundances derived in the present work ( [ o / h ] between @xmath84 and @xmath85 ) is consistent with the mean stellar magnesium abundance [ mg / h ] = @xmath5 . however , the mean nebular oxygen abundance is 0.62 dex lower than the oxygen abundance ( [ o / h ] = @xmath6 ) in wlm 15 . we note that the mean stellar [ mg / h ] in supergiant stars agrees with the nebular [ o / h ] in both ngc 6822 @xcite and sextans a @xcite . measurements of additional supergiants in wlm will be crucial in confirming the result in wlm 15 , and whether there are any spatial inhomogeneities in metallicity among stars . @xcite reported two peaks in the distribution ; the west maximum lies just north of hm 9 , whereas the second maximum lies approximately 1 east of hm 7 . figure [ fig_wlm ] shows that regions along the eastern edge of the galaxy in the vicinity of the east peak have nonzero reddening values derived from observed balmer emission - line ratios . this agrees with the discovery of small regions or patches found to have redder @xmath86 and @xmath87 colors than the rest of the galaxy @xcite . this is thought to be extinction internal to wlm , because the foreground extinction is known to be small ( see table [ table_wlm ] ) . we note that the bright regions ( i.e. , hm 7 and hm 9 ) on the western side of the galaxy do not exhibit any appreciable reddening . a comparison of figs . [ fig_h2rs ] and [ fig_wlm ] shows that the two supergiant stars measured by @xcite are located next to the eastern peak . in fact , the supergiant wlm 15 with the high oxygen abundance is located ( in projection ) near region hm 18 , where the average reddening was found to be @xmath88 0.5 mag . we have shown that the present nebular oxygen abundance agrees with previous measurements , and additional explanations are required to explain the discrepancy between the nebular and stellar oxygen abundances . @xcite reasoned that large changes in stellar parameters would not reconcile the stellar abundance with the nebular abundance . three alternative scenarios were considered to explain the discrepancy : lowering the nebular abundance through dilution by the infall of metal - poor gas , the depletion of ism oxygen onto dust grains , and the possibility of large spatial inhomogeneities in the metallicity . the first two possibilities were shown to be unlikely . the third possibility may be possible with the discovery of the second peak . however , we have shown that there are no significant variations in the nebular oxygen abundance . nevertheless , two questions arise . are the nonzero reddenings measured in regions along the eastern side of the galaxy related to the second peak seen by @xcite ? because of their proximity , is there any relationship between the the metallicity of wlm 15 and the highly reddened regions in the southeast ? we note that determining the oxygen abundance for the supergiant is not sensitive to internal reddening @xcite . based on the apparent spatial correlation of the peak with regions of redder color , nonzero ( internal ) extinctions in regions to the east and southeast , and the location of the supergiant wlm 15 , these suggest that something unusual may be happening in this more metal - rich part of the galaxy . unfortunately , regions hm 18 , 18a , and 18b are underluminous and exhibit very weak nebular emission ; there are few additional bright regions in the area . deeper spectroscopy of these regions may prove illuminating . however , measurements of additional blue supergiants in the center and the eastern side of the galaxy ( e.g. , wlm 30 ; @xcite ) could show whether the stellar abundances remain high with respect to the nebular values and whether the stellar abundances are also spatially homogeneous . a new search for molecular gas along the east and southeast sides of the galaxy ( especially in the vicinity of wlm 15 ) would also be timely . while @xcite did not detect co , their pointings missed regions of interest on the eastern side of the galaxy . optical spectra of 13 regions were obtained in wlm , and oxygen abundances were derived in nine regions . was measured in bright regions hm 7 and hm 9 . the resulting direct oxygen abundance for hm 7 is in agreement with previously published values . our @xmath89 detection of in hm 9 confirms the lower signal - to - noise measurements reported by @xcite and @xcite . for the remaining regions , oxygen abundances derived with bright - line methods are accurate to about 0.2 dex . we adopt for wlm a mean nebular oxygen abundance 12@xmath0log(o / h ) = @xmath3 , which corresponds to [ o / h ] = @xmath4 , or 15% of the solar value . the upper end of the range of derived nebular oxygen abundances just agrees with the mean stellar magnesium abundance reported by @xcite , but the present mean nebular result is still 0.62 dex lower than the oxygen abundance derived for the a - type supergiant wlm 15 . significant reddening values derived from observed balmer emission - line ratios were found in regions on the eastern side of the galaxy near one of the peaks discovered by @xcite . there may be a relationship between the location of the east peak , regions of redder color ( higher extinction ) , large reddenings derived from balmer emission - line ratios in regions along the eastern side of the galaxy , and the location of wlm 15 . we thank the anonymous referee for comments which improved the presentation of this paper . h. l. thanks dale jackson for a copy of his figures . we are grateful to eso for awarded telescope time , and lisa germany and the staff at eso la silla for their help in acquiring the spectra . h. l. and e. d. s. acknowledge partial support from a nasa ltsarp grant nag 59221 and from the university of minnesota . k. a. v. thanks the national science foundation for support through a career award ast 9984073 . for their one - year visit , e. d. s. and k. a. v. thank the institute of astronomy , university of cambridge for their hospitality and support . this research has made use of nasa s astrophysics data system , and 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 . garnett , d. r. 2004 , in cosmochemistry : the melting pot of the elements , xiii canary islands winter school of astrophysics , eds . c. esteban , r. j. garca - lpez , a. herrero , & f. snchez ( cambridge : cambridge university press ) , 171 gilmore , g. , & wyse , r. f. g. 1991 , , 367 , l55 krawchuk , c. a. p. , mccall , m. l. , komljenovic , m. , kingsburgh , r. , richer , m. g. , & stevenson , c. 1997 , in iau symp . 180 , planetary nebulae , ed . h. j. habing and h. j. g. l. m. lamers ( dordrecht : kluwer ) , 116 massey , p. , hodge , p. w. , holmes , s. , jacoby , g. , king , n. l. , olsen , k. , smith , c. , & saha , a. 2002 , baas , 34 , 1272 matteucci , f. , 2003 , , 284 , 539 mccall , m. l. , rybski , p. m. , & shields , g. a. 1985 , , 57 , 1 tolstoy , e. , & venn , k. a. 2004 , in highlights of astronomy , vol . 13 , as presented at the xxvth general assembly of the iau , sydney , australia , 1425 august 2003 , ed . o. engvold , ( san francisco : astron . pacific ) , in press ( astro - ph/0402295 ) venn , k. a. , irwin , m. , shetrone , m. d. , tout , c. a. , hill , v. , & tolstoy , e. 2004a , , 128 , 1177 venn , k. a. , tolstoy , e. , kaufer , a. , & kudritzki , r. p. 2004b , in carnegie observatories astrophysics series , vol . 4 : origin and evolution of the elements , ed . a. mcwilliam & m. rauch ( pasadena : carnegie observatories , http://www.ociw.edu/ociw/symposia/series/ + symposium4/proceedings.html ; also astro - ph/0305188 ) wiese , w. l. , fuhr , j. r. , & deters , t. m. 1996 , atomic transition probabilities of carbon , nitrogen , and oxygen : a critical data compilation ( american chemical society for the national institute of standards and technology ) ccc property & value & references + type & ib(s)m & + alternate names & ddo 221 , ugca 444 & + distance & @xmath90 mpc & 1 + linear to angular scale at this distance & 4.6 pc arsec@xmath12 & 2 + @xmath91 & @xmath92 & 3 + @xmath93 & 0.037 & 4 + @xmath94 & @xmath95 jy km s@xmath12 & 5 + @xmath96 & @xmath97 km s@xmath12 & 6 + @xmath98 $ ] & @xmath99 ( @xmath100 0.26 ) & 7 + @xmath101 $ ] , wlm 15@xmath102 & @xmath103 ( @xmath100 0.05 ) & 7 + @xmath104 $ ] , @xmath102 & @xmath105 & 2 + ccc + total area & + field of view & + pixel size & + image scale & + gain & + read - noise ( rms ) & + + length & + width & + & grating # 11 & grating # 7 + groove density & 300 lines mm@xmath12 & 600 lines mm@xmath12 + blaze @xmath42 ( 1st order ) & 4000 & 3800 + dispersion & 2.04 pixel@xmath12 & 0.96 pixel@xmath12 + effective @xmath42 range & 33807520 & 32705240 + lccccccc & date & & & @xmath106 & & & rms + region & ( ut 2003 ) & grating & @xmath107 & ( s ) & @xmath108 & & ( mag ) + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) + hm 2 & 28 aug & # 11 & 1 @xmath72 1200 & 1200 & 1.24 & no & 0.034 + hm 2 & 31 aug & # 7 & 3 @xmath72 1200 & 3600 & 1.21 & no & 0.025 + hm 7 & 26 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.06 & yes & 0.030 + hm 8 & 26 aug & # 11 & 7 @xmath72 1200 & 8400 & 1.20 & no & 0.030 + hm 9 & 26 aug & # 11 & 7 @xmath72 1200 & 8400 & 1.20 & yes & 0.030 + hm 9 & 31 aug & # 7 & 3 @xmath72 1200 & 3600 & 1.21 & no & 0.025 + hm 12 & 26 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.06 & no & 0.030 + hm 12 & 28 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.08 & no & 0.034 + hm 16 nw & 27 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.04 & no & 0.029 + hm 16 se & 27 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.04 & no & 0.029 + hm 17 & 28 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.03 & no & 0.034 + hm 18 & 28 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.03 & no & 0.034 + hm 18a & 28 aug & # 11 & 1 @xmath72 1200 & 1200 & 1.06 & no & 0.034 + hm 18b & 28 aug & # 11 & 1 @xmath72 1200 & 1200 & 1.06 & no & 0.034 + hm 19 & 27 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.04 & no & 0.029 + hm 19 & 28 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.08 & no & 0.034 + hm 21 & 28 aug & # 11 & 3 @xmath72 1200 & 3600 & 1.08 & no & 0.034 + rccccccc & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 + @xmath117 & @xmath118 & & & & & @xmath119 & @xmath120 + @xmath121 & @xmath122 & & & & & @xmath123 & @xmath124 + @xmath125 & @xmath126 & & & & & @xmath127 & @xmath128 + @xmath129\;3869 $ ] & @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath135 & @xmath136 + @xmath137 & @xmath138 & & & & & @xmath139 & @xmath140 + @xmath141 & @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath147 & @xmath148 + @xmath149 & @xmath150 & & & & & @xmath151 & @xmath152 + @xmath153\;4068 $ ] & @xmath154 & & & & & @xmath155 & @xmath155 + @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 & @xmath163 + @xmath164 & @xmath165 & @xmath166 & @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 + @xmath172\;4363 $ ] & @xmath173 & & & @xmath174 & @xmath175 & @xmath176 & @xmath177 + @xmath178 & @xmath179 & & & & & @xmath180 & @xmath180 + @xmath181 & @xmath182 & & & & & @xmath183 & @xmath184 + @xmath185 + { \rm he\;i}\;4713 $ ] & @xmath186 & & & & & @xmath187 & @xmath187 + @xmath185\;4740 $ ] & @xmath188 & & & & & @xmath189 & @xmath189 + @xmath190 & 0.000 & @xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath196 + @xmath172\;4959 $ ] & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 & @xmath203 + @xmath172\;5007 $ ] & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 & @xmath209 & @xmath210 + @xmath211 & @xmath212 & & & & & @xmath213 & @xmath214 + @xmath215\;6300 $ ] & @xmath216 & & & & & @xmath217 & @xmath218 + @xmath215\;6363 $ ] & @xmath219 & & & & & @xmath220 & @xmath221 + @xmath222 & @xmath223 & & & @xmath224 & @xmath225 & @xmath226 & @xmath227 + @xmath228\;6583 $ ] & @xmath229 & & & & & @xmath230 & @xmath231 + @xmath232 & @xmath233 & & & & & @xmath234 & @xmath235 + @xmath153\;6716 $ ] & @xmath236 & & & & & @xmath237 & @xmath238 + @xmath153\;6731 $ ] & @xmath239 & & & & & @xmath240 & @xmath241 + @xmath242 & @xmath243 & & & & & @xmath244 & @xmath245 + @xmath246\;7136 $ ] & @xmath247 & & & & & @xmath248 & @xmath249 + @xmath250\;7320 $ ] & @xmath251 & & & & & @xmath252 & @xmath253 + @xmath250\;7330 $ ] & @xmath254 & & & & & @xmath255 & @xmath256 + & & & + & & & + & & & + & & & + & & & + & & & + rccccccc & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 + @xmath129\;3869 $ ] & @xmath130 & @xmath263 & @xmath264 & @xmath265 & @xmath266 & @xmath267 & @xmath268 + @xmath141 & @xmath142 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 & @xmath274 + @xmath156 & @xmath157 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 + @xmath164 & @xmath165 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 + @xmath172\;4363 $ ] & @xmath173 & @xmath287 & @xmath288 & @xmath289 & @xmath290 & & + @xmath190 & 0.000 & @xmath291 & @xmath292 & @xmath291 & @xmath292 & @xmath293 & @xmath294 + @xmath172\;4959 $ ] & @xmath197 & @xmath295 & @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath300 + @xmath172\;5007 $ ] & @xmath204 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & @xmath306 + & & & + & & & + & & & + & & & + & & & + & & & + & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath312 + @xmath129\;3869 $ ] & @xmath130 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 + @xmath137 & @xmath138 & @xmath319 & @xmath320 & @xmath321 & @xmath322 & @xmath323 & @xmath324 + @xmath141 & @xmath142 & @xmath325 & @xmath326 & @xmath327 & @xmath328 & @xmath329 & @xmath330 + @xmath156 & @xmath157 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 + @xmath164 & @xmath165 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 + @xmath172\;4363 $ ] & @xmath173 & @xmath343 & @xmath344 & @xmath345 & @xmath346 & @xmath347 & @xmath348 + @xmath181 & @xmath182 & & & @xmath349 & @xmath350 & @xmath351 & @xmath352 + @xmath190 & 0.000 & @xmath353 & @xmath292 & @xmath292 & @xmath195 & @xmath354 & @xmath355 + @xmath172\;4959 $ ] & @xmath197 & @xmath356 & @xmath357 & @xmath358 & @xmath359 & @xmath360 & @xmath361 + @xmath172\;5007 $ ] & @xmath204 & @xmath362 & @xmath363 & @xmath364 & @xmath365 & @xmath366 & @xmath367 + @xmath211 & @xmath212 & @xmath368 & @xmath369 & @xmath370 & @xmath371 & @xmath372 & @xmath373 + @xmath222 & @xmath223 & @xmath374 & @xmath375 & @xmath376 & @xmath377 & @xmath378 & @xmath379 + @xmath228\;6583 $ ] & @xmath229 & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 & @xmath385 + @xmath153\;6716,6731 $ ] & @xmath386 & @xmath387 & @xmath388 & @xmath389 & @xmath390 & @xmath391 & @xmath392 + @xmath246\;7136 $ ] & @xmath247 & @xmath393 & @xmath394 & @xmath395 & @xmath396 & @xmath397 & @xmath398 + & & & + & & & + & & & + & & & + & & & + & & & + rccccccc & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath399 & @xmath400 & @xmath401 & @xmath402 & @xmath403 & @xmath404 + @xmath129\;3869 $ ] & @xmath130 & & & & & @xmath405 & @xmath406 + @xmath156 & @xmath157 & & & & & @xmath407 & @xmath408 + @xmath164 & @xmath165 & @xmath409 & @xmath410 & @xmath411 & @xmath412 & @xmath413 & @xmath414 + @xmath190 & 0.000 & @xmath415 & @xmath416 & @xmath417 & @xmath418 & @xmath419 & @xmath420 + @xmath172\;4959 $ ] & @xmath197 & @xmath421 & @xmath422 & @xmath423 & @xmath424 & @xmath425 & @xmath426 + @xmath172\;5007 $ ] & @xmath204 & @xmath427 & @xmath428 & @xmath429 & @xmath430 & @xmath431 & @xmath432 + @xmath211 & @xmath212 & & & & & @xmath433 & @xmath434 + @xmath222 & @xmath223 & @xmath435 & @xmath436 & @xmath437 & @xmath438 & @xmath439 & @xmath440 + @xmath228\;6583 $ ] & @xmath229 & & & @xmath441 & @xmath442 & & + @xmath153\;6716,6731 $ ] & @xmath386 & @xmath443 & @xmath444 & @xmath445 & @xmath446 & @xmath447 & @xmath448 + & & & + & & & + & & & + & & & + & & & + & & & + & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath449 & @xmath450 & @xmath451 & @xmath452 & @xmath453 & @xmath454 + @xmath141 & @xmath142 & @xmath455 & @xmath456 & & & & + @xmath156 & @xmath157 & @xmath457 & @xmath458 & & & & + @xmath164 & @xmath165 & @xmath459 & @xmath460 & @xmath461 & @xmath462 & @xmath463 & @xmath464 + @xmath172\;4363 $ ] & @xmath173 & @xmath465 & @xmath466 & & & & + @xmath190 & 0.000 & @xmath467 & @xmath468 & @xmath469 & @xmath470 & @xmath471 & @xmath472 + @xmath172\;4959 $ ] & @xmath197 & @xmath473 & @xmath474 & @xmath475 & @xmath476 & & + @xmath172\;5007 $ ] & @xmath204 & @xmath477 & @xmath478 & @xmath479 & @xmath480 & @xmath481 & @xmath482 + @xmath211 & @xmath212 & @xmath483 & @xmath484 & & & & + @xmath222 & @xmath223 & @xmath485 & @xmath486 & @xmath487 & @xmath488 & @xmath489 & @xmath490 + @xmath228\;6583 $ ] & @xmath229 & @xmath491 & @xmath492 & & & & + @xmath153\;6716,6731 $ ] & @xmath386 & @xmath493 & @xmath494 & @xmath495 & @xmath496 & @xmath497 & @xmath498 + & & & + & & & + & & & + & & & + & & & + & & & + rccccccc & & & & + & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath499 & @xmath500 & & + @xmath190 & 0.000 & @xmath501 & @xmath502 & & + @xmath222 & @xmath223 & @xmath503 & @xmath504 & & + @xmath153\;6716,6731 $ ] & @xmath386 & @xmath505 & @xmath506 & & + & & & + & & & + & & & + & & & + & & & + & & & + & & & & + & & & & & & & + @xmath109\;3727 $ ] & @xmath110 & @xmath507 & @xmath508 & @xmath509 & @xmath510 & @xmath511 & @xmath512 + @xmath129\;3869 $ ] & @xmath130 & & & @xmath513 & @xmath514 & & + @xmath141 & @xmath142 & & & & & @xmath515 & @xmath516 + @xmath156 & @xmath157 & & & @xmath483 & @xmath517 & @xmath518 & @xmath519 + @xmath164 & @xmath165 & @xmath520 & @xmath521 & @xmath522 & @xmath523 & @xmath524 & @xmath525 + @xmath190 & 0.000 & @xmath526 & @xmath527 & @xmath528 & @xmath355 & @xmath355 & @xmath526 + @xmath172\;4959 $ ] & @xmath197 & @xmath529 & @xmath530 & @xmath531 & @xmath532 & @xmath533 & @xmath534 + @xmath172\;5007 $ ] & @xmath204 & @xmath535 & @xmath536 & @xmath537 & @xmath538 & @xmath539 & @xmath540 + @xmath222 & @xmath223 & @xmath541 & @xmath542 & @xmath543 & @xmath544 & @xmath545 & @xmath546 + @xmath228\;6583 $ ] & @xmath229 & & & @xmath547 & @xmath548 & @xmath549 & @xmath550 + @xmath153\;6716,6731 $ ] & @xmath386 & & & @xmath551 & @xmath552 & @xmath553 & @xmath554 + & & & + & & & + & & & + & & & + & & & + & & & + lccccccc property & hm 2 & hm 2 & hm 7 & hm 8 & hm 9 ap1 & hm 9 ap2 & hm 9 ap3 + & ( gr # 7 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 7 ) & ( gr # 7 ) & ( gr # 7 ) + @xmath58(o@xmath11 ) ( k ) & & @xmath555 & @xmath556 & & @xmath557 & @xmath558 & + @xmath58(o@xmath55 ) ( k ) & & @xmath559 & @xmath560 & & @xmath561 & @xmath562 & + o@xmath55/h @xmath563 & & @xmath564 & @xmath565 & & @xmath566 & @xmath567 & + o@xmath11/h @xmath563 & & @xmath568 & @xmath569 & & @xmath570 & @xmath571 & + o / h @xmath563 & & @xmath572 & @xmath573 & & @xmath574 & @xmath575 & + 12@xmath0log(o / h ) & & @xmath576 & @xmath577 & & @xmath578 & @xmath579 & + 12@xmath0log(o / h ) m91 & 8.17 & 8.11 & 7.84 & 8.01 & 8.03 & 8.06 & 8.15 + 12@xmath0log(o / h ) p00 & 7.92 & 7.85 & 7.67 & 8.18 & 7.92 & 7.99 & 8.23 + ar@xmath11/h @xmath580 & & & @xmath581 & & & & + ar@xmath60/h @xmath580 & & & @xmath582 & & & & + icf(ar ) & & & 1.06 & & & & + ar / h @xmath580 & & & @xmath583 & & & & + log(ar / o ) & & & @xmath584 & & & & + n@xmath55/o@xmath55 & & & @xmath585 & & & & + log(n / o ) & & & @xmath586 & & & & + ne@xmath11/o@xmath11 & & & @xmath587 & & & & + log(ne / o ) & & & @xmath588 & & & & + lcccccc property & hm 9 ap1 & hm 9 ap2 & hm 9 ap3 & hm 12 & hm 12 & hm 16 nw + & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) + @xmath58(o@xmath11 ) ( k ) & @xmath589 & @xmath590 & @xmath591 & & & @xmath592 + @xmath58(o@xmath55 ) ( k ) & @xmath593 & @xmath594 & @xmath595 & & & @xmath596 + o@xmath55/h ( @xmath597 ) & @xmath598 & @xmath599 & @xmath600 & & & @xmath601 + o@xmath11/h ( @xmath597 ) & @xmath602 & @xmath603 & @xmath604 & & & @xmath605 + o / h ( @xmath597 ) & @xmath606 & @xmath607 & @xmath397 & & & @xmath608 + 12@xmath0log(o / h ) & @xmath609 & @xmath610 & @xmath611 & & & @xmath612 + 12@xmath0log(o / h ) m91 & 8.04 & 8.04 & 8.01 & 8.16 & 8.00 & 8.21 + 12@xmath0log(o / h ) p00 & 7.85 & 7.86 & 7.84 & 8.48 & 7.96 & 8.29 + ar@xmath11/h @xmath580 & @xmath613 & @xmath614 & @xmath615 & & & + icf(ar ) & 1.50 & 1.48 & 1.48 & & & + ar / h @xmath580 & @xmath616 & @xmath617 & @xmath618 & & & + log(ar / o ) & @xmath619 & @xmath620 & @xmath621 & & & + n@xmath55/o@xmath55 & & @xmath622 & @xmath623 & & & + log(n / o ) & & @xmath624 & @xmath625 & & & + ne@xmath11/o@xmath11 & @xmath626 & @xmath627 & @xmath628 & & & + log(ne / o ) & @xmath629 & @xmath630 & @xmath631 & & & + + property & hm 17 & hm 19 & hm 19 & hm 21 + & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) & ( gr # 11 ) + 12@xmath0log(o / h ) m91 & 8.17 & 8.24 & 7.64 & 8.14 + 12@xmath0log(o / h ) p00 & & 8.40 & 8.12 & 8.28 + ccccccc & & & + property & hm 2 & hm 9 & hm 7 & hm 9 & hm 7 & hm 9 + o / h ( @xmath597 ) & @xmath632 & @xmath633 & @xmath634 & @xmath345 & @xmath573 & @xmath635 + 12@xmath0log(o / h ) & @xmath636 & @xmath637 & @xmath638 & @xmath639 & @xmath1 & @xmath2 + log(n / o ) & & @xmath640 & @xmath641 & @xmath642 & @xmath586 & @xmath73 + log(ne / o ) & @xmath643 & @xmath644 & @xmath645 & @xmath646 & @xmath588 & @xmath647 + log(ar / o ) & & @xmath648 & @xmath649 & @xmath650 & @xmath584 & @xmath81 +

we extend the relativistic mean field theory model of sugahara and toki ( tm1 ) by adding new couplings suggested by modern effective field theories . an improved set of parameters ( tm1 * ) is developed with the goal to test the ability of the models based on effective field theory to describe the properties of finite nuclei and , at the same time , to be consistent with the trends of dirac brueckner hartree fock calculations at densities away from the saturation region . we compare our calculations with other relativistic nuclear force parameters for various nuclear phenomena . * effects of new non - linear couplings in relativistic effective field theory * + m. del estal , m. centelles , x. vias and s.k . patra + _ departament destructura i constituents de la matria , facultat de fsica , + universitat de barcelona , diagonal _ 647 _ , e-_08028 _ barcelona , spain _ _ pacs : _ 21.60.-n , 21.10.dr , 21.65.+f , 21.30.fe relativistic mean field approach , effective field theory , non - linear self - interactions , dirac brueckner hartree fock , nuclear structure

nucl-th9909010

in the recent decades relativistic quantum field theory has been very successful for the description of the nuclear many - body problem @xcite . the relativistic models take care _ ab initio _ of many natural phenomena which are practically absent or have to be included in an _ ad hoc _ manner in the non - relativistic formalism . specifically , the relativistic mean field ( rmf ) treatment of quantum hadrodynamics ( qhd ) @xcite has become a popular way to deal with the nuclear physics problems . the original linear @xmath0 model of walecka @xcite was complemented with cubic and quartic non - linearities of the @xmath1 meson @xcite ( non - linear @xmath0 model ) to improve the incompressibility and the finite nuclei results . since these models were proposed to be renormalizable , the scalar self - interactions were limited to a quartic polynomial and scalar - vector and vector - vector interactions were not allowed . very recently , however , inspired by effective field theory ( eft ) , furnstahl , serot and tang @xcite abandoned the idea of renormalizability and extended the rmf theory by allowing other non - linear scalar - vector and vector - vector self - interactions in addition to tensor couplings @xcite . the effective field theory contains all the non - renormalizable couplings consistent with the underlying symmetries of qcd . since one has to deal with an effective lagrangian with an infinite number of terms it is imperative to develop a suitable expansion scheme . in the nuclear many - body problem the scalar ( @xmath2 ) and vector ( @xmath3 ) meson fields are normally small as compared with the nucleon mass ( @xmath4 ) and they vary slowly in finite nuclei . this means that the ratios @xmath5 , @xmath6 , @xmath7 and @xmath8 are the useful expansion parameters @xcite . the concept of naturalness @xcite , i.e. , that all the coupling constants written in an appropriate dimensionless form should be of the order of unity , is used to avoid ambiguities in the expansion . then one can estimate the contributions coming from different terms by counting powers in the expansion parameters and truncating the lagrangian at a given level of accuracy . for the truncation to be consistent , the coupling constants should exhibit naturalness and none should be arbitrarily dropped out to the given order without additional symmetry arguments . if the eft lagrangian is truncated at fourth order one recovers the standard non - linear @xmath0 model plus some additional couplings @xcite , with thirteen free parameters . in ref.@xcite these parameters have been fitted ( sets g1 and g2 ) to reproduce twenty - nine finite nuclei observables ( binding energies , charge form factors and spin - orbit splittings of magic nuclei ) . the fits display naturalness and the results are not dominated by the last terms retained . this evidence confirms the utility of the eft concepts and of the naturalness assumption , and shows that truncating the effective lagrangian at the first lower orders is justified . the eft approach has also been helpful to elucidate the empirical success of the usual non - linear @xmath0 models that have less free parameters . it has been shown that the mean field phenomenology of bulk and single - particle nuclear observables does not constrain all of the parameters of the eft model unambiguosly . that is , the constants in the eft model are underdetermined by the observables included in the fits and several parameter sets with low @xmath9 can be found @xcite . an analysis of the particular impact of each one of the new couplings arising in eft on the determination of the saturation properties of nuclear matter and on the nuclear surface properties has been carried out in ref . recent applications of the eft - based models include studies of pion - nucleus scattering @xcite , the nuclear spin - orbit force @xcite and asymmetric nuclear matter at finite temperature @xcite . on a more microscopic level , it is well known that non - relativistic brueckner goldstone calculations based on realistic nn potentials are not able to give the right saturation density and binding energy of infinite nuclear matter at the same time ( coester line ) @xcite . to obtain relatively correct values an additional repulsive part has to be added . this can be achieved by working in the relativistic framework : the dirac brueckner hartree fock ( dbhf ) theory @xcite introduces an extra density dependence that allows to fit the nn phase shifts and to approach the empirical equilibrium point of nuclear matter . the large scalar and time - like vector self - energies of the dbhf calculations show two interesting features : a rather small effect of the two - body correlations and a weak momentum dependence , at least for not very high densities . this suggests fitting the nuclear matter dbhf self - energies by a much simpler rmf approach . this strategy was carried out in the past using the @xmath0 model with scalar self - interactions @xcite , and also including a quartic vector self - interaction @xcite ( as proposed by bodmer @xcite ) . the outcoming parameter sets did not properly reproduce the properties of finite nuclei . the saturation properties were close to those of dbhf and it is known that they are not accurate enough , in spite of the significant improvement over the non - relativistic bhf results . refs . @xcite showed that the success of the usual rmf model with only scalar self - interactions for describing the saturation point and the data for finite nuclei is not followed by a proper description of the trends of the dbhf scalar and vector self - energies . this is caused mainly by a too restrictive treatment of the @xmath10meson term : while in the standard rmf model the vector potential increases linearly with density and gets stronger , in dbhf it bends down with density ( see figure 1 later ) . moreover , the scalar potential overestimates the dbhf result at high density in order to compensate for the strong repulsion in the vector channel . this is the reason for providing the wrong sign in the coupling constant of the @xmath11 term in most of the successful rmf parameter sets . furthermore , the equation of state becomes much steeper and soon separates from the dbhf tendency when the density grows ( see figure 2 ) . adding a quartic vector self - interaction remarkably improves the behaviour of the vector and scalar potentials , softens the equation of state and brings about a positive sign for the @xmath11 coupling @xcite . in particular , sugahara and toki @xcite took into account the non - linear term ( @xmath12 ) in the @xmath13 vector field and fitted the free parameters to the data for several nuclei . even with inclusion of the @xmath12 term they could not get with a single parametrization , and at the same time , a positive coupling constant of the @xmath11 term and a quality for nuclei along the periodic table similar to that of nl1 @xcite . thus they constructed two parameter sets , tm1 and tm2 , both with a positive @xmath11 coupling constant . the tm2 set was designed for charge numbers @xmath14 and the tm1 set for larger @xmath15 . tm1 was also applied to calculate the equation of state and the structure of neutron stars and supernovae @xcite . apart from giving good results for finite nuclei of @xmath16 , tm1 agreed with the trends of the nuclear matter dbhf calculations much better than the conventional non - linear @xmath0 sets ( such as nl1 @xcite or nl3 @xcite ) owing to the vector self - interaction . > from the point of view of effective field theory @xcite the models of refs . @xcite that include up to a quartic vector self - interaction have the drawback that the coefficients of some couplings , which should otherwise be present in the effective lagrangian truncated at fourth order , have been put equal to zero without theoretical justification . this fact motivates us to include the remaining terms and to study their effect on nuclear matter and finite nuclei . we will show that it is possible to extend the tm1 set to describe the finite nuclei observables with @xmath17 and to obtain a description of nuclear matter that follows the dbhf tendencies better than the conventional non - linear @xmath0 models . to do that we will investigate the effects of the new couplings from eft keeping the equilibrium properties fixed to those of tm1 . it should be pointed out that , actually , the equilibrium properties of tm1 lie in the range for which a reasonable description of finite nuclei properties can be achieved , provided that the eft parameters of the model are natural @xcite . here one has to note that the specific values of tm1 vary from the dbhf result . however , the rmf sets with only scalar self - interactions which give good saturation properties deviate sharply from dbhf at high density . thus , for our purposes , we believe that a good way to settle the parametrization is to remain close to the empirical value near saturation and follow an equation of state similar to that of the dbhf theory . in the next section we shall detail our strategy . we emphasize that our goal is not to produce a new optimal set of parameters intended to compete with well - established conventional sets like nl3 @xcite , which already are very successful for nuclei both at and away from the line of @xmath18-stability . instead , we wish to learn the possibilities of the new eft models for describing finite nuclei and for simultaneously tuning the behaviour with density of the scalar and vector self - energies . in doing this we want to ascertain whether the comparison with dbhf can provide useful constraints on the new couplings . for our study , the determination of the parameters of the model through a least - squares fitting procedure , by calculating nuclear properties repeatedly until obtaining a best fit , would make the connections between the resulting parameters and the considered nuclear observables more obscure . the paper is organized as follows . section 2 is devoted to a summary of the mean field equations and to fit part of the parameters of the effective lagrangian to nuclear matter data . we compare our results with the predictions of other parametrizations available in the literature . in the third section the remaining parameters of the effective lagrangian are obtained by imposing that our mean field approach reproduces the experimental data for some selected nuclei . a bcs - type pairing correlation is added in section 4 to calculate non - magic even - even nuclei . the extended parameter set is tested in some applications to nuclear structure phenomena like isotopic / isotonic energy differences , the isotopic change in the charge radius and nuclei with large neutron / proton excess . finally , the summary and concluding remarks are given in section 5 . the rmf treatment of the qhd models automatically includes the spin - orbit force , the finite range and the density dependence of the nuclear interaction . the rmf model has the advantage that , with the proper relativistic kinematics and with the meson properties already known or fixed from the properties of a small number of finite nuclei , gives excellent results for binding energies , root - mean - square radii , quadrupole and hexadecapole deformations and other properties of spherical and deformed nuclei @xcite . the quality of the results is comparable to that found in non - relativistic nuclear structure calculations with effective skyrme @xcite or gogny @xcite forces . in recent years the effective field theory approach to qhd has been studied extensively . the theory and the equations for finite nuclei and nuclear matter can be found in the literature @xcite and we shall only outline the formalism here . we start from ref . @xcite where the field equations were derived from an energy density functional containing dirac baryons and classical scalar and vector mesons . although this energy functional can be obtained from the effective lagrangian in the hartree approximation @xcite , it can also be considered as an expansion in terms of ratios of the meson fields and their gradients to the nucleon mass of a general energy density functional that contains the contributions of correlations within the spirit of density functional theory . according to refs . @xcite this energy density functional for finite nuclei can be written as @xmath19 + w({r } ) + \frac{1}{2}\tau_3 r({r } ) + \frac{1+\tau_3}{2 } a ( { r } ) \nonumber \\[3 mm ] & & - \frac{i \beta\mbox{\boldmath$\alpha$}}{2m}\!\cdot\ ! \left ( f_v \mbox{\boldmath$\nabla$ } w({r } ) + \frac{1}{2}f_\rho\tau_3 \mbox{\boldmath$\nabla$ } r({r } ) \right ) \bigg\ } \varphi_\alpha ( r ) \nonumber \\[3 mm ] & & \null + \left ( \frac{1}{2 } + \frac{\kappa_3}{3!}\frac{\phi({r})}{m } + \frac{\kappa_4}{4!}\frac{\phi^2({r})}{m^2}\right ) \frac{m_s^2}{g_s^2 } \phi^2({r } ) - \frac{\zeta_0}{4 ! } \frac{1 } { g_v^2 } w^4 ( { r } ) \nonumber \\[3 mm ] & & \null + \frac{1}{2g_s^2}\left ( 1 + \alpha_1\frac{\phi({r})}{m}\right ) \left ( \mbox{\boldmath $ \nabla$}\phi({r})\right)^2 - \frac{1}{2g_v^2}\left ( 1 + \alpha_2\frac{\phi({r})}{m}\right ) \left ( \mbox{\boldmath $ \nabla$ } w({r } ) \right)^2 \nonumber \\[3 mm ] & & \null - \frac{1}{2}\left(1 + \eta_1 \frac{\phi({r})}{m } + \frac{\eta_2}{2 } \frac{\phi^2 ( { r})}{m^2 } \right ) \frac{m_v^2}{g_v^2 } w^2 ( { r } ) - \frac{1}{2e^2 } \left ( \mbox{\boldmath $ \nabla$ } a({r})\right)^2 \nonumber \\[3 mm ] & & \null - \frac{1}{2g_\rho^2 } \left ( \mbox{\boldmath $ \nabla$ } r({r})\right)^2 - \frac{1}{2 } \left ( 1 + \eta_\rho \frac{\phi({r})}{m } \right ) \frac{m_\rho^2}{g_\rho^2 } r^2({r } ) \ , , \label{eqfn1}\end{aligned}\ ] ] where the index @xmath20 runs over all occupied states of the positive energy spectrum , @xmath21 , @xmath22 , @xmath23 and @xmath24 . except for the terms with @xmath25 and @xmath26 , the functional ( [ eqfn1 ] ) is of fourth order in the expansion . following refs . @xcite , we retain the fifth - order terms @xmath25 and @xmath26 because their contribution to the nuclear surface energy is numerically of the same magnitude as the contribution from the quartic scalar term . one can see that the new terms concentrate on the isoscalar channel and that the expansion with respect to the isovector meson is shorter ( the @xmath27 coupling is of third order ) . higher non - linear couplings of the @xmath28 meson are not considered because the expectation value of the @xmath28 field is typically an order of magnitude smaller than that of the @xmath13 field @xcite , and they only have a marginal impact on the usual properties studied for terrestrial nuclei . for example , in calculations of the high - density equation of state , mller and serot @xcite found the effects of a quartic @xmath28 meson coupling ( @xmath29 ) to be appreciable only in stars made of pure neutron matter . a surface contribution @xmath30 was tested in ref.@xcite and it was found to have absolutely negligible effects . we should note , nevertheless , that very recently it has been shown that couplings of the type @xmath31 and @xmath32 are useful to modify the neutron radius in heavy nuclei while making very small changes to the proton radius and the binding energy @xcite . the dirac equation corresponding to the energy density ( [ eqfn1 ] ) becomes @xmath33 + w(r ) + \frac{1}{2 } \tau_3 r(r ) + \frac{1 + \tau_3}{2}a(r ) \nonumber \\[3 mm ] & & \null \left . - \frac{i\beta \mbox{\boldmath$\alpha$}}{2 m } \!\cdot\ ! \left [ f_v \mbox{\boldmath$\nabla$ } w(r ) + \frac{1}{2}f_{\rho } \tau_3 \mbox{\boldmath$\nabla$ } r(r ) \right ] \right\ } \varphi_\alpha ( r ) = \varepsilon_\alpha \ , \varphi_\alpha ( r ) \ , . \label{eqfn12}\end{aligned}\ ] ] the mean field equations for @xmath2 , @xmath3 , @xmath34 and @xmath35 are given by @xmath36 & & \null + { g_s^2 \over 2 m } \left(\eta_1+\eta_2{\phi(r)\over m}\right ) { m_v^2\over g_v^2 } w^2 ( r ) + { \eta_{\rho } \over 2m}{g_s^2 \over g{_\rho}^2 } { m_\rho^2 } r^2 ( r ) \nonumber \\[3 mm ] & & \null + { \alpha_1 \over 2 m } [ ( \mbox{\boldmath $ \nabla$}\phi(r))^2 + 2\phi(r)\delta \phi(r ) ] + { \alpha_2 \over 2 m } { g_s^2\over g_v^2 } ( \mbox{\boldmath $ \nabla$}w(r))^2 \ , , \label{eqfn2 } \\[3 mm ] -\delta w(r ) + m_v^2 w(r ) & = & g_v^2 \left ( \rho(r ) + \frac{f_v}{2 } \rho_{\rm t}(r ) \right ) -\left ( \eta_1+{\eta_2\over 2}{\phi(r)\over m } \right ) { \phi(r ) \over m } m_v^2 w(r ) \nonumber \\[3 mm ] & & \null -{1\over 3!}\zeta_0 w^3(r ) + { \alpha_2 \over m } [ \mbox{\boldmath $ \nabla$}\phi(r ) \cdot\mbox{\boldmath $ \nabla$}w(r ) + \phi(r)\delta w(r ) ] \ , , \label{eqfn3 } \\[3 mm ] -\delta r(r ) + m_{\rho}^2 r(r ) & = & { 1 \over 2 } g_{\rho}^2 \left ( \rho_{3}(r ) + { 1 \over 2 } f_{\rho}\rho_{\rm t,3}(r ) \right ) - \eta_\rho { \phi ( r ) \over m } m_{\rho}^2 r(r ) \ , , \label{eqfn4 } \\[3 mm ] -\delta a(r ) & = & e^2 \rho_{\rm p}(r ) \ , , \label{eqfn5 } \end{aligned}\ ] ] where the baryon , scalar , isovector , proton and tensor densities are @xmath37 \rho_s(r ) & = & \sum_\alpha \varphi_\alpha^\dagger(r ) \beta \varphi_\alpha(r ) \ , , \label{eqfn7 } \\[3 mm ] \rho_3 ( r ) & = & \sum_\alpha \varphi_\alpha^\dagger(r ) \tau_3 \varphi_\alpha(r ) \ , , \label{eqfn8 } \\[3 mm ] \rho_{\rm p}(r ) & = & \sum_\alpha \varphi_\alpha^\dagger(r ) \left ( \frac{1 + \tau_3}{2 } \right ) \varphi_\alpha(r ) \ , , \label{eqfn9 } \\[3 mm ] \rho_{\rm t}(r ) & = & \sum_\alpha \frac{i}{m } \mbox{\boldmath$\nabla$ } \!\cdot\ ! \left [ \varphi_\alpha^\dagger(r ) \beta \mbox{\boldmath$\alpha$ } \varphi_\alpha(r ) \right ] \ , , \label{eqfn10 } \\[3 mm ] \rho_{\rm t,3}(r ) & = & \sum_\alpha \frac{i}{m } \mbox{\boldmath$\nabla$ } \!\cdot\ ! \left [ \varphi_\alpha^\dagger(r ) \beta \mbox{\boldmath$\alpha$ } \tau_3 \varphi_\alpha(r ) \right ] \ , . \label{eqfn11 } \end{aligned}\ ] ] in the context of density functional theory it is possible to parametrize the exchange and correlation effects through local potentials ( kohn sham potentials ) , as long as those contributions be small enough that can be considered as minor perturbations to the potentials @xcite . as it is known , this is the case with the local meson fields . the hartree values are the ones that control the dynamics in the relativistic dbhf calculations . therefore , the meson fields can also be interpreted as kohn sham potentials . equations ( [ eqfn2])([eqfn5 ] ) thus correspond to the kohn sham equations in the relativistic case @xcite and in this sense they include effects beyond the hartree approach through the non - linear couplings @xcite . for infinite nuclear matter all of the gradients of the fields in eqs . ( [ eqfn1])([eqfn5 ] ) vanish and only the @xmath38 , @xmath39 , @xmath40 , @xmath41 and @xmath42 non - linear couplings remain . due to the fact that the solution of symmetric nuclear matter in mean field depends on the ratios @xmath43 and @xmath44 @xcite , we have seven unknown parameters . by imposing the values of the saturation density , total energy , incompressibility modulus and effective mass , we still have three free parameters ( the value of @xmath45 is fixed from the bulk symmetry energy coefficient @xmath46 ) . a possible starting point for our study of the effects of the new terms in the eft energy density , as mentioned in the introduction , is the tm1 parametrization @xcite . first , because it nicely agrees with the dbhf calculations with the bonn - a potential @xcite for a wide range of densities . and second , because it provides good results when applied to finite nuclei calculations , even far away from the @xmath18-stability line . our aim is to study the effects of the new couplings in the description of nuclear matter and finite nuclei and , at the same time , to improve the tm1 parametrization . then , instead of determining the whole set of parameters by a least - squares fit , we will follow a step - by - step strategy , similar to the one used to determine the parameter sets in refs . @xcite in the relativistic framework , or to determine the skyrme skm * parametrization @xcite in the non - relativistic case . according to this strategy we first fix the saturation properties to be those of tm1 and then introduce the coupling @xmath42 and the new scalar - vector non - linear couplings @xmath40 and @xmath41 . this way we can make sure , broadly speaking , that we have the same behaviour of the equation of state around the saturation point as with tm1 . the addition of the couplings @xmath42 , @xmath40 and @xmath41 can not be done with complete freedom once the saturation properties have been fixed . including these extra couplings is translated into a modification of the other coefficients which , eventually , may be driven to non - natural values . an enlarged discussion of this effect can be found in ref . @xcite . to keep all the coefficients within natural values we find that @xmath47 , @xmath48 and @xmath49 is a good choice . it furthermore produces an equation of state and self - energies in better agreement with dbhf than tm1 ( and also contributes to improve the results for @xmath50o that is one of the weak points of tm1 , see section 3 ) . the values of the coupling constants along with the saturation properties are collected in table 1 . we have denoted this set of parameters as tm1*. we can see that @xmath39 is positive and that all the coefficients are natural , i.e. , @xmath51 . the fact that @xmath39 takes a positive value is very gratifying . otherwise the energy spectrum has no lower bound and instabilities in calculations of the equation of state and of finite systems may occur @xcite . figure 1 displays the scalar @xmath52 and vector @xmath53 potentials as a function of the nuclear matter density calculated with tm1 * , tm1 ( that contains a quartic vector self - interaction but not the new couplings ) and with the generalized sets g1 and g2 of ref . @xcite ( effective field theory model ) , in comparison with the dbhf result . we also show the results obtained with the nl3 parameter set @xcite that we have chosen as a representative of the usual non - linear @xmath0 parametrizations with only scalar self - interactions . note that the @xmath39 term of nl3 bears a negative sign ( table 1 ) . figure 2 shows the equation of state for the different approaches . the dbhf predictions are believed to be realistic up to a density , typically , around twice the saturation density @xcite . > from figures 1 and 2 it is clear that the cubic and quartic self - interactions play a crucial role in following the dbhf results at high density . the standard non - linear @xmath0 sets such as nl3 completely fail in doing so : the vector potential grows almost like a straight line and gives a much too stiff equation of state . the quartic vector self - interaction brings down the vector potential and makes the equation of state softer . this softening of the high - density equation of state is needed to be consistent with the observed neutron star masses @xcite . by construction tm1 * gives the same saturation properties as tm1 . however , including the meson interactions @xmath40 and @xmath41 we have been able to reproduce the dbhf results with tm1 * better than with tm1 , for moderate and high densities . we have checked that if one tries to reproduce the dbhf results setting @xmath54 this favours large non - natural values of @xmath42 . if we set @xmath54 and @xmath42 is small ( roughly @xmath55 ) then @xmath39 remains negative . only by introducing the extra constants @xmath40 and @xmath41 one can agree better with dbhf , have @xmath56 and a not very large @xmath42 value . nevertheless , we have found that the contributions of the third - order term @xmath57 and of the fourth - order term @xmath58 in the energy density are far more important than the contribution of the quartic term @xmath59 . in fact , we underline that in our calculation @xmath41 is compatible with a vanishing value , indicating that the comparison with dbhf serves to fix only two couplings of the triad ( @xmath60 ) . on the other hand , the generalized g1 and g2 sets ( that were obtained by fitting finite nuclei observables different from the ones used in tm1 ) also show a good agreement with the dbhf results . from figures 1 and 2 one can see that the results obtained with g1 are similar to those of tm1 , while the predictions of g2 are closer to dbhf . in finite nuclei the contributions from the couplings @xmath25 and @xmath26 between the scalar field and the gradients of the vector and scalar fields , as well as the tensor couplings @xmath61 and @xmath62 of the @xmath13 and @xmath28 mesons to the nucleon , do not vanish . therefore , we have in principle four more parameters to adjust , plus the masses of the @xmath1 , @xmath13 and @xmath28 mesons ( or , equivalently , the coupling constants @xmath63 , @xmath64 and @xmath65 ) . in accordance with our strategy , we will fix the meson masses of tm1 * to the same values of tm1 : @xmath66 mev , @xmath67 mev and @xmath68 mev ( the nucleon mass is @xmath69 mev ) . in this way we do not mask the influence of the terms that we want to study . in our numerical calculation of finite nuclei we have transformed the dirac equation ( [ eqfn12 ] ) into a schrdinger - like equation by eliminating the small component of the wave function . this equation is solved by using a standard code for non - relativistic skyrme hartree fock calculations @xcite . in the calculations performed with the improved tm1 set ( tm1 * ) we use the same center - of - mass correction for energies and charge radii as sugahara and toki @xcite used for tm1 : @xmath70 where @xmath71 mev , @xmath72 is the harmonic oscillator parameter and 0.64 @xmath73 takes into account the finite size correction of the proton @xcite . we obtain the coupling constants @xmath25 , @xmath26 , @xmath61 , @xmath42 , @xmath40 and @xmath41 ( the last three combined with the nuclear matter calculation as explained in section 2 ) by imposing that the total energy , the charge radius and the 1@xmath74 splitting for neutrons and protons of the symmetric nucleus @xmath50o be as close as possible to the experimental values . to deal with asymmetric nuclei the @xmath75 , @xmath76 and @xmath77 couplings are needed . following ref . @xcite we fix the volume asymmetry coefficient @xmath78 to be 36.9 mev . actually this value corresponds to the difference between the neutron and nuclear matter dbhf energies per particle @xcite calculated at the nuclear matter saturation density , which is known to be a good approach for estimating @xmath46 @xcite . then we determine @xmath27 and @xmath77 so that the energy of @xmath79pb be the experimental one . the tensor coupling @xmath62 happens to be useless in our fitting : its contribution , as previously reported @xcite , is negligible and we have taken @xmath80 for tm1*. this is not the case for the coupling @xmath27 , whose influence is noticeable @xcite . as a final step in our fitting procedure we have to check that the values of all the parameters are natural . the whole set of parameters of tm1 * is given in table 1 . we should like to discuss the systematics of the finite nuclei properties with the new couplings in some more detail . the bulk parameters ( @xmath60 ) only have a slight influence on the binding energies ( @xmath81 ) and charge radii ( @xmath82 ) , and practically no effect on the spin - orbit splittings ( @xmath83 ) . the incidence of @xmath41 is again negligible compared to @xmath42 and @xmath40 . however , if ( @xmath60 ) are given the wrong values then it may be impossible to correct the results for @xmath81 and @xmath82 with only the surface parameters ( @xmath84 ) . thus , one first needs a reasonable ansatz for ( @xmath60 ) to be able to get acceptable values for @xmath81 and @xmath82 . it was not obvious a priori that the ( @xmath60 ) values favoured by the comparison with dbhf would fall into this category . if all other parameters are kept fixed , decreasing @xmath25 makes @xmath81 and @xmath85 larger and @xmath82 smaller ( we define @xmath81 to be positive ) . the coupling @xmath26 has just the opposite effect . for the same change in @xmath25 as in @xmath26 , the modifications on @xmath81 , @xmath82 and @xmath85 are roughly twice larger with @xmath25 than with @xmath26 . once a set of ( @xmath60 ) values is specified , @xmath61 serves to bring the strength of @xmath85 closer to the desired value . then the couplings @xmath25 and @xmath26 are used for the fine tuning of the @xmath81 , @xmath82 and @xmath85 values . we point out that after specifying ( @xmath60 ) almost the same @xmath81 , @xmath82 and @xmath85 are obtained with many distinct families of ( @xmath86 ) values . in principle , we have realized that making @xmath25 small or negative and readjusting @xmath26 to recover the same binding energies and spin - orbit splittings , helps one to eventually get slightly larger radii . that is , from the interplay of @xmath25 and @xmath26 it is possible to achieve some change in the value of @xmath82 relative to the value of @xmath81 , but the effect is not very significant . we are led to conclude that at least one of the three couplings ( @xmath84 ) is not singled out by the properties analyzed , and that a correlation exists between these surface parameters and the bulk parameters ( @xmath87 ) . as a first test of the full tm1 * parametrization we have calculated the surface energy coefficient @xmath88 and the surface thickness @xmath89 of the density profile ( standard 90%-10% fall - off distance of the nuclear density ) in semi - infinite nuclear matter . the results are shown in table 2 . the surface energy obtained with tm1 * lies within the region of empirical values , whereas the surface thickness @xmath89 is slightly small @xcite . the energies , charge radii and spin - orbit splittings of the magic nuclei @xmath50o and @xmath79pb used in our fit as well as the values for @xmath90ca , @xmath91ca and @xmath92zr , which were included in the fit of tm1 @xcite , are also displayed in table 2 . we show the experimental values and the results obtained with the sets tm1 * , tm1 and nl3 @xcite ( non - linear model with only scalar self - interactions ) , and with the generalized parameter sets g1 and g2 of ref . @xcite . in addition to the couplings listed in table 1 for g1 and g2 , these sets have a few more parameters related with the electromagnetic structure of the pion and the nucleon ( see ref . @xcite ) , which we have taken into account for table 2 . the tm1 results for @xmath50o are given here for completeness , as we recall that tm1 was devised for heavier nuclei @xcite . concerning the influence of the centre - of - mass motion on the energy and the charge radius , it should be noted that different parametrizations use , in general , different prescriptions . due to the fact that the centre - of - mass corrections are included in the fit of the parameters , we report in table 2 the values we have obtained with the same prescription as the authors used in their original works . the tm1 * calculations for magic nuclei displayed in table 2 reproduce the experimental energies within @xmath93 and the charge radii and spin - orbit splittings with a similar quality to the successful nl3 @xcite , g1 or g2 @xcite parametrizations . in order to check the ability of tm1 * for describing nuclei far from the stability line , we have calculated the energy and charge radius of some drip - line ( double - closed shell ) nuclei , namely @xmath94ni , @xmath95ni , @xmath96sn and @xmath97sn . table 2 shows that all the forces considered here produce similar results for the energy per particle and the charge radius of finite nuclei , which agree well with experiment . the single - particle energies of neutrons and protons are compared with the experimental data in figures 3a and 3b for the @xmath79pb nucleus with the tm1 * , tm1 and nl3 sets . one can see that all these parametrizations qualitatively describe the experimental values . although the nuclear matter properties are equal in tm1 and tm1 * , the spectra are slightly different mainly due to the tensor coupling @xmath61 present in tm1 * , which has a noticeable influence in the spin - orbit potential @xcite . to describe even - even nuclei other than double magic nuclei we introduce the pairing correlation in the bcs approximation with a constant gap @xmath98 , as in earlier calculations @xcite . it is to be kept in mind that the seniority pairing recipe is not appropriate for exotic nuclei near the drip lines because the coupling to the continuum is not treated properly . the fact that continuum states become significantly populated as one approaches the drip lines can be taken into account in the relativistic hartree - plus - bogoliubov ( rhb ) method @xcite . however , it has been pointed out in ref . @xcite that a qualitative estimation of the drip lines can be obtained within the bcs scheme by taking into account some quasi - bound states owing to their centrifugal barrier which mocks up the influence of the continnum . in order to be as consistent as possible with tm1 here we take the gap energy @xmath99 mev , that corresponds to the widely used phenomenological formula of bohr and mottelson @xcite . in practice we have found that the same gap energy is obtained by fitting the sn isotopic energy difference @xcite @xmath100_{\rm bcs } - [ e - e(^{116 } { \rm sn } ) ] _ { \rm exp . } \label{eqfn22}\ ] ] calculated with tm1*. we restrict the number of active shells to the occupied shells contained in a major harmonic oscillator shell above and below of the last closed shell . when the nuclei approach the drip lines there are not bound single - particle levels above the chemical potential . in this case we take the bound - state contributions as well as those coming from quasi - bound states at positive energies @xcite . in table 3 we report the energy and charge radii of @xmath101ni , @xmath102sn and @xmath103pb that were used in the tm1 fit . tm1 * shows an agreement with experiment similar to that found for tm1 . apart from the results presented in tables 2 and 3 , we have compared the energy given by tm1 * for several light nuclei of @xmath104 with the results given by tm2 @xcite ( as tm2 was designed for @xmath105 ) and with the tm1 results . tm1 * improves the tm1 results in this region and the quality of the energies is similar to that of tm2 . in the following we will calculate isotopic and isotonic energy differences , isotopic shifts in charge radii and two - neutron and two - proton separation energies near and away from the @xmath18-stability line , to examine whether tm1 * is also acceptable for these properties in comparison with experiment and with other relativistic sets . we have calculated the isotopic energy differences @xmath106 for several sn and pb isotopes ( referred to @xmath107sn and @xmath79pb , respectively ) with the tm1 * , tm1 and nl3 parametrizations . the results are shown in figures 4a and 4b . in the case of the sn isotopes there are some differences between the nl3 results and those of tm1 or tm1*. the nl3 isotopic energy differences appreciably deviate from the experiment for sn isotopes with a neutron number @xmath108 larger than 66 , while the tm1 or tm1 * results remain close to the experimental values . if we compare with the non - relativistic calculations performed in ref . @xcite with the skyrme forces sly4 and skm * , the nl3 results qualitatively behave as those of skm * , whereas the tm1 and tm1 * predictions are closer to those of sly4 . the results for the pb isotopes are shown in figure 4b , and for @xmath109 isotones ( referred to @xmath97sn ) in figure 4c . the tm1 * , tm1 and nl3 sets show different trends for the lead isotopes . tm1 predicts better @xmath110 values over the other parameter sets . for the @xmath111 isotopes tm1 and tm1 * predict an arch structure similar to the one found with the sly4 interaction @xcite , while nl3 shows a structure more similar to the skm * force @xcite . for @xmath112 , @xmath110 increases as a function of @xmath108 for the three relativistic sets , similarly to the sly4 calculation @xcite . the @xmath113 isotone energy differences found with nl3 , tm1 and tm1 * show a completely different behaviour compared with sly4 and skm*. in the relativistic case @xmath110 decreases with increasing @xmath15 up to @xmath114 and increases afterwards . the largest separation with respect to the experimental value corresponds to @xmath114 , although the difference is more pronounced for tm1 and tm1 * than for nl3 . in the non - relativistic calculations @xcite one finds an arch structure with sly4 , with the largest difference with experiment corresponding to @xmath114 ( although it is positive in this case ) , and a monotonous increasing of @xmath110 as a function of @xmath15 with skm*. this qualitatively different behaviour in the @xmath113 isotopic chain could be caused by the different pairing interaction used in the present calculation ( constant gap ) and in the non - relativistic calculations of ref . @xcite where a density - dependent zero - range pairing force ( more similar to a constant strength ) was considered . in the past years the isotopic shifts in charge radii have been studied for the isotopic chain of pb nuclei using various techniques @xcite . non - linear @xmath0 calculations with a constant gap pairing interaction in general reproduce the experimentally observed kink in the isotopic shifts about @xmath79pb @xcite . however , the standard non - relativistic ( zero range or finite range ) forces are not able to describe this kink . only by improving the pairing interaction and by taking into account some terms usually not considered in the skyrme functional and the two - body center - of - mass correction , the non - relativistic results agree with the experimental observation @xcite . here we have calculated the pb isotopic shifts with the tm1 * parameter set . in figure 5 the result is compared with the prediction of the tm1 and nl3 sets , and also with the experimental data . all these parameter sets yield qualitatively similar results and reproduce the experimental kink reasonably well . notice that these pb isotopic shifts are not included in the tm1 and tm1 * fits . we have evaluated the two - neutron @xmath115 and two - proton @xmath116 separation energies from the calculated energies using @xcite @xmath117 the @xmath115 values for the illustrative cases of @xmath118 and 50 as well as the @xmath116 value for @xmath113 with the tm1 * , nl3 and tm1 sets are presented in figures 6a , 6b and 6c , respectively . the experimental data are also given for comparison . on the wole , the @xmath115 and @xmath116 values obtained from tm1 * agree well with the experimental observation and also with the predictions of tm1 and nl3 ( except for a slight discrepancy for some specific cases ) . in concrete , for ca isotopes the shell effects at @xmath119 and 28 are well reproduced by the three relativistic sets . another shell effect is predicted at @xmath120 although no experimental information is available to confirm it . something similar happens for sn isotopes at @xmath121 and at @xmath113 where the calculated results qualitatively agree with the experimental value . with respect to the isotone chain of @xmath113 no experimental information exists to confirm the shell effect at @xmath122 . in this case the relativistic sets are not able to quantitatively reproduce the experimental @xmath116 energies in the @xmath12358 region , due maybe to the adopted pairing scheme . the @xmath115 value decreases with increasing the neutron number and vanishes at the neutron - drip line . similarly , the @xmath116 value decreases with increasing proton number as the proton - drip line is reached . as we have mentioned , although the bcs approach to pairing correlations is not well suited for dealing with the drip lines @xcite , an estimate can be given with the present bcs calculation that takes into account some continuum effects through the quasi - bound levels @xcite . the authors of ref.@xcite discussed the unability of tm1 for describing zr isotopes with @xmath108 larger than 82 , while they could be described with the nl1 parametrization . we have found that to be able to describe these nuclei , which have a chemical potential close to zero , it is crucial for the bcs calculation to take into account the quasi - bound levels @xmath124 , @xmath125 and @xmath126 that lie a few mev above the fermi level . in this way we have estimated the neutron - drip line for zr at @xmath127 . similarly , we have estimated the neutron - drip line for ca , sn and pb isotopes at @xmath128 62 , 164 and 264 , respectively , for all the analysed parameter sets . these bcs estimates are in good agreement with the results of the non - relativistic interactions sly4 and skm * reported in ref . @xcite , where almost the same technique was used for dealing with the pairing correlations . for @xmath113 isotones we find the proton - drip line at @xmath129 , which corresponds to @xmath130w in agreement with experimental information @xcite . we have explored whether the parameter set tm1 @xcite can be improved by adding new couplings that stem from the modern effective field theory approach to relativistic nuclear phenomenology . we have been concerned with analyzing the possibilities of the new couplings to ensure a reasonable agreement with the density dependence of the scalar and vector components of the dbhf self - energies , while performing well for finite nuclei . the extended parameter set has been called tm1*. it is able to reproduce ground - state properties of spherical nuclei for @xmath17 with a quality similar to conventional sets like nl1 or nl3 , and with the appealing feature of having a positive quartic scalar self - coupling . this could not be achieved with the set tm1 which had to be restricted to @xmath15 larger than 20 in order to keep @xmath39 positive @xcite . it is important to note that this limitation seems to be common to any set of parameters containing only a quartic vector self - interaction on top of the standard non - linear @xmath0 model . to check this point we have performed calculations with the recently proposed nl - sv1 and nl - sv2 parameter sets @xcite that include a quartic vector self - coupling ( like tm1 ) . for light nuclei we find a good agreement with experiment when we use the nl - sv1 set which has a negative @xmath39 coupling , whereas this is not the case with the nl - sv2 set where @xmath39 is positive . in comparison with the dbhf results in nuclear matter the extended set tm1 * shows a significant improvement over tm1 due to the addition of the @xmath40 and @xmath41 couplings . the latter couplings ( at least @xmath40 ) are very helpful to bring the vector and scalar potentials closer towards the dbhf calculations as the density grows . to the end of computing finite nuclei we have introduced the @xmath61 , @xmath25 , @xmath26 , @xmath27 and @xmath62 parameters on top of the set that describes nuclear matter . we remark that the new parameters have a minor influence on the investigated properties of finite nuclei . however , they allow the full tm1 * force to improve the agreement with experiment for double - closed shell nuclei compared with the starting tm1 parameters and to obtain better results for light - mass nuclei , which was a shortcoming of the tm1 set . we also have tested the tm1 * force for isotopic energy differences , isotopic changes in charge radii and two - neutron and two - proton separation energies . nuclei near the drip lines have been explored for some particular cases by taking into account quasi - bound states in the bcs calculation following the method of ref . it should be mentioned that by including all of the relevant couplings in the energy density expansion compatible with the eft approach to qhd , as developed in refs . @xcite , the tm1 * model is more consistent with our current understanding of effective field theories . nevertheless , we have seen that some of the new couplings of the eft model remain underdetermined in spite of the information taken into account about the equation of state and the self - energies at higher densities . in conclusion , the relativistic mean field approach extended by the new non - linear meson self - interactions and tensor couplings based upon effective field theory , allows one to reproduce at the same time the trends of microscopic dbhf calculations up to relatively high densities and various finite nuclei properties . in the low - density domain ( that corresponds to the finite nuclei region ) the main properties are almost fixed by the nuclear matter properties around saturation , and then the new parameters have only a small contribution . however , as the density increases the vector - vector and scalar - vector meson interactions play an important role in providing enough flexibility to the model to be able to follow the tendency of the dbhf calculations . extended sets like tm1 * may be more useful for systems having relatively higher density and temperature , whereas they will serve the same purpose for normal systems as the conventional parameter sets . to further constrain the new eft parameters additional observables will be required . nuclear phenomena involving currents could prove helpful for couplings such as @xmath25 and @xmath26 that imply the derivatives of the fields . on the side of the isovector channel , information from many - body dbhf calculations of asymmetric and neutron matter as well as data on neutron radii and the neutron skin thickness should be relevant . the authors would like to acknowledge support from the dgicyt ( spain ) under grant pb98 - 1247 and from dgr ( catalonia ) under grant 1998sgr-00011 . m.d.e . acknowledges financial support from the cirit ( catalonia ) . thanks the spanish education ministry grant sb97-ol174874 for financial support and the departament destructura i constituents de la matria of the university of barcelona for kind hospitality . 99 = -1.5 mm r. machleidt , adv . nucl . phys . * 19 * , 189 ( 1989 ) . c. j. horowitz and b. d. serot , phys . b * 137 * , 283 ( 1984 ) . m. r. anastasio , l. s. celenza , w. s. pong , and c. m. shakin , phys . 100 * , 327 ( 1983 ) . r. brockmann and r. machleidt , phys . b * 149 * , 283 ( 1984 ) . b. ter haar and r. malfliet , phys . rep . * 149 * , 207 ( 1987 ) . b. d. serot and j. d. walecka , adv . * 16 * , 1 ( 1986 ) . b. d. serot and j. d. walecka , int . j. of mod . e * 6 * , 515 ( 1997 ) . j. d. walecka , ann . ( n.y . ) * 83 * , 491 ( 1974 ) . j. boguta and a. r. bodmer , nucl . * a292 * , 413 ( 1977 ) . r. j. furnstahl , b. d. serot , and h. b. tang , nucl . * a598 * , 539 ( 1996 ) . r. j. furnstahl , b. d. serot , and h. b. tang , nucl . * a615 * , 441 ( 1997 ) . h. mller and b. d. serot , nucl . * a606 * , 508 ( 1996 ) . j. j. rusnak and r. j. furnstahl , nucl . a627 * , 495 ( 1997 ) . b. c. clark , r. j. furnstahl , l. k. kerr , j. rusnak , and s. hama , phys . b * 427 * , 231 ( 1998 ) . r. j. furnstahl and b. d. serot , nucl . * a671 * , 447 ( 2000 ) . m. del estal , m. centelles , and x. vias , nucl . * a650 * , 443 ( 1999 ) . r. j. furnstahl , j. j. rusnak , and b. d. serot , nucl . * a632 * , 607 ( 1998 ) . wang , phys . c * 61 * , 054904 ( 2000 ) . f. coester , s. cohen , b. d. day , and c. m. vincent , phys . c * 1 * 769 ( 1970 ) . a. amorim and j. a. tjon , phys . lett . * 68 * , 772 ( 1992 ) . s. gmuca , j. phys . g * 17 * , 1115 ( 1991 ) . m. rashdan , phys . b * 395 * , 141 ( 1997 ) . s. gmuca , z. phys . a * 342 * , 387 ( 1992 ) ; nucl . * a547 * , 447 ( 1992 ) . a. r. bodmer , nucl . * a526 * , 703 ( 1991 ) ; a. r. bodmer and c. e. price , nucl . * a505 * , 123 ( 1989 ) . y. sugahara and h. toki , nucl . * a579 * , 557 ( 1994 ) . reinhard , m. rufa , j. maruhn , w. greiner , and j. friedrich , z. phys . a * 323 * , 13 ( 1986 ) . k. sumiyoshi , h. kuwabara , and h. toki , nucl . a581 * , 725 ( 1995 ) . g. a. lalazissis , j. knig , and p. ring , phys . c * 55 * , 540 ( 1997 ) . y. k. gambhir , p. ring , and a. thimet , ann . ( n.y . ) * 198 * , 132 ( 1990 ) . m. rufa , p. -g . reinhard , j. a. maruhn , w. greiner , and m. r. strayer , phys . c * 38 * , 390 ( 1988 ) . s. k. patra and c. r. praharaj , phys . c * 44 * , 2552 ( 1991 ) . d. vautherin and d. m. brink , phys . c * 5 * , 626 ( 1972 ) . j. decharg and d. gogny , phys . c * 21 * , 1568 ( 1980 ) . c. j. horowitz and j. piekarewicz , astro - ph/0010227 ( 2000 ) . w. kohn and l. j. sham , phys . a * 140 * , 1133 ( 1965 ) . c. speicher , r. m. dreizler , and e. engel , ann . ( n.y . ) * 213 * , 413 ( 1992 ) . r. brockmann and r. machleidt , phys . c * 42 * , 1965 ( 1990 ) . c. j. horowitz and b. d. serot , nucl . * a368 * , 503 ( 1981 ) . a. bouyssy , s. marcos , and pham van thieu , nucl . phys . * a422 * , 541 ( 1984 ) . j. bartel , p. quentin , m. brack , c. guet , and h. -b . hkansson , nucl . * a386 * , 79 ( 1982 ) . g. baym , phys . rev . * 117 * , 886 ( 1960 ) ; b. m. waldhauser , j. a. maruhn , h. stcker , and w. greiner , phys . c * 38 * , 1003 ( 1988 ) ; p. -g . reinhard , z. phys . a * 329 * , 257 ( 1988 ) . j. w. negele , phys . c * 1 * , 1260 ( 1970 ) . g. q. li , r. machleidt , and r. brockmann , phys . c * 45 * , 2782 ( 1992 ) . l. engvik , m. hjorth - jensen , r. machleidt , h. mther , and a. polls , nucl . * a627 * , 85 ( 1997 ) . m. centelles and x. vias , nucl . * a563 * , 173 ( 1993 ) ; m. centelles , m. del estal , and x. vias , nucl . a635 * , 193 ( 1998 ) . t. gonzalez - llarena , j. l. egido , g. a. lalazissis , and p. ring , phys . b * 379 * , 13 ( 1996 ) . g. a. lalazisis , d. vretenar , and p. ring , phys . c * 57 * , 2294 ( 1998 ) . m. m. sharma , a. r. farhan , and s. mythili , phys . c * 61 * , 054306 ( 2000 ) . e. chabanat , p. bonche , p. haensel , j. meyer , and r. schaeffer , nucl . phys . * a635 * , 231 ( 1998 ) . a. bohr and b. r. mottelson , `` nuclear structure '' , vol . i ( w.a . benjamin , new york , 1969 ) , ch . 2 . n. tajima , p. bonche , h. flocard , p. -h . heenen , and m. s. weiss , nucl . * a551 * , 434 ( 1993 ) . m. sharma , g. a. lalazissis , and p. ring , phys . b * 317 * , 9 ( 1993 ) . j. dobaczewski , t. r. werner , j. f. berger , c. r. chenn , and j. deharge , phys . c * 53 * , 2809 ( 1996 ) ; j. terasaki , p. -h . heenen , h. flocard , and p. bonche , nucl . * a600 * , 371 ( 1996 ) . r. d. page , p. j. woods , r. a. cunningham , t. davidson , n. j. davis , s. hofmann , a. n. james , k. livingston , p. j. sellin , and a. c. shotter , phys . rev . lett . * 68 * , 1287 ( 1992 ) . figure 1 . : : scalar @xmath52 and vector @xmath53 potentials against the nuclear matter density as obtained in a dbhf calculation with the bonn - a potential @xcite and with the relativistic mean field parametrizations tm1 * , tm1 @xcite , g1 and g2 @xcite . : : equation of state for the same cases as in figure 1 . : : the single - particle energies for @xmath79pb obtained by various relativistic mean - field parametrizations are compared with the experimental data for neutrons ( a ) and protons ( b ) . : : the isotopic energy difference obtained with the tm1 * parameter set is compared with the tm1 and nl3 calculations for sn isotopes ( a ) and pb isotopes ( b ) . plot ( c ) shows the isotonic energy difference for @xmath113 . : : the isotopic shifts in charge radii for the @xmath131 chain . figure 6 . : : the calculated separation energies are compared with the experimental data : ( a ) two - neutron separation energy @xmath115 for @xmath132 , ( b ) two - neutron separation energy for @xmath122 , and ( c ) two - proton separation energy @xmath116 for @xmath113 . .various parameter sets for the relativistic energy density functional and the corresponding saturation properties . the coupling constants are dimensionless . [ cols="^,^,^,^,^,^,^,^,^,^,^",options="header " , ]

we adapt the notion of generating functions for lagrangian submanifolds to symplectic microgeometry . we show that a symplectic micromorphism always admits a global generating function . as an application , we describe hamiltonian flows as special symplectic micromorphisms whose local generating functions are the solutions of hamilton - jacobi equations . we obtain a purely categorical formulation of the temporal evolution in classical mechanics .

1103.0672

this article is a continuation of @xcite , in which we introduced the notion of symplectic microfolds and symplectic micromorphisms between them . recall that a symplectic microfold is essentially the same thing as a germ @xmath0 $ ] of symplectic manifolds @xmath1 around a lagrangian submanifold @xmath2 , called the core . a symplectic micromorphism from @xmath0 $ ] to @xmath3 $ ] is a germ around the graph of a smooth map @xmath4 , the core map , of a canonical relation of a symplectic manifold @xmath5 is the manifold @xmath1 endowed with the opposite symplectic form @xmath6 . ] @xmath7 containing the graph of @xmath8 and satisfying a certain transversality condition ; we review the definitions in section [ subsec - definitions ] below . in contrast with their macroworld counterparts ( i.e. canonical relations ) , symplectic micromorphisms always compose well , forming thus , with the symplectic microfolds as objects , an honest symmetric monoidal category : the microsymplectic category , which can be thought as the appropriate `` microworld '' analog to the symplectic `` category '' of symplectic manifolds and canonical relations @xcite . this makes the microsymplectic category a natural setting for questions related to the functorial behavior of quantization schemes in symplectic geometry . by `` quantization scheme '' , we mean here any well - defined correspondence between a certain type of geometric structures in the symplectic realm ( such as poisson structures ) and a certain type of algebraic structures in the realm of analysis ( such as @xmath9-algebras ) . two important ingredients involved in these quantization questions have been the generating functions and the fourier integral operators associated with a given lagrangian submanifold ( see @xcite ) . in this paper , we extend the notion of generating functions to the microworld . our first main result is that _ any _ symplectic micromorphism admits a _ global _ generating function . this is the best possible case when it comes to quantization via fourier integral operators ( as will be shown in a sequel @xcite ) . a second result ( theorem [ thm : normal form ] ) states that the underlying lagrangian submicrofold of a symplectic micromorphism can always be decomposed into a fibration by graphs of actual smooth map germs . the main example here is the cotangent lift @xmath10 of a smooth map @xmath4 , whose underlying lagrangian submicrofold is the germ of the lagrangian submanifold @xmath11 around the graph of @xmath8 . when @xmath8 is not a diffeomorphism , @xmath12 is not the graph of a symplectomorphism , although the cotangent maps @xmath13 to @xmath8 at each @xmath14 _ are _ actual maps . this gives us a decomposition @xmath15 of @xmath12 associated with the lagrangian fibration of @xmath16 by its cotangent fibers . more generally , a similar decomposition holds for general symplectic micromorphisms from @xmath0 $ ] to @xmath3 $ ] , which is uniquely associated with the data of a lagrangian fibration of @xmath3 $ ] . this special geometry of the underlying lagrangian submicrofolds may be of help in questions related to the continuity of their corresponding fourier integral operators . namely , in various instances ( @xcite,@xcite,@xcite ) , the continuity in @xmath17-spaces of some classes of fios has been related to their wave - fronts being local graphs . from a different perspective , this decomposition is reminiscent of the notion of co - morphisms of chen and liu introduced in @xcite in the context of lie groupoids . we derive the existence of global generating functions for symplectic micromorphisms from a more general result : the _ equivalence theorem _ for clean lagrangian submicrofolds ( theorem [ thm : equivalence theorem ] ) . it states that the data of a germ @xmath18 $ ] of a lagrangian submanifold @xmath19 that intersects the zero section _ cleanly _ in @xmath20 is equivalent to the data @xmath21 of a symplectomorphism germ , when it is clear from the context , we write @xmath22 to denote the microfold @xmath23 $ ] , where @xmath24 is the zero section of @xmath22 . moreover , we also identify @xmath24 with @xmath2 and the submanifolds @xmath25 with the corresponding submanifolds of @xmath24 , yielding the notation @xmath26 $ ] . ] @xmath27 $ ] ( of the type prescribed by the lagrangian embedding theorem @xcite applied to the conormal bundle @xmath28 and satisfying an extra condition ) together with a function germ @xmath29 around the zero section ( and vanishing on it , as well as its differential ) . actually , we prove that for each @xmath30 there exists a unique @xmath31 such that @xmath32 . this defines the global generating function @xmath31 of the clean lagrangian submicrofold @xmath18 $ ] associated with the symplectomorphism germ @xmath30 . it turns out that the symplectic micromorphisms from @xmath33 to @xmath16 with core map @xmath8 are nothing but the _ clean _ lagrangian submicrofolds in @xmath34 with core @xmath35 . while giving another ( and simpler ) characterization of symplectic micromorphisms , the equivalence theorem also tells us that they admit global generating functions ( associated with special symplectomorphism germs @xmath30 as before ) . another application of the equivalence theorem comes from considering the restriction of a symplectic micromorphism to a local chart . the restriction is again a symplectic micromorphism , but now we have a canonical symplectomorphism germ @xmath30 coming from the affine structure of the local chart . this allows us to define the _ local _ generating function of the symplectic micromorphism in the local chart as the global generating function of its restriction . we conclude this paper by associating with any hamiltonian flow on a cotangent bundle a symplectic micromorphism , the evolution micromorphism , which encodes the dynamics for asymptotically short times . we show that the local generating functions of the evolution micromorphism coincide with the solutions of the hamilton - jacobi equation for the flow in local charts . moreover , by considering a particular monoid object in the microsymplectic category , the energy monoid @xmath36 , we show that the @xmath36-modules are essentially the same thing as germs of hamiltonian flows with time - independent hamiltonian . this gives us a categorical formulation for the temporal evolution in classical mechanics . finally , we briefly describe how symmetries in classical mechanics can be formalized using the language of symplectic microgeometry . this approach to symmetry , which will be developed in full details somewhere else , is very close in spirit to the work of benenti on the hamilton - jacobi equation for an hamiltonian action ( @xcite ) . a.s.c . acknowledges partial support from snf grant 200020_131813/1 . thanks pedro de m. rios for useful discussions on generating families and acknowledges partial support from nwo grant 613.000.602 carried out at utrecht university and from fapesp grant 2010/15069 - 8 hosted by the icmc of sao paulo university at sao carlos . acknowledges partial support from nsf grant dms-0707137 and the hospitality of the institut mathmatique de jussieu . in this section , we start by recalling some basic definitions concerning the geometry of manifold germs or microfolds ( see @xcite for more details ) . then we focus on the geometry of lagrangian submanifolds in cotangent bundles around their intersection with the zero section . this geometry is captured by the notion of a lagrangian submicrofold @xmath18 $ ] in @xmath37 $ ] , that is , a germ @xmath38 $ ] of lagrangian submanifolds @xmath19 around @xmath25 . the main result of this section is the equivalence theorem ( theorem [ thm : equivalence theorem ] ) which states that the lagrangian submicrofolds @xmath18 $ ] that intersect the zero section _ cleanly _ in @xmath20 coincide with certain deformations of the conormal microbundle @xmath39 $ ] . moreover , the cleanliness assumption is enough to define a notion of global generating function for the lagrangian submicrofold . a * microfold * is an equivalence class @xmath40 $ ] of manifold pairs @xmath41 , where @xmath2 is a closed submanifold of @xmath42 ; two pairs @xmath43 and @xmath44 are equivalent if there exists a third one @xmath45 for which @xmath46 is simultaneously an open submanifold of both @xmath47 and @xmath48 . the manifold @xmath2 is called the * core * of the microfold @xmath40 $ ] . in other words , @xmath40 $ ] is a manifold germ around @xmath2 . a morphism @xmath49:[g , a]\to [ h , b]$ ] between microfolds is an equivalence class of smooth maps @xmath50 between representatives , where two such maps are equivalent if there is a common neighborhood of @xmath2 on which they are both defined , and equal . the morphism is an isomorphism when there is a representative map which is a diffeomorphism . any such isomorphism induces a diffeomorphism between the cores . a * microbundle * is a microfold @xmath23 $ ] obtained from a vector bundle @xmath51 by taking the germ of @xmath22 around the zero section @xmath24 . ( every microfold is diffeomorphic to a microbundle . ) when clear from the context , we will write @xmath22 instead of @xmath23 $ ] . throughout , we will use the canonical identifications @xmath52 where the * vertical * * bundle * @xmath53 of a vector bundle @xmath54 is the subbundle of @xmath55 whose fiber at @xmath56 is the kernel of @xmath57 . if @xmath20 is a submanifold of @xmath2 and @xmath51 a vector bundle , we will write @xmath58 to denote the restriction of @xmath22 along @xmath20 . a * submicrofold * @xmath59 $ ] of @xmath40 $ ] is a microfold for which there are representatives @xmath60 and @xmath41 such that @xmath61 is a submanifold of @xmath2 and @xmath62 is a submanifold of @xmath42 , and @xmath63 . we say that @xmath59 $ ] is a * clean * submicrofold if the intersection of @xmath62 with the core @xmath2 is clean ( i.e. @xmath64 ) . a * symplectic microfold * is a microfold @xmath0 $ ] such that @xmath1 is a symplectic manifold and @xmath2 is a lagrangian submanifold . a * lagrangian splitting * of a symplectic microfold @xmath0 $ ] is a lagrangian subbundle @xmath65 of @xmath66 that is transverse to @xmath2 . a * lagrangian submicrofold * @xmath18 $ ] of @xmath0 $ ] is a submicrofold such that @xmath67 is lagrangian in @xmath1 . a * symplectic micromorphism * from @xmath0 $ ] to @xmath3 $ ] is a lagrangian submicrofold @xmath68 \subset [ \overline{m}\times n , a\times b],\ ] ] whose core is the graph of a smooth map @xmath4 and that satisfies the following _ transversality condition _ : we require @xmath69 to be transverse along @xmath35 to a ( and hence any ) lagrangian subbundle of the form @xmath70 where @xmath30 is a lagrangian splitting of @xmath3 $ ] . to distinguish symplectic micromorphisms from other lagrangian submicrofolds , we use the special notation @xmath71,\phi):[m , a]\rightarrow [ n , b ] $ ] instead of @xmath72 $ ] . the definition of symplectic micromorphisms above differs from the one given in ( * ? 3.1 ) but is equivalent to it as stated in ( * ? ? * cor 3.1 ) . it is better suited to our purposes since we will mostly be dealing here with symplectic micromorphisms @xmath71,\phi):{{t}^*}a\rightarrow { { t}^*}b$ ] between cotangent microbundles . in this case , we have a canonical lagrangian subbundle @xmath73 where @xmath74 is the vertical subbundle of @xmath16 . the symplectic micromorphisms from @xmath33 to @xmath16 are thus the lagrangian submicrofolds @xmath72 $ ] that are transverse to @xmath75 . the symplectic microbundles and symplectic micromorphisms between them form a symmetric monoidal category ; see @xcite . let us recall here that the tensor product of two microfolds is derived from the usual cartesian product of manifolds in the obvious way : @xmath76\otimes [ n , b ] : = [ m\times n , a\times b].\ ] ] in turn , the tensor product of two symplectic micromorphisms is given by the tensor product of their underlying submicrofolds as above . we now define two _ a priori _ different classes of lagrangian submicrofolds @xmath18\subset{{t}^*}a$ ] , each of which has the conormal microbundle @xmath77 as one of its members : the class of _ strongly transverse lagrangian submicrofolds _ and the class of _ conormal microbundle deformations_. theorem [ thm : equivalence theorem ] proves that these classes coincide with one another and with the class of clean lagrangian submicrofolds . [ def : w - distribution]let @xmath20 be a submanifold of @xmath2 and let @xmath78 be a complementary subbundle to @xmath79 in @xmath80 . we define the subbundle @xmath81 of @xmath82 by setting @xmath83 where @xmath84 is the annihilator of @xmath85 ( i.e. the subspace of the covectors in @xmath86 that vanish on @xmath85 ) for each @xmath87 . [ lem : w - distributions]all lagrangian subbundles of @xmath82 of the form are transverse to @xmath77 along @xmath20 . one checks that the symplectic orthogonal @xmath88 is contained in @xmath89 , and we conclude @xmath89 is a lagrangian subspace by dimension count . to see that @xmath90 is transverse to @xmath77 , we observe that @xmath91 . by definition , @xmath85 is transverse to @xmath92 . this implies that their respective annihilator @xmath93 and @xmath94 are also transverse in @xmath95 . we say that a lagrangian submicrofold @xmath18 $ ] of @xmath33 is * strongly transverse * if it is transverse to all lagrangian subbundles of the form . the conormal bundle @xmath77 is strongly transverse . let @xmath96 be a one - form on @xmath2 , which we regard as a map @xmath97 . its image @xmath98 is a lagrangian submanifold of @xmath33 if and only if @xmath96 is closed . we say that @xmath98 is a * projectable * lagrangian submanifold of @xmath33 since the restriction of the canonical projection @xmath99 to @xmath98 is a diffeomorphism . conversely , all projectable lagrangian submanifolds of @xmath33 are of this form . now if @xmath96 vanishes on some submanifold @xmath25 , then @xmath98 contains @xmath20 ( or , more precisely , contains the corresponding submanifold of the zero section @xmath24 ) , and we can consider the induced lagrangian submicrofold @xmath100 $ ] , which depends only on the germ of @xmath96 around @xmath20 . the class of lagrangian submicrofolds @xmath18 $ ] in @xmath33 whose image through some special type of symplectomorphism germ from @xmath101 $ ] to @xmath102 is projectable will be very important for us in the sequel : [ def : conormal deformation ] we say that a lagrangian submicrofold @xmath18 $ ] is a * deformation * of @xmath77 ( or a conormal microbundle deformation ) if , for all symplectomorphism germs @xmath103:{{t}^*}n^{*}c\longrightarrow[{{t}^*}a , n^{*}c ] \label{eq : lag . embedding germ}\ ] ] fixing the core and such that the image by @xmath104 of the vertical distribution along @xmath20 in @xmath102 is of the form , there is a germ @xmath105 $ ] around @xmath20 of a closed one - form @xmath106 vanishing on @xmath20 and such that @xmath107 where @xmath108 is the symplectomorphism germ on @xmath109 $ ] obtained from @xmath110 by fiber translation as illustrated in figure [ deformation ] . is equivalent to @xmath111 . @xmath33 at 20 180 @xmath112 at 110 230 @xmath113 at 470 230 @xmath114 $ ] at 280 180 @xmath115 at 550 180 @xmath67 at 187 160 @xmath116 at 400 130 @xmath2 at 187 95 @xmath20 at 123 103 @xmath20 at 460 100 @xmath112 at 370 95 @xmath116 at 540 80 @xmath117 at 180 10 , title="fig : " ] [ thm : equivalence theorem ] let @xmath18 $ ] be a lagrangian submicrofold of @xmath37 $ ] . then the following statements are equivalent : \(1 ) @xmath18 $ ] is clean , \(2 ) @xmath18 $ ] is strongly transverse , \(3 ) @xmath18 $ ] is a deformation of @xmath77 . we first recall that , for three subspaces , @xmath118 and @xmath42 , of a given vector space , we have @xmath119 we start by showing that cleanliness is equivalent to strong transversality . @xmath120 : let @xmath81 be a lagrangian subbundle as in . using @xmath121 and identity , we see that the cleanliness assumption is equivalent to @xmath122 since @xmath123 is contained in @xmath124 . the transversality of @xmath85 and @xmath92 implies that the two terms in the r.h.s . of intersect only in @xmath125 . therefore we can conclude that @xmath126 by taking the symplectic orthogonal of , we obtain @xmath127 where the last equality comes from the identity @xmath128 whose first term is contained in @xmath129 while its second term is contained in @xmath124 . @xmath130 : the strong transversality assumption implies that @xmath67 is transverse to all lagrangian subbundles @xmath81 as in . choose one . the symplectic orthogonal of the transversality condition yields @xmath131 and , because @xmath132 is contained in @xmath129 , we have that @xmath124 intersects @xmath133 only in @xmath125 . now , using this together with , we obtain @xmath134 which proves that the intersection @xmath135 is clean . the equivalence between @xmath136 and @xmath137 is almost clear from the definitions . namely , the tangent map of a symplectomorphism germ @xmath30 as in definition [ def : conormal deformation ] maps the vertical bundle in @xmath102 along @xmath20 to a lagrangian subbundle @xmath81 of the form . clearly , a lagrangian submicrofold @xmath18 $ ] of @xmath33 is transverse to @xmath81 if and only if @xmath138 $ ] is transverse to the vertical distribution along @xmath20 . by continuity , this is equivalent to the existence of a ( small enough ) representative @xmath67 such that @xmath139 is transverse to the vertical distribution , or , in other words , such that @xmath139 is projectable onto the zero section in @xmath102 . with this in mind , the implication @xmath140 is clear . the converse follows from the lagrangian embedding theorem which guarantees the existence of a @xmath30 that sends the vertical distribution to any lagrangian subbundle along @xmath77 and transverse to it in @xmath33 . [ cor : equivalence]let @xmath18\subset{{t}^*}a$ ] be a lagrangian submicrofold . then the following statements are equivalent : \(1 ) @xmath18 $ ] is clean , ( 2 ) @xmath18 $ ] is transverse to a lagrangian subbundle @xmath81 as in , ( 3 ) @xmath18 $ ] is the image of @xmath77 by a symplectomorphism germ fixing the core as in . a closer look at the @xmath130 part in the proof of theorem [ thm : equivalence theorem ] shows that we actually use the weaker version @xmath141 of @xmath136 to prove @xmath142 . so we have a cyclic sequence of implications @xmath143 showing that @xmath18 $ ] is clean iff it is transverse to a lagrangian distribution @xmath90 . similarly , we have that @xmath144 . by the relative poincar lemma , the one - form germ @xmath105 $ ] in definition [ def : conormal deformation ] is exact , that is , there is a function germ @xmath145:n^{*}c\rightarrow{\mathbb{r}}$ ] such that @xmath146 . from now on , we will remove the ambiguity in the choice of @xmath145 $ ] by requiring that it vanishes on @xmath20 . we will call @xmath145 $ ] the * generating function * of the clean lagrangian submicrofold @xmath18 $ ] _ associated with the symplectomorphism germ _ @xmath30 . consider a smooth function @xmath147 . the image @xmath148 of its differential @xmath149 is a lagrangian submanifold @xmath33 . the * critical set * of @xmath31 is the set @xmath150 where @xmath151 is the zero section in @xmath33 . we would like to characterize the class of smooth functions on @xmath2 for which @xmath152 $ ] is a clean lagrangian submicrofold . for this to make sense , we need @xmath153 to be a submanifold to be submanifolds of possibly different dimensions . ] of @xmath2 , in which case we call it the * critical submanifold * of @xmath31 . the cleanliness of @xmath152 $ ] is related to the following notion : we say that the critical submanifold @xmath153 is * nondegenerate * if @xmath154 for all @xmath155 , where @xmath156 is the hessian of @xmath31 at @xmath157 ( which we see as a linear map from @xmath158 to @xmath159 ) . a function @xmath147 whose critical submanifold is nondegenerate is called a * morse - bott function*. ( if the components of @xmath160 are isolated points , @xmath31 is a morse function . ) let @xmath147 be a smooth function whose critical set @xmath153 is a submanifold of @xmath2 . then the lagrangian submicrofold @xmath152 $ ] is clean if and only if @xmath153 is nondegenerate . consider the tangent map @xmath161 and let @xmath162 . on the one hand , we have that @xmath163 and , on the other hand , we know that @xmath164 . so the intersection of @xmath165 with @xmath166 consists of the vectors @xmath167 such that @xmath168 , which , thanks to our assumption on @xmath169 , is exactly the subspace @xmath170 . consider a function @xmath171 . we see its differential as the map @xmath172 , where @xmath173 is the gradient of @xmath31 . the critical set @xmath153 are the points in @xmath174 where the gradient vanishes . the hessian at @xmath175 is the jacobian matrix @xmath176 regarded as a linear map from @xmath174 to @xmath174 . in case @xmath153 is a submanifold , it is nondegenerate when the jacobian matrix vanishes only on vectors tangent to @xmath153 . in particular , if @xmath177 is transverse to the zero section in @xmath178 , @xmath153 is a discrete collection of points , and the nondegeneracy condition for @xmath153 corresponds to the nondegeneracy of the jacobian matrix at the critical points . among the polynomial functions @xmath179 on the real line , the only one that yields a clean lagrangian submicrofold @xmath180 $ ] is the quadratic one . we say that the smooth function germ @xmath181:[a , c]\rightarrow[{\mathbb{r}},0]$ ] is a * morse - bott * * germ * with critical submanifold @xmath20 if there is a representative @xmath182 $ ] having @xmath20 has its nondegenerate critical submanifold . in other words , the lagrangian submicrofold @xmath183 $ ] is clean if and only if @xmath181:[a , c]\rightarrow[{\mathbb{r}},0]$ ] is a morse - bott germ . suppose that the lagrangian submicrofold @xmath18 $ ] of @xmath33 intersects the zero section transversally in @xmath20 , that is , @xmath184 for all @xmath87 . of course , this implies that @xmath18 $ ] is clean , but now we also have that @xmath20 must be discrete , since two transverse lagrangian submanifolds can intersect only in isolated points . for instance , if @xmath20 is reduced to a single point @xmath157 , @xmath185 $ ] is clean iff it is transverse to @xmath186 . therefore the clean lagrangian submicrofolds whose core is a single point correspond precisely to the symplectic micromorphisms from @xmath187 $ ] to the cotangent bundle of the one point manifold . theorem [ thm : equivalence theorem ] tells us that @xmath185 $ ] is a deformation of the conormal microbundle of @xmath157 , that is , @xmath188 $ ] . in other words , @xmath67 is the image of @xmath159 by a symplectomorphism germ around @xmath157 fixing this point and the zero section . in the cotangent bundle @xmath189 of the real line , consider the clean lagrangian submicrofolds @xmath190 $ ] with the origin as core . the transversality of @xmath190 $ ] tells us that the projection @xmath191 maps a representative @xmath67 diffeomorphically onto a neighborhood of @xmath192 in the fiber @xmath193 . now suppose further that our clean lagrangian submicrofold is the image of @xmath177 for a morse - bott germ @xmath181:[{\mathbb{r}},0]\rightarrow[{\mathbb{r}},0]$ ] . then the projection @xmath194 maps @xmath148 diffeomorphically onto a neighborhood of @xmath192 in @xmath195 . so the class of morse - bott germs @xmath181:[{\mathbb{r}},0]\rightarrow[{\mathbb{r}},0]$ ] with @xmath196 corresponds to the lagrangian submanifold germs through the origin that are projectable simultaneously on both the @xmath197-fiber and @xmath157-fiber . in this section , we show that the lagrangian submicrofold underlying a symplectic micromorphism is clean . as a consequence of the equivalence theorem for clean lagrangian submicrofolds , we find that a symplectic micromorphism is always a deformation of the cotangent lift of its core . this allows us to associate a global generating function ( depending on the choice of some symplectomorphism germ ) with any symplectic micromorphism . this determines its local form in terms of local generating functions in admissible local charts ( theorem [ thm : local form ] ) . finally , we prove a theorem ( theorem [ thm : normal form ] ) that gives a decomposition of the underlying lagrangian submicrofold of a symplectic micromorphism as a fibration over its core , the fibers of which are actual graphs of smooth map germs . following @xcite , we call * schwartz transform * the symplectomorphism @xmath198 the schwartz transform gives a one - to - one correspondence between the canonical relations from @xmath33 to @xmath16 and the lagrangian submanifolds of @xmath34 . this remains true in the microworld : [ pro : sympl . micromorph . as trans . lags.]the schwartz transform induces a one - to - one correspondence between the symplectic micromorphisms from @xmath33 to @xmath16 with core @xmath4 and the clean lagrangian submicrofolds in @xmath34 with core @xmath35 . thanks to corollary [ cor : equivalence ] , we only need to show that the distribution @xmath75 in is of the form @xmath90 as in definition [ def : w - distribution ] for some @xmath199 . to begin , we observe that the distribution @xmath200 along @xmath35 is transverse to @xmath201 in @xmath202 . its annihilator is @xmath203 , and we verify that @xmath90 coincides with @xmath75 when restricted to @xmath35 . the proposition above allows us to apply theorem [ thm : equivalence theorem ] to symplectic micromorphisms @xmath71,\phi):{{t}^*}a\rightarrow{{t}^*}b$ ] . the * cotangent lift * @xmath10 of the core map plays the role of the conormal microbundle since , by definition , @xmath204 the conormal bundle @xmath205 is the image of the lagrangian embedding @xmath206 of @xmath207 into @xmath34 given by @xmath208 thus the cotangent lift can be described by the representative @xmath209 this allows us to identify the cotangent lift @xmath12 and the conormal bundle @xmath210 with the pullback bundle @xmath207 . we can now apply the equivalence theorem to the case of symplectic micromorphisms . namely , theorem [ thm : equivalence theorem ] tells us that the data of a symplectomorphism germ @xmath211,\label{eq : k - phi}\ ] ] whose restriction to the zero section coincides with @xmath206 and which satisfies the condition in definition [ def : conormal deformation ] , allows us to describe a symplectic micromorphism @xmath71,\phi)$ ] from @xmath33 to @xmath16 in the two following ways : * @xmath212=r_{\phi}^{v}({{t}^*}\phi)$ ] for a symplectomorphism germ @xmath213\rightarrow\big[\overline{{{t}^*}a}\times{{t}^*}b,\,{\operatorname{gr}}\phi\big],\ ] ] which is uniquely determined by @xmath214 and @xmath71,\phi)$ ] . in other words , each symplectic micromorphism is a deformation in this sense of the cotangent lift of its core . * @xmath212=k_{\phi}^{-1}({{\operatorname{im}}}[df])$ ] for a smooth function germ @xmath215:[\phi^{*}({{t}^*}a),z_{b}]\rightarrow[{\mathbb{r}},0]\ ] ] whose critical submanifold is the zero section @xmath216 of @xmath207 . the germ @xmath181 $ ] is uniquely determined by @xmath214 and @xmath71,\phi)$ ] , and it is called the * global generating function * of @xmath71,\phi)$ ] associated with @xmath214 . as before , the two descriptions are related : namely , @xmath217 where @xmath218 is again the symplectomorphism germ obtained by fiber translation with the one form germ @xmath177 . [ exa : local gen fct 1 ] let @xmath8 be a smooth map from an open subset @xmath219 to an open subset @xmath220 . in this case , we have a canonical symplectomorphism germ @xmath221\\ \big((v_{1},p_{2}),(p_{1},x_{2})\big ) & \mapsto & \big(p_{1},\phi(x_{2})+v_{1},\,({{t}^*}_{x_{2}}\phi)p_{1}+p_{2},x_{2}\big)\end{aligned}\ ] ] since we can identify @xmath222 with @xmath223 . therefore , a symplectic micromorphism @xmath71,\phi)$ ] from @xmath224 to @xmath225 is completely determined by the germ of a function @xmath215:[{\mathbb{r}}_{p_{1}}^{k}\times u_{2},\{0\}\times u_{2}]\rightarrow[{\mathbb{r}},0]\quad\textrm{s.t.}\quad\partial_{p}f(0,x_{2})=0.\label{eq : local gen . fcts}\ ] ] in very explicit terms , a representative @xmath226of the symplectic micromorphism can be described as the set of points in @xmath227 of the form @xmath228 where @xmath229 runs in a suitable neighborhood of the zero section in @xmath230 . here , we see that the symplectomorphism germ @xmath231\rightarrow\big[\overline{{{t}^*}u_{1}}\times{{t}^*}u_{2},\,{\operatorname{gr}}\phi\big]\ ] ] given by the formula @xmath232 maps @xmath12 diffeomorphically onto @xmath71,\phi)$ ] as prescribed by the equivalence theorem . example [ exa : local gen fct 1 ] sets us on the way toward a notion of local generating function for symplectic micromorphisms . an * admissible local chart * for a symplectic micromorphism @xmath71,\phi):{{t}^*}a\rightarrow{{t}^*}b$ ] is a local chart of @xmath233 with domain of the form @xmath234 , where both factors of @xmath235 are the domains of coordinate patches @xmath236 and @xmath237 such that @xmath238 . we define the * restriction * of @xmath71,\phi)$ ] to @xmath234 to be the symplectic micromorphism @xmath239,\phi_{u})$ ] from @xmath224 to @xmath225 obtained as the image @xmath240,\phi_{u}):=({{t}^*}\chi_{1}\times{{t}^*}\chi_{2})\big(\big[v\cap{{t}^*}(a\times b)_{|\chi(u)},{\operatorname{gr}}\phi\cap\chi(u)\big]\big),\ ] ] where @xmath241 . since the restriction of a symplectic micromorphism to an admissible local chart is a symplectic micromorphism and admits a generating function as in example [ exa : local gen fct 1 ] , we immediately obtain the following `` local form '' theorem : [ thm : local form](local form ) . let @xmath71,\phi):{{t}^*}a\rightarrow{{t}^*}b$ ] be symplectic micromorphism , and let @xmath234 be an admissible local chart . then there is a representative @xmath242 $ ] of the restriction to the local chart such that @xmath243 where @xmath199 is a suitable neighborhood of the zero section in @xmath230 . the * local generating function * @xmath244 is of the form @xmath245 where @xmath31 is a representative of the function germ as in . let @xmath10 be the cotangent lift of a smooth map @xmath4 . in an admissible local chart @xmath234 , the cotangent lift admits the local generating function @xmath246 however , the global generating function @xmath181:\phi^{*}({{t}^*}a)\rightarrow[{\mathbb{r}},0]$ ] associated by the equivalence theorem with any symplectomorphism germ is always zero for cotangent lifts . let @xmath247 and @xmath248 be three sets and let @xmath249 and @xmath250 be two binary relations . we say that the composition @xmath251 is * monic * if , for all @xmath252 , the set @xmath253 is a singleton . in the microsymplectic world , the corresponding definition is the following : let @xmath0 $ ] , @xmath3 $ ] and @xmath254 $ ] be three symplectic microfolds . the composition of the symplectic micromorphisms @xmath76\overset{([v],\phi)}{\longrightarrow}[n , b]\overset{([w],\psi)}{\longrightarrow}[p , c]\ ] ] is * monic * if there are representatives @xmath255 $ ] and @xmath256 $ ] whose composition , as binary relations , is monic . our goal here is to show that the composition of symplectic micromorphisms is _ always _ monic . by the lagrangian embedding theorem , it is enough to see this for symplectic micromorphisms between cotangent microbundles : @xmath257,\phi)}{\longrightarrow}{{t}^*}b\overset{([w],\psi)}{\longrightarrow}{{t}^*}c.\ ] ] first of all , for all @xmath87 and for all representatives @xmath255 $ ] and @xmath256 $ ] , we have that @xmath258 and @xmath259 . we need to check that the composition remains monic in a neighborhood of @xmath260 . to do this , we can go to local coordinates and consider admissible local charts @xmath261 such that @xmath262 , @xmath263 and @xmath264 . by theorem [ thm : local form ] , we can express the restrictions of these symplectic micromorphisms in terms of local generating functions : @xmath265 where @xmath266 is a neighborhood of @xmath267 in @xmath268 and @xmath269 is a neighborhood of @xmath270 in @xmath271 . a point @xmath272 is of the form @xmath273 for @xmath274 such that @xmath275 the following lemma shows that @xmath276 is reduced to a single point . for all @xmath277 with @xmath278 small enough , the following system @xmath279 has a unique solution @xmath280 . this follows from a straightforward application of the implicit function theorem to the function @xmath281 around the point @xmath282 since @xmath283 and @xmath284 . putting everything together , we obtain the following proposition : [ pro : simplicity ] the composition of symplectic micromorphisms is always monic . as a byproduct , we also get a composition formula for local generating functions . we start by noticing that the unique point @xmath274 in @xmath276 is also the unique critical point of the function @xmath285 where @xmath278 and @xmath286 are held fixed . if we denote by @xmath287 the function @xmath288@xmath289 evaluated at its critical point , we obtain the following result : in the notation as above , the function @xmath290 is the local generating function of the composition @xmath291 . we see this by noticing that @xmath292 where @xmath274 is the unique critical point of @xmath293 . a cotangent lift @xmath10 has the nice property that , even though it is not itself the graph of a map when @xmath8 is not a diffeomorphism , the intersection of its underlying lagrangian submicrofold with the coisotropic submanifold @xmath294 is the graph of a map : @xmath295 to make sense of this map - like property for general symplectic micromorphisms , we need the following definitions : a * lagrangian fibration * @xmath296 of a symplectic microfold @xmath0 $ ] is a collection , smoothly parametrized by @xmath2 , of transverse lagrangian submicrofolds ( in the sense of paragraph [ sec : trans . lag . subs ] ) @xmath297\subset[m , a],\quad x\in a.\ ] ] a lagrangian fibration of @xmath0 $ ] * along a smooth map * @xmath4 is a collection @xmath296 smoothly indexed by @xmath61 of transverse lagrangian submicrofolds @xmath298\subset[m , a],\quad y\in b.\ ] ] the _ vertical _ lagrangian fibration in a cotangent microbundle is given by the germ of its fibers at @xmath192 . consider the symplectic micromorphism @xmath299,\psi_{|a}^{-1})$ ] from @xmath0 $ ] to @xmath3 $ ] coming from a symplectomorphism germ @xmath300\rightarrow[n , b]$ ] between two symplectic microfolds . for any lagrangian fibration @xmath301 of @xmath3 $ ] , we obtain a corresponding lagrangian fibration @xmath296 of @xmath0 $ ] along @xmath302 by setting @xmath303 if we denote by @xmath304 the restriction of @xmath305 to @xmath306 , we obtain the identities @xmath307 which are similar to the ones we had in the case of cotangent lifts , except that here the lagrangian fibration @xmath301 , ( which now plays the role of the vertical distribution for cotangent microbundles ) is not canonical . [ thm : normal form ] let @xmath71,\phi):[m , a]\rightarrow[n , b]$ ] be a symplectic micromorphism between two symplectic microfolds . then , for any lagrangian fibration @xmath301 of @xmath3 $ ] , there exists a unique corresponding lagrangian fibration @xmath296 along @xmath8 such that @xmath308 for suitable representatives , and where @xmath309:[\mathcal{f}_{y}a,\{\phi(y)\}]\rightarrow[\mathcal{f}_{y}b,\{y\}]$ ] is a collection of smooth map germs indexed by @xmath61 . [ rem : composition ] the lagrangian fibration @xmath301 of @xmath3 $ ] in the theorem above gives rise to the collection of symplectic micromorphisms @xmath310 , c_y\big ) : [ n , b]\longrightarrow e,\quad y\in b,\ ] ] where @xmath311 is the cotangent microbundle of the one - point manifold @xmath312 . the core maps are the constant functions @xmath313 that map the unique point of the core of @xmath22 to each @xmath314 . the corresponding lagrangian fibration @xmath296 along @xmath8 is then obtained from the symplectic micromorphism @xmath71,\phi)$ ] by composition @xmath315 , c_{\phi(y)}\big ) = \big(\big[\mathcal{f}_yb\big ] , c_y\big)\circ ( [ v],\phi ) .\label{eq : lag fib as mic}\ ] ] the uniqueness of the decomposition is immediate from remark [ rem : composition ] . as for the existence , let @xmath301 be a lagrangian fibration of @xmath3 $ ] and consider the lagrangian fibration @xmath296 along @xmath8 as defined by . we denote by @xmath316 the intersection of @xmath226 with @xmath317 . by remark [ rem : composition ] , we have that @xmath318 and that , for appropriate representatives , @xmath319 we need to show that @xmath316 is the graph of a map . for this , we consider the composition of the symplectic micromorphisms @xmath76\overset{([v],\phi)}{\longrightarrow}[n , b]\overset{\mathcal{f}_{y}b}{\longrightarrow}e.\ ] ] a point @xmath320 is of the form @xmath321 , where @xmath322 . since , by proposition [ pro : simplicity ] , the composition of two symplectic micromorphisms is alway monic the set @xmath276 is a singleton whose unique point is in @xmath323 . we denote this point by @xmath324 , and this gives us a map @xmath325 whose graph is , by definition , @xmath316 . [ exa : local normal form]let @xmath71,\phi):{{t}^*}u_{1}\rightarrow{{t}^*}u_{2}$ ] be a symplectic micromorphism between cotangent microbundles over the open subsets @xmath220 and @xmath219 . the local form theorem tells us that there is a representative @xmath255 $ ] that can be described by a generating function @xmath181 $ ] as in . thus @xmath326 is the locus of points of the form @xmath327 where @xmath328 is fixed while @xmath278 is free to vary in a neighborhood of @xmath192 in @xmath329 . now , for each @xmath330 , we may define the symplectomorphism germ @xmath331 & \overset{r_{x_{2}}}{\longrightarrow } & [ { { t}^*}u_{1},\phi(x_{2})],\\ ( p_{1},x_{1 } ) & \mapsto & x_{1}+\partial_{p}f(p_{1},x_{2}).\end{aligned}\ ] ] the image of the cotangent bundle fiber over @xmath332 by @xmath333 defines a fiber of our lagrangian fibration of @xmath224 along the core map . explicitly , it is @xmath334 where @xmath199 is a suitable neighborhood of @xmath192 in @xmath335 . @xmath336 is a lagrangian submanifold of @xmath224 that intersects the zero section transversally in @xmath332 . now the map @xmath337 from @xmath338 to @xmath339 is given by the formula @xmath340 consider the symplectic micromorphism @xmath71,\phi)$ ] from @xmath341 to itself whose core is the constant function @xmath342 and whose generating function is given by @xmath343 . the map @xmath344 parametrizes a representative @xmath255 $ ] . consider the straight lines @xmath345 in @xmath346 through the origin . the intersection of @xmath226 with @xmath347 is the graph of the map @xmath348 that folds @xmath349 at the origin into a half - line and maps this ray linearly into the half - line parallel to the @xmath197-axis and passing through @xmath350 . let @xmath351 be a symplectomorphism germ that fixes the core ; i.e. , the core of the corresponding symplectic micromorphism is the identity map on @xmath2 . for each @xmath352 , @xmath353 defines two lagrangian distributions , @xmath354 the forward and backward images of the cotangent fibers via @xmath353 . clearly , the restriction of @xmath353 to the backward distribution yields the decomposition @xmath355 associated with the cotangent fiber distribution . now , consider a symplectic micromorphism of the form @xmath356,\phi)={\operatorname{gr}}[r_{a}^{-1}]\circ{{t}^*}\phi\circ{\operatorname{gr}}[r_{b}].\ ] ] where @xmath353 and @xmath357 are symplectomorphism germs on respectively @xmath33 and @xmath357 of @xmath16 fixing the cores , and where @xmath10 is a cotangent lift . then we obtain a decomposition given by the following diagram : t^*_(x_2)a & ^t^*_x_2 & t^*_x_2 b + ^r_a^-1 & & _ r_b + _ ( x_2)a & _ _ x_2 & _ in this section , we explain how symplectic microgeometry is a natural framework for the hamilton - jacobi theory of hamiltonian flows through their local generating functions . we show that there is a canonical symplectic micromorphism , @xmath358 the evolution micromorphism , that encodes the short - time dynamics of an hamiltonian system @xmath359 . the local generating function of @xmath360 in an admissible local chart coincides with the solution of the hamilton - jacobi equation for the generating function of the hamiltonian flow in this chart . from a different perspective , the evolution micromorphism allows us to formulate the short - time evolution in classical mechanics in a purely categorical way . namely , we show that @xmath360 turns @xmath361 into a module over @xmath362 and that all @xmath362-modules arise from hamiltonian flows ( with possibly _ time_-_dependent _ hamiltonians ) . consider a hamiltonian system @xmath359 . the time evolution @xmath363 generated by @xmath62 is the flow of the hamiltonian vector field @xmath364 . it produces a lagrangian submanifold @xmath365 of @xmath366 , which we call the * evolution submanifold * , and which is defined as @xmath367 where @xmath368 is the maximal interval on which @xmath369 is defined . the core @xmath195 of the @xmath362-factor in the product @xmath366 above corresponds to the possible energy levels @xmath370 of the system and not to the possible times . to remind us of this fact , we will denote this core by @xmath371 instead of @xmath195 . a point @xmath372 in the fiber represents the time , in accordance with the physical time - energy duality . we see that the points in @xmath365 with @xmath378 and @xmath379 lie in the graph of @xmath377 , and therefore it makes sense to consider the lagrangian submicrofold @xmath380 $ ] . a straightforward check shows that , if restricted to sufficiently small times and momenta , the evolution submanifold intersects the zero section cleanly in @xmath381 . this yields the following result : let @xmath359 be an hamiltonian system . then the germ of the evolution submanifold around the graph of yields a symplectic micromorphism @xmath382,j\big):{{t}^*}{\mathcal{e}}\otimes{{t}^*}q & \longrightarrow & { { t}^*}q,\label{eq : evolution micromorphism}\end{aligned}\ ] ] which we will refer to as the * evolution micromorphism * of the hamiltonian system . in the discussion above , we started with a global hamiltonian @xmath383 and we obtained from it its evolution micromorphism @xmath384,j)$ ] , which encodes the short - time dynamics of the hamiltonian system . by doing so , we lost some information since two different global hamiltonians will yield the same evolution submicrofold if their germs around the zero section in @xmath361 coincide . of course , it would have sufficed to start with a hamiltonian function germ @xmath385:[{{t}^*}q , z_q]\rightarrow { \mathbb{r}}$ ] and then consider the induced germ of flow to define the evolution submicrofold . we can always choose the admissible charts for the evolution micromorphism to be of the form @xmath386 where @xmath387 is a coordinate patch of @xmath376 . the local form theorem tells us that there exist a unique local generating function @xmath388 of the form @xmath389 which is defined for sufficiently small times and momenta , and such that @xmath390 where the variables @xmath391 in range in a suitable neighborhood of the zero section in @xmath392 , is a representative of the restriction of evolution micromorphism to the local chart . comparing with , we see that @xmath393 satisfies the following partial differential equation @xmath394 with initial condition @xmath395 . equation is known as the * hamilton - jacobi equation * ( for generating functions of type @xmath396 ) of the hamiltonian system @xmath397 . here , the existence of a solution is an immediate consequence of the local form theorem applied to the evolution micromorphism . now suppose that we are given a generating function @xmath388 that is nondegenerate in the following sense : @xmath398 in this case , the implicit function theorem applied to the function @xmath399 around the point @xmath400 guarantees that the following implicit system @xmath401 has a unique solution @xmath402 for each @xmath403 with @xmath404 and @xmath197 small enough . this generates a flow @xmath405 on a neighborhood of the zero section of @xmath234 . the hamilton - jacobi theorem tells us that , if @xmath393 further satisfies , then this flow coincides with the flow generated on the local chart @xmath234 by the hamiltonian flow of @xmath62 . for the local generating function of the evolution micromorphism , the nondegeneracy is clear from , and the fact that the implicit equations and generate the flow is immediate from a comparison of with . to conclude this paragraph , we will exhibit some nice relations between the hamiltonian and the local generating functions of its evolution micromorphism . from the hamilton - jacobi equation together with , we obtain @xmath406 moreover , a straightforward taylor expansion in the time variable yields @xmath407 our next task is to characterize the symplectic micromorphisms from @xmath408 to @xmath361 that come from the evolution submanifolds of ( possibly time - dependent ) hamiltonian systems . we proceed by imposing obvious conditions in terms of their local generating functions . let @xmath409,j):{{t}^*}{\mathcal{e}}\otimes{{t}^*}q\rightarrow{{t}^*}q$ ] be a symplectic micromorphism with core map @xmath410 . the general form for its local generating function in an admissible local chart as in is @xmath411 where @xmath31 is some function that vanishes , as well as its derivatives in the @xmath404 and @xmath197 directions , when @xmath412 . this implies in general that @xmath413 because of the vanishing of @xmath31 . now the nondegeneracy condition is satisfied iff @xmath414 is a local diffeomorphism . if this is the case , equations - define , as before , a flow @xmath415 on @xmath234 such that @xmath416 if we also want @xmath417 to be the identity map when @xmath378 , we need to further impose that @xmath418 , or equivalently , that @xmath419 differentiating this last equation with respect to @xmath197 and setting @xmath420 , we are left we no choice but to require that @xmath421 . now we can define the _ time - dependent _ hamiltonian @xmath422 where @xmath423 . this way , the generating function satisfies by definition the time - dependent hamilton - jacobi equation @xmath424 with the same initial condition for @xmath393 as in the time - independent case . as a consequence , @xmath425 is now of the form with the hamiltonian @xmath62 replaced with the time - dependent one @xmath426 . the hamilton - jacobi theorem now tells us now that @xmath417 is the flow of @xmath426 in @xmath234 . because our local charts are of the form @xmath427 where @xmath428 is a local coordinate patch @xmath429 , the hamiltonian flows that are induced on the coordinate patches @xmath430 coincide on their overlaps . thus this defines a time - dependent hamiltonian system @xmath431 ( where @xmath426 is only defined for small times and momenta ) whose evolution microfold is precisely @xmath409,j)$ ] . let us sum up what we have proven so far : [ pro : time - independent hamiltonian]a symplectic micromorphism @xmath432 is the evolution micromorphism of a ( possibly time - dependent ) hamiltonian flow if and only if @xmath433 for some smooth function @xmath434 . we want to categorize the evolution micromorphisms at a purely categorical level . to give an indication as where we are aiming at , observe that the unitality condition @xmath435 for the symplectic micromorphism @xmath436 ( @xmath437 is the unique symplectic micromorphism from the cotangent microbundle of the point to @xmath36 ) implies ] where @xmath438 is the core of @xmath439 and where the projection @xmath440 is the core of @xmath441 . if we denote the two components of the core map of @xmath442 by @xmath443 , we see that this last equation is satisfied if and only if @xmath444 . ] that @xmath445 for some function @xmath434 . therefore , thanks to proposition [ pro : time - independent hamiltonian ] , the unitality condition , which is purely categorical , already singles out the class of @xmath442 coming from evolution micromorphisms of possibly _ time - dependent _ hamiltonian systems . the time - independent case is more subtle , and requires us to look at @xmath36 in a different way . first of all , consider the lie algebra @xmath446of the time translation group @xmath447 , which is the abelian lie algebra on @xmath195 . we can identify @xmath448 with the dual of @xmath446 . as the dual of a ( trivial ) lie algebra , @xmath448 can be seen as a poisson manifold endowed with the zero poisson structure . we call @xmath446 the _ lie algebra of time _ and @xmath448 the _ poisson manifold of energy_. the cotangent microbundle @xmath449 is a symplectic groupoid ( see @xcite ) with source and target maps coinciding with the bundle projection ; the space of composable pairs is @xmath450 , and the groupoid product is the addition of times in a fiber of constant energy : @xmath451 one verifies that the graph of the groupoid product is a symplectic micromorphism @xmath452,\delta_{\mathcal{e}}):{{t}^*}{\mathcal{e}}\otimes{{t}^*}{\mathcal{e } } & \longrightarrow & { { t}^*}{\mathcal{e}},\end{aligned}\ ] ] where the core map @xmath453 is the diagonal map on @xmath448 . it is easy to see that @xmath454 further satisfies the following associativity and unitality equations @xmath455 where @xmath288@xmath441 is the unique symplectic micromorphism from the cotangent bundle of the point to @xmath36 . in other words , @xmath456 is a monoid object in the microsymplectic category . [ pro : classical mechanics ] a symplectic micromorphism @xmath436 is the evolution micromorphism of a time - independent hamiltonian system if and only if @xmath457 in other words , if and only if @xmath458 is a @xmath36-module in the microsymplectic category . let @xmath458 be a @xmath36-module . we have already seen that the unitality condition tells us that @xmath442 is the evolution micromorphism of a time - dependent hamiltonian system @xmath431 . therefore there is a representative of the form with @xmath62 replaced with @xmath426 . a direct computation with binary relations gives us a representative of the l.h.s of , @xmath459 as well as a representative of its r.h.s . , @xmath460 requiring the equality of both sides is equivalent to imposing that @xmath461 and @xmath462 for all @xmath463 and @xmath404 . in other words , the associativity equation holds iff @xmath442 is the evolution micromorphism of a time - independent hamiltonian system . it is straightforward to generalize the proposition above to general @xmath36-modules @xmath464\rightarrow[p , q]\ ] ] in the microsymplectic category by using a symplectomorphism germ @xmath465:{{t}^*}q\rightarrow[p , q]$ ] coming from the lagrangian embedding theorem . it is possible to generalize the previous scheme to a general hamiltonian action of a lie group @xmath42 on @xmath361 with momentum map @xmath466 . in this case , we define the symmetry submanifold to be @xmath467 where @xmath468 is the maximal neighborhood of @xmath192 in the lie algebra @xmath469 on which the exponential mapping @xmath470 is a diffeomorphism on its image . taking the germ of @xmath471 around the graph of @xmath472 , yields a symplectic micromorphism @xmath473 now , thanks to the exponential mapping , we can define a generating function germ from @xmath474 to @xmath195 via the formula @xmath475 where @xmath476 is the canonical paring between the lie algebra and its dual . this generating function germ defines a symplectic micromorphism @xmath477 from @xmath478 to @xmath479 . one can show that @xmath480 is a monoid and that @xmath481 a @xmath479-module . this situation will be treated in full details elsewhere . as a final comment , let us mention that our approach to hamiltonian flows and symmetries through symplectic micromorphisms is close in spirit to the work of almeida and rios @xcite , benenti @xcite and zakrzewski @xcite on generalized versions of the hamilton - jacobi theory for classical symmetries . m. ruzhansky , singularities of affine fibrations in the theory of regularity of fourier integral operators ( russian ) , _ uspekhi mat . * 55 * ( 2000 ) , 99170 ; translation in russian _ math . surveys _ * 55 * ( 2000 ) , 93161 .

we consider the multiple - access communication problem in a distributed setting for both the additive white gaussian noise channel and the discrete memoryless channel . we propose a scheme called distributed rate splitting to achieve the optimal rates allowed by information theory in a distributed manner . in this scheme , each real user creates a number of virtual users via a power / rate splitting mechanism in the-user gaussian channel or via a random switching mechanism in the-user discrete memoryless channel . at the receiver , all virtual users are successively decoded . compared with other multiple - access techniques , distributed rate splitting can be implemented with lower complexity and less coordination . furthermore , in a symmetric setting , we show that the rate tuple achieved by this scheme converges to the maximum equal rate point allowed by the information - theoretic bound as the number of virtual users per real user tends to infinity . when the capacity regions are asymmetric , we show that a point on the dominant face can be achieved asymptotically . finally , when there is an unequal number of virtual users per real user , we show that differential user rate requirements can be accommodated in a distributed fashion . multiple access , rate splitting , successive decoding , stripping , interference cancellation , aloha .

cs0605041

we consider the basic multiple - access communication problem in a distributed setting . in gallager s survey paper @xcite , it is pointed out that the multiple - access problem has been studied from a number of different perspectives , each having its own advantages and shortcomings . in the data networking community , a well - known distributed multiple access scheme is aloha @xcite . in aloha , it is assumed that a collision " happens whenever more than one user transmit simultaneously . those packets involved in a collision are discarded and retransmitted according to some retransmission probability . the collision channel model , however , does not accurately describe the underlying physical multiple - access channel . it is well known that there exist coding techniques which can decode multiple users messages when simultaneous transmissions occur . indeed , more sophisticated models such as signal capture @xcite , spread aloha @xcite , and multi - packet reception @xcite have been developed to enhance aloha . even these improved schemes , however , are not optimal from the viewpoint of information theory . in the information theory literature , the capacity regions of various multiple - access channel ( mac ) models have been characterized ( see @xcite ) . rate splitting multiple - access techniques ( or generalized time - sharing ) are presented in @xcite to achieve every point in the gaussian or the discrete memoryless mac capacity region using only single - user codes . these schemes , however , require a pre - defined decoding order , which makes distributed implementation difficult . finally , in the spread spectrum community , cdma techniques are adopted . here , users are decoded regarding all other users signals as interference . this , however , is not optimal from the information theoretic viewpoint . to address some of the shortcomings mentioned above , medard _ et al . _ @xcite use information - theoretic techniques to analyze different notions of capacity for time - slotted aloha systems . a coding / decoding scheme which combines rate splitting and superposition coding is constructed . this scheme allows some bits to be reliably received even when collision occurs , and more bits to be reliably received in the absence of collisions . shamai @xcite proposes a similar scheme to apply a broadcast strategy to multiple - access channel under static fading where the fading coefficients are not available to the transmitters or the receiver . to implement the scheme in @xcite , however , a pre - defined decoding order is required , as in @xcite . in @xcite , cheng proposes a distributed scheme called `` stripping cdma '' for the @xmath1 out of @xmath2 gaussian mac . here , no pre - defined decoding order is required . it is shown in @xcite that stripping cdma is asymptotically optimal , although the optimal operating parameters are not specified . in this paper , we investigate distributed multiple - access schemes based on the idea of rate splitting for both the @xmath0-user additive white gaussian noise mac and the @xmath0-user discrete memoryless mac . we characterize the optimal operating parameters as well as the asymptotic optimality of these schemes from the viewpoint of information theory . assume that every user has an infinite backlog of bits to send , and that every user knows the total number of users @xmath0 . we propose a distributed scheme , called _ distributed rate splitting _ ( drs ) , to achieve the optimal communication rates allowed by information theory . in this scheme , each real user creates a number of virtual users via a power / rate splitting mechanism in the @xmath0-user gaussian channel or via a random switching mechanism in the @xmath0-user discrete memoryless channel . at the receiver , all virtual users are successively decoded . a possible advantage of the drs scheme is that it can be implemented with lower complexity when compared with multiple - access schemes such as joint coding and less coordination when compared with time - sharing and rate splitting . in sections [ sec : gau ] and [ sec : dmc ] , we focus first on symmetric situations where the channel capacity regions are symmetric and every real user creates the same number of virtual users . in this case , the drs scheme entails the following . each user @xmath3 creates @xmath1 virtual users indexed by @xmath4 , @xmath5 . the virtual user class @xmath6 consists of users @xmath7 ( i.e. we have altogether @xmath1 virtual user classes and there are @xmath0 virtual users in each class ) . in the @xmath0-user gaussian mac , virtual users are created via a power / rate splitting mechanism . the signal transmitted by a real user is the superposition of all its virtual users signals . the receiver receives the sum of the virtual users signals plus noise . all virtual users are then successively decoded in increasing order of their class . that is , all virtual users in class @xmath6 , @xmath8 , are decoded before any virtual user in @xmath9 , where @xmath10 , is decoded . in contrast to @xcite , the optimal operating parameters , such as power and rate , are explicitly specified for any finite @xmath1 . in the @xmath0-user discrete memoryless mac , virtual users with the same input distribution as the real users are created , and the transmitted signal of a real user is determined by a random switch . the receiver successively decodes all virtual users in increasing order of their class given the side information of already decoded virtual users . the optimal switch is found for any finite @xmath1 for the @xmath11-user case . finally , it is shown that for both channel models , the rate tuple achieved by the drs scheme converges to the maximum equal rate point allowed by the information - theoretic bound as the number of virtual users per real user tends to infinity . next , in section [ sec : variation ] , we consider more general situations where the capacity regions can be asymmetric and real users may generate different numbers of virtual users . for the case of asymmetric capacity regions , new operating parameters are specified for any finite number of virtual users per real user . we show that the drs scheme still can achieve a point on the dominant face as the number of virtual users per real user tends to infinity . for the case of unequal number of virtual users per real user , we present a variation of drs which supports differential user rate requirements in a distributed manner . in this new scheme , each user @xmath3 , independently from other users , generates @xmath12 virtual users according to its own rate requirement . all virtual users are then decoded reliably at the receiver . furthermore , as each real user generates more virtual users , the rate tuple achieved under this variation of drs converges the maximum equal rate point on the dominant face . we first examine a gaussian mac with a symmetric capacity region . later in section [ sec : variation ] , we consider the asymmetric case . consider an @xmath0-user gaussian mac where each transmitter has transmission power @xmath13 and the receiver has noise variance @xmath14 . the capacity region @xmath15 is the set of @xmath16 satisfying to denote @xmath17 throughout . ] @xmath18 where @xmath19 is the cardinality of the set @xmath20 . the _ dominant face _ @xmath21 is the subset of rate tuples which gives equality in for @xmath22 . for this symmetric setting , it is easy to see that the maximum common rate that every user can achieve is achievable if for any @xmath23 , there exists an @xmath24 multiple - access code with overall error probability @xmath25 , where @xmath26 is the block length . ] is @xmath27 . it is well - known that rate tuples on the dominant face other than the vertices can not be achieved via standard successive decoding @xcite . note that the optimal rate tuple @xmath28 , called the _ maximum equal rate point _ , is such a point . for the two - user gaussian mac , the maximum equal rate point is shown in fig . [ fig : twogaussian ] . currently , three methods are known to achieve general points on the dominant face : joint encoding / decoding , time - sharing , and rate - splitting . joint encoding / decoding is not practical because of its high complexity @xcite . in time - sharing , all @xmath0 users need to coordinate their transmissions . therefore , some communication overhead is required . the rate - splitting method in @xcite achieves every point in @xmath29 via a generalized successive decoding scheme . for the two - user case , user @xmath30 creates two virtual users , say @xmath31 and @xmath32 , by splitting its power @xmath13 into @xmath33 and @xmath34 and setting @xmath35 , @xmath36 . user @xmath11 does not split its power and sets its rate to @xmath37 . the decoding order is @xmath38 . in order to achieve the maximum equal rate point , we solve @xmath39 , yielding @xmath40 . thus , both time - sharing and rate splitting require some coordination among users . in this paper , we focus on _ distributed _ multiple - access communication schemes . in particular , we introduce the _ distributed rate splitting _ ( drs ) scheme . the drs scheme offers the possibility of multiple - access communication with lower complexity when compared with joint coding , and communication with less coordination when compared with the time - sharing or rate splitting method . moreover , we show that drs can achieve the maximum equal rate point of the mac capacity region asymptotically . we now formally present the drs scheme . in this scheme , each user creates @xmath1 virtual users by splitting its power @xmath13 into @xmath41 , where @xmath42 is the power allocated to the @xmath43th virtual user and @xmath44 . each user then assigns transmission rate @xmath45 to virtual user @xmath43 . note that the proposed drs scheme is symmetric , i.e. all @xmath0 users split their powers and set their rates in the same way . the signal transmitted by a user is the superposition of its virtual users signals . as defined in section [ sec : intro ] , virtual user class @xmath6 consists of all virtual users indexed by @xmath43 . the receiver receives the sum of all virtual users signals plus noise . all virtual users are then successively decoded in increasing order of their class . to illustrate the drs scheme , consider the case @xmath46 . each real user splits its power @xmath13 into @xmath33 and @xmath34 . notice there are two major differences between our scheme and the traditional rate splitting scheme in @xcite . first , in our scheme , all real users split in the same way , whereas there is at least one user who does not split in the traditional rate splitting scheme . second , virtual users in the same class , ( i.e. with the same index @xmath43 ) , are allocated the same rate in our scheme , whereas all virtual users have different rates according to the pre - defined decoding order in the traditional rate splitting scheme . is not optimal . later in this section , we demonstrate the asymptotic optimality of drs by taking @xmath1 to infinity . ] these differences are illustrated in fig . [ fig : difference ] . since we assume the receiver uses successive decoding method , some virtual user must be decoded first . without loss of generality we assume one of the @xmath33 virtual users is decoded first . for the case @xmath46 , we show that there is a unique way for a real user to split its power in order to maximize its total throughput . for @xmath47 and for a fixed @xmath33 , each real user s throughput is maximized by setting @xmath48 and @xmath49 . _ proof : _ the @xmath33 virtual user who is decoded first must have @xmath50 ( i.e. the virtual user regards all other virtual users as interference ) in order to be decoded successfully . due to symmetry , all other @xmath33 virtual users must have the same @xmath51 . then the problem of maximizing each real user s throughput reduces to @xmath52 , subject to ( i ) @xmath53 , ( ii ) one of the @xmath33 virtual users is decoded first and ( iii ) @xmath54 note that @xmath55 is maximized when the interference plus noise faced by all the @xmath56 virtual users is minimized , and the only way to minimize the interference plus noise faced by all the @xmath56 virtual users is to decode all the @xmath33 virtual users before decoding any @xmath56 virtual user . therefore , the minimum interference plus noise faced by any @xmath56 virtual user is @xmath57 . hence , the maximum rate associated with a @xmath56 virtual user is @xmath58 . @xmath59 + using the drs scheme with @xmath47 , each user can strictly increase its throughput relative to the case where users do not split their powers and decode against each other as noise . this is easily verified by observing that for any @xmath60 , @xmath61 now consider the case where each user creates more than two virtual users @xmath62 . here , we show that each user s throughput increases further . [ le : morevuser ] given a drs scheme with @xmath1 virtual users per real user , where @xmath63 are the virtual users powers , it is possible to strictly increase the throughput via an @xmath64 virtual user system with powers @xmath65 , where @xmath66 . _ proof : _ suppose that every user splits its power into @xmath1 virtual users : @xmath67 subject to @xmath44 , where @xmath1 is an arbitrary integer and @xmath42 is the power of @xmath43th virtual user . since virtual user @xmath1 is decoded last , following the reasoning in the proof of lemma @xmath30 , we have @xmath68 . we now split the virtual user with power @xmath69 into two new virtual users with powers @xmath70 and @xmath71 , where @xmath72 . we set @xmath73 now each real user has @xmath74 virtual users . notice that we do not change the power and decoding order of any of the other virtual users ( i.e. virtual users @xmath75 ) . from a real user s view point , the virtual user with @xmath76 is decoded second to last _ among all virtual users generated by this real user _ and the virtual user with @xmath77 is decoded last . thus , all virtual users can be decoded and from , @xmath78 . therefore , every real user with @xmath1 virtual users can strictly increase its throughput by splitting its power among @xmath74 virtual users . @xmath59 + before we examine the asymptotic behavior of drs , we solve the problem of how to split a user s power _ optimally _ among a fixed number of virtual users . the main difficulty here is that the objective function is not concave . in order to find the optimal splitting method , we prove the following lemma . [ lemma : maximization ] consider the following optimization problem : @xmath79 subject to @xmath42+@xmath80 = @xmath81 and @xmath82 , where @xmath83 , @xmath0 and @xmath81 are positive constants and @xmath84 . the unique solution to is also the unique solution to @xmath85 , where @xmath86 . _ proof : _ substitute @xmath87 into the objective function , we have @xmath88setting @xmath89 subject to @xmath90 , the unique solution is @xmath91 . thus , @xmath92 is the unique stationary point of @xmath93 . we can also verify that @xmath94 and @xmath95 . so @xmath96 is the unique solution to our maximization problem . we can directly solve @xmath97 subject to @xmath98 and @xmath82 . the unique solution is also @xmath96 . @xmath59 + we now present the optimal splitting method . theorem [ thm : nece ] states a necessary condition for the optimal splitting method , and theorem [ thm : unique ] implies there is a unique optimal splitting method . in corollary [ cor : optimal ] , we formally present the optimal splitting method and the required power levels . [ thm : nece ] let each real user split its power into @xmath1 virtual users . let @xmath42 be the power allocated to the @xmath43th virtual user and @xmath99 @xmath100 . if @xmath101 maximizes @xmath102 and satisfies @xmath103 , @xmath104 for @xmath105 , then @xmath106 , for all @xmath43 . that is , the optimal power split must lead to equal transmission rates for all virtual users . _ proof : _ we use a perturbation argument . suppose @xmath107 maximizes @xmath108 and satisfies @xmath109 , @xmath110 and the resulting @xmath111 is not the same for all @xmath43 . then we can find a pair of virtual users @xmath112 , where virtual user @xmath43 and @xmath113 are decoded at the @xmath43th and @xmath114th places respectively , and @xmath115 . without loss of generality , let us consider the case where @xmath116 @xmath117 . by the definition of @xmath45 , we have @xmath118 we can verify that if we change @xmath119 to @xmath120 and @xmath121 to @xmath122 , where @xmath123 is a small positive number , then the first term of decreases and the second term of increases . let @xmath124 be the solution to @xmath125 ( the existence of @xmath126 can be demonstrated ) . let @xmath127 and @xmath128 . notice that @xmath129 . since the maximization considered in lemma [ lemma : maximization ] has a unique solution , @xmath130 . this contradicts our assumption that @xmath107 maximizes + @xmath108 . therefore , the theorem follows . @xmath59 + by theorem [ thm : nece ] , if @xmath131 maximizes @xmath132 and satisfies @xmath133 , @xmath134 for @xmath105 , then we must have @xmath135 where @xmath136 . therefore , if we show that @xmath137 for all @xmath138 , has a unique solution , then there is _ at most _ one feasible solution to the maximization problem . [ thm : unique ] the set of equations : @xmath139 , subject to @xmath140 and @xmath141 @xmath142 , has a unique solution . _ proof : _ see appendix i. [ cor : optimal ] if a real user splits its power @xmath13 into @xmath1 virtual users , then the unique way to maximize this user s throughput is to set @xmath143 $ ] for @xmath144 . _ proof : _ since the constraint region @xmath145 is a simplex and @xmath108 is continuous , there exists at least one solution . we denote one solution by @xmath146 . by the necessary condition stated in theorem [ thm : nece ] , @xmath146 must satisfy @xmath147 . moreover , by the uniqueness property stated in theorem [ thm : unique ] , @xmath146 is the unique solution to @xmath148 subject to @xmath149 next , we plug @xmath42 into the expression of @xmath150 . let @xmath151 , we are able to verify that @xmath152 , which is independent of @xmath43 . hence , the corollary follows . @xmath59 + we now examine the asymptotic behavior of the drs scheme . we first demonstrate the interesting fact that the rate tuple converges to the maximum equal rate point for a general power split as long as all virtual users powers go to @xmath153 as @xmath154 . this implies a convergence result for the optimal power split . we then analyze the rate of convergence under the optimal power split . [ thm : genesplit ] given any power split @xmath155 , a sufficient condition for @xmath156 is @xmath157 as @xmath158 . _ @xmath159where @xmath160 , @xmath161 . the equality in is justified as follows . note that , @xmath162 . hence , for sufficiently small positive @xmath33 , @xmath163 . now , we examine the error term @xmath164 \right| \nonumber \\ & \leq & \sum_{k=1}^l \left| \log \left ( 1 + \frac{p_k}{mp - m\sum_{j < k}p_j - p_k + n } \right ) \right . \nonumber \\ & & \hspace{1 cm } \left . - \left(\frac{p_k}{mp - m\sum_{j < k}p_j - p_k + n } \right ) \right| \nonumber \\ & \leq & \sum_{k=1}^l \left(\frac{p_k}{mp - m\sum_{j < k}p_j - p_k + n } \right)^2 \label{eq : small } \\ & \leq & \sum_{k=1}^l \left(\frac{1}{n}\right)^2 p_k^2 \label{inq : powerconv}\\ & \leq & \left(\frac{1}{n}\right)^2 \max_{k } p_k \sum_{k=1}^l p_k \\ & = & p \left(\frac{1}{n}\right)^2 \max_{k } p_k\end{aligned}\ ] ] where inequality in holds because @xmath165 when @xmath166 , and the inequality in follows from the fact that @xmath167 . since @xmath168 as @xmath154 , the error term goes to zero in the limit . this justifies the equality in . using the capacity bound , we also have @xmath169 . therefore , @xmath170 @xmath59 + note that our optimal power split satisfies the sufficient condition in theorem [ thm : genesplit ] . therefore , its convergence is assured . [ cor : conv ] if each real user adopts the optimal splitting method specified in corollary [ cor : optimal ] , then @xmath171 next , we examine the rate of convergence to the maximum equal rate point under the optimal power split . define the error term @xmath172 \equiv r^ * - \sum_{k=1}^l r_k$ ] , we analyze how fast this error term tends to @xmath153 as @xmath154 . we prove the following : = \omega\left(g\left[n\right]\right)$ ] if there exist positive constants @xmath173 and @xmath174 such that @xmath175 \geq c_1 g\left[n\right ] $ ] for all @xmath176 , and @xmath177 = o(g[n])$ ] if there are positive constants @xmath178 and @xmath174 , such that @xmath179 \leq c_2 g[n]$ ] for all @xmath180 . finally , @xmath177 = \theta \left(g[n]\right)$ ] if @xmath177 = \omega \left(g[n]\right)$ ] and @xmath177 = o \left(g[n]\right)$ ] . ] [ thm : speed ] @xmath181 = \theta \left(\frac{1}{l}\right)$ ] . _ proof : _ let @xmath182 . @xmath183 \nonumber \\ & = & \lim_{l \rightarrow \infty } l \left ( \frac{1}{2m}\log \left(a\right ) - \frac{l}{2 } \log \left ( \frac{m}{m-1 + a^{-\frac{1}{l}}}\right)\right ) \nonumber \\ & = & \lim_{y \rightarrow 0 } \frac{\frac{y\log\left(a\right)}{2m}-\frac{1}{2 } \log \left(\frac{m}{m-1+a^{-y}}\right)}{y^2}\\ & = & \lim_{y \rightarrow 0 } \frac{\frac{\log\left(a\right)}{2m}-\frac{1}{2}\frac{a^{-y}\log\left(a\right)}{m-1+a^{-y}}}{2y } \label{eq : l1}\\ & = & \frac{\left(m-1\right)\left(\log\left(a\right)\right)^2}{4m^2}.\label{eq : l2}\end{aligned}\ ] ] note that equalities in and can be verified by lhospital s rule . consequently , given any @xmath23 , there exists a positive integer @xmath174 such that for all @xmath184 , we have @xmath185 - \frac{\left(m-1\right)\left(\log\left(a\right)\right)^2}{4m^2}\right| < \varepsilon$ ] . this implies @xmath186 < \frac{\left(m-1\right)\left(\log\left(a\right)\right)^2}{4m^2 } + \varepsilon$ ] . therefore , we can choose a small enough @xmath123 such that @xmath187 . let @xmath188 , we have @xmath189 \leq c_2 $ ] . this implies that there exists a positive integer @xmath174 and for all @xmath184 , we have @xmath190 \leq \frac{c_2}{l}$ ] . @xmath59 + finally , we note that all virtual users in one virtual user class can be decoded _ in parallel_. thus , the decoding delay of drs is proportional to the number of virtual users @xmath1 and independent of the number of real users . since @xmath1 is controlled by the designer , drs offers a tradeoff between the throughput of a real user and the decoding delay . in fig . [ fig : simulation ] , we present some numerical simulations illustrating the tradeoff between the number of virtual users and the throughput for both the high and low snr regimes . a system with @xmath191 real users is used in the simulations . an @xmath0-user discrete memoryless mac is defined in terms of @xmath0 discrete input alphabets @xmath192 , @xmath193 , an output alphabet @xmath194 and a stochastic matrix @xmath195 with entries @xmath196 . for any product input distribution @xmath197 , let the achievable region @xmath198 $ ] be the set of @xmath199 satisfying @xmath200 where @xmath201 and @xmath202 . the capacity region of the asynchronous mac is @xcite @xcite @xmath203.\ ] ] we fix the input product distribution @xmath204 and focus on achieving the desired operating point in @xmath205 $ ] . in this section , we consider only discrete memoryless channels satisfying the following symmetry condition : @xmath206 @xmath207 such that @xmath208 . later , in section [ sec : variation ] , we consider the more general asymmetric case . we further assume that for @xmath209 , if @xmath210 , then @xmath211 . under our symmetric setting , the maximum common rate that every user can achieve is @xmath212 . in the gaussian mac , virtual users are created via a power / rate splitting mechanism . for the discrete memoryless mac , we adopt the random switching mechanism of @xcite where virtual users with the same input distribution as the real users are created and the transmitted signal of a real user is determined by a random switch . we first consider the two - user discrete memoryless mac @xmath213 , and illustrate the random switching mechanism . the optimal random switches and the asymptotic behavior of drs under the optimal switching are presented . we then examine the @xmath0-user case @xmath214 , and present a sufficient condition for the random switching mechanism to converge to the information theoretic upper bound . finally , we investigate the rate of convergence for a simple suboptimal random switch . consider a two - user mac , @xmath215 . for a fixed input product distribution @xmath216 , the achievable region is given by : @xmath217 under our symmetry assumption ( cf ) , we have @xmath218 , @xmath219 , @xmath220 , and the optimal rate tuple is @xmath221 let us consider the random switching mechanism for this channel . we first consider the case where each real user generates two virtual users . later , we consider the case where the number of virtual users per real user goes to infinity . we split by means of two switches , as shown in fig . [ fig : switch ] . each switch has two inputs , @xmath222 and @xmath223 and one output @xmath224 . switch @xmath3 is controlled by a random variable @xmath225 with @xmath226 . the output is given by @xmath227 if @xmath228 , and @xmath229 if @xmath230 . the switching random variables @xmath231 are independent of the channel inputs . we also assume that @xmath231 are available at the receiver . in practice , one would generate @xmath232 and @xmath233 at the transmitters and at the receiver , e.g. by means of a pseudorandom sequence generator . assign to the channel inputs @xmath234 , @xmath235 , @xmath236 and @xmath237 the probability mass function @xmath238 . notice that @xmath239 and @xmath240 are independent and each has the same probability mass function as the random variable @xmath241 for @xmath242 . in successive decoding for the discrete memoryless mac , the signals of decoded virtual users are used as side information to aid the decoding process of subsequent virtual users . the first constituent decoder observes the output @xmath243 and tries to decode @xmath244 and @xmath236 . the second constituent decoder is informed of the decision about @xmath245 made by the previous constituent decoder and tries to decode @xmath235 and @xmath237 . without loss of generality , let us focus on real user 1 . @xmath246where the second equality follows from the independence between @xmath234 and @xmath232 , and the last equality follows from the fact that when @xmath247 , @xmath234 is independently of the output @xmath248 and @xmath233 . similarly , we have @xmath249.\end{aligned}\ ] ] it can be verified that in the two - user discrete memoryless mac , both real users throughput can be strictly increased by splitting their inputs via a random switch , relative to the case where they do not split , ( i.e. @xmath250 for @xmath251 ) . next , we show that by generating more virtual users , the throughput of each real user increases further . for @xmath252 , consider a distributed rate splitting scheme with @xmath1 virtual users per real user . the random switch for user @xmath3 is controlled by @xmath253 , where @xmath254 for @xmath144 . it is possible to strictly increase the throughput via an @xmath64 virtual user system by splitting the @xmath1th virtual user into two virtual users . _ proof : _ without loss of generality , we consider user @xmath30 . for the @xmath43th virtual user , we have @xmath255 \nonumber \\ & = & \lambda_k \left [ \left(\sum_{j < k } \lambda_j \right)i \left(x_1;y , x_2 \right ) \right . \nonumber \\ & & \hspace{2 cm } + \left . \left ( 1 - \sum_{j < k } \lambda_j \right ) i\left(x_1;y \right ) \right ] , \label{eq : lelast}\end{aligned}\]]where equality in is due to the independence between @xmath256 and @xmath232 , and equality in follows from the fact that when @xmath257 , @xmath256 is independent of the output @xmath248 and all the other random variables . finally , equality in holds because when @xmath258 , one of the random variables @xmath259 is the switch output , and when @xmath260 , none of them is the switch output . therefore , @xmath261.\ ] ] now let us split @xmath262 into @xmath263 and @xmath264 by using a switch controlled by a binary random variable @xmath265 with @xmath266 . we have @xmath267 where @xmath268 . hence , @xmath269 \\ & = & \alpha \overline{\alpha } \lambda_l^2 \left[i \left(x_1;y , x_2 \right ) - i \left(x_1;y \right ) \right ] \\ & \geq & 0\end{aligned}\ ] ] with strict inequality if @xmath270 . @xmath59 + before we examine the asymptotic behavior of the drs scheme for @xmath252 , we first solve the problem of how to find the optimal switches for a fixed number of virtual users per real user . for @xmath271 , if a real user has @xmath1 virtual users , then the optimal random variable to control the switch for user @xmath3 is @xmath272 with @xmath273 for @xmath144 and @xmath242 . _ proof : _ we use a perturbation argument . suppose the random variables @xmath274 with @xmath275 maximize @xmath276 , where @xmath277 and @xmath278 . moreover , suppose there exists @xmath279 such that @xmath280 . let @xmath279 be the first element which is not equal to @xmath281 . we consider the pair @xmath282 . we have @xmath283 @xmath284.\end{gathered}\ ] ] therefore , @xmath285i(x_1;y , x_2 ) \\ & & + \left[\left(\lambda_k + \lambda_{k+1}\right)\frac{l - k+1}{l } - \lambda_k \lambda_{k+1 } \right ] i(x_1;y).\end{aligned}\ ] ] first , we consider the case where @xmath286 . we let @xmath287 and @xmath288 for @xmath289 . we have @xmath290i(x_1;y , x_2 ) \\ + \left [ \left(\lambda_k + \lambda_{k+1 } \right)\frac{l - k+1}{l } - \left(\lambda_k-\varepsilon \right ) \left(\lambda_{k+1}+\varepsilon \right ) \right ] i(x_1;y ) . \end{gathered}\ ] ] thus , @xmath291 + @xmath292 - @xmath293 + @xmath294 = @xmath295 + @xmath296 @xmath123 -@xmath297 @xmath298-@xmath299 notice that the second term of the r.h.s . expression is positive and there exists @xmath123 such that @xmath300 + @xmath301 @xmath123-@xmath302 . this , however , contradicts our assumption that the random variables @xmath274 with @xmath303 , maximize @xmath276 . we obtain similar contradictions for the case where @xmath304 . @xmath59 + we now examine the asymptotic behavior of the drs scheme in the two - user discrete memoryless mac . [ thm : twoconv ] for @xmath252 , if both real users adopt @xmath305 for @xmath8 and @xmath306 , then @xmath307 for @xmath308 . moreover , if we define the error term @xmath172 \equiv r^ * - \sum_{k=1}^l r_{x_{ik}}$ ] , then @xmath172 = \theta ( \frac{1}{l } ) $ ] for @xmath306 . _ proof : _ without loss of generality , we consider real user @xmath30 . since @xmath309 we have @xmath310 \\ & = & \lim_{l \rightarrow \infty}\left[\frac{l+1}{2l}i(x_1;y ) + \frac{l-1}{2l}i(x_1;yx_2)\right ] \\ & = & \frac{1}{2 } \left[i(x_1;y ) + i(x_1;y , x_2 ) \right ] \\ & = & \frac{1}{2}i(x_1,x_2;y),\end{aligned}\]]where the last equality follows from the fact that @xmath219 in the symmetric setting . next , we examine the rate of convergence . we have the error term @xmath172 = \frac{1}{2}i(x_1;y ) + \frac{1}{2}i(x_1;y , x_2 ) - ( \frac{l+1}{2l}i(x_1;y ) + \frac{l-1}{2l}i(x_1;y , x_2))$ ] . thus , @xmath311 = \frac{1}{2l}\left ( i(x_1;y , x_2 ) - i(x_1;y ) \right),\ ] ] which implies @xmath312 = \frac{1}{2}\left ( i(x_1;y , x_2 ) - i(x_1;y ) \right ) \triangleq \frac{c}{2}.\ ] ] for any @xmath313 , there exists an @xmath174 such that for all @xmath184 , @xmath314 - \frac{c}{2}|<\varepsilon$ ] . hence , @xmath315 < \frac{c}{2}+\varepsilon $ ] . we can choose @xmath123 small enough such that @xmath316 . this implies @xmath172 = \theta ( \frac{1}{l } ) $ ] . @xmath59 + there are @xmath0 real users and each real user creates @xmath1 virtual users . we have @xmath0 switches @xmath317 to do the splitting with probabilities @xmath254 for @xmath318 and @xmath319 . we assume the receiver also knows @xmath317 . this may require common randomness to exist between all transmitters and the receiver . due to symmetry , we focus on one user , say user @xmath30 . for @xmath144 , we set @xmath320 , \label{eq : mlast}\end{aligned}\]]where @xmath321 and @xmath322 . equality in is due to the independence between @xmath256 and @xmath232 . equality in holds because when @xmath257 , @xmath256 is independent of the output @xmath248 and all the other random variables . the first term in follows from @xmath323 = @xmath324 . the second summation term in follows from the fact that the probability of @xmath3 switching random variables among @xmath325 having values less than @xmath43 is @xmath326 . it can be verified that real user @xmath30 with @xmath1 virtual users can strictly increase its total throughput via an @xmath74 virtual user system . in order to maximize the total throughput of real user 1 for fixed @xmath1 , we need to find the optimal @xmath327 to maximize @xmath328 . this is a non - convex optimization problem and appears to be difficult . we are able to verify that for the general @xmath0-user case ( unlike the two - user case ) , random switches with a uniform distribution are in general suboptimal . nevertheless , it is possible to generalize the asymptotic result of theorem [ thm : twoconv ] . we first demonstrate the fact that the convergence result holds for a general switch controlled by @xmath253 , where @xmath329 for @xmath8 , as long as @xmath330 as @xmath154 . we then analyze the rate of convergence for a particular suboptimal switch , the uniform switch . [ thm : mconv ] for a general random switch controlled by @xmath253 , where @xmath331 for @xmath144 , a sufficient condition for @xmath332 is @xmath330 as @xmath154 for @xmath318 . _ proof : _ without loss of generality , let us examine real user @xmath30 . @xmath333 \\ & = & \lim_{l \rightarrow \infty}\sum_{i=0}^{m-1}c_i i_i,\end{aligned}\]]where @xmath334 for @xmath335 . it is sufficient to prove @xmath336 for all @xmath3 . for @xmath335 , @xmath337where @xmath160 , @xmath338 and @xmath339 is the beta function . the term @xmath340 can be shown to converge to @xmath341 as @xmath154 by a similar argument . @xmath59 + next , we analyze a particular suboptimal switch , the uniform switch . since the uniform switch satisfies the sufficient condition in theorem [ thm : mconv ] , the convergence result holds . the next lemma presents its rate of convergence . consider an @xmath0-user discrete memoryless mac . let each real user have @xmath1 virtual users and each switch be controlled by an i.i.d . random variable @xmath342 with @xmath343 , @xmath8 . define the error term @xmath172 \equiv \frac{1}{m}i(x_1,x_2, ... ,x_m;y ) - \sum_{k=1}^l r_{x_{ik } } $ ] . then @xmath172 = o ( \frac{1}{l } ) $ ] for all @xmath3 . _ proof : _ in the uniform switch setting , @xmath343 for @xmath318 and @xmath144 . without loss of generality , we examine the total throughput of real user @xmath30 . @xmath344,\end{gathered}\]]where @xmath345 for @xmath346 . we denote @xmath347 , where @xmath348 is defined in the proof of theorem [ thm : mconv ] for @xmath349 , and @xmath350for @xmath335 . therefore , the error term can be calculated as follows , @xmath351 & = & \left| \frac{1}{m } i(x_1,x_2, ... \sum_{i=0}^{m-1}c_i i_i \right|\\ & = & \left| \sum_{i=0}^{m-1 } \left(\frac{1}{m } - c_i \right ) i_i \right| \\ & \leq & \sum_{i=0}^{m-1 } \left| \frac{1}{m } - c_i \right| i_i \\ & \leq & m \left[\max_i \left(\left| \frac{1}{m } - c_i \right| i_i \right ) \right].\end{aligned}\ ] ] note that @xmath352 is maximized at @xmath353 . for @xmath335 , it can be verified that @xmath354 @xmath355 multiplying both sides of the above two inequalities by @xmath356 , we have @xmath357 therefore , @xmath358 \frac{1}{l } \\ & \equiv & \alpha \frac{1}{l},\end{aligned}\ ] ] so @xmath172 \leq m \alpha \frac{1}{l}$ ] , where @xmath359 . for the term @xmath340 , a similar argument can be used to show that @xmath360 . therefore , @xmath172 = o \left ( \frac{1}{l } \right)$ ] . in section [ sec : gau ] and section [ sec : dmc ] , we imposed two symmetry constraints . the first is that the capacity region for the gaussian mac and the achievable rate region for the discrete memoryless mac are symmetric . the second is that users generate the same number of virtual users . in this section , we describe two variations of drs . the first variation is presented in section [ sec : unequalpower ] , where we relax the symmetric region constraint . in this case , we show that as the number of virtual user per real user tends to infinity , the rate tuple achieved under drs approaches a point on the dominant face . the second variation is presented in section [ sec : differentvuser ] , where each real user may generate a different number of virtual users . the main advantage of this variation is that it can accommodate different user rate requirements in a distributed fashion . in this section , we consider the case where real users in a gaussian mac may have different transmission powers ( i.e. the capacity region may not be symmetric ) . we assume that user @xmath3 has transmission power @xmath361 and the power vector @xmath362 is known to all users . we also assume that all real users split their powers into @xmath1 virtual users according to the common power splitting rule defined by the vector @xmath363 , where @xmath364 @xmath365 and @xmath366 . the power vector for the virtual users generated by user @xmath3 is @xmath367 for @xmath368 . for any real user with @xmath1 virtual users , the unique way to maximize this user s throughput is to set + + @xmath369 $ ] + + for @xmath144 . moreover , if all real users adopt this power allocation rule , then @xmath370for @xmath318 . _ proof : _ by replacing @xmath42 by @xmath371 for @xmath144 , we can use arguments similar to those in section [ sec : gau ] to prove the following : 1 . given a drs scheme with @xmath1 virtual users per real user , it is possible to strictly increase the throughput via an @xmath64 virtual user system . 2 . under the optimal power split , all virtual users generated by real user @xmath3 must have the same rate for @xmath372 . ( virtual users generated by different real users may have different rates . ) 3 . for any real user with @xmath1 virtual users , the unique way to maximize this user s throughput is to set + + @xmath373 $ ] + + for @xmath144 . if all real users adopt this power allocation rule , then @xmath374for @xmath318 @xmath59 + we illustrate this achievable point on the dominant face for a two - user gaussian mac in fig . [ fig : asy_reg ] . in section [ sec : dmc ] , we considered the symmetric setting ( cf ): @xmath375 @xmath207 such that @xmath208 . in this section , we relax this constraint and consider an asymmetric achievable region . we require only that for @xmath376 , if @xmath377 , then @xmath378 . the @xmath0 switches @xmath379 have probabilities @xmath380 for @xmath318 and @xmath381 . consider a general random switch controlled by @xmath253 , where @xmath380 , @xmath381 . define @xmath382 . if @xmath383 as @xmath154 , then @xmath384 for @xmath318 . _ proof : _ we can replace @xmath385 by @xmath386 and use arguments similar to those for theorem [ thm : mconv ] to prove the above lemma . @xmath59 + in this section , we retain the assumption that every user has the same transmission power @xmath13 , but we do not require all real users to create the same number of virtual users . that is , user @xmath3 and user @xmath387 create @xmath12 and @xmath388 virtual users independently , where @xmath12 may not be equal to @xmath388 . the signal transmitted by a real user is the superposition of all its virtual users signals . we also assume in this section that user @xmath3 transmits the number @xmath12 in a header message to the receiver . the receiver receives the sum of @xmath389 signals plus noise . we now describe a protocol which allows each user to split its power and set its rates independently , and allows the receiver to decode all virtual users one by one via a generalized successive decoding mechanism . recall that for the gaussian mac , successive decoding works as follows . users are decoded one after another regarding all other users that have not been decoded as interference , and the signals of decoded users are subtracted from the overall received signal . _ protocol @xmath30 _ : for user @xmath3 , the power split and rate allocation rule are defined as follows : for @xmath390 , @xmath391,\\ r_{ik } & = \frac{1}{2 } \log \left(1 + \frac{p_{ik}}{m\left(p - \sum_{j < k } p_{ij } \right ) - p_{ik } + n } \right ) . \ ] ] note that the power split and rate allocation rule in _ protocol @xmath30 _ are the same as that discussed in section [ sec : gau ] . the generalized successive decoding algorithm is given by the following pseudo - program . note that after a virtual user is decoded , its signal is subtracted from the overall received signal . * decode * virtual users @xmath392 in any order or in parallel . * set * @xmath393 * while * ( some virtual users are not decoded ) , find the minimal element in @xmath394 , say the @xmath3th entry ; decode the subsequent virtual user of user @xmath3 ; update the @xmath3th entry of @xmath394 : @xmath395 , where @xmath396 is the power of the virtual user being decoded in the previous step ; * end * [ le : adrs ] if all @xmath0 users adopt the power split and the rate allocation rule described in _ protocol @xmath30 _ , then for any @xmath397 , @xmath398 , the decoder can decode all virtual users one by one following the decoding algorithm . _ proof : _ by the rate allocation rule , @xmath399 for @xmath400 . thus , each of them can tolerate the maximum amount of interference plus noise , @xmath401 . it is then easy to see that virtual users @xmath402 can be decoded reliably in any order or in parallel . now , we set @xmath403 . in the first run of the * while * loop , if @xmath404 is the minimum entry in @xmath405 , the receiver decodes virtual user @xmath406 . by the rate allocation rule , @xmath407 this implies that the maximum amount of interference plus noise that virtual user @xmath406 can tolerate is @xmath408 . however , the real amount of interference plus noise it faces is @xmath409 , which is smaller than or equal to what it can tolerate because @xmath410 . therefore , virtual user @xmath406 can be decoded reliably at the receiver . suppose the decoding process succeeds in the @xmath411th run of the * while * loop . now , @xmath412 . let us consider the @xmath413th run of the * while * loop . suppose the @xmath3th entry , @xmath414 , is the minimum entry in @xmath394 . the receiver decodes the subsequent virtual user of user @xmath3 , denoted by @xmath415 . by the rate allocation rule , @xmath416 this implies the maximum amount of interference plus noise that virtual user @xmath415 can tolerate is @xmath417 = @xmath418 . however , the real amount of interference plus noise it faces is @xmath419 , which is smaller than or equal to what virtual user @xmath415 can tolerate because @xmath420 . therefore , virtual user @xmath415 can be decoded reliably . hence , the lemma follows by induction . @xmath59 + to illustrate the decoding algorithm , let us carefully examine a three - user example shown in fig . [ fig : examgau ] . the shaded regions correspond to the virtual users @xmath421 , which are decoded in any order . suppose we decode `` @xmath422 '' first . by the rate allocation rule , @xmath423 , which means the maximum amount interference plus noise that virtual user `` @xmath422 '' can tolerate is @xmath424 . this is exactly the amount of interference plus noise it faces . therefore , `` @xmath422 '' can be decoded reliably and we can subtract the signal of virtual user `` @xmath422 '' from the overall received signal . similarly , @xmath425 can be decoded reliably and subtracted from the overall received signal . now @xmath426 . the subsequent decoding order is illustrated by the numbers in fig . [ fig : examgau ] . in the first run of the * while * loop , since @xmath427 is the minimum in @xmath394 , the receiver decodes virtual user `` @xmath428 '' . by the rate allocation rule , @xmath429 , which implies the maximum amount interference plus noise it can tolerate is @xmath430 . however , the real interference plus noise it faces is @xmath431 , which is smaller than what it can tolerate since @xmath432 . so virtual user `` @xmath428 '' can be decoded reliably and subtracted from the received signal . by searching for the minimum entry in @xmath394 in each run , we always decode a virtual user that can tolerate more interference than what it really faces . this assures the validity of our decoding algorithm . all of the other virtual users can be decoded in a similar fashion . by lemma [ le : morevuser ] and lemma [ le : adrs ] , user @xmath3 can choose any @xmath397 , independently from other users , and have all virtual users decoded reliably at the receiver . therefore , user @xmath3 can choose @xmath12 according to its own service requirement . for example , if user @xmath3 wants to send low rate voice communication packets , it can set @xmath433 , which corresponds to the basic cdma scheme . if user @xmath3 wants to send high rate stream video , it can set @xmath12 equal to a large value in order to get higher throughput at the expense of higher coding complexity . thus , this variation of drs provides an explicit way for end users to trade off throughput and coding complexity , making differential rate requirements achievable in a distributed manner . finally , corollary [ cor : conv ] and theorem [ thm : speed ] demonstrate the asymptotic optimality of this scheme and its rate of convergence . in this section , we describe a variation of the drs scheme for the discrete memoryless mac which supports differential rate requirements to end users in a distributed manner . in this scheme , we adopt the uniform switch , but we do not require every user to have the same number of virtual users . we split by means of @xmath0 independent switches . without loss of generality , let us consider user @xmath3 . if user @xmath3 has @xmath12 virtual users , then switch @xmath3 has @xmath12 inputs , @xmath434 for @xmath390 , and one output @xmath224 . switch @xmath3 is controlled by a uniform random variable @xmath435 with @xmath436 for @xmath437 . the output is given by : @xmath438 , if @xmath439 . we now describe the protocol for the discrete memoryless mac which allows user @xmath3 to choose @xmath12 independently . we show the asymptotic optimality of this variation of drs under the protocol . _ protocol 2 _ : for user @xmath3 with @xmath12 virtual users , the switch @xmath3 is controlled by a uniform random variable @xmath253 where @xmath436 for @xmath437 . the rate allocation rule is defined as follows : @xmath440.\end{gathered}\ ] ] the decoding algorithm is given by the following pseudo - program . note that after a virtual user is decoded , its signal is used as side information to aid the decoding process of subsequent virtual users . * decode * virtual users @xmath392 in any order or in parallel . * set * @xmath441 * while * ( some virtual users are not decoded ) , find the minimal element in @xmath442 , say the @xmath3th entry ; decode the subsequent virtual user of user @xmath3 ; update the @xmath3th entry of @xmath442 : @xmath443 ; * end * if all @xmath0 users adopt the rate allocation rule in _ protocol 2 _ , then for any @xmath397 @xmath444 , the decoder can decode all virtual users one by one following the decoding algorithm . _ proof _ : virtual users @xmath392 can be decoded in any order if and only if @xmath445 for all @xmath3 . this is true because under our rate allocation rule , @xmath446 . we set @xmath447 in the first run of the * while * loop , if @xmath448 is the minimum entry of @xmath442 , the receiver decodes virtual user @xmath406 . by the rate allocation rule @xmath449 \label{eq : i2}\end{aligned}\ ] ] virtual user @xmath406 can be decoded reliably if @xmath450 this mutual information can be simplified in the same way as described in equations - . recall the definition @xmath382 . @xmath451 we can verify that the @xmath452 term in is less than or equal to the corresponding @xmath452 term in equation . this follows from @xmath453since @xmath454 for all @xmath455 . similarly , it can be verified that the second term in is less than or equal to the corresponding term in equation . therefore , virtual user @xmath406 can be decoded reliably . suppose the decoding process succeeds in the @xmath411th run of the * while * loop . now , @xmath456 . let us consider the @xmath413th run of the * while * loop . suppose the @xmath3th entry , @xmath457 , is the minimum entry in @xmath442 . the receiver decodes the subsequent virtual user of user @xmath3 , denoted by @xmath458 . by the rate allocation rule @xmath459.\end{aligned}\ ] ] again , we can simplify the mutual information @xmath460 and show that it is great than or equal to @xmath461 , which implies that virtual user @xmath458 can be decoded reliably . hence , the lemma follows by induction . @xmath59 + let us illustrate the decoding algorithm by the following example . we consider a two - user discrete memoryless mac where user @xmath30 creates @xmath11 virtual users and user @xmath11 creates @xmath462 virtual users . random switch @xmath30 is controlled by @xmath232 where @xmath463 , and random switch @xmath11 is controlled by @xmath233 where @xmath464 for @xmath465 . by the rate allocation rule , the virtual users rates can be simplified as follows @xmath466\end{aligned}\ ] ] @xmath467\\ r_{x_{23 } } & = & \frac{1}{3}\left[\frac{1}{3 } i(x_1;y ) + \frac{2}{3}i(x_1;y , x_2 ) \right].\end{aligned}\ ] ] we first decode @xmath468 in any order . suppose we decode `` @xmath469 '' first . virtual user `` @xmath469 '' can be decoded reliably if @xmath470 . the condition holds because @xmath471 virtual user `` @xmath422 '' can be decoded similarly . now , we set @xmath472 . in the first run of the * while * loop , the receiver decodes `` @xmath473 '' since @xmath474 . let us calculate the mutual information between @xmath237 and @xmath248 given @xmath475 and previously decoded @xmath476 . @xmath477 the third equality is due to our symmetric assumption . therefore , virtual user `` @xmath473 '' can be decoded . by searching for the minimum entry in @xmath442 in each run , we always decode a virtual user whose rate is smaller than or equal to the corresponding mutual information . this guarantees the correctness of our decoding algorithm . virtual users @xmath478 can be decoded reliably at the receiver in a similar fashion . the asymptotic optimality of this scheme in the discrete memoryless mac can be demonstrated by theorem [ thm : mconv ] . in this paper , we take an information - theoretic approach to the problem of distributed multiple - access communication . we present a distributed rate splitting scheme whereby each real user creates a number of virtual users and all virtual users are successively decoded at the receiver . one possible advantage of distributed rate splitting is that it can be implemented with lower complexity when compared with joint coding schemes , and less coordination among users when compared with either time - sharing or rate splitting . for the symmetric @xmath0-user gaussian mac , each real user creates the same number of virtual users via a power / rate splitting mechanism . the transmitted signal of a real user is the superposition of all its virtual users signals . for the symmetric @xmath0-user discrete memoryless mac , each real user creates the same number of virtual users via a random switching mechanism , and the transmitted signal of a real user is determined by a random switch . all virtual users are successively decoded at the receiver . it is shown that drs can achieve the maximum equal rate point for both channel models as the number of virtual users per real user tends to infinity . finally , we present two variations of the drs scheme . for the case of asymmetric capacity regions , we show that a point on the dominant face can be achieved asymptotically . for the case of an unequal number of virtual users , we show that different user rates requirements can be accommodated independently in a distributed manner . _ proof : _ we use induction on the number of virtual users . for @xmath46 , the original problem reduces to : @xmath479 subject to @xmath480 and @xmath481 . the unique solution is @xmath482 for @xmath483 , suppose @xmath484 uniquely solves @xmath485 for any @xmath486 , subject to @xmath487 and @xmath488 . let us consider the @xmath387 virtual users case . given any tuple @xmath489 such that @xmath490 , we can fix @xmath80 , so @xmath491 . we now solve @xmath492 subject to @xmath493@xmath42 = @xmath494 , @xmath42 @xmath495 for @xmath43=@xmath496 for fixed @xmath80 , by the induction hypothesis , we have a unique solution @xmath497@xmath498 , ... , @xmath499@xmath500 which solves equation ( [ eq : hminusone ] ) . let @xmath501 @xmath502 . we are able to verify that @xmath503 is a strictly decreasing function of @xmath80 , and @xmath504 = @xmath505 is a strictly increasing function of @xmath80 . moreover , the function values at boundary points satisfy @xmath506 and @xmath507 . so there exists a unique @xmath508 such that @xmath509 . hence , we conclude @xmath510 is the unique solution to the @xmath387 virtual users case . the theorem follows by induction . edmund yeh gratefully acknowledges useful discussions with prof . shlomo shamai on this topic . m. medard , j. huang , a. goldsmith , s. meyn , and t. coleman , `` capacity of time - slotted aloha packetized multiple - access systems over the awgn channel , '' _ ieee trans . wireless commun . . 3 , pp . 486499 , mar . 2004 . r. cheng , `` stripping cdma an asymptotically optimal coding scheme for l - out - of - k white gaussian channels , '' in _ proc . ieee global telecommunications conference _ , sorrento , italy , nov . 1996 , pp . 142146 . jian cao jian cao received the b.s . degree in electrical and electronics engineering with first class honor from the hong kong university of science and technology , hong kong , in 2002 , the m.s . degree in electrical engineering from yale university , new haven , in 2003 , and is currently working toward the ph.d . degree in electrical engineering at yale . his research interests include information theory and queueing theory . edmund yeh edmund yeh received his b.s . in electrical engineering with distinction from stanford university in 1994 , m.phil in engineering from the university of cambridge in 1995 , and ph.d . in electrical engineering and computer science from mit in 2001 . since july 2001 , he has been on the faculty at yale university , new haven , connecticut , where he is currently an associate professor of electrical engineering and computer science . yeh is a recipient of the army research office ( aro ) young investigator program ( yip ) award ( 2003 ) , the winston churchill scholarship ( 1994 ) , the national science foundation and office of naval research fellowships ( 1994 ) for graduate study , the frederick e. terman award from stanford university ( 1994 ) and the barry m. goldwater scholarship from the united states congress ( 1993 ) . yeh is a member of phi beta kappa , tau beta pi , and ieee . he has been visiting faculty at mit , princeton university , university of california at berkeley , and swiss federal institute of technology , lausanne .

the force field by lenosky and coworkers is the latest force field for silicon which is one of the most studied materials . it has turned out to be highly accurate in a large range of test cases . the optimization and parallelization of this force field using openmp and fortan90 is described here . the optimized program allows us to handle a very large number of silicon atoms in large scale simulations . since all the parallelization is hidden in a single subroutine that returns the total energies and forces , this subroutine can be called from within a serial program in an user friendly way . -0.75 in

cond-mat0201475

title of program : siliconiap computer hardware and operating system : any shared memory computer running under unix or linux programming language : fortran90 with openmp compiler directives memory requirements : roughly 150 words per atom no . of bits in a word : 64 no . of processors used : tested on up to 4 processors has the code been vectorized or parallelized : parallelized withe openmp no . of bytes in distributed program , including test data , etc : 50 000 distribution format : compressed tar file keywords : silicon , interatomic potential , force field , molecular dynamics nature of physical problem : condensed matter physics method of solution : interatomic potential restrictions on the complexity of the problem : none typical running time : 30 @xmath0sec per step and per atom on a compaq dec alpha unusual features of the program : none due to its technological importance , silicon is one of the most studied materials . for small system sizes ab - initio density functional calculations @xcite are the preferred approach . unfortunately this kind of calculation becomes unfeasible for larger systems required to study problems such as interfaces or extended defects . for this type of calculations one resorts to force fields which are several orders of magnitude faster . recent progress in the development of force fields has demonstrated that they can be a reliable tool for such studies . a highly accurate silicon force field has been developed by lenosky and coworkers @xcite . its transferability has been demonstrated by extensive tests containing both bulk and cluster systems @xcite . its accuracy is in part due to the fact that second nearest neighbor interactions are included . this makes it unfortunately somewhat slower than force fields containing only nearest neighbor interactions . in the following a highly optimized parallel implementation of this force field will be presented that allows large scale calculations with this force field . the parallelization is achieved by using openmp , an emerging industry standard for medium size shared memory parallel computers . molecular dynamics calculations @xcite have also been parallelized on distributed memory supercomputers @xcite . this approach is considerably more complex than the one presented here . since few researches have access to massively parallel supercomputers and are willing to overcome the complexities of doing molecular dynamics on such machines , medium scale parallelization @xcite of molecular dynamics has an important place in practice . user friendliness was one of the major design goals in the development of this routine . using fortran90 made it possible to hide all the complexities in an object oriented fashion from the user . the calling sequence is just .... call lenosky(nat , alat , rxyz , fxyz , ener , coord , ener_var , coord_var , count ) .... on input the user has to specify the number of atoms , @xmath1 , the vector @xmath2 containing the 3 lattice constant of the orthorhombic periodic volume and the atomic positions @xmath3 . the program then returns the total energy , @xmath4 , the forces , @xmath5 , the average coordination number , the variation of the energy per atom and of the coordination number as well as an counter that is increased in each call . in particular the user has not to supply any verlet list . since the calculation of the forces is typically much more expensive than the update of the atomic positions in molecular dynamics or geometry optimizations , we expect that the subroutine will be called in most cases from within a serial program . in case the user is on a shared memory machine the subroutine will then nevertheless be executed in parallel if the program is compiled with the appropriate openmp options . in addition the subroutine can of course also be used on a serial machine . in this case all the parallelization directives are considered by the compiler to be comments . the verlet list gives all the atoms that are contained within the potential cutoff distance @xmath6 of any given atom . typically the verlet list consists of two integer arrays . the first array , called @xmath7 in this work , points to the first / last neighbor position in the second array @xmath8 that contains the numbering of the atoms that are neighbors . a straightforward implementation for a non - periodic system containing @xmath1 atoms is shown below . in this simple case the search through all atoms is sequential with respect to their numbering and it is redundant to give both the starting positions @xmath9 and the ending position @xmath10 , since @xmath11 . but in the more complicated linear scaling algorithm to be presented below , both will be needed . .... indc=0 do 10 iat=1,nat c starting position lsta(1,iat)=indc+1 do 20 jat=1,nat if ( jat.ne.iat ) then xrel1= rxyz(1,jat)-rxyz(1,iat ) xrel2= rxyz(2,jat)-rxyz(2,iat ) xrel3= rxyz(3,jat)-rxyz(3,iat ) rr2=xrel1**2 + xrel2**2 + xrel3**2 if ( rr2 .le . cut**2 ) then indc = indc+1 c nearest neighbor numbers lstb(indc)=jat endif endif 20 continue c ending position lsta(2,iat)=indc 10 continue .... this straightforward implementations has a quadratic scaling with respect to the numbers of atoms . due to this scaling the calculation of the verlet list starts to dominate the linear scaling calculation of the energies and forces for system sizes of more than 10 000 atoms . it is therefore good practice to calculate the verlet list with a modified algorithm that has linear scaling @xcite as well . to do this one first subdivides the system into boxes that have a side length that is equal to or larger than @xmath6 and then finds all the atoms that are contained in each box . the cpu time for this first step is less than 1 percent of the entire verlet list calculation . hence this part was not parallelized . it could significantly affect the parallel performance according to amdahls law @xcite only if more than 50 processors are used . the largest smp machines at our disposal had however only 4 processors . to implement periodic boundary conditions all atoms within a distance @xmath6 of the boundary of the periodic volume are replicated on the opposite part as shown in figure [ period ] . this part is equally well less than 1 percent of the cpu time for the verlet list for a 8000 atom system . being a surface term it becomes even smaller for larger systems . consequently it was nt parallelized either . ( -1.5,0.5 ) after these two preparing steps one has to search only among all the atoms in the reference cell containing the atom for which one wants to find its neighbors as well as all the atoms in the cells neighboring this reference cell ( 26 cells in 3 dimensions ) . this implies that starting and ending positions @xmath7 for the atoms @xmath12 to @xmath1 are not calculated in a sequential way , necessitating , as mentioned before , separate starting and ending positions in the array @xmath7 . the corresponding parallel code is shown below . the indices @xmath13 refer to the cells , @xmath14 contains the number of atoms in cell @xmath13 and @xmath15 their numbering . the array @xmath16 saves the relative positions and distances that will again be needed in the loop calculating the forces and energies . each thread has its own starting position @xmath17 in the shared memory space @xmath8 and these starting positions are uniformly distributed . this approach allows the different threads to work independently . the resulting speedup is much higher than the one that one would obtain by calculating in the parallel version an array @xmath8 that is identical to the one from the serial version . if there are on the average more neighbors than expected ( 24 by default ) the allocated space becomes too small . in this case the array @xmath8 is deallocated and a new larger version is allocated . this check for sufficient memory requires some minimal amount of coordination among the processors and is implemented by a critical section . if a reallocation is necessary , a message is written into an file to alert the user of the inefficiency due to the need of a second calculation of the verlet list . .... allocate(lsta(2,nat ) ) nnbrx=24 2345 nnbrx=3*nnbrx/2 allocate(lstb(nnbrx*nat),rel(5,nnbrx*nat ) ) indlstx=0 ! $ omp parallel & ! $ omp private(iat , cut2,iam , ii , indlst , l1,l2,l3,myspace , npr ) & ! $ omp shared ( indlstx , nat , nn , nnbrx , ncx , ll1,ll2,ll3,icell , lsta , lstb , lay , & ! $ omp rel , rxyz , cut , myspaceout ) npr=1 ! $ npr = omp_get_num_threads ( ) iam=0 ! $ iam = omp_get_thread_num ( ) cut2=cut**2 myspace=(nat*nnbrx)/npr if ( iam.eq.0 ) myspaceout = myspace ! verlet list , relative positions indlst=0 do 6000,l3=0,ll3 - 1 do 6000,l2=0,ll2 - 1 do 6000,l1=0,ll1 - 1 do 6600,ii=1,icell(0,l1,l2,l3 ) iat = icell(ii , l1,l2,l3 ) if ( ( ( iat-1)*npr)/nat .eq . iam ) then lsta(1,iat)=iam*myspace+indlst+1 call sublstiat(iat , nn , ncx , ll1,ll2,ll3,l1,l2,l3,myspace , & rxyz , icell , lstb(iam*myspace+1),lay , rel(1,iam*myspace+1),cut2,indlst ) lsta(2,iat)=iam*myspace+indlst endif 6600 continue 6000 continue ! $ omp critical indlstx = max(indlstx , indlst ) ! $ omp end critical ! $ omp end parallel if ( indlstx.gt.myspaceout ) then write(10 , * ) count,'nnbrx too small ' , nnbrx deallocate(lstb , rel ) goto 2345 endif subroutine sublstiat(iat , nn , ncx , ll1,ll2,ll3,l1,l2,l3,myspace , & rxyz , icell , lstb , lay , rel , cut2,indlst ) implicit real*8 ( a - h , o - z ) dimension rxyz(3,nn),lay(nn),icell(0:ncx,-1:ll1,-1:ll2,-1:ll3 ) , & lstb(0:myspace-1),rel(5,0:myspace-1 ) do 6363,k3=l3 - 1,l3 + 1 do 6363,k2=l2 - 1,l2 + 1 do 6363,k1=l1 - 1,l1 + 1 do 6363,jj=1,icell(0,k1,k2,k3 ) jat = icell(jj , k1,k2,k3 ) if ( jat.eq.iat ) goto 6363 xrel= rxyz(1,iat)-rxyz(1,jat ) yrel= rxyz(2,iat)-rxyz(2,jat ) zrel= rxyz(3,iat)-rxyz(3,jat ) rr2=xrel**2 + yrel**2 + zrel**2 if ( rr2 .le . cut2 ) then indlst = min(indlst , myspace-1 ) lstb(indlst)=lay(jat ) tt = sqrt(rr2 ) tti=1.d0/tt rel(1,indlst)=xrel*tti rel(2,indlst)=yrel*tti rel(3,indlst)=zrel*tti rel(4,indlst)=tt rel(5,indlst)=tti indlst= indlst+1 endif 6363 continue return end .... the computationally most important part taking some 80 percent of the cpu time is the calculation of the energies and forces . the energy expression for the lenosky force field is given by @xmath18 all the functions ( @xmath19 , @xmath20 , @xmath21 , @xmath22 , @xmath23 ) in this energy expression are given by cubic splines . the subroutine for evaluating the cubic spline is listed below . the case that the argument is outside the cubic spline interval @xmath24 $ ] is rare and unimportant for performance considerations . the important cubic spline case is characterized by many dependencies . in the case of such dependencies the latency of the functional unit pipeline comes into play and reduces the attainable speed @xcite . a latency of some 20 cycles comes from the first two statements ( tt=(x - tmin)*hi ; klo = tt ) alone , requiring arithmetic operations and a floating point to integer conversion . for this reason the calculation of tt was taken out of the ( most likely occurring ) else block to overlap its evaluation with the evaluation of the if clauses . to further speed up the evaluation of the splines the structure of the energy expression [ energy ] was exploited . in the computationally most important loop over @xmath25 and @xmath26 two splines ( @xmath27 and @xmath28 ) have to be evaluated . inlining by hand the subroutine splint for both evaluations and calculating alternatingly one step of the first spline evaluation and one step of the second spline evaluation introduces two independent streams . this reduces the effect of latencies and boosts speed . compilers are not able to do these complex type of optimizations . the best performance after these optimizations was obtained with low level compiler optimization flags ( -o3 -qarch = pwr3 -qtune - pwr3 on ibm power3 , -o2 on the compaq dec alpha , -o2 -xw on intel pentium4 ) .... subroutine splint(ya , y2a , tmin , tmax , hsixth , h2sixth , hi , n , x , y , yp ) implicit real*8 ( a - h , o - z ) dimension y2a(0:n-1),ya(0:n-1 ) ! interpolate if the argument is outside the cubic spline interval [ tmin , tmax ] tt=(x - tmin)*hi if ( x.lt.tmin ) then yp = hi*(ya(1)-ya(0 ) ) - & ( y2a(1)+2.d0*y2a(0 ) ) * hsixth y = ya(0 ) + ( x - tmin)*yp else if ( x.gt.tmax ) then yp = hi*(ya(n-1)-ya(n-2 ) ) + & ( 2.d0*y2a(n-1)+y2a(n-2 ) ) * hsixth y = ya(n-1 ) + ( x - tmax)*yp ! otherwise evaluate cubic spline else klo = tt khi = klo+1 ya_klo = ya(klo ) y2a_klo = y2a(klo ) b = tt - klo a=1.d0-b ya_khi = ya(khi ) y2a_khi = y2a(khi ) b2=b*b y = a*ya_klo yp = ya_khi - ya_klo a2=a*a cof1=a2 - 1.d0 cof2=b2 - 1.d0 y = y+b*ya_khi yp = hi*yp cof3=3.d0*b2 cof4=3.d0*a2 cof1=a*cof1 cof2=b*cof2 cof3=cof3 - 1.d0 cof4=cof4 - 1.d0 yt1=cof1*y2a_klo yt2=cof2*y2a_khi ypt1=cof3*y2a_khi ypt2=cof4*y2a_klo y = y + ( yt1+yt2)*h2sixth yp = yp + ( ypt1 - ypt2 ) * hsixth endif return end .... the final single processor performance for the entire subroutine is 460 mflops on a compaq dec alpha at 833 mhz , 300 mflops on a ibm power3 at 350 mhz and 550 mflops on a pentium 4 . in order to obtain a high parallel speedup in this central part of the subroutine the threads are completely decoupled . this was done by introducing private copies for each thread to accumulate the energies @xmath29 and forces @xmath30 . the global energy and force are summed up in an additional loop at the end of the parallel region in a critical section . .... ! $ omp parallel & ! $ omp private(iam , npr , iat , iat1,iat2,lot , istop , tcoord , tcoord2 , & ! $ omp tener , tener2,txyz , f2ij , f3ij , f3ik , npjx , npjkx ) & ! $ omp shared ( nat , nnbrx , lsta , lstb , rel , ener , ener2,fxyz , coord , coord2,istopg ) npr=1 ! $ npr = omp_get_num_threads ( ) iam=0 ! $ iam = omp_get_thread_num ( ) npjx=300 ; npjkx=3000 istopg=0 if ( npr.ne.1 ) then ! parallel case ! create temporary private scalars for reduction sum on energies and ! temporary private array for reduction sum on forces ! $ omp critical allocate(txyz(3,nat),f2ij(3,npjx),f3ij(3,npjkx),f3ik(3,npjkx ) ) ! $ omp end critical if ( iam.eq.0 ) then ener=0.d0 ener2=0.d0 coord=0.d0 coord2=0.d0 do 121,iat=1,nat fxyz(1,iat)=0.d0 fxyz(2,iat)=0.d0 121 fxyz(3,iat)=0.d0 endif lot = nat / npr+.999999999999d0 iat1=iam*lot+1 iat2=min((iam+1)*lot , nat ) call subfeniat(iat1,iat2,nat , lsta , lstb , rel , tener , tener2 , & tcoord , tcoord2,nnbrx , txyz , f2ij , npjx , f3ij , npjkx , f3ik , istop ) ! $ omp critical ener = ener+tener ener2=ener2+tener2 coord = coord+tcoord coord2=coord2+tcoord2 istopg = istopg+istop do 8093,iat=1,nat fxyz(1,iat)=fxyz(1,iat)+txyz(1,iat ) fxyz(2,iat)=fxyz(2,iat)+txyz(2,iat ) fxyz(3,iat)=fxyz(3,iat)+txyz(3,iat ) 8093 continue ! $ omp end critical deallocate(txyz , f2ij , f3ij , f3ik ) else ! serial case iat1=1 iat2=nat allocate(f2ij(3,npjx),f3ij(3,npjkx),f3ik(3,npjkx ) ) call subfeniat(iat1,iat2,nat , lsta , lstb , rel , ener , ener2 , & coord , coord2,nnbrx , fxyz , f2ij , npjx , f3ij , npjkx , f3ik , istop ) deallocate(f2ij , f3ij , f3ik ) endif ! $ omp end parallel if ( istopg.gt.0 ) stop ' dimension error ( see warning above ) ' ener_var = ener2/nat-(ener / nat)**2 coord = coord / nat coord_var = coord2/nat - coord**2 deallocate(rxyz , icell , lay , lsta , lstb , rel ) end subroutine subfeniat(iat1,iat2,nat , lsta , lstb , rel , tener , tener2 , & tcoord , tcoord2,nnbrx , txyz , f2ij , npjx , f3ij , npjkx , f3ik , istop ) implicit real*8 ( a - h , o - z ) dimension lsta(2,nat),lstb(nnbrx*nat),rel(5,nnbrx*nat),txyz(3,nat ) dimension f2ij(3,npjx),f3ij(3,npjkx),f3ik(3,npjkx ) initialize data ........ ! create temporary private scalars for reduction sum on energies and tener=0.d0 tener2=0.d0 tcoord=0.d0 tcoord2=0.d0 istop=0 do 121,iat=1,nat txyz(1,iat)=0.d0 txyz(2,iat)=0.d0 121 txyz(3,iat)=0.d0 ! calculation of forces , energy do 1000,iat = iat1,iat2 dens2=0.d0 dens3=0.d0 jcnt=0 jkcnt=0 coord_iat=0.d0 ener_iat=0.d0 do 2000,jbr = lsta(1,iat),lsta(2,iat ) jat = lstb(jbr ) jcnt = jcnt+1 if ( jcnt.gt.npjx ) then write(6 , * ) ' warning : enlarge npjx ' istop=1 endif fxij = rel(1,jbr ) fyij = rel(2,jbr ) fzij = rel(3,jbr ) rij = rel(4,jbr ) sij = rel(5,jbr ) ! coordination number calculated with soft cutoff between first and ! second nearest neighbor if ( rij.le.2.36d0 ) then coord_iat = coord_iat+1.d0 else if ( rij.ge.3.83d0 ) then else x=(rij-2.36d0)*(1.d0/(3.83d0 - 2.36d0 ) ) coord_iat = coord_iat+(2*x+1.d0)*(x-1.d0)**2 endif ! pairpotential term call splint(cof_phi , dof_phi , tmin_phi , tmax_phi , & hsixth_phi , h2sixth_phi , hi_phi,10,rij , e_phi , ep_phi ) ener_iat = ener_iat+(e_phi*.5d0 ) txyz(1,iat)=txyz(1,iat)-fxij*(ep_phi*.5d0 ) txyz(2,iat)=txyz(2,iat)-fyij*(ep_phi*.5d0 ) txyz(3,iat)=txyz(3,iat)-fzij*(ep_phi*.5d0 ) txyz(1,jat)=txyz(1,jat)+fxij*(ep_phi*.5d0 ) txyz(2,jat)=txyz(2,jat)+fyij*(ep_phi*.5d0 ) txyz(3,jat)=txyz(3,jat)+fzij*(ep_phi*.5d0 ) ! 2 body embedding term call splint(cof_rho , dof_rho , tmin_rho , tmax_rho , & hsixth_rho , h2sixth_rho , hi_rho,11,rij , rho , rhop ) dens2=dens2+rho f2ij(1,jcnt)=fxij*rhop f2ij(2,jcnt)=fyij*rhop f2ij(3,jcnt)=fzij*rhop ! 3 body embedding term call splint(cof_fff , dof_fff , tmin_fff , tmax_fff , & hsixth_fff , h2sixth_fff , hi_fff,10,rij , fij , fijp ) do 3000,kbr = lsta(1,iat),lsta(2,iat ) kat = lstb(kbr ) if ( kat.lt.jat ) then jkcnt = jkcnt+1 if ( jkcnt.gt.npjkx ) then write(6 , * ) ' warning : enlarge npjkx ' istop=1 endif ! begin optimized version rik = rel(4,kbr ) if ( rik.gt.tmax_fff ) then fikp=0.d0 ; fik=0.d0 gjik=0.d0 ; gjikp=0.d0 ; sik=0.d0 costheta=0.d0 ; fxik=0.d0 ; fyik=0.d0 ; fzik=0.d0 else if ( rik.lt.tmin_fff ) then fxik = rel(1,kbr ) fyik = rel(2,kbr ) fzik = rel(3,kbr ) costheta = fxij*fxik+fyij*fyik+fzij*fzik sik = rel(5,kbr ) fikp = hi_fff*(cof_fff(1)-cof_fff(0 ) ) - & ( dof_fff(1)+2.d0*dof_fff(0 ) ) * hsixth_fff fik = cof_fff(0 ) + ( rik - tmin_fff)*fikp tt_ggg=(costheta - tmin_ggg)*hi_ggg if ( costheta.gt.tmax_ggg ) then gjikp = hi_ggg*(cof_ggg(8 - 1)-cof_ggg(8 - 2 ) ) + & ( 2.d0*dof_ggg(8 - 1)+dof_ggg(8 - 2 ) ) * hsixth_ggg gjik = cof_ggg(8 - 1 ) + ( costheta - tmax_ggg)*gjikp else klo_ggg = tt_ggg khi_ggg = klo_ggg+1 cof_ggg_klo = cof_ggg(klo_ggg ) dof_ggg_klo = dof_ggg(klo_ggg ) b_ggg = tt_ggg - klo_ggg a_ggg=1.d0-b_ggg cof_ggg_khi = cof_ggg(khi_ggg ) dof_ggg_khi = dof_ggg(khi_ggg ) b2_ggg = b_ggg*b_ggg gjik = a_ggg*cof_ggg_klo gjikp = cof_ggg_khi - cof_ggg_klo a2_ggg = a_ggg*a_ggg cof1_ggg = a2_ggg-1.d0 cof2_ggg = b2_ggg-1.d0 gjik = gjik+b_ggg*cof_ggg_khi gjikp = hi_ggg*gjikp cof3_ggg=3.d0*b2_ggg cof4_ggg=3.d0*a2_ggg cof1_ggg = a_ggg*cof1_ggg cof2_ggg = b_ggg*cof2_ggg cof3_ggg = cof3_ggg-1.d0 cof4_ggg = cof4_ggg-1.d0 yt1_ggg = cof1_ggg*dof_ggg_klo yt2_ggg = cof2_ggg*dof_ggg_khi ypt1_ggg = cof3_ggg*dof_ggg_khi ypt2_ggg = cof4_ggg*dof_ggg_klo gjik = gjik + ( yt1_ggg+yt2_ggg)*h2sixth_ggg gjikp = gjikp + ( ypt1_ggg - ypt2_ggg ) * hsixth_ggg endif else fxik = rel(1,kbr ) tt_fff = rik - tmin_fff costheta = fxij*fxik fyik = rel(2,kbr ) tt_fff = tt_fff*hi_fff costheta = costheta+fyij*fyik fzik = rel(3,kbr ) klo_fff = tt_fff costheta = costheta+fzij*fzik sik = rel(5,kbr ) tt_ggg=(costheta - tmin_ggg)*hi_ggg if ( costheta.gt.tmax_ggg ) then gjikp = hi_ggg*(cof_ggg(8 - 1)-cof_ggg(8 - 2 ) ) + & ( 2.d0*dof_ggg(8 - 1)+dof_ggg(8 - 2 ) ) * hsixth_ggg gjik = cof_ggg(8 - 1 ) + ( costheta - tmax_ggg)*gjikp khi_fff = klo_fff+1 cof_fff_klo = cof_fff(klo_fff ) dof_fff_klo = dof_fff(klo_fff ) b_fff = tt_fff - klo_fff a_fff=1.d0-b_fff cof_fff_khi = cof_fff(khi_fff ) dof_fff_khi = dof_fff(khi_fff ) b2_fff = b_fff*b_fff fik = a_fff*cof_fff_klo fikp = cof_fff_khi - cof_fff_klo a2_fff = a_fff*a_fff cof1_fff = a2_fff-1.d0 cof2_fff = b2_fff-1.d0 fik = fik+b_fff*cof_fff_khi fikp = hi_fff*fikp cof3_fff=3.d0*b2_fff cof4_fff=3.d0*a2_fff cof1_fff = a_fff*cof1_fff cof2_fff = b_fff*cof2_fff cof3_fff = cof3_fff-1.d0 cof4_fff = cof4_fff-1.d0 yt1_fff = cof1_fff*dof_fff_klo yt2_fff = cof2_fff*dof_fff_khi ypt1_fff = cof3_fff*dof_fff_khi ypt2_fff = cof4_fff*dof_fff_klo fik = fik + ( yt1_fff+yt2_fff)*h2sixth_fff fikp = fikp + ( ypt1_fff - ypt2_fff ) * hsixth_fff else klo_ggg = tt_ggg khi_ggg = klo_ggg+1 khi_fff = klo_fff+1 cof_ggg_klo = cof_ggg(klo_ggg ) cof_fff_klo = cof_fff(klo_fff ) dof_ggg_klo = dof_ggg(klo_ggg ) dof_fff_klo = dof_fff(klo_fff ) b_ggg = tt_ggg - klo_ggg b_fff = tt_fff - klo_fff a_ggg=1.d0-b_ggg a_fff=1.d0-b_fff cof_ggg_khi = cof_ggg(khi_ggg ) cof_fff_khi = cof_fff(khi_fff ) dof_ggg_khi = dof_ggg(khi_ggg ) dof_fff_khi = dof_fff(khi_fff ) b2_ggg = b_ggg*b_ggg b2_fff = b_fff*b_fff gjik = a_ggg*cof_ggg_klo fik = a_fff*cof_fff_klo gjikp = cof_ggg_khi - cof_ggg_klo fikp = cof_fff_khi - cof_fff_klo a2_ggg = a_ggg*a_ggg a2_fff = a_fff*a_fff cof1_ggg = a2_ggg-1.d0 cof1_fff = a2_fff-1.d0 cof2_ggg = b2_ggg-1.d0 cof2_fff = b2_fff-1.d0 gjik = gjik+b_ggg*cof_ggg_khi fik = fik+b_fff*cof_fff_khi gjikp = hi_ggg*gjikp fikp = hi_fff*fikp cof3_ggg=3.d0*b2_ggg cof3_fff=3.d0*b2_fff cof4_ggg=3.d0*a2_ggg cof4_fff=3.d0*a2_fff cof1_ggg = a_ggg*cof1_ggg cof1_fff = a_fff*cof1_fff cof2_ggg = b_ggg*cof2_ggg cof2_fff = b_fff*cof2_fff cof3_ggg = cof3_ggg-1.d0 cof3_fff = cof3_fff-1.d0 cof4_ggg = cof4_ggg-1.d0 cof4_fff = cof4_fff-1.d0 yt1_ggg = cof1_ggg*dof_ggg_klo yt1_fff = cof1_fff*dof_fff_klo yt2_ggg = cof2_ggg*dof_ggg_khi yt2_fff = cof2_fff*dof_fff_khi ypt1_ggg = cof3_ggg*dof_ggg_khi ypt1_fff = cof3_fff*dof_fff_khi ypt2_ggg = cof4_ggg*dof_ggg_klo ypt2_fff = cof4_fff*dof_fff_klo gjik = gjik + ( yt1_ggg+yt2_ggg)*h2sixth_ggg fik = fik + ( yt1_fff+yt2_fff)*h2sixth_fff gjikp = gjikp + ( ypt1_ggg - ypt2_ggg ) * hsixth_ggg fikp = fikp + ( ypt1_fff - ypt2_fff ) * hsixth_fff endif endif ! end optimized version tt = fij*fik dens3=dens3+tt*gjik t1=fijp*fik*gjik t2=sij*(tt*gjikp ) f3ij(1,jkcnt)=fxij*t1 + ( fxik - fxij*costheta)*t2 f3ij(2,jkcnt)=fyij*t1 + ( fyik - fyij*costheta)*t2 f3ij(3,jkcnt)=fzij*t1 + ( fzik - fzij*costheta)*t2 t3=fikp*fij*gjik t4=sik*(tt*gjikp ) f3ik(1,jkcnt)=fxik*t3 + ( fxij - fxik*costheta)*t4 f3ik(2,jkcnt)=fyik*t3 + ( fyij - fyik*costheta)*t4 f3ik(3,jkcnt)=fzik*t3 + ( fzij - fzik*costheta)*t4 endif 3000 continue 2000 continue dens = dens2+dens3 call splint(cof_uuu , dof_uuu , tmin_uuu , tmax_uuu , & hsixth_uuu , h2sixth_uuu , hi_uuu,8,dens , e_uuu , ep_uuu ) ener_iat = ener_iat+e_uuu ! only now ep_uu is known and the forces can be calculated , lets loop again jcnt=0 jkcnt=0 do 2200,jbr = lsta(1,iat),lsta(2,iat ) jat = lstb(jbr ) jcnt = jcnt+1 txyz(1,iat)=txyz(1,iat)-ep_uuu*f2ij(1,jcnt ) txyz(2,iat)=txyz(2,iat)-ep_uuu*f2ij(2,jcnt ) txyz(3,iat)=txyz(3,iat)-ep_uuu*f2ij(3,jcnt ) txyz(1,jat)=txyz(1,jat)+ep_uuu*f2ij(1,jcnt ) txyz(2,jat)=txyz(2,jat)+ep_uuu*f2ij(2,jcnt ) txyz(3,jat)=txyz(3,jat)+ep_uuu*f2ij(3,jcnt ) ! 3 body embedding term do 3300,kbr = lsta(1,iat),lsta(2,iat ) kat = lstb(kbr ) if ( kat.lt.jat ) then jkcnt = jkcnt+1 txyz(1,iat)=txyz(1,iat)-ep_uuu*(f3ij(1,jkcnt)+f3ik(1,jkcnt ) ) txyz(2,iat)=txyz(2,iat)-ep_uuu*(f3ij(2,jkcnt)+f3ik(2,jkcnt ) ) txyz(3,iat)=txyz(3,iat)-ep_uuu*(f3ij(3,jkcnt)+f3ik(3,jkcnt ) ) txyz(1,jat)=txyz(1,jat)+ep_uuu*f3ij(1,jkcnt ) txyz(2,jat)=txyz(2,jat)+ep_uuu*f3ij(2,jkcnt ) txyz(3,jat)=txyz(3,jat)+ep_uuu*f3ij(3,jkcnt ) txyz(1,kat)=txyz(1,kat)+ep_uuu*f3ik(1,jkcnt ) txyz(2,kat)=txyz(2,kat)+ep_uuu*f3ik(2,jkcnt ) txyz(3,kat)=txyz(3,kat)+ep_uuu*f3ik(3,jkcnt ) endif 3300 continue 2200 continue tener = tener+ener_iat tener2=tener2+ener_iat**2 tcoord = tcoord+coord_iat tcoord2=tcoord2+coord_iat**2 1000 continue return end .... in addition to the energy and the forces the program still returns the coordination number as well as the variance of the energy per atom and the coordination number . the coordination number is calculated using a soft cutoff between the first and second nearest neighbor distance . these extra calculations are very cheap and not visible as an increase in the cpu time table [ speedup_dec ] shows the final overall speedups obtained by the program . the results were obtained for an 8000 atom system , but the cpu time per call and atom is nearly independent of system size . . timings in @xmath0sec for a combined evaluation of the forces and the energy per particle as well as the corresponding speedups ( in parentheses ) on an ibm sp3 based on a 375 mhz power3 processor , on a compaq sc 232 based on a 833mhz ev67 processor and on an intel pentium4 biprocessor at 2 ghz [ speedup_dec ] [ cols="<,^,^,^",options="header " , ] obtaining such high speedups was not straightforward . only the compaq fortran90 compiler was able to use in the original version of the program the openmp parallel do directive to obtain a good speedup . both the ibm compiler and the intel compiler failed . in order to get the performances of table [ speedup_dec ] , it was necessary to encapsulate the workload of the different threads into the subroutines sublstias and subfen , which amounts to doing the parallelization quasi by hand . using allocatable arrays in connection with openmp turned also out to be tricky . because of these problems , the parallelization was much more painful that one might expect for a shared memory model . the results show that simulations for very large silicon systems are feasible on relatively cheap serial or parallel computers accessible to a large number of researches .